This action might not be possible to undo. Are you sure you want to continue?
https://www.scribd.com/doc/92740259/CyclostationarityHalfaCenturyofResearch
11/04/2012
text
original
Signal Processing 86 (2006) 639–697
Review
Cyclostationarity: Half a century of research
William A. Gardner
a
, Antonio Napolitano
b,Ã
, Luigi Paura
c
a
Statistical Signal Processing Incorporated (SSPI), 1909 Jefferson Street, Napa, CA 94559, USA
b
Universita` di Napoli ‘‘Parthenope’’, Dipartimento per le Tecnologie, via Acton 38, I80133 Napoli, Italy
c
Universita` di Napoli ‘‘Federico II’’, Dipartimento di Ingegneria Elettronica e delle Telecomunicazioni, via Claudio 21,
I80125 Napoli, Italy
Received 22 December 2004; accepted 29 June 2005
Available online 6 September 2005
Abstract
In this paper, a concise survey of the literature on cyclostationarity is presented and includes an extensive
bibliography. The literature in all languages, in which a substantial amount of research has been published, is included.
Seminal contributions are identiﬁed as such. Citations are classiﬁed into 22 categories and listed in chronological order.
Both stochastic and nonstochastic approaches for signal analysis are treated. In the former, which is the classical one,
signals are modelled as realizations of stochastic processes. In the latter, signals are modelled as single functions of time
and statistical functions are deﬁned through inﬁnitetime averages instead of ensemble averages. Applications of
cyclostationarity in communications, signal processing, and many other research areas are considered.
r 2005 Elsevier B.V. All rights reserved.
Keywords: Cyclostationarity; Almostcyclostationary time series; Almostcyclostationary processes; Bibliography
Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 641
2. General treatments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 642
3. General properties and structure of stochastic processes and timeseries . . . . . . . . . . . . . . . . . . . . . . . . . . 642
3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 642
3.2. Stochastic processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643
3.2.1. Continuoustime processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643
3.2.2. Discretetime processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 645
www.elsevier.com/locate/sigpro
01651684/$  see front matter r 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.sigpro.2005.06.016
Ã
Corresponding author. Tel.: +39 081 768 37 81; fax: +39 081 768 31 49.
Email addresses: wag@statsig.com (W.A. Gardner), antonio.napolitano@uniparthenope.it (A. Napolitano),
luigi.paura@unina.it (L. Paura).
ARTICLE IN PRESS
3.3. Time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 646
3.3.1. Continuoustime time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 646
3.3.2. Discretetime time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 648
3.4. Link between the stochastic and fractionoftime approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 649
3.5. Complex processes and time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 649
3.6. Linear ﬁltering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 650
3.6.1. Structure of linear almostperiodically timevariant systems . . . . . . . . . . . . . . . . . . . . . . . . 650
3.6.2. Input/output relations in terms of cyclic statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 650
3.7. Product modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 651
3.8. Supports of cyclic spectra of band limited signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 652
3.9. Sampling and aliasing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653
3.10. Representations by stationary components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 654
3.10.1. Continuoustime processes and time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 654
3.10.2. Discretetime processes and time series. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 655
4. Ergodic properties and measurement of characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 656
4.1. Estimation of the cyclic autocorrelation function and the cyclic spectrum . . . . . . . . . . . . . . . . . . . . 656
4.2. Two alternative approaches to the analysis of measurements on time series . . . . . . . . . . . . . . . . . . . 657
5. Manufactured signals: modelling and analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 658
5.1. General aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 658
5.2. Examples of communication signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659
5.2.1. Double sideband amplitudemodulated signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659
5.2.2. Pulseamplitudemodulated signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659
5.2.3. Directsequence spreadspectrum signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 660
6. Natural signals: modelling and analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 661
7. Communications systems: analysis and design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 661
7.1. General aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 661
7.2. Cyclic Wiener ﬁltering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 661
8. Synchronization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 663
8.1. Spectral line generation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 663
9. Signal parameter and waveform estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 663
10. Channel identiﬁcation and equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 664
10.1. General aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 664
10.2. LTIsystem identiﬁcation with noisymeasurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 664
10.3. Blind LTIsystem identiﬁcation and equalization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665
10.4. Nonlinearsystem identiﬁcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665
11. Signal detection and classiﬁcation, and source separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665
12. Periodic AR and ARMA modelling and prediction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 666
13. Higherorder statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 666
13.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 666
13.2. Higherorder cyclic statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 667
13.2.1. Continuoustime time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 667
13.2.2. Discretetime time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 669
14. Applications to circuits, systems, and control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 670
15. Applications to acoustics and mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 670
16. Applications to econometrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
17. Applications to biology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
18. Level crossings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
19. Queueing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
20. Cyclostationary random ﬁelds. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
21. Generalizations of cyclostationarity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
21.1. General aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 640
1. Introduction
Many processes encountered in nature arise
from periodic phenomena. These processes,
although not periodic functions of time, give rise
to random data whose statistical characteristics
vary periodically with time and are called cyclosta
tionary processes [2.5]. For example, in telecom
munications, telemetry, radar, and sonar
applications, periodicity is due to modulation,
sampling, multiplexing, and coding operations. In
mechanics it is due, for example, to gear rotation.
In radio astronomy, periodicity results from
revolution and rotation of planets and on pulsa
tion of stars. In econometrics, it is due to
seasonality; and in atmospheric science it is
due to rotation and revolution of the earth.
The relevance of the theory of cyclostationarity
to all these ﬁelds of study and more was ﬁrst
proposed in [2.5].
Widesense cyclostationary stochastic processes
have autocorrelation functions that vary periodi
cally with time. This function, under mild reg
ularity conditions, can be expanded in a Fourier
series whose coefﬁcients, referred to as cyclic
autocorrelation functions, depend on the lag para
meter; the frequencies, called cycle frequencies, are
all multiples of the reciprocal of the period of
cyclostationarity [2.5]. Cyclostationary processes
have also been referred to as periodically correlated
processes [2.18,3.5]. More generally, if the frequen
cies of the (generalized) Fourier series expansion of
the autocorrelation function are not commensu
rate, that is, if the autocorrelation function is an
almostperiodic function of time, then the process
is said to be almostcyclostationary [3.26] or,
equivalently, almostperiodically correlated [3.5].
The almostperiodicity property of the autocorre
lation function is manifested in the frequency
domain as correlation among the spectral com
ponents of the process that are separated by
amounts equal to the cycle frequencies. In contrast
to this, widesense stationary processes have
autocorrelation functions that are independent
of time, depending on only the lag para
meter, and all distinct spectral components are
uncorrelated.
As an alternative, the presence of periodicity in
the underlying datagenerating mechanism of a
phenomenon can be described without modelling
the available data as a sample path of a stochastic
process but, rather, by modelling it as a single
function of time [4.31]. Within this nonstochastic
framework, a timeseries is said to exhibit second
order cyclostationarity (in the wide sense), as ﬁrst
deﬁned in [2.8], if there exists a stable quadratic
timeinvariant transformation of the timeseries
that gives rise to ﬁnitestrength additive sinewave
components.
In this paper, a concise survey of the literature
(in all languages in which a substantial amount of
research has been published) on cyclostationarity
is presented and includes an extensive bibliography
and list of issued patents. Citations are classiﬁed
into 22 categories and listed, for each category, in
chronological order. In Section 2, general treat
ments and tutorials on the theory of cyclostatio
narity are cited. General properties of processes
and timeseries are presented in Section 3. In
Section 4, the problem of estimating statistical
functions is addressed. Models for manufactured
and natural signals are considered in Sections 5
and 6, respectively. Communications systems and
related problems are treated in Sections 7–11.
Speciﬁcally, the analysis and design of commu
nications systems is addressed in Section 7, the
problem of synchronization is addressed in Section
8, the estimation of signal parameters and wave
forms is addressed in Section 9, the identiﬁcation
and equalization of channels is addressed in
Section 10, and the signal detection and classiﬁca
tion problems and the problem of source separa
tion are addressed in Section 11. Periodic
autoregressive (AR) and autoregressive moving
average (ARMA) modelling and prediction are
treated in Section 12. In Section 13, theory and
ARTICLE IN PRESS
21.2. Generalized almostcyclostationary signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 672
21.3. Spectrally correlated signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 672
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673
Further references . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 697
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 641
applications of higherorder cyclostationarity are
presented. We address applications to circuits,
systems, and control in Section 14, to acoustics
and mechanics in Section 15, to econometrics in
Section 16, and to biology in Section 17. Applica
tions to the problems of level crossing and
queueing are addressed in Sections 18 and 19,
respectively. Cyclostationary random ﬁelds are
treated in Section 20. In Section 21, some classes
of nonstationary signals that extend the class of
almostcyclostationary signals are considered. Fi
nally, some miscellaneous references are listed
[22.1–22.8]. Further references only indirectly
related to cyclostationarity are [23.1–23.15].
To assist readers in ‘‘going to the source’’,
seminal contributions—if known— are identiﬁed
within the literature published in English. In some
cases, identiﬁed sources may have been preceded in
the literature of another language, most likely
Russian. For the most part, the subject of this
survey developed independently in the literature
published in English.
2. General treatments
General treatments on cyclostationarity are in
[2.1–2.18]. The ﬁrst extensive treatments of the
theory of cyclostationary processes can be found
in the pioneering works of Hurd [2.1] and Gardner
[2.2]. In [4.13,2.5,2.11], the theory of secondorder
cyclostationary processes is developed mainly with
reference to continuoustime stochastic processes,
but discretetime is the focus in [4.13]. Discrete
time processes are treated more generally in [2.17]
in a manner largely analogous to that in [2.5]. The
statistical characterization of cyclostationary time
series in the nonstochastic framework is intro
duced and treated in depth [2.6,2.8,2.12,2.14] with
reference to continuoustime signals and in [2.15]
for both continuous and discretetime signals.
The case of complex signals is introduced and
treated in depth in [2.8,2.9]. Finally, a rigorous
treatment of periodically correlated processes
within the framework of harmonizable processes
is given in [2.18].
The theory of higherorder cyclostationarity in
the nonstochastic framework is introduced in
[13.4,13.5], and treated in depth in [13.9,13.13,
13.14,13.17]. An analogous treatment for stochas
tic processes is given in [13.10].
3. General properties and structure of stochastic
processes and timeseries
3.1. Introduction
General properties of cyclostationary processes
(see [3.1–3.91]) are derived in terms of the Fourier
series expansion of the autocorrelation function.
The frequencies, called cycle frequencies, are
multiples of the reciprocal of the period of
cyclostationarity and the coefﬁcients, referred to
as cyclic autocorrelation functions, are continuous
functions of the lag parameter. Cyclostationary
processes are characterized in the frequency
domain by the cyclic spectra, which are the Fourier
transforms of the cyclic autocorrelation functions.
The cyclic spectrum at a given cycle frequency
represents the density of correlation between two
spectral components of the process that are
separated by an amount equal to the cycle
frequency. Almostcyclostationary processes have
autocorrelation functions that can be expressed in
a (generalized) Fourier series whose frequencies
are possibly incommensurate.
The ﬁrst contributions to the analysis of the
general properties of cyclostationary stochastic
processes are in the Russian literature [3.1–3.3,
3.5,3.7–3.9]. The problem of spectral analysis is
mainly treated in [3.10,3.11,3.32,3.35,3.51,3.53,
3.61,3.68,3.69,3.80,3.88]. The stationarizing effects
of random shifts are examined in [3.20,3.26,3.57]
and the important impact of random shifts on
cycloergodic properties is exposed in [2.15]. The
problem of ﬁltering is considered in [3.9,3.13,3.67].
The mathematical link between the stochastic
and nonstochastic approaches, which was ﬁrst
established in [2.5,2.12], is rigorously treated in
[3.87]. Wavelet analysis of cyclostationary pro
cesses is addressed in [3.70,3.81].
For furtherrelated references, see the general
treatments in [2.1,2.2,2.4–2.8,2.11–2.13,2.15,2.18],
and also [4.13,4.31,4.41,13.2,13.3,13.6,16.3,18.3,
21.9,21.14].
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 642
3.2. Stochastic processes
3.2.1. Continuoustime processes
Let us consider a continuoustime realvalued
stochastic process {x(t; o); t c R; o c O], with ab
breviated notation x(t) when this does not create
ambiguity, deﬁned on a probability space
(O; F; P), where O is the sample space, equipped
with the sﬁeld F, and P is a probability measure
deﬁned on the elements of F.
The process x(t) is said to be Nthorder
cyclostationary in the strict sense [2.2,2.5] if its
Nthorder distribution function
F
x(t÷t
1
)···x(t÷t
N÷1
)x(t)
(x
1
; . . . ; x
N÷1
; x
N
)
9P{x(t ÷ t
1
)px
1
; . . . ; x(t ÷ t
N÷1
)
px
N÷1
; x(t)px
N
] (3:1)
is periodic in t with some period, say T
0
:
F
x(t÷t
1
÷T
0
)···x(t÷t
N÷1
÷T
0
)x(t÷T
0
)
(x
1
; . . . ; x
N÷1
; x
N
)
= F
x(t÷t
1
)···x(t÷t
N÷1
)x(t)
(x
1
; . . . ; x
N÷1
; x
N
)
\t c R \(t
1
; . . . ; t
N÷1
) c R
N÷1
\(x
1
; . . . ; x
N
) c R
N
. (3:2)
The process x(t) is said to be secondorder
cyclostationary in the wide sense [2.2,2.5] if its
mean E{x(t)] and autocorrelation function
R
x
(t; t)9E{x(t ÷t)x(t)] (3.3)
are periodic with some period, say T
0
:
E{x(t ÷ T
0
)] = E{x(t)], (3.4)
R
x
(t ÷ T
0
; t) = R
x
(t; t) (3.5)
for all t and t. Therefore, by assuming that the
Fourier series expansion of R
x
(t; t) is convergent
to R
x
(t; t), we can write
R
x
(t; t) =
÷o
n=÷o
R
n=T
0
x
(t)e
j2p(n=T
0
)t
, (3.6)
where the Fourier coefﬁcients
R
n=T
0
x
(t)9
1
T
0
_
T
0
=2
÷T
0
=2
R
x
(t; t)e
÷j2p(n=T
0
)t
dt (3.7)
are referred to as cyclic autocorrelation functions
and the frequencies {n=T
0
]
ncZ
are called cycle
frequencies. Widesense cyclostationary processes
have also been called periodically correlated
processes (see e.g., [2.1,3.3,3.59]). As ﬁrst shown
in [3.23], x(t) and its frequencyshifted version
x(t)e
j2pnt=T
0
are correlated. The widesense station
ary processes are the special case of cyclostation
ary processes for which R
n=T
0
x
(t)c0 only for
n = 0. It can be shown that if x(t) is cyclosta
tionary with period T
0
, then the stochastic process
x(t ÷ y), where y is a random variable that is
uniformly distributed in [0; T
0
) and is statistically
independent of x(t), is wide sense stationary
[2.5,3.20,3.26,3.57].
A more general class of stochastic processes is
obtained if the autocorrelation function R
x
(t; t) is
almost periodic in t for each t.
A function z(t) is said to be almost periodic if it is
the limit of a uniformly convergent sequence of
trigonometric polynomials in t [23.2,23.3,23.4,
Paragraphs 24–25,23.6,23.7, Part 5, 23.11]:
z(t) =
acA
z
a
e
j2pat
, (3.8)
where A is a countable set, the frequencies a c A
are possibly incommensurate, and the coefﬁcients
z
a
are given by
z
a
9¸z(t)e
÷j2pat
)
t
= lim
T÷o
1
T
_
T=2
÷T=2
z(t)e
÷j2pat
dt. (3:9)
Such functions are called almostperiodic in the
sense of Bohr [23.3, Paragraphs 84–92] or,
equivalently, uniformly almost periodic in t in
the sense of Besicovitch [23.2, Chapter 1].
The process x(t) is said to be Nthorder almost
cyclostationary in the strict sense [2.2,2.5] if its Nth
order distribution function (3.1) is almostperiodic
in t. With reference to the autocorrelation proper
ties, a continuoustime realvalued stochastic pro
cess x(t) is said to be almostcyclostationary (ACS)
in the wide sense [2.2,2.5,3.26] if its autocorrelation
function R
x
(t; t) is an almost periodic function of t
(with frequencies not depending on t). Thus, it is
the limit of a uniformly convergent sequence of
trigonometric polynomials in t:
R
x
(t; t) =
acA
R
a
x
(t)e
j2pat
, (3.10)
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 643
where
R
a
x
(t)9 lim
T÷o
1
T
_
T=2
÷T=2
R
x
(t; t)e
÷j2pat
dt (3.11)
is the cyclic autocorrelation function at cycle
frequency a. As ﬁrst shown in [3.26], if x(t) is an
ACS process, then the process x(t) and its
frequencyshifted version x(t)e
j2pat
are correlated
when a c A. Widesense almostcyclostationary
processes have also been called almostperiodically
correlated processes (see e.g., [2.1,3.5,3.59]). The
widesense cyclostationary processes are obtained
as a special case of the ACS processes when A ¬
{n=T
0
]
ncZ
for some T
0
.
Let A
t
be the set
A
t
9{a c R : R
a
x
(t)a0]. (3.12)
In [3.59], it is shown that the ACS processes are
characterized by the following conditions:
(1) The set
A9
_
tcR
A
t
(3.13)
is countable.
(2) The autocorrelation function R
x
(t; t) is uni
formly continuous in t and t.
(3) The timeaveraged autocorrelation function
R
0
x
(t)9¸R
x
(t; t))
t
is continuous for t = 0
(and, hence, for every t).
(4) The process is meansquare continuous, that is
sup
tcR
E{[x(t ÷ t) ÷x(t)[
2
] ÷ 0 as t ÷ 0.
(3.14)
More generally, a stochastic process x(t) is said to
exhibit cyclostationarity at cycle frequency a if
R
a
x
(t)c0 [2.5]. In such a case, the autocorrelation
function can be expressed as
R
x
(t; t) =
acA
R
a
x
(t)e
j2pat
÷ r
x
(t; t), (3.15)
where the function
acA
R
a
x
(t)e
j2pat
is not necessarily continuous in t and the term
r
x
(t; t) does not contain any ﬁnitestrength addi
tive sinewave component:
¸r
x
(t; t)e
÷j2pat
)
t
= 0 \a c R. (3.16)
In the special case where lim
[t[÷o
r
x
(t; t) = 0, x(t)
is said to be asymptotically almost cyclostationary
[3.26].
Let us now consider the secondorder (wide
sense) characterization of ACS processes in the
frequency domain. Let
X(f )9
_
R
x(t)e
÷j2pft
dt (3.17)
be the stochastic process obtained from Fourier
transformation of the ACS process x(t), where the
Fourier transform is assumed to exist, at least in
the sense of distributions (generalized functions)
[23.10], with probability 1. By using (3.10), in the
sense of distributions we obtain
E{X(f
1
)X
+
(f
2
)] =
acA
S
a
x
(f
1
)d(f
2
÷ f
1
÷a),
(3.18)
where d(·) is the Dirac delta, superscript + is
complex conjugation, and
S
a
x
(f )9
_
R
R
a
x
(t)e
÷j2pf t
dt (3.19)
is referred to as the cyclic spectrum at cycle
frequency a. Therefore, for an ACS process,
correlation exists between spectral components
that are separated by amounts equal to the cycle
frequencies. The support in the (f
1
; f
2
) plane of the
spectral correlation function E X(f
1
)X
+
(f
2
)
_ _
con
sists of parallel lines with unity slope. The density
of spectral correlation on this support is described
by the cyclic spectra S
a
x
(f ), a c A, which can be
expressed as [2.5]
S
a
x
(f ) = lim
Df ÷0
lim
T÷o
1
T
_
T=2
÷T=2
E{DfX
1=Df
(t; f )
X
+
1=Df
(t; f ÷ a)] dt, (3:20)
where the order of the two limits cannot be
reversed, and
X
Z
(t; f )9
_
t÷Z=2
t÷Z=2
x(s)e
÷j2pfs
ds. (3.21)
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 644
Therefore, the cyclic spectrum S
a
x
(f ) is also called
the spectral correlation density function. It repre
sents the timeaveraged statistical correlation (with
zero lag) of two spectral components at frequen
cies f and f ÷a, as the bandwidth approaches
zero. For a = 0, the cyclic spectrum reduces to the
power spectrum or spectral density function S
0
x
(f )
and (3.19) reduces to the Wiener–Khinchin Rela
tion. Consequently, when (3.19) and (3.20) was
discovered in [2.5] it was dubbed the Cyclic
Wiener–Khinchin Relation.
In contrast, for widesense stationary processes
the autocorrelation function does not depend on t,
E{x(t ÷ t)x(t)] = R
0
x
(t) (3.22)
and, equivalently, no correlation exists between
distinct spectral components,
E{X(f
1
)X
+
(f
2
)] = S
0
x
(f
1
)d(f
2
÷ f
1
). (3.23)
A covariance (or autocorrelation) function
E{x(t
1
)x
+
(t
2
)] is said to be harmonizable if it can
be expressed as
E{x(t
1
)x
+
(t
2
)] =
_
R
2
e
j2p(f
1
t
1
÷f
2
t
2
)
dg(f
1
; f
2
), (3.24)
where g(f
1
; f
2
) is a covariance of bounded varia
tion on R R and the integral is a Fourier–
Stieltjes transform [23.5]. Moreover, a second
order stochastic process x(t) is said to be
harmonizable if there exists a secondorder sto
chastic process w(f ) with covariance function
E{w(f
1
)w
+
(f
2
)] = g(f
1
; f
2
) of bounded variation
on R R such that
x(t) =
_
R
e
j2pft
dw(f ) (3.25)
with probability one. In [23.5], it is shown that a
necessary condition for a stochastic process to be
harmonizable is that it be secondorder contin
uous. Moreover, it is shown that a stochastic
process is harmonizable if and only if its covar
iance is harmonizable. If a stochastic process
is harmonizable, in the sense of distributions
[23.10], we have dw(f ) = X(f ) df and dg(f
1
; f
2
) =
E{dw(f
1
) dw
+
(f
2
)] = E{X(f
1
)X
+
(f
2
)] df
1
df
2
, and
w(f ) is the indeﬁnite integral of X(f ) or the
integrated Fourier transform of x(t) [23.4]. There
fore, if the stochastic process is harmonizable and
ACS, then it follows that [2.1,2.18,3.59]
dg(f
1
; f
2
) =
acA
S
a
x
(f
1
)d(f
2
÷ f
1
÷a) df
1
df
2
.
(3.26)
Finally, note that symmetric deﬁnitions of cyclic
autocorrelation function and cyclic spectrum have
been widely used (see, e.g., [2.5,2.8,2.11]). They are
linked to the asymmetric deﬁnitions (3.11) and
(3.20) by the relationships
lim
T÷o
1
T
_
T=2
÷T=2
E{x(t ÷t=2)x(t ÷t=2)]
e
÷j2pat
dt = R
a
x
(t)e
÷jpat
(3:27)
and
lim
Df ÷0
lim
T÷o
1
T
_
T=2
÷T=2
Df E{X
1=Df
(t; f ÷a=2)
X
+
1=Df
(t; f ÷a=2)] dt = S
a
x
(f ÷ a=2), (3:28)
respectively.
3.2.2. Discretetime processes
Let us consider a discretetime realvalued
stochastic process {x(n; o); n c Z; o c O], with
abbreviated notation x(n) when this does not
create ambiguity. The stochastic process x(n) is
said to be secondorder almostcyclostationary in
the wide sense [2.2,2.5,4.13] if its autocorrelation
function
¯
R
x
(n; m)9E{x(n ÷ m)x(n)] (3.29)
is an almostperiodic function of the discretetime
parameter n. Thus, it can be expressed as
¯
R
x
(n; m) =
¯ac
¯
A
¯
R
¯a
x
(m)e
j2p¯an
, (3.30)
where
¯
R
¯a
x
(m)9 lim
N÷o
1
2N ÷ 1
N
n=÷N
¯
R
x
(n; m)e
÷j2p¯an
(3.31)
is the cyclic autocorrelation function at cycle
frequency ¯a and
¯
A9{¯a c [÷
1
2
;
1
2
) :
¯
R
¯a
x
(m)c0] (3.32)
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 645
is a countable set. Note that the cyclic autocorre
lation function
¯
R
¯a
x
(m) is periodic in ¯a with period
1. Thus, the sum in (3.30) can be equivalently
extended to the set
¯
A
1
9{¯a c [0; 1) :
¯
R
¯a
x
(m)c0].
In general, the set
¯
A (or
A
1
) contains possibly
incommensurate cycle frequencies ¯a. In the special
case where
¯
A
1
¬ {0; 1=N
0
; . . . ; (N
0
÷1)=N
0
] for
some integer N
0
, the autocorrelation function
¯
R
x
(n; m) is periodic in n with period N
0
and the
process x(n) is said to be cyclostationary in the wide
sense. If N
0
= 1 then x(n) is widesense stationary.
Let
¯
X(n)9
ncZ
x(n)e
÷j2pnn
(3.33)
be the stochastic process obtained from Fourier
transformation (in a generalized sense) of the ACS
process x(n). By using (3.30), in the sense of
distributions it can be shown that [3.59]
E{
¯
X(n
1
)
¯
X
+
(n
2
)]
=
¯ac
¯
A
¯
S
¯a
x
(n
1
)
‘cZ
d(n
2
÷n
1
÷¯a ÷‘), (3:34)
where
¯
S
¯a
x
(n)9
mcZ
¯
R
¯a
x
(m)e
÷j2pnm
(3.35)
is the cyclic spectrum at cycle frequency ¯a. The
cyclic spectrum
¯
S
¯a
x
(n) is periodic in both n and ¯a
with period 1.
In [3.3,4.41], it is shown that discretetime
cyclostationary processes are harmonizable.
3.3. Time series
3.3.1. Continuoustime time series
A statistical analysis framework for time series
that is an alternative to the classical stochastic
process framework is the fractionoftime (FOT)
probability framework ﬁrst introduced for ACS
time series in [2.5,2.8,2.12]; see also [2.15]. In the
FOT probability framework, signals are modelled
as single functions of time (time series) rather than
sample paths of stochastic processes. This ap
proach turns out to be more appropriate when an
ensemble of realizations does not exist and would
have to be artiﬁcially introduced just to create a
mathematical model, that is, the stochastic pro
cess. Such a model can be unnecessarily abstract
when there is only a single time series at hand.
In the FOT probability approach, probabilistic
parameters are deﬁned through inﬁnitetime
averages of a single time series (and functions of
this time series) rather than through expected
values or ensemble averages of a stochastic
process. Moreover, starting from this single time
series, a (possibly time varying) probability dis
tribution function can be constructed and this
leads to an expectation operation and all the
associated familiar probabilistic concepts and
parameters, such as stationarity, cyclostationarity,
nonstationarity, independence, mean, variance,
moments, cumulants, etc. For comprehensive
treatments of the FOT probability framework see
[2.8,2.12,2.15] and, for more mathematical rigor
on foundations and existence proofs, see [23.15].
The extension of the Wold isomorphism to
cyclostationary sequences was ﬁrst introduced in
[2.5], treated more in depth in [2.12], and ﬁnally—
with mathematical rigor—in [3.87].
The timevariant FOT probability framework is
based on the decomposition of functions of a time
series into their (possibly zero) almostperiodic
components and residual terms. Starting from
such a decomposition, the expectation operator is
deﬁned. Speciﬁcally, for any ﬁniteaveragepower
time series x(t), let us consider the decomposition
x(t)9x
ap
(t) ÷ x
r
(t), (3.36)
where x
ap
(t) is an almostperiodic function and
x
r
(t) a residual term not containing ﬁnitestrength
additive sinewave components; that is,
¸x
r
(t)e
÷j2pat
)
t
¬ 0 \a c R. (3.37)
The almostperiodic component extraction operator
E
{a]
{·] is deﬁned to be the operator that extracts all
the ﬁnitestrength additive sinewave components
of its argument, that is,
E
{a]
{x(t)]9x
ap
(t). (3.38)
Let x(t), t c R, be a realvalued continuoustime
time series (a single function of time) and let us
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 646
assume that the set G
1
of frequencies (for every x)
of the almostperiodic component of the function
of t 1
{x(t)px]
is countable, where
1
{x(t)px]
9
1; t : x(t)px;
0; t : x(t)4x
_
(3.39)
is the indicator function of the set {t c R : x(t)px].
In [2.12] it is shown that the function of x
F
{a]
x(t)
(x)9E
{a]
{1
{x(t)px]
] (3.40)
for any t is a valid cumulative distribution func
tion except for the rightcontinuity property (in
the discontinuity points). That is, it has values
in [0; 1], is non decreasing, F
{a]
x(t)
(÷o) = 0, and
F
{a]
x(t)
(÷o) = 1. Moreover, its derivative with res
pect to x, denoted by f
{a]
x(t)
(x), is a valid probability
density function and, for any wellbehaved func
tion g(·), it follows that
E
{a]
{g(x(t))] =
_
R
g(x)f
{a]
x(t)
(x) dx (3.41)
which reveals that E
{a]
{·] is the expecta
tion operator with respect to the distribution
function F
{a]
x(t)
(x) for the time series x(t). The
result (3.41), ﬁrst introduced in [2.8], is referred
to as the fundamental theorem of temporal expecta
tion, by analogy with the corresponding funda
mental theorem of expectation from probability
theory.
For an almostperiodic signal x(t), we have
x(t) ¬ x
ap
(t) and, hence,
E
{a]
{x(t)] = x(t). (3.42)
That is, the almostperiodic functions are the
deterministic signals in the FOT probability
framework. All the other signals are the random
signals. Note that the term ‘‘random’’ here is not
intended to be synonymous with ‘‘stochastic’’. In
fact, the adjective stochastic is adopted, as usual,
when an ensemble of realizations or sample paths
exists, whereas the adjective random is used in
reference to a single function of time.
Analogously, a secondorder characterization
for the realvalued time series x(t) can be obtained
by using the almostperiodic component extraction
operator as the expectation operator. Speciﬁcally,
let us assume that the set G
2
of frequencies (for
every x
1
; x
2
and t) of the almostperiodic compo
nent of the function of t 1
{x(t÷t)px
1
]
1
{x(t)px
2
]
is
countable. Then, the function of x
1
and x
2
F
{a]
x(t÷t)x(t)
(x
1
; x
2
)
9E
{a]
{1
{x(t÷t)px
1
]
1
{x(t)px
2
]
] (3:43)
is a valid secondorder joint cumulative distribu
tion function for every ﬁxed t and t, except for the
rightcontinuity property (in the discontinuity
points) with respect to x
1
and x
2
. Moreover, the
secondorder derivative, with respect to x
1
and x
2
,
of F
{a]
x(t÷t)x(t)
(x
1
; x
2
), denoted by f
{a]
x(t÷t)x(t)
(x
1
; x
2
), is
a valid secondorder joint probability density
function [2.12]. Furthermore, it can be shown that
the function
R
x
(t; t)9E
{a]
{x(t ÷t)x(t)] (3.44)
is a valid autocorrelation function and can be
characterized by
R
x
(t; t) =
_
R
2
x
1
x
2
f
{a]
x(t÷t)x(t)
(x
1
; x
2
) dx
1
dx
2
=
acA
R
a
x
(t)e
j2pat
, (3:45)
where A is a countable set and
R
a
x
(t)9¸x(t ÷t)x(t)e
÷j2pat
)
t
(3.46)
is the (nonstochastic) cyclic autocorrelation func
tion at cycle frequency a (a c R).
The classiﬁcation of the kind of nonstationarity
of a time series is made on the basis of the elements
contained in the set A. In general, the set A can
contain incommensurate cycle frequencies a and,
in such a case, the time series is said to be wide
sense almost cyclostationary. In the special case
where A ¬ {k=T
0
]
kcZ
the time series x(t) is said to
be widesense cyclostationary. If the set A contains
only the element a = 0, then the time series x(t) is
said to be widesense stationary.
Finally, note that the periodic component with
period T
0
of an almostperiodic time series (or lag
product time series) z(t) can be extracted by
exploiting the synchronized averaging identity
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 647
introduced in [2.5,2.8]:
÷o
k=÷o
z
k=T
0
e
j2p(k=T
0
)t
= lim
N÷o
1
2N ÷ 1
N
n=÷N
z(t ÷nT
0
), (3:47)
where the Fourier coefﬁcients z
k=T
0
are deﬁned
according to (3.9).
ACS time series are characterized in the spectral
domain by the (nonstochastic) cyclic spectrum or
spectral correlation density function at cycle fre
quency a:
S
a
x
(f )9 lim
Df ÷0
lim
T÷o
1
T
_
T=2
÷T=2
Df
X
1=Df
(t; f )X
+
1=Df
(t; f ÷a) dt, (3:48)
where X
1=Df
(t; f ) is deﬁned according to (3.21) and
the order of the two limits cannot be reversed. The
Cyclic Wiener Relation introduced in [2.8]
S
a
x
(f ) =
_
R
R
a
x
(t)e
÷j2pf t
dt (3.49)
links the cyclic autocorrelation function to the
cyclic spectrum. This relation generalizes that for
a = 0, which was ﬁrst dubbed the Wiener Relation
in [2.5] to distinguish it from the Khinchin Relation
(3.19), which is frequently called the Wiener–
Khinchin Relation.
3.3.2. Discretetime time series
The characterization of discretetime time series
(sequences) is similar to that of continuoustime
time series. We consider here only the widesense
secondorder characterization.
Let x(n), n c Z, be a realvalued discretetime
time series. Let us assume that the set
¯
G
2
of
frequencies (for every x
1
; x
2
and m) of the almost
periodic component of the function of
n 1
{x(n÷m)px
1
]
1
{x(n)px
2
]
is countable. Then, the
function of x
1
and x
2
F
{¯a]
x(n÷m)x(n)
(x
1
; x
2
)
9E
{¯a]
{1
{x(n÷m)px
1
]
1
{x(n)px
2
]
] (3:50)
is a valid secondorder joint cumulative distribu
tion function for every ﬁxed n and m, except for
the rightcontinuity property (in the discontinuity
points) with respect to x
1
and x
2
. In (3.50), E
{¯a]
{·]
denotes the discretetime almostperiodic compo
nent extraction operator, which is deﬁned analo
gously to its continuoustime counterpart. The
secondorder derivative with respect to x
1
and x
2
of F
{¯a]
x(n÷m)x(n)
(x
1
; x
2
), denoted by f
{¯a]
x(n÷m)x(n)
(x
1
; x
2
),
is a valid secondorder joint probability density
function [2.12]. Furthermore, it can be shown that
the function
¯
R
x
(n; m)9E
{¯a]
{x(n ÷ m)x(n)] (3.51)
is a valid autocorrelation function and can be
characterized by
¯
R
x
(n; m) =
_
R
2
x
1
x
2
f
{¯a]
x(n÷m)x(n)
(x
1
; x
2
) dx
1
dx
2
=
¯ac
¯
A
¯
R
¯a
x
(m)e
j2p¯an
, (3:52)
where
¯
A9{¯a c [÷
1
2
;
1
2
) :
¯
R
¯a
x
(m)c0] (3.53)
is a countable set and
¯
R
¯a
x
(m)9 lim
N÷o
1
2N ÷1
N
n=÷N
x(n ÷m)x(n)e
÷j2p¯an
(3.54)
is the (nonstochastic) discretetime cyclic autocor
relation function at cycle frequency ¯a.
Obviously, as in the stochastic framework, the
cyclic autocorrelation function
¯
R
¯a
x
(m) is periodic in
¯a with period 1. Thus, the sum in (3.52) can be
equivalently extended to the set
¯
A
1
9{¯a c [0; 1) :
¯
R
¯a
x
(m)c0]. Moreover, as in the continuoustime
case, the classiﬁcation of the kind of nonstationarity
of discretetime time series is made on the basis of
the elements contained in set
¯
A. That is, in general,
set
¯
A can contain incommensurate cycle frequencies
¯a and, in such a case, the time series is said to be
widesense almost cyclostationary. In the special case
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 648
where
¯
A
1
¬ {0; 1=N
0
; . . . ; (N
0
÷ 1)=N
0
] for some
integer N
0
, the time series x(n) is said to be wide
sense cyclostationary. If the set
¯
A contains only the
element ¯a = 0, then the time series x(n) is said to be
widesense stationary.
Discretetime ACS time series are characterized
in the spectral domain by the (nonstochastic)
cyclic spectrum or spectral correlation density
function at cycle frequency ¯a:
¯
S
¯a
x
(n)9 lim
Dn÷0
lim
N÷o
1
2N ÷1
N
n=÷N
Dn
X
¸1=Dn¡
(n; n)X
+
¸1=Dn¡
(n; n ÷¯a), (3:55)
where
X
2M÷1
(n; n)9
n÷M
k=n÷M
x(k)e
÷j2pnk
(3.56)
and ¸·¡ denotes the closest odd integer. The Cyclic
Wiener relation [2.5]
¯
S
¯a
x
(n) =
mcZ
¯
R
¯a
x
(m)e
÷j2pnm
(3.57)
links the cyclic autocorrelation function to the
cyclic spectrum.
3.4. Link between the stochastic and fractionof
time approaches
In the timevariant nonstochastic framework
(for ACS time series), the almostperiodic func
tions play the same role as that played by the
deterministic functions in the stochasticprocess
framework, and the expectation operator is the
almostperiodic component extraction operator.
Therefore, in the case of processes exhibiting
suitable ergodicity properties, results derived in
the FOT probability approach for time series can
be interpreted in the classical stochastic process
framework by substituting, in place of the almost
periodic component extraction operator E
{a]
{·],
the statistical expectation operator E{·]. In fact, by
assuming that {x(t; o); t c R; o c O] is a stochastic
process satisfying appropriate ergodicity proper
ties (see Section 4), the stochastic and FOT
autocorrelation functions are identical for almost
all sample paths x(t),
R
x
(t; t)9E{x(t ÷ t)x(t)]
= E
{a]
{x(t ÷ t)x(t)]9R
x
(t; t) (3:58)
and, hence, so too are their cyclic components and
frequencydomain counterparts,
R
a
x
(t) = R
a
x
(t), (3.59)
S
a
x
(f ) = S
a
x
(f ). (3.60)
Analogous equivalences hold in the discretetime
case.
The link between the two approaches is treated
in depth in [2.5,2.8,2.9,2.11,2.12,2.15,3.87,4.31,
23.15]. In the following, most results are presented
in the FOT probability framework.
3.5. Complex processes and time series
The case of complexvalued processes and time
series, ﬁrst treated in depth in [2.9], is extensively
treated in [2.6,2.8,3.43,13.13,13.14,13.20,13.25].
Several results with reference to higherorder
statistics are reported in Section 13.
Let x(t) be a zeromean complexvalued con
tinuoustime time series. Its widesense character
ization can be made in terms of the two second
order moments
R
xx
+ (t; t)9E
{a]
{x(t ÷t)x
+
(t)]
=
acA
xx
+
R
a
xx
+ (t)e
j2pat
, (3:61)
R
xx
(t; t)9E
{a]
{x(t ÷t)x(t)]
=
bcA
xx
R
b
xx
(t)e
j2pbt
(3:62)
which are called the autocorrelation function and
conjugate autocorrelation function, respectively.
The Fourier coefﬁcients
R
a
xx
+ (t)9¸x(t ÷ t)x
+
(t)e
÷j2pat
)
t
, (3.63)
R
b
xx
(t)9¸x(t ÷t)x(t)e
÷j2pbt
)
t
(3.64)
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 649
are referred to as the cyclic autocorrelation function
at cycle frequency a and the conjugate cyclic
autocorrelation function at conjugate cycle fre
quency b, respectively. The Fourier transforms of
the cyclic autocorrelation function and conjugate
cyclic autocorrelation function
S
a
xx
+ (f )9
_
R
R
a
xx
+ (t)e
÷j2pf t
dt, (3.65)
S
b
xx
(f )9
_
R
R
b
xx
(f )e
÷j2pf t
dt (3.66)
are called the cyclic spectrum and the conjugate
cyclic spectrum, respectively.
Let us denote by the superscript (+) an optional
complex conjugation. In the following, when it
does not create ambiguity, both functions
deﬁned in (3.63) and (3.64) are simultaneously
represented by
R
a
xx
(+)
(t)9 x(t ÷t)x
(+)
(t)e
÷j2pat
¸ _
t
; a c A
xx
(+) .
(3.67)
Analogously, both functions deﬁned in (3.65) and
(3.66) are simultaneously represented by
S
a
xx
(+)
(f )9
_
R
R
a
xx
(+)
(t)e
÷j2pf t
dt. (3.68)
It should be noted that, in [2.6,2.8,2.9], for
example, a different conjugation notation in the
subscripts is adopted to denote the autocorrelation
and the conjugate autocorrelation function. Spe
ciﬁcally, R
xx
(t; t) = E
{a]
{x(t ÷ t=2)x
+
(t ÷ t=2)] and
R
xx
+ (t; t) = E
{a]
{x(t ÷t=2)x(t ÷t=2)]. Moreover,
an analogous notation is adopted in these refer
ences and others for the (conjugate) cyclic auto
correlation function and the (conjugate) cyclic
spectrum.
3.6. Linear ﬁltering
The linear almostperiodically timevariant ﬁl
tering of ACS signals introduced in [2.2] and
followed by [2.5] for cyclostationary processes and
generalized in [2.8,2.9] to ACS signals, is also
considered in followon work in [7.60,9.51,13.9,
13.14,13.20] as well as the early Russian work in
[3.9]. Properties of linear periodically timevarying
systems are analyzed in [23.8,23.12,23.13,23.14].
More general timevariant linear ﬁltering is ad
dressed in [21.10,21.11].
3.6.1. Structure of linear almostperiodically time
variant systems
A linear timevariant system with input x(t),
output y(t), impulseresponse function h(t; u), and
input–output relation
y(t) =
_
R
h(t; u)x(u) du (3.69)
is said to be linear almostperiodically timevariant
(LAPTV) if the impulseresponse function admits
the Fourier series expansion
h(t; u) =
scG
h
s
(t ÷ u)e
j2psu
, (3.70)
where G is a countable set.
By substituting (3.70) into (3.69) we see that the
output y(t) can be expressed in the two equivalent
forms [2.8,9.51]:
y(t) =
scG
h
s
(t) Q [x(t)e
j2pst
] (3.71a)
=
scG
[g
s
(t) Q x(t)]e
j2pst
, (3.71b)
where
g
s
(t)9h
s
(t)e
÷j2pst
. (3.72)
From (3.71a) it follows that a LAPTV systems
performs a linear timeinvariant ﬁltering of
frequencyshifted version of the input signal.
For this reason LAPTV ﬁltering is also referred
to as frequencyshift (FRESH) ﬁltering [7.31].
Equivalently, from (3.71b) it follows that a
LAPTV systems performs a frequency shift
of linear timeinvariant ﬁltered versions of the
input.
In the special case for which G ¬ {k=T
0
]
kcZ
for
some period T
0
, the system is said to be linear
periodically timevariant (LPTV). If G contains
only the element s = 0, then the system is linear
timeinvariant (LTI).
3.6.2. Input/output relations in terms of cyclic
statistics
Let x
i
(t), i = 1; 2, t c R, be two possibly
complex ACS continuoustime time series with
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 650
secondorder (conjugate) crosscorrelation func
tion
R
x
1
x
(+)
2
(t; t)9E
{a]
{x
1
(t ÷ t)x
(+)
2
(t)]
=
acA
12
R
a
x
1
x
(+)
2
(t)e
j2pat
, (3:73)
where
R
a
x
1
x
(+)
2
(t)9¸x
1
(t ÷ t)x
(+)
2
(t)e
÷j2pat
)
t
(3.74)
is the (conjugate) cyclic crosscorrelation between
x
1
and x
2
at cycle frequency a and
A
12
9{a c R : R
a
x
1
x
(+)
2
(t)c0] (3.75)
is a countable set. If the set A
12
contains at
least one nonzero element, then the time series
x
1
(t) and x
2
(t) are said to be jointly ACS. Note
that, in general, set A
12
depends on whether (+)
is conjugation or not and can be different from
the sets A
11
and A
22
(both deﬁned according to
(3.75)).
Let us consider now two linear LAPTV systems
whose impulseresponse functions admit the Four
ier series expansions
h
i
(t; u) =
s
i
cG
i
h
s
i
(t ÷ u)e
j2ps
i
u
; i = 1; 2. (3.76)
The (conjugate) crosscorrelation of the outputs
y
i
(t) =
_
R
h
i
(t; u)x
i
(u) du; i = 1; 2 (3.77)
is given by
R
y
1
y
(+)
2
(t; t)9E
{a]
{y
1
(t ÷ t)y
(+)
2
(t)]
=
acA
12
s
1
cG
1
s
2
cG
2
[R
a
x
1
x
(+)
2
(t)e
j2ps
1
t
]
Q
t
r
a÷s
1
÷(÷)s
2
s
1
s
2
(+)
(t)e
j2p(a÷s
1
÷(÷)s
2
)t
, (3:78)
where Q
t
denotes convolution with respect to t, (÷)
is an optional minus sign that is linked to (+), and
r
g
s
1
s
2
(+)
(t)9
_
R
h
s
1
(t ÷ s)h
(+)
s
2
(s)e
÷j2pgs
ds. (3.79)
Thus,
R
b
y
1
y
(+)
2
(t)9¸y
1
(t ÷t)y
(+)
2
(t)e
÷j2pbt
)
t
=
s
1
cG
1
s
2
cG
2
[R
b÷s
1
÷(÷)s
2
x
1
x
(+)
2
(t)e
j2ps
1
t
]
Q
t
r
b
s
1
s
2
(+)
(t), (3:80)
S
b
y
1
y
(+)
2
(f )9
_
R
R
b
y
1
y
(+)
2
(t)e
÷j2pf t
dt
=
s
1
cG
1
s
2
cG
2
S
b÷s
1
÷(÷)s
2
x
1
x
(+)
2
(f ÷ s
1
)
H
s
1
(f )H
(+)
s
2
((÷)(b ÷ f )), (3:81)
where
H
s
i
(f )9
_
R
h
s
i
(t)e
÷j2pf t
dt (3.82)
and, in the sums in (3.80) and (3.81), only those
s
1
c G
1
and s
2
c G
2
such that b ÷s
1
÷ (÷)s
2
c
A
12
give nonzero contribution.
Eqs. (3.80) and (3.81) can be specialized to
several cases of interest. For example, if
x
1
= x
2
= x, h
1
= h
2
= h, y
1
= y
2
= y, and (+) is
conjugation, then we obtain the input–output
relations for LAPTV systems in terms of cyclic
autocorrelation functions and cyclic spectra:
R
b
yy
+ (t) =
s
1
cG
s
2
cG
[R
b÷s
1
÷s
2
xx
+ (t)e
j2ps
1
t
]
Q
t
r
b
12
(t), (3:83)
S
b
yy
+ (f ) =
s
1
cG
s
2
cG
S
b÷s
1
÷s
2
xx
+ (f ÷ s
1
)
H
s
1
(f )H
+
s
2
(f ÷ b), (3:84)
where
r
b
12
(t)9
_
R
h
s
1
(t ÷s)h
+
s
2
(s)e
÷j2pbs
ds. (3.85)
For further examples and applications, see Sec
tions 7 and 10.
3.7. Product modulation
Let x(t) and c(t) be two ACS signals with
(conjugate) autocorrelation functions
R
xx
(+) (t; t) =
a
x
cA
xx
(+)
R
a
x
xx
(+)
(t)e
j2pa
x
t
, (3.86)
R
cc
(+) (t; t) =
a
c
cA
cc
(+)
R
a
c
cc
(+)
(t)e
j2pa
c
t
. (3.87)
If x(t) and c(t) are statistically independent in the
FOT probability sense, then their joint probability
density function factors into the product of the
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 651
marginal probability densities [2.8,2.9,2.12] and
the (conjugate) autocorrelation functions of the
product waveform
y(t) = c(t)x(t) (3.88)
also factor,
R
yy
(+) (t; t) = R
cc
(+) (t; t)R
xx
(+) (t; t). (3.89)
Therefore, the (conjugate) cyclic autocorrelation
function and the (conjugate) cyclic spectrum of
y(t) are [2.5,2.8]:
R
a
yy
(+)
(t) =
a
x
cA
xx
(+)
R
a
x
xx
(+)
(t)R
a÷a
x
cc
(+)
(t), (3.90)
S
a
yy
(+)
(f ) =
a
x
cA
xx
(+)
_
R
S
a
x
xx
(+)
(l)S
a÷a
x
cc
(+)
(f ÷ l) dl,
(3.91)
where, in the sums, only those (conjugate) cycle
frequencies a
x
such that a ÷ a
x
c A
cc
(+) give non
zero contribution.
Observe that, if c(t) is an almostperiodic
function
c(t) =
gcG
c
g
e
j2pgt
, (3.92)
then
R
cc
(+) (t; t) = c(t ÷ t)c
(+)
(t)
=
g
1
cG
g
2
cG
c
g
1
c
(+)
g
2
e
(÷)j2pg
2
t
e
j2p(g
1
÷(÷)g
2
)t
(3:93)
and
R
a
cc
(+)
(t) =
gcG
c
g
c
(+)
(÷)(a÷g)
e
j2p(a÷g)t
, (3.94)
S
a
cc
(+)
(f ) =
gcG
c
g
c
(+)
(÷)(a÷g)
d(f ÷ g ÷a). (3.95)
3.8. Supports of cyclic spectra of band limited
signals
By specializing (3.81) to the case x
1
= x
2
= x,
h
1
= h
2
= h (LTI), and y
1
= y
2
= y, we obtain
S
a
yy
(+)
(f ) = S
a
xx
(+)
(f )H(f )H
(+)
((÷)(a ÷ f )). (3.96)
From (3.96) it follows that the support in the (a; f )
plane of the (conjugate) cyclic spectrum of y(t) is
such that
supp[S
a
yy
(+)
(f )]9{(a; f ) c R R : S
a
yy
(+)
(f )a0]
_ {(a; f ) c R R : H(f )
H
(+)
((÷)(a ÷ f ))a0]. (3:97)
Let x(t) be a strictly bandlimited lowpass signal
with monolateral bandwidth B; that is, S
0
xx
+ (f ) ¬ 0
for f e(÷B; B). Then,
x(t) ¬ x(t) Q h
LPF
(t), (3.98)
where h
LPF
(t) is the impulseresponse function of
the ideal lowpass ﬁlter with harmonicresponse
function:
H
LPF
(f ) = rect
f
2B
_ _
9
1; [f [pB;
0; [f [4B:
_
(3.99)
Accounting for (3.97), we have (see Fig. 1)
supp[S
a
xx
(+)
(f )] _ {(a; f ) c R R : H
LPF
(f )
H
LPF
(f ÷ a)a0], (3:100)
where the fact that rect(f ) is real and even is
used.
ARTICLE IN PRESS
B −B
2B
−2B
f
Fig. 1. Cyclicspectrum support of a lowpass signal.
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 652
Let x(t) be a strictly bandlimited highpass
signal; that is, S
0
xx
+ (f ) ¬ 0 for f c (÷b; b). Then,
x(t) ¬ x(t) Q h
HPF
(t), (3.101)
where h
HPF
(t) is the impulseresponse function of
the ideal highpass ﬁlter with harmonicresponse
function
H
HPF
(f ) = 1 ÷rect
f
2b
_ _
. (3.102)
From (3.97), we have (see Fig. 2)
supp[S
a
xx
(+)
(f )] _ {(a; f ) c R R : H
HPF
(f )
H
HPF
(f ÷a)a0]. (3:103)
Finally, let x(t) be a strictly bandlimited band
pass signal; that is, S
0
xx
+ (f ) ¬ 0 for f e(÷B; ÷b)C
(b; B), where 0oboB. Then,
x(t) ¬ x(t) Q h
BPF
(t), (3.104)
where h
BPF
(t) is the impulseresponse function of
the ideal bandpass ﬁlter with harmonicresponse
function
H
BPF
(f ) = H
LPF
(f )H
HPF
(f ), (3.105)
where H
LPF
(f ) and H
HPF
(f ) are given by (3.99)
and (3.102), respectively. Therefore, accounting
for (3.97), we have (see Fig. 3)
supp[S
a
xx
(+)
(f )]
_ {(a; f ) c R R : H
BPF
(f )H
BPF
(f ÷ a)a0]
= {(a; f ) c R R : H
LPF
(f )H
LPF
(f ÷ a)a0]
¨ {(a; f ) c R R : H
HPF
(f )
H
HPF
(f ÷ a)a0]. (3:106)
It is noted that supports for symmetric deﬁnitions
of cyclic spectra (3.28) are reported in [2.6,2.8].
3.9. Sampling and aliasing
Let x(n) be the sequence obtained by uniformly
sampling, with period T
s
= 1=f
s
, the continuous
time signal x
a
(t):
x(n)9x
a
(t)[
t=nT
s
. (3.107)
The (conjugate) autocorrelation function of the
discretetime signal x(n) can be shown to be the
sampled version of the (conjugate) autocorrelation
function of the continuoustime signal x
a
(t):
E
{¯a]
{x(n ÷m)x
(+)
(n)]
= E
{a]
{x
a
(t ÷t)x
a
(t)][
t=nT
s
;t=mT
s
. (3:108)
ARTICLE IN PRESS
b −b
2b
−2b
f
Fig. 2. Cyclicspectrum support of a highpass signal.
b −b
2b
−2b
B −B
2B
−2B
f
Fig. 3. Cyclicspectrum support of a bandpass signal.
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 653
However, the (conjugate) cyclic autocorrelation
functions of x(n) are not sampled versions of the
(conjugate) cyclic autocorrelation functions of
x
a
(t) because of the presence of aliasing in the
cyclefrequency domain. Consequently, for the
(conjugate) cyclic spectra, aliasing in both the
spectralfrequency and cyclefrequency domains
occurs. Speciﬁcally, the (conjugate) cyclic auto
correlation functions and the (conjugate) cyclic
spectra of x(n) can be expressed in terms of the
(conjugate) cyclic autocorrelation functions and
the (conjugate) cyclic spectra of x
a
(t) by the
relations [2.5,2.8,13.20,13.23]:
¯
R
¯a
xx
(+) (m) =
pcZ
R
a÷pf
s
x
a
x
(+)
a
(t)[
t=mT
s
;a=¯af
s
, (3.109)
¯
S
¯a
xx
(+) (n) =
1
T
s
pcZ
qcZ
S
a÷pf
s
x
a
x
(+)
a
(f ÷ qf
s
)[
f =nf
s
;a=¯af
s
.
(3.110)
Let x(t) be a strictly bandlimited lowpass signal
with monolateral bandwidth B. From (3.100), it
follows that
supp[S
a
xx
(+)
(f )]
_ {(a; f ) c R R : [f [pB; [a ÷ f [pB]
_ {(a; f ) c R R : [f [pB; [a[p2B]. (3:111)
Thus, the support of each replica in (3.110) is
contained in the set
{(a; f ) c R R : [f ÷ qf
s
[pB; [a ÷ pf
s
[p2B]
and, consequently, a sufﬁcient condition for
assuring that the replicas in (3.110) do not
overlap is
f
s
X4B. (3.112)
In such a case, only the replica with p = 0 gives
nonzero contribution in the base support region
in (3.109) and (3.110). Thus, the (conjugate)
cyclic autocorrelation functions and the cyclic
spectra of the continuoustime signal x
a
(t) are
amplitude and/or time or frequencyscaled ver
sions of the (conjugate) cyclic autocorrelation
functions and cyclic spectra of the discretetime
signal x(n) [13.20]:
R
a
x
a
x
(+)
a
(t)[
t=mT
s
=
¯
R
¯a
xx
(+) (m)[
¯a=a=f
s
; [a[p
f
s
2
;
0 otherwise;
_
¸
_
¸
_
(3:113)
S
a
x
a
x
(+)
a
(f )
=
T
s
¯
S
¯a
xx
(+) (n)[
n=f =f
s
;¯a=a=f
s
; [a[p
f
s
2
; [f [p
f
s
2
;
0 otherwise:
_
_
_
(3:114)
The discretetime signal obtained by sampling,
with period T
s
, a cyclostationary continuoustime
signal with cyclostationarity period T
0
= KT
s
, K
integer, is a discretetime cyclostationary signal
with cyclostationarity period K. If, however,
T
0
= KT
s
÷ , with 0ooT
s
and incommensu
rate with T
s
, then the discretetime signal is
almostcyclostationary [13.23]. In [3.44], common
pitfalls arising in the application of the stationary
signal theory to time sampled cyclostationary
signals are examined.
3.10. Representations by stationary components
3.10.1. Continuoustime processes and time series
A continuoustime widesense cyclostationary
signal (process or timeseries) x(t) can be expressed
in terms of singularly and jointly wide
sense stationary signals with nonoverlapping
spectral bands [2.5,3.22,3.23,3.77,4.41]. That is,
if T
0
is the period of cyclostationarity, and we
deﬁne
¯ x
k
(t)9[x(t)e
÷j2p(k=T
0
)t
] Q h
0
(t), (3.115)
where h
0
(t)9(1=T
0
) sinc(t=T
0
) is the impulse
response function of an ideal lowpass ﬁlter
with monolateral bandwidth 1=(2 T
0
), then
x(t) can be expressed by the harmonic series
representation:
x(t) =
÷o
k=÷o
¯ x
k
(t)e
j2p(k=T
0
)t
, (3.116)
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 654
where
E
{a]
{¯ x
k
(t ÷t)¯ x
+
h
(t)]
=
_
1=(2 T
0
)
÷1=(2 T
0
)
S
(h÷k)=T
0
xx
+ f ÷
k
T
0
_ _
e
j2pf t
df (3:117)
which is independent of t. This result reﬂects the
fact that the Fourier transforms
¯
X
k
(f ) of the
signals ¯ x
k
(t) have nonoverlapping support of
width 1=T
0
¯
X
k
f ÷
k
T
0
_ _
=
X(f ); [f ÷ k=T
0
[p1=(2 T
0
);
0 otherwise;
_
(3:118)
therefore, since only spectral components of x(t)
with frequencies separated by an integer multiple
of 1=T
0
can be correlated, then the only pairs of
spectral components in ¯ x
k
(t) and ¯ x
h
(t) that can be
correlated are those with the same frequency f.
Thus, there is no spectral crosscorrelation at
distinct frequencies in ¯ x
k
(t) and ¯ x
h
(t).
It follows that a continuoustime widesense
cyclostationary scalar signal x(t) is equivalent to
the inﬁnitedimensional vectorvalued widesense
stationary signal
[. . . ; ¯ x
÷k
(t); . . . ; ¯ x
÷1
(t); ¯ x
0
(t); ¯ x
1
(t); . . . ; ¯ x
k
(t); . . .].
Another representation of a cyclostationary signal
by stationary components is the translation series
representation in terms of any complete orthonor
mal set of basis functions [3.23]. One example uses
the Karhunen–Loe` ve expansion of the cyclosta
tionary signal x(t) on the intervals t c [nT
0
; (n ÷
1)T
0
) for all integers n, where T
0
is the period of
cyclostationarity.
A harmonic series representation can also be
obtained for an ACS process x(t), provided that it
belongs to the subclass of the almostperiodically
unitary processes [3.65]. In this case,
x(t) =
÷o
k=÷o
¯ x
k
(t)e
j2pl
k
t
, (3.119)
where the frequencies l
k
are possibly incommen
surate and the processes ¯ x
k
(t) are singularly and
jointly widesense stationary but not necessarily
band limited. No translation series representation
with stationary components has been demon
strated to exist for ACS processes.
3.10.2. Discretetime processes and time series
A discretetime widesense cyclostationary sig
nal (process or timeseries) x(n) can be expressed in
terms of a ﬁnite number of singularly and jointly
widesense stationary signals ¯ x
k
(n) with non over
lapping bands [2.1,2.5,2.17,4.41,12.3,12.30]. That
is, if N
0
is the period of cyclostationarity, and we
deﬁne
¯ x
k
(n)9[x(n)e
÷j2p(k=N
0
)n
] Q h
0
(n), (3.120)
where h
0
(n)9(1=N
0
) sinc(n=N
0
) is the ideal low
pass ﬁlter with monolateral bandwidth 1=(2N
0
),
then x(n) can be expressed by the harmonic series
representation:
x(n) =
N
0
÷1
k=0
¯ x
k
(n)e
j2p(k=N
0
)n
. (3.121)
Therefore, a discretetime widesense cyclostation
ary scalar signal x(n) is equivalent to the N
0

dimensional vectorvalued widesense stationary
signal
[¯ x
0
(n); ¯ x
1
(n); . . . ; ¯ x
N
0
÷1
(n)].
A further decomposition of a discretetime cyclos
tationary signal can be obtained in terms of sub
sampled components. Let x(n) be a discretetime
realvalued widesense cyclostationary timeseries
with period N
0
. The subsampled (or decimated)
timeseries:
x
i
(n)9x(nN
0
÷i); i = 0; . . . ; N
0
÷ 1 (3.122)
constitute what is called the polyphase decomposi
tion of x(n). Given the set of time series x
i
(n),
i = 0; . . . ; N
0
÷ 1, the original signal x(n) can be
reconstructed by using the synthesis formula:
x(n) =
N
0
÷1
i=0
‘cZ
x
i
(‘)d
n÷i÷‘N
0
, (3.123)
where d
g
is the Kronecker delta (d
g
= 1 for g = 0
and d
g
= 0 for ga0). The signal x(n) is wide
sense cyclostationary with period N
0
if and
only if the set of subsampled x
i
(n) are jointly
widesense stationary [3.3]. In fact, due to the
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 655
cyclostationarity of x(n):
E
{a]
{x
i
(n ÷m)x
k
(n)]
= E
{a]
{x((n ÷m)N
0
÷i)x(nN
0
÷k)]
= E
{a]
{x(mN
0
÷ i)x(k)], (3:124)
which is independent of n.
4. Ergodic properties and measurement of
characteristics
4.1. Estimation of the cyclic autocorrelation
function and the cyclic spectrum
Ergodic properties and measurements of char
acteristics are treated in [4.1–4.61]. Consistent
estimates of secondorder statistical functions of
an ACS stochastic process can be obtained provided
that the stochastic process has ﬁnite or ‘‘effectively
ﬁnite’’ memory. Such a property is generally
expressed in terms of mixing conditions or summa
bility of second and fourthorder cumulants. Under
such mixing conditions, the cyclic correlogram
R
a
x
(t; t
0
; T)9
1
T
_
t
0
÷T=2
t
0
÷T=2
x(t ÷ t)x(t)e
÷j2pat
dt (4.1)
is a consistent estimator of the cyclic autocorrelation
function R
a
x
(t) (see (3.11)). Moreover,
ﬃﬃﬃﬃ
T
_
[R
a
x
(t; t
0
; T) ÷R
a
x
(t)]
is an asymptotically (T ÷ o) zeromean complex
normal random variable for each t and t
0
.
Consistency for estimators of the cyclic autocorrela
tion function for cyclostationary and/or ACS
processes has been addressed in [2.1,3.1,4.3,4.13,
4.15,4.23,4.24,4.36,4.42,13.18]. The ﬁrst treatment
for ACS processes is in [4.13].
In the frequency domain, the cyclic periodogram
I
a
x
(t; f )9
1
T
X
T
(t; f )X
+
T
(t; f ÷a), (4.2)
where X
T
(t; l) is deﬁned according to (3.21), is
an asymptotically unbiased but not consistent
estimator of the cyclic spectrum S
a
x
(f ) (see (3.19)
and (3.20)). However, under the abovemen
tioned mixing conditions, the frequencysmoothed
cyclic periodogram
S
a
x
T
(t
0
; f )
Df
9
1
Df
_
f ÷Df =2
f ÷Df =2
1
T
X
T
(t
0
; l)X
+
T
(t
0
; l ÷ a) dl (4:3)
is a consistent estimator of the cyclic spectrum
S
a
x
(f ). Moreover,
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
T Df
_
[S
a
x
T
(t; f )
Df
÷S
a
x
(f )]
is an asymptotically (T ÷ o, Df ÷ 0, with
T Df ÷ o) zeromean complex normal random
variable for each f and t
0
. The ﬁrst detailed study
of the variance of estimators of the cyclic spectrum
is given in [2.8] and is based on the FOT
framework. Consistency for estimators of the
cyclic spectrum has been addressed in [2.1,4.14,
4.20,4.30,4.35,4.43,4.44,4.55,4.59,13.12].
In [2.5,2.8,4.17], it is shown that the time
smoothed cyclic periodogram:
S
a
x
1=Df
(t; f )
T
9
1
T
_
t÷T=2
t÷T=2
Df
X
1=Df
(s; f )X
+
1=Df
(s; f ÷a) ds (4:4)
is asymptotically equivalent to the frequency
smoothed cyclic periodogram in the sense that
lim
Df ÷0
lim
T÷o
S
a
x
T
(t; f )
Df
= lim
Df ÷0
lim
T÷o
S
a
x
1=Df
(t; f )
T
, (4:5)
where the order of the two limits on each side
cannot be reversed. Note that both the time
smoothed and frequencysmoothed cyclic period
ograms exhibit spectral frequency resolution on
the order of Df and a cycle frequency resolution
on the order of 1=T [4.17]. The problem of cyclic
leakage in the estimate of a cyclic statistic at cycle
frequency a arising from cyclic statistics at cycle
frequencies different from a, ﬁrst observed in [2.8],
is addressed in [2.8,2.9,4.17]. In particular, it is
shown that a strong stationary spectrally over
lapping noise component added to a cyclostation
ary signal degrades the performance (bias and
variance) of the estimators of cyclic statistics at
nonzero cycle frequencies because of the leakage
from the zero cycle frequency.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 656
A spectral correlation analyzer can be realized
by frequency shifting the signal x(t) by two
amounts differing by a, passing such frequency
shifted versions through two lowpass ﬁlters h
Df
(t)
with bandwidth Df and unity passband height,
and then correlating the output signals (see Fig. 4).
The spectral correlation density function S
a
x
(f ) is
obtained by normalizing the output by Df , and
taking the limit as the correlation time T ÷ o
and the bandwidth Df ÷ 0, in this order [4.17].
Reliable spectral estimates with reduced com
putational requirements can be obtained by using
nonlinear transformations of the data [4.39,4.51].
Strictsense ergodic properties of ACS processes,
referred to as cycloergodicity in the strict sense,
were ﬁrst treated in depth in [4.13]; see also [2.15].
A survey of estimation problems is given in [4.41].
Cyclostationary feature measurements in the non
stochastic approach are treated in considerable
depth in [4.17] and also in [4.31]. Problems arising
from the presence of jitter in measurements are
addressed in [4.53,4.61]. Computationally efﬁcient
digital implementations of cyclic spectrum analy
zers are developed and analyzed in [4.25,4.32,
4.38,4.45,4.48].
For measurements of cyclic higherorder statis
tics see [4.50,4.56,4.57,13.7,13.9,13.12,13.14,13.18,
13.29].
For further references, see the general treat
ments [2.1,2.2,2.5,2.8,2.9,2.11,2.15,2.18] and also
see [3.22,3.35,3.39,3.59,3.72,11.10,11.13,11.20,12.49,
12.51]. Measurements on cyclostationary random
ﬁelds are treated in [20.3,20.5]. The problem of
measurement of statistical functions for more
general classes of nonstationary signals is considered
in [21.1,21.5,21.12,21.14,21.15].
4.2. Two alternative approaches to the analysis of
measurements on time series
In the FOT probability approach, probabilistic
parameters are deﬁned through inﬁnitetime
averages of functions of a single time series (such
as products of time and frequencyshifted ver
sions of the time series) rather than through
expected values or ensemble averages of a stochas
tic process. Estimators of the FOT probabilistic
parameters are obtained by considering ﬁnitetime
averages of the same quantities involved in the
inﬁnitetime averages. Therefore, assuming the
abovementioned limits exist (that is, the inﬁnite
time averages exist), their asymptotic estimators
converge by deﬁnition to the true values, which are
exactly the inﬁnitetime averages, without the
necessity of requiring ergodicity properties as in
the stochastic process framework. Thus, in the
FOT probability framework, the kind of conver
gence of the estimators to be considered as the
datarecord length approaches inﬁnity is the
convergence of the function sequence of the
ﬁnitetime averages (indexed by the datarecord
length). Therefore, unlike the stochastic process
framework where convergence must be deﬁned, for
example, in the ‘‘stochastic meansquare sense’’
[3.59,4.41,13.18,13.26] or ‘‘almost sure sense’’
[4.35,4.56] or ‘‘in distribution’’, the convergence
in the FOT probability framework must be
considered ‘‘pointwise’’, in the ‘‘temporal mean
square sense’’ [2.8,2.9,2.31], or in the ‘‘sense of
generalized functions (distributions)’’ [23.9].
By following the guidelines in [2.8,2.9,23.1], let
us consider the convergence of time series in the
temporal meansquare sense (t.m.s.s.). Given a
ARTICLE IN PRESS
x (t)
× h
∆f
(t)
×
e
−j2πft
× h
∆f
(t) (⋅)*
e
−j2π( f−)t
〈⋅〉
T
∆f S
x
1/∆ f
(t,f )
T
Fig. 4. Spectral correlation analyzer.
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 657
time series z(t) (such as a lag product of another
time series), we deﬁne
z
b
(t)
T
9
1
T
_
t÷T=2
t÷T=2
z(u)e
÷j2pbu
du, (4.6)
z
b
9 lim
T÷o
1
T
_
÷T=2
÷T=2
z(u)e
÷j2pbu
du (4.7)
and we assume that
lim
T÷o
z
b
(t)
T
= z
b
(t:m:s:s:) \b c R, (4.8)
that is,
lim
T÷o
¸[z
b
(t)
T
÷ z
b
[
2
)
t
= 0 \b c R. (4.9)
It can be shown that, if the time series z(t) has
ﬁniteaveragepower (i.e., ¸[z(t)[
2
)
t
oo), then the
set B9{b c R : z
b
a0] is countable, the series
bcB
[z
b
[
2
is summable [2.9] and, accounting for
(4.8), it follows that
lim
T÷o
bcB
z
b
(t)
T
e
j2pbt
=
bcB
z
b
e
j2pbt
(t:m:s:s:).
(4.10)
The magnitude and phase of z
b
are the amplitude
and phase of the ﬁnitestrength additive complex
sinewave with frequency b contained in the time
series z(t). Moreover, the righthand side in (4.10)
is just the almostperiodic component contained in
the time series z(t).
The function z
b
(t)
T
is an estimator of z
b
based
on the observation {z(u); u c [t ÷T=2; t ÷ T=2]]. It
is worthwhile to emphasize that, in the FOT
probability framework, probabilistic functions are
deﬁned in terms of the almostperiodic component
extraction operation, which plays the same role as
that played by the statistical expectation operation
in the stochastic process framework [2.8,2.15].
Therefore,
bias{z
b
(t)
T
]9E
{a]
{z
b
(t)
T
] ÷ z
b
’ ¸z
b
(t)
T
)
t
÷ z
b
, (4:11)
var{z
b
(t)
T
]9E
{a]
{[z
b
(t)
T
÷E
{a]
{z
b
(t)
T
][
2
]
’ ¸[z
b
(t)
T
÷¸z
b
(t)
T
)
t
[
2
)
t
, (4:12)
where the approximation becomes exact equality
in the limit as T ÷ o. Thus, unlike the stochastic
process framework where the variance accounts
for ﬂuctuations of the estimates over the ensemble
of sample paths, in the FOT probability frame
work the variance accounts for the ﬂuctuations of
the estimates in the time parameter t, viz., the
central point of the ﬁnitelength time series
segment adopted for the estimation. Therefore,
the assumption that the estimator asymptotically
approaches the true value (the inﬁnitetime aver
age) in the meansquare sense is equivalent to the
statement that the estimator is meansquare
consistent in the FOT probability sense. In fact,
from (4.9), (4.11), and (4.12) it follows that
¸[z
b
(t)
T
÷ z
b
[
2
)
t
’ var{z
b
(t)
T
]
÷ [bias{z
b
(t)
T
][
2
(4:13)
and this approximation become exact as T ÷ o.
In such a case, estimates obtained by using
different time segments asymptotically do not
depend on the central point of the segment.
5. Manufactured signals: modelling and analysis
5.1. General aspects
Cyclostationarity in manmade communications
signals is due to signal processing operations used
in the construction and/or subsequent processing
of the signal, such as modulation, sampling,
scanning, multiplexing and coding operations
[5.1–5.23].
The analytical cyclic spectral analysis of math
ematical models of analog and digitally modulated
signals was ﬁrst carried out in [2.5] for stochastic
processes and in [5.9,5.10] for nonstochastic time
series. The effects of multiplexing are considered in
[5.13]. Continuousphase frequencymodulated
signals are treated in [5.12,5.18,5.19,5.23]. The
effects of timing jitter on the cyclostationarity
properties of communications signals are ad
dressed in [2.8,5.17,5.21].
On this general subject, see the general treat
ments [2.5,2.8,2.9,2.11,2.13], and also see [2.15,
6.11,7.22].
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 658
5.2. Examples of communication signals
In this section, two fundamental examples of
cyclostationary communication signals are con
sidered and their widesense cyclic statistics are
described. The derivations of the cyclic statistics
can be accomplished by using the results of
Sections 3.6 and 3.7. Then a third example of a
more sophisticated communication signal is con
sidered.
5.2.1. Double sideband amplitudemodulated
signal
Let x(t) be the (realvalued) double sideband
amplitudemodulated (DSBAM) signal:
x(t)9s(t) cos(2pf
0
t ÷f
0
). (5.1)
The cyclic autocorrelation function and cyclic
spectrum of x(t) are [5.9]:
R
a
x
(t) =
1
2
R
a
s
(t) cos (2pf
0
t)
÷
1
4
{R
a÷2f
0
s
(t)e
÷j2pf
0
t
e
÷j2f
0
÷ R
a÷2f
0
s
(t)e
j2pf
0
t
e
j2f
0
], (5:2)
S
a
x
(f ) =
1
4
{S
a
s
(f ÷f
0
) ÷ S
a
s
(f ÷ f
0
)
÷S
a÷2f
0
s
(f ÷ f
0
)e
÷j2f
0
÷S
a÷2f
0
s
(f ÷ f
0
)e
j2f
0
], (5:3)
respectively. If s(t) is a widesense stationary signal
then R
a
s
(t) = R
0
s
(t)d
a
and
R
a
x
(t) =
1
2
R
0
s
(t) cos(2pf
0
t); a = 0;
1
4
R
0
s
(t)e
±j2pf
0
t
e
±j2f
0
; a = ±2f
0
;
0 otherwise;
_
¸
_
¸
_
(5.4)
S
a
x
(f ) =
1
4
{S
0
s
(f ÷ f
0
) ÷S
0
s
(f ÷f
0
)]; a = 0;
1
4
S
0
s
(f ( f
0
)e
±j2f
0
; a = ±2f
0
;
0 otherwise:
_
¸
_
¸
_
(5.5)
Thus, x(t) is cyclostationary with period 1=(2 f
0
).
In Fig. 5(a) the magnitude of the cyclic
autocorrelation function R
a
x
(t), as a function of a
and t, and in Fig. 5(b) the magnitude of the cyclic
spectrum S
a
x
(f ), as a function of a and f, are
reported for the DSBAM signal (5.1) with
stationary modulating signal s(t) having triangular
autocorrelation function.
5.2.2. Pulseamplitudemodulated signal
Let x(t) be the complexvalued pulseamplitude
modulated (PAM) signal:
x(t)9
kcZ
a
k
q(t ÷kT
0
), (5.6)
where q(t) is a complexvalued square integrable
pulse and {a
k
]
kcZ
, a
k
c C, is an ACS sequence
whose cyclostationarity is possibly induced by
framing, multiplexing, or coding [13.21].
The (conjugate) cyclic autocorrelation fun
ction and (conjugate) cyclic spectrum of x(t)
ARTICLE IN PRESS
Fig. 5. (a) Magnitude of the cyclic autocorrelation function
R
a
x
(t), as a function of a and t, and (b) magnitude of the cyclic
spectrum S
a
x
(f ), as a function of a and f, for the DSBAM signal
(5.1).
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 659
are [5.9,5.10]:
R
a
xx
(+)
(t) =
1
T
0
mcZ
[
¯
R
¯a
aa
(+) (m)]
¯a=aT
0
r
a
qq
(+)
(t ÷ mT
0
), (5:7)
S
a
xx
(+)
(f ) =
1
T
0
[
¯
S
¯a
aa
(+) (n)]
n=fT
0
;¯a=aT
0
Q(f )Q
(+)
((÷)(a ÷ f )), (5:8)
respectively, where
¯
R
¯a
aa
(+) (m)9 lim
N÷o
1
2N ÷1
N
k=÷N
a
k÷m
a
(+)
k
e
÷j2p¯ak
,
(5.9)
¯
S
¯a
aa
(+) (n)9
mcZ
¯
R
¯a
aa
(+) (m)e
÷j2pnm
(5.10)
are the (conjugate) cyclic autocorrelation function
and the (conjugate) cyclic spectrum, respectively,
of the sequence {a
k
]
kcZ
,
Q(f )9
_
R
q(t)e
÷j2pft
dt (5.11)
and
r
a
qq
(+)
(t)9q(t) Q [q
(+)
(÷t)e
j2pat
]
=
_
R
q(t ÷ t)q
(+)
(t)e
÷j2pat
dt. (5:12)
If the sequence {a
k
]
kcZ
is widesense stationary
and white, then
¯
R
¯a
aa
+ (m) =
¯
R
0
aa
+ (0)d
(¯a mod 1)
d
m
, (5.13)
where mod denotes the modulo operation. In this
case, the cyclic autocorrelation function and the
cyclic spectrum of x(t) become
R
a
xx
+ (t) =
¯
R
0
aa
+ (0)
T
0
d
(aT
0
mod 1)
r
a
qq
+ (t), (5.14)
S
a
xx
+ (f ) =
¯
R
0
aa
+ (0)
T
0
d
(aT
0
mod 1)
Q(f )Q
+
(f ÷ a),
(5.15)
respectively. Thus, x(t) exhibits cyclostationarity
with cycle frequencies a = k=T
0
, k c Z; that is, x(t)
is cyclostationary with period T
0
.
In Fig. 6(a) the magnitude of the cyclic
autocorrelation function R
a
x
(t), as a function of a
and t, and in Fig. 6(b) the magnitude of the cyclic
spectrum S
a
x
(f ), as a function of a and f, are
reported for the PAM signal (5.6) with stationary
modulating sequence {a
k
]
kcZ
, and rectangular
pulse q(t)9rect((t ÷T
0
=2)=T
0
). In this case,
(5.12) reduces to
r
a
qq
(t) = e
÷jpa(T
0
÷t)
rect
t
2T
0
_ _
1 ÷
[t[
T
0
_ _
T
0
sinc(a(T
0
÷ [t[)). (5:16)
5.2.3. Directsequence spreadspectrum signal
Let x(t) be the directsequence spreadspectrum
(DSSS) baseband PAM signal
x(t)9
kcZ
a
k
q(t ÷kT
0
), (5.17)
ARTICLE IN PRESS
Fig. 6. (a) Magnitude of the cyclic autocorrelation function
R
a
x
(t), as a function of a and t, and (b) magnitude of the cyclic
spectrum S
a
x
(f ), as a function of a and f, for the PAM signal
(5.6).
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 660
where {a
k
]
kcZ
, a
k
c C, is an ACS sequence, T
0
is
the symbol period, and
q(t)9
N
c
÷1
n=0
c
n
p(t ÷ nT
c
) (5.18)
is the spreading waveform. In (5.18), {c
0
; . . . ;
c
N
c
÷1
] is the N
c
length spreading sequence (code)
with c
n
c C, and T
c
is the chip period such that
T
0
= N
c
T
c
.
The (conjugate) cyclic autocorrelation function
and (conjugate) cyclic spectrum of x(t) are
[2.8,8.24]:
R
a
xx
(+)
(t) =
1
T
0
mcZ
[
¯
R
¯a
aa
(+) (m)]
¯a=aT
0
r
a
pp
(+)
(t ÷ mT
0
) Q g
a
cc
(+)
(t), (5:19)
S
a
xx
(+)
(f ) =
1
T
0
[
¯
S
¯a
aa
(+) (n)]
n=fT
0
;¯a=aT
0
P(f )P
(+)
((÷)(a ÷f ))
¯
G
¯a
cc
(+) (n)[
n=fT
c
;¯a=aT
c
, (5:20)
where
g
a
cc
(+)
(t)9
N
c
÷1
n
1
=0
N
c
÷1
n
2
=0
c
n
1
c
(+)
n
2
e
÷j2pan
2
T
c
d(t ÷ (n
1
÷n
2
)T
c
) (5:21)
and
¯
G
¯a
cc
(+) (n)9C(n)C
(+)
((÷)(¯a ÷ n)) (5.22)
with
C(n)9
N
c
÷1
n=0
c
n
e
÷j2pnn
. (5.23)
6. Natural signals: modelling and analysis
Cyclostationarity occurs in data arising from a
variety of natural (not manmade) phenomena due
to the presence of periodic mechanisms in the
phenomena [6.1–6.27]. In climatology and atmo
spheric science, cyclostationarity is due to rotation
and revolution of the earth [6.1,6.2,6.9,6.12,6.13,
6.21–6.24]. Applications in hydrology are consid
ered in [6.3,6.11,6.14,6.16,6.17,6.19,6.25,6.26]; for
periodic ARMA modelling and the predic
tion problem in hydrology, see [12.16,12.20,12.21,
12.23]; for the modelling of ocean waves as a
twodimensional cyclostationary random ﬁeld
see [20.1].
A patent on a speech recognition technique
exploiting cyclostationarity is [6.27].
7. Communications systems: analysis and design
7.1. General aspects
Cyclostationarity properties of modulated sig
nals can be suitably exploited in the analysis and
design of communications systems (see [7.1–7.101])
since the signals involved are typically ACS (see
Section 5).
The cyclostationary nature of interference in
communications systems is characterized in
[7.5,7.6,7.8,7.9,7.18,7.25,7.47,7.98]. The problem
of optimum ﬁltering (cyclic Wiener ﬁltering) of
ACS signals is addressed in Section 7.2.
The problems of synchronization, signal para
meter and waveform estimation, channel identiﬁ
cation and equalization, and signal detection and
classiﬁcation are treated in Sections 8–11.
On the analysis and design of communications
systems, see the general treatments [2.5,2.8,
2.11,2.13], and also see [3.23,3.47,5.9,5.10,5.14,
10.7,22.8].
7.2. Cyclic Wiener ﬁltering
The problem of optimum linear ﬁltering consists
of designing the linear transformation of the data
x(t) that minimizes the meansquared error of the
ﬁlter output relative to a desired signal, say d(t). In
the case of complex data, the optimum ﬁlter is
obtained by processing both x(t) and x
+
(t), leading
to the linearconjugatelinear (LCL) structure
[2.8,2.9]. If d(t) and x(t) are jointly ACS signals,
the optimum ﬁltering is referred to (as ﬁrst
suggested in [7.31]) as cyclic Wiener ﬁltering (and
also frequencyshift (FRESH) ﬁltering). FRESH
ﬁltering consists of periodically or almostperiodi
cally timevariant ﬁltering of x(t) and x
+
(t) and
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 661
adding the results, where the frequency shifts are
chosen in accordance with the cycle frequencies of
x(t) and d(t) (e.g., cycle frequencies of the signal of
interest and interference, x(t) = d(t) ÷n(t) where
n(t) is the interference) [7.19,7.21,7.31]. The
problem of optimum LPTV and LAPTV ﬁltering
was ﬁrst addressed in [2.2,3.23] and, in terms of
cyclic autocorrelations and cyclic spectra in
[2.5,2.8].
Let
´
d(t)9y(t) ÷ y
c
(t)
=
_
R
h(t; u)x(u) du ÷
_
R
h
c
(t; u)x
+
(u) du (7:1)
be the LCL estimate of the desired signal d(t)
obtained from the data x(t). To minimize the
meansquared error
E
{a]
{[
´
d(t) ÷d(t)[
2
] (7.2)
a necessary and sufﬁcient condition is that the
error signal be orthogonal to the data (orthogon
ality condition [2.5]); that is,
E
{a]
{[
´
d(t ÷ t) ÷ d(t ÷ t)]x
+
(t)] = 0
\t c R \t c R, (7:3a)
E
{a]
{[
´
d(t ÷ t) ÷ d(t ÷ t)]x(t)] = 0
\t c R \t c R. (7:3b)
If x(t) and d(t) are singularly and jointly ACS,
E
{a]
{x(t ÷t)x
(+)
(t)] =
acA
xx
(+)
R
a
xx
(+)
(t)e
j2pat
, (7.4)
E
{a]
{d(t ÷ t)x
(+)
(t)] =
gcF
dx
(+)
R
g
dx
(+)
(t)e
j2pgt
, (7.5)
then it follows that the optimum ﬁlters are
LAPTV:
h(t; u) =
scG
h
s
(t ÷u)e
j2psu
, (7.6)
h
c
(t; u) =
ZcG
c
h
c
Z
(t ÷u)e
j2pZu
. (7.7)
By substituting (7.1), (7.6), and (7.7) into (7.3a)
and (7.3b), we obtain the system of simultaneous
ﬁlter design [7.31,9.51]:
scG
e
j2pst
[R
g÷s
xx
+ (t) Q h
s
(t)]
÷
ZcG
c
e
j2pZt
[R
Z÷g
xx
(t)
+
Q h
c
Z
(t)]
= R
g
dx
+ (t) \g c F
dx
+ , (7:8a)
scG
e
j2pst
[R
g÷s
xx
(t) Q h
s
(t)]
÷
ZcG
c
e
j2pZt
[R
Z÷g
xx
+ (t)
+
Q h
c
Z
(t)]
= R
g
dx
(t) \g c F
dx
, (7:8b)
where the fact that R
b
x
+
x
+ (t) = R
÷b
xx
(t)
+
and
R
a
x
+
x
(t) = R
÷a
xx
+ (t)
+
is used and, in (7.8a), for each
g c F
dx
+ , the sums are extended to all frequency
shifts s c G and Z c G
c
such that g ÷s c A
xx
+ and
Z ÷g c A
xx
; and, in (7.8b), for each g c F
dx
, the
sums are extended to all frequency shifts s c G
and Z c G
c
such that g ÷ s c A
xx
and Z ÷ g c A
xx
+ .
Eqs. (7.8a) and (7.8b) can be reexpressed in the
frequency domain:
scG
S
g÷s
xx
+ (f ÷s)H
s
(f ÷ s)
÷
ZcG
c
S
Z÷g
xx
(Z ÷ f )
+
H
c
Z
(f ÷Z)
= S
g
dx
+ (f ) \g c F
dx
+ , (7:9a)
scG
S
g÷s
xx
(f ÷s)H
s
(f ÷ s)
÷
ZcG
c
S
Z÷g
xx
+ (Z ÷ f )
+
H
c
Z
(f ÷ Z)
= S
g
dx
(f ) \g c F
dx
, (7:9b)
where H
s
(f ) and H
c
Z
(f ) are the Fourier transforms
of h
s
(t) and h
c
Z
(t), respectively.
Applications of FRESH ﬁltering to interference
suppression have been considered in [7.56,7.58,
7.68,7.73,7.76,7.77,7.79–7.81,7.85,7.94,7.98].
Adaptive FRESH ﬁltering is addressed in [7.11,
7.14,7.19,7.20,7.21,7.23,7.24,7.27,7.30,7.46,7.57,7.69,
7.74,7.75,7.81].
Cyclic Wiener ﬁltering or FRESH ﬁltering can
be recognized to be linked to FSE’s, RAKE ﬁlters,
adaptive demodulators, and LMMSE despreaders
for recovery of baseband symbol data from PAM,
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 662
GMSK/GFSK, and DSSS signals, and for separa
tion of baseband symbol streams in overlapped
signal environments, as well as multicarrier or
direct frequency diversity spread spectrum systems
for adaptive transmission and combining. In
particular, on blind and nonblind adaptive demo
dulation of PAM signals and LMMSE blind
despreading of shortcode DSSS/CDMA signals
using FSE’s and RAKE ﬁltering structures, see
[7.12,7.15–7.17,7.28,7.29,7.36,7.38,7.39,7.44,7.48,
7.50,7.51,7.53,7.59,7.61,7.65–7.67]. On direct fre
quencydiverse multicarrier transceivers, see
[7.32–7.34,7.52].
Patents of inventions for the analysis and design
of communications systems exploiting cyclostatio
narity are [7.35,7.37,7.42,7.45,7.70,7.83,7.86,7.87,
7.92,7.95,7.96,7.99–7.101]. Patents of inventions
exploiting the FRESH ﬁltering are [7.55,7.93].
8. Synchronization
8.1. Spectral line generation
Let x(t) be a realvalued secondorder wide
sense ACS time series. According to the results of
Section 3.3, the secondorder lag product x(t ÷
t)x(t) can be decomposed into the sum of its
almostperiodic component and a residual term
‘
x
(t; t) not containing any ﬁnitestrength additive
sinewave component (see (3.44) and (3.46)):
x(t ÷t)x(t) = E
{a]
{x(t ÷ t)x(t)] ÷ ‘
x
(t; t)
=
acA
R
a
x
(t)e
j2pat
÷ ‘
x
(t; t), (8:1)
where
¸‘
x
(t; t)e
÷j2pat
)
t
¬ 0 \a c R. (8.2)
For communications signals, the cycle frequen
cies a c A are related to parameters such as
sinewave carrier frequency, pulse rate, symbol
rate, frame rate, sampling frequency, etc. (see
Section 5). Therefore, the extraction of the almost
periodic component in (8.1) leads to a signal
suitable for synchronization purposes. For exam
ple, if x(t) is the binary PAM signal deﬁned in (5.6)
with q(t) real and duration limited to an interval
strictly less than T
0
, and a
k
c {÷1; 1], then it
follows that
x
2
(t) =
kcZ
q
2
(t ÷ kT
0
) (8.3)
and ‘
x
(t; 0) = 0. Therefore, the synchronization
signal x
2
(t) is periodic with period T
0
.
More generally, by deﬁnition, ACS signals
enable spectral lines to be generated by passage
through a stable nonlinear timeinvariant trans
formation (see Sections 10.4 and 13). That is,
quadratic or higherorder nonlinear timeinvariant
transformations of an ACS signal give rise to time
series containing ﬁnitestrength additive sinewave
components whose frequencies are the second or
higherorder cycle frequencies of the original
signal. All synchronization schemes can be recog
nized to exploit the second or higherorder
cyclostationarity features of signals [8.13]. Cyclo
stationarity properties are exploited for synchro
nization in [8.1–8.36].
The spectral analysis of timing waveforms with
regenerated spectral lines is treated in
[8.1,8.8,8.11,8.13–8.15]. Phaselock loops are ana
lyzed in [8.2,8.5–8.7]. Blind or nondataaided
synchronization algorithms are described in
[8.20–8.26,8.28–8.33,8.35,8.36]. See also [2.8,5.1].
Patents on synchronization techniques exploit
ing cyclostationarity are [8.17,8.27,8.34].
9. Signal parameter and waveform estimation
Cyclostationarity properties can be exploited to
design signal selective algorithms for signal para
meter and waveform estimation [9.1–9.100]. In
fact, if the desired and interfering signals have
different cyclic parameters such as carrier fre
quency or baud rate, then they exhibit cyclosta
tionarity at different cycle frequencies and,
consequently, parameters of the desired signal
can be extracted by estimating cyclic statistics of
the received data, consisting of desired signal plus
interfering signal, at a cycle frequency exhibited by
the desired signal but not by the interference.
This signal selectivity by exploitation of cyclosta
tionarity was ﬁrst suggested in [4.8]. Parameters
that can be estimated in the presence of inter
ference and/or high noise include carrier frequency
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 663
and phase, pulse rate and phase, signal power
level, modulation indices, bandwidths, time and
frequencydifference of arrival, direction of arri
val, and so on.
Signal selective timedifferenceofarrival esti
mation algorithms are considered in [9.11,9.12,
9.19,9.31,9.33,9.36–9.38,9.54,9.56,9.72,9.75,9.83].
In [9.84,9.90], both timedifferenceofarrival and
frequencydifferenceof arrival are estimated. Ar
ray processing problems, including spatial ﬁltering
and direction ﬁnding, are treated in [7.15,9.7,9.9,
9.10,9.15,9.17,9.18,9.20–9.22,9.26–9.30,9.32,9.34,
9.39–9.44,9.47–9.50,9.53,9.57–9.65,9.68,9.73,9.74,
9.79,9.80,9.82,9.86–9.88,9.91,9.94,9.95,9.97,9.99,9.100].
The polynomialphase signal parameter estima
tion is addressed in [9.67,9.71,9.78,9.93]. Harmo
nics in additive and multiplicative noise are
considered in [9.55,9.66,9.69,9.70,9.76,9.77]. But,
there are difﬁculties in application of some of this
work—difﬁculties associated with lack of ergodic
properties of stochasticprocess models adopted
for polynomialphase signals and signals with
coupled harmonics.
On this subject see the general treatments
[2.2,2.5,2.8,2.9,2.11], and also see [3.23,3.60,5.12,
7.1,7.2,7.7,7.19,7.21,7.24,8.24,11.1,21.6].
Patents of inventions on signal parameter and
waveform estimation exploiting cyclostationarity
are [9.16,9.45,9.46,9.52,9.81,9.96].
10. Channel identiﬁcation and equalization
10.1. General aspects
Cyclostationaritybased techniques have been
exploited for channel identiﬁcation and equaliza
tion [10.1–10.57]. Linear and nonlinear systems,
and timeinvariant, periodically and almostperi
odically timevariant systems, have been consid
ered. Also, techniques for noisy input/output
measurement, and blind adaptation algorithms,
have been developed for LTI systems.
On this subject see the general treatments
[2.2,2.5,2.8,2.9,2.11], and also see [7.25,10.36,8.24,
9.59,13.20,13.28,14.11].
Patents of inventions on blind system identiﬁca
tion exploiting cyclostationarity are [10.3,10.6,
10.10,10.11,10.39,10.56].
10.2. LTIsystem identiﬁcation with noisy
measurements
Cyclostationaritybased techniques can be
exploited in channel identiﬁcation and equaliza
tion problems in order to separate the desired and
disturbance contributions in noisy input/output
measurements, provided that there is at least one
cycle frequency of the desired signal that is not
shared by the disturbance.
Let us consider the problem of estimating the
impulseresponse function h(t) or, equivalently, the
harmonicresponse function:
H(f )9
_
R
h(t)e
÷j2pft
dt (10.1)
of an LTI system with input/output relation
y(t) = h(t) Q x(t) (10.2)
on the basis of the observed noisy signals v(t) and
z(t)
v(t) = x(t) ÷ n(t), (10.3)
z(t) = y(t) ÷ m(t), (10.4)
where x(t), n(t), and m(t) are zeromean time
series.
By assuming x
1
= x
2
= x, y
1
= y, y
2
= x, h
1
= h
(LTI), and h
2
= d in (3.77), and choosing (+) to be
conjugation, (3.81) specializes to
S
a
yx
+ (f ) = S
a
xx
+ (f )H(f ). (10.5)
Therefore, from the model (10.3) and (10.4), we
obtain
S
a
vv
+ (f ) = S
a
xx
+ (f ) ÷ S
a
nn
+ (f ), (10.6)
S
a
zv
+ (f ) = S
a
yx
+ (f ) ÷ S
a
mn
+ (f )
= S
a
xx
+ (f )H(f ) ÷S
a
mn
+ (f ) (10:7)
provided that x(t) is uncorrelated with both n(t)
and m(t).
Eqs. (10.6) and (10.7) reveal the ability of
cyclostationaritybased algorithms to be signal
selective. In fact, under the assumption that n(t)
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 664
does not exhibit cyclostationarity at cycle fre
quency a (i.e., S
a
nn
+ (f ) ¬ 0) and n(t) and m(t) do
not exhibit joint cyclostationarity with cycle
frequency a (i.e., S
a
mn
+ (f ) ¬ 0), the harmonic
response function for the system is given by
H(f ) =
S
a
yx
+ (f )
S
a
xx
+ (f )
=
S
a
zv
+ (f )
S
a
vv
+ (f )
. (10.8)
That is, H(f ) can be expressed in terms of the
cyclic spectra of the noisy input and output
signals. Therefore, (10.8) provides a system identi
ﬁcation formula that is intrinsically immune to the
effects of noise and interference as ﬁrst observed in
[2.8,10.5]. Thus, this identiﬁcation method is
highly tolerant to disturbances in practice, pro
vided that a sufﬁciently long integration time is
used for the cyclic spectral estimates.
The identiﬁcation of LTI systems based on noisy
input/output measurements is considered in [2.5,
2.8,9.11,9.12,9.36,9.37,10.5,10.12,10.14,10.18,10.20,
10.24,10.55].
10.3. Blind LTIsystem identiﬁcation and
equalization
By specializing (3.84) to the case of the LTI
system (10.2), we get
S
a
yy
+ (f ) = S
a
xx
+ (f )H(f )H
+
(f ÷ a) (10.9)
which, for a = 0, reduces to the input/output
relationship in terms of power spectra:
S
0
yy
+ (f ) = S
0
xx
+ (f )[H(f )[
2
. (10.10)
From (10.9) and (10.10) it follows that input/output
relationships for LTI systems in terms of cyclic
statistics, unlike those in terms of autocorrelation
functions and power spectra, preserve phase in
formation of the harmonicresponse function H(f ).
Thus, cyclostationarity properties of the output
signal are suitable to be exploited for recovering
both phase and magnitude of the system harmonic
response function as ﬁrst observed in [10.7].
Cyclostationarity properties are exploited for
blind identiﬁcation (without measurements of the
system input) of linear systems and for blind
equalization techniques in [9.59,10.1,10.7,10.15,
10.16,10.19,10.21,10.22,10.25,10.26,10.28–10.33,
10.36–10.38,10.42–10.45,10.47,10.48,10.50–10.53].
10.4. Nonlinearsystem identiﬁcation
Let y(t) the output signal of a Volterra system
excited by the input signal x(t):
y(t) =
÷o
n=1
_
R
n
k
n
(t
1
; . . . ; t
n
)x(t ÷ t
1
) · · ·
x(t ÷ t
n
) dt
1
· · · dt
n
. (10:11)
Accounting for the results of Sections 3.3 and 13,
the input lagproduct waveform can be decom
posed into the sum of an almostperiodic compo
nent, referred to as the temporal moment function,
and a residual term not containing any ﬁnite
strength additive sinewave component (see (13.3)):
x(t ÷ t
1
) · · · x(t ÷ t
n
) =
acA
x
R
a
x
(s)e
j2pat
÷‘
x
(t; s).
(10.12)
Thus, identiﬁcation and equalization techniques
for Volterra systems excited by ACS signals make
use of higherorder cyclostationarity properties. In
fact, there are potentially substantial advantages
to using cyclostationary input signals, relative to
stationary input signals, for purposes of Volterra
system modelling and identiﬁcation as ﬁrst ob
served in [10.13].
Both timeinvariant and almostperiodically
timevariant nonlinear systems are treated in
[10.4,10.8,10.13,10.23,10.35,10.41,10.46,13.16].
11. Signal detection and classiﬁcation, and source
separation
Signal detection techniques designed for cyclos
tationary signals take account of the periodicity or
almost periodicity of the signal autocorrelation
function [11.1–11.39]. Singlecycle and multicycle
detectors exploit one or multiple cycle frequencies,
respectively. The detection problem for additive
Gaussian noise is addressed in [11.2,11.5,11.7,
11.9–11.11,11.15,11.19,11.21,11.26,11.27,11.32],
and for nonGaussian noise, in [11.16,11.18,
11.24,11.25]. The problem of signal detection in
cyclostationary noise is treated in [11.2,11.8,
11.12,11.17]. Tests for the presence of cyclostatio
narity are proposed in [11.14,11.20,11.31,12.31] by
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 665
exploiting the asymptotic properties of the cyclic
correlogram and in [11.6,11.13] by exploiting the
properties of the support of the spectral correla
tion function. Modulation classiﬁcation techni
ques are proposed in [11.23,11.28–11.30]; see also
[13.22]. The problem of cyclostationary source
separation is considered in [11.34,11.35,11.37,
11.39].
On this subject see the general treatments
[2.5,2.11,2.13], and also see [3.60,7.72,7.74,9.13,
9.14,9.19,11.11,13.30,13.31].
Patents on detection and signal recognition
exploiting cyclostationarity are [11.22,11.36,11.38].
See also the URL http://www.sspitech.
com for information on general purpose, auto
matic, communicationsignal classiﬁcation software
systems.
12. Periodic AR and ARMA modelling and
prediction
Periodic autoregressive (AR) and autoregressive
moving average (ARMA) (discretetime) systems
are characterized by input/output relationships
described by difference equations with periodically
timevarying coefﬁcients and system orders
[12.36,12.37]:
P(n)
k=0
a
k
(n)y(n ÷k) =
Q(n)
m=0
b
m
(n)x(n ÷m), (12.1)
where x(n) and y(n) are the input and output
signals, respectively, and the coefﬁcients a
k
(n) and
b
m
(n) and the orders P(n) and Q(n) are periodic
functions with the same period N
0
. Thus, periodic
ARMA systems are a special case of discretetime
periodically timevarying systems. When they are
excited by a stationary or cyclostationary input
signal x(n), they give rise to a cyclostationary
output signal y(n). When they are excited by an
almostcyclostationary input signal, they give rise
to an almostcyclostationary output signal. The
problem of ﬁtting an AR model to data y(n) is
equivalent to the problem of solving for a linear
predictor for y(n), where x(n) is the modelﬁtting
error time series.
Periodic AR and ARMA systems are treated in
[12.1–12.53]. The modelling and prediction pro
blem is treated in [12.1–12.8,12.12–12.16,12.19,
12.21,12.23–12.26,12.28–12.30,12.32,12.34–12.37,
12.41,12.43,12.45,12.46,12.50,12.52]. The para
meter estimation problem is addressed in [12.9,
12.11,12.17,12.22,12.27,12.31,12.36,12.38–12.40,
12.47–12.49,12.51].
On the subject of periodic AR and ARMA
modelling and prediction, see the general treat
ments [2.5,2.8,2.11,2.13], and also see [3.30,3.69,
14.23,15.3] for the prediction problem; [4.3,4.5,
4.12,4.27,4.29] for the parameter estimation pro
blem; [6.2,6.9,6.13,6.16] for modelling of atmo
spheric and hydrologic signals; [16.1,16.2,16.5,
16.7] for applications to econometrics; and [21.4]
for application to modelling helicopter noise.
13. Higherorder statistics
13.1. Introduction
As ﬁrst deﬁned in [13.4], a signal x(t) is said to
exhibit higherorder cyclostationarity (HOCS) if
there exists a homogeneous nonlinear transfor
mation of x(t) of order greater than two such that
the output of this transformation contains ﬁnite
strength additive sinewave components. Motiva
tions to study HOCS properties of signals include
the following:
(1) Signals not exhibiting secondorder cyclosta
tionarity can exhibit HOCS [13.13].
(2) Narrowband ﬁltering can destroy second
order (widesense) cyclostationarity. In fact,
let us consider the input/output relationship
for LTI systems in terms of cyclic spectra
(10.9). If the bandwidth of H(f ) is smaller than
the smallest nonzero secondorder cycle fre
quency of the input signal x(t), then the output
signal y(t) does not exhibit secondorder wide
sense cyclostationarity. The signal y(t), how
ever, can exhibit HOCS (see also [3.13]).
(3) The exploitation of HOCS can be useful for
signal classiﬁcation [13.22]. Speciﬁcally, differ
ent communication signals, even if they
exhibit the same secondorder cyclostationarity
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 666
properties, can exhibit different cyclic features
of higher order. Moreover, different behaviors
can be obtained for different conjugation
conﬁgurations [13.9,13.14,13.17].
(4) Exploitation of HOCS can be useful for many
estimation problems, as outlined below.
On the subject of higherorder cyclostationarity
and its applications see [13.1–13.31]; in addition,
see [3.81] for the wavelet decomposition; [4.56] for
estimation issues; [5.23] for the cyclic higherorder
properties of CPM signals; [7.74] for blind
adaptive detection; [8.8] for the spectral analysis
of PAM signals; [9.55,9.61,9.67,9.69–9.71,9.80,
9.85,9.93,9.99] for application to signal parameter
and waveform estimation; [10.13,10.23,10.24,
10.34,10.40,10.41,10.46] for applications to non
linear system identiﬁcation and equalization;
[11.20] for applications to signal detection;
[11.28–11.30] for applications to signal classiﬁca
tion; [11.34] for applications to source sepa
ration; [15.18,15.21,15.23] for applications in
acoustics and mechanics; [21.9–21.11,21.13] for
the higherorder characterization of generalized
almostcyclostationary signals.
13.2. Higherorder cyclic statistics
In this section, cyclic higherorder (joint) statis
tics for both continuoustime and discretetime
time series are presented in the FOT probability
framework as ﬁrst introduced for continuous time
in [13.4,13.5] and developed in [13.9,13.13,13.14,
13.17], and, for discrete and continuous time, in
[13.20]. For treatments within the stochastic
framework, see [13.10,13.12,13.15,13.18,13.26] or
extend to higherorder statistics the link between
the two frameworks discussed in Section 3.4.
13.2.1. Continuoustime time series
Let us consider the column vector
x(t)9[x
(+)
1
1
(t); . . . ; x
(+)
N
N
(t)]
T
whose components
are N not necessarily distinct complexvalued
continuoustime timeseries and (+)
k
represents
optional complex conjugation of the kth signal
x
k
(t). The N timeseries exhibit joint Nthorder
widesense cyclostationarity with cycle frequency
aa0 if at least one of the Nthorder cyclic temporal
crossmoment functions (CTCMFs)
R
a
x
(s)9
N
k=1
x
(+)
k
k
(t ÷ t
k
)e
÷j2pat
_ _
t
, (13.1)
where s9[t
1
; . . . ; t
N
]
T
is not identically zero. Thus,
N time series exhibit widesense joint Nthorder
cyclostationarity with cycle frequency aa0 if, for
some s, the lag product waveform
L
x
(t; s)9
N
k=1
x
(+)
k
k
(t ÷t
k
) (13.2)
contains a ﬁnitestrength additive sinewave com
ponent with frequency a, whose amplitude and
phase are the magnitude and phase of R
a
x
(s),
respectively.
The Nthoder temporal crossmoment function
(TCMF) is deﬁned by
R
x
(t; s)9E
{a]
{L
x
(t; s)]
=
acA
x
R
a
x
(s)e
j2pat
(13:3)
where A
x
is the countable set (not depending on s)
of the Nthorder cycle frequencies of the time
series x
(+)
1
1
(t ÷t
1
); . . . ; x
(+)
N
N
(t ÷ t
N
) (for the given
conjugation conﬁguration).
The Ndimensional Fourier transform of the
CTCMF
S
a
x
(f )9
_
R
N
R
a
x
(s)e
÷j2pf
T
s
ds, (13.4)
where f 9[f
1
; . . . ; f
N
]
T
, is called the Nthorder
cyclic spectral crossmoment function (CSCMF)
and can be written as
S
a
x
(f ) = S
a
x
(f
/
)d(f
T
1 ÷ a), (13.5)
where 1 is the vector [1; . . . ; 1]
T
, and prime denotes
the operator that transforms a vector
u9[u
1
; . . . ; u
K
]
T
into the reduceddimension ver
sion u
/
9[u
1
; . . . ; u
K÷1
]
T
. The function S
a
x
(f
/
), re
ferred to as the reduceddimension CSCMF (RD
CSCMF), can be expressed as
S
a
x
(f
/
) =
_
R
N÷1
R
a
x
(s
/
)e
÷j2pf
/
T
s
/
ds
/
, (13.6)
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 667
where
R
a
x
(s
/
)9R
a
x
(s)[
t
N
=0
(13.7)
is the reduceddimension CTCMF (RDCTCMF).
The RDCSCMF can also be expressed as
S
a
x
(f
/
) = lim
T÷o
lim
Z÷o
1
Z
_
t÷Z=2
t÷Z=2
1
T
X
(+)
N
N;T
(u; (÷)
N
(a ÷ f
/
T
1))
N÷1
k=1
X
(+)
k
k;T
(u; (÷)
k
f
k
) du, (13:8)
where
X
k;T
(t; f
k
)9
_
t÷T=2
t÷T=2
x
k
(s)e
÷j2pf
k
s
ds (13.9)
and (÷)
k
denotes an optional minus sign that is
linked to the optional conjugation (+)
k
. Eq. (13.8)
reveals that N timeseries exhibit joint Nthorder
widesense cyclostationarity with cycle frequency a
(i.e., R
a
x
(s)c0 or, equivalently, S
a
x
(f
/
)c0) if and
only if the Nthorder temporal crossmoment of
their spectral components at frequencies f
k
, whose
sum is equal to a, is nonzero. In fact, by using
h
B
(t)9Brect(Bt) (13.10)
with B91=T and rect(t) = 1 if [t[p
1
2
and rect(t) =
0 if [t[4
1
2
, Eq. (13.8) can be rewritten as
S
a
x
(f
/
) = lim
B÷0
lim
Z÷o
1
Z
_
t÷Z=2
t÷Z=2
1
B
N÷1
[(x
(+)
N
N
(u)
e
÷j2p(a÷f
/
T
1)u
) Q h
B
(u)]
N÷1
k=1
[(x
(+)
k
k
(u)e
÷j2pf
k
u
) Q h
B
(u)] du (13:11)
which is the Nthorder temporal crossmoment of
lowpassﬁltered versions of the frequencyshifted
signals x
(+)
k
k
(t)e
÷j2pf
k
t
when the sum of the fre
quency shifts is equal to a and the bandwidth B
approaches zero. Such a property is the general
ization to the order N42 of the spectral correla
tion property of signals that exhibit secondorder
cyclostationarity. Moreover, relation (13.4) be
tween the Nthorder cyclic spectral moment
function (13.5), (13.8) and Nthorder cyclic tem
poral moment (13.1) is, at ﬁrst pointed out in
[13.4], the Nthorder generalization of the Cyclic
Wiener Relation.
Let us note that in general the function R
a
x
(s
/
) is
not absolutely integrable because it does not in
general decay as s
/
 ÷ o, but rather it oscillates.
Thus, S
a
x
(f
/
) can contain impulses and, conse
quently, S
a
x
(f ) can contain products of impulses.
In [13.13], it is shown that the RDCSCMF S
b
x
(f
/
)
can contain impulsive terms if the vector f with
f
N
= b ÷
N÷1
k=1
f
k
lies on the bsubmanifold; i.e.,
if there exists at least one partition {m
1
; . . . ; m
p
] of
{1; . . . ; N] with p41 such that each sum a
m
i
=
kcm
i
f
k
is a [m
i
[thorder cycle frequency of x(t),
where [m
i
[ is the number of elements in m
i
.
In the spectralfrequency domain, a wellbe
haved function can be introduced starting from the
Nthorder temporal crosscumulant function
(TCCF):
C
x
(t; s)9cum{x
(+)
k
k
(t ÷t
k
); k = 1; . . . ; N]
= (÷j)
N
@
N
@o
1
· · · @o
N
log
e
E
{a]
exp j
N
k=1
o
k
x
(+)
k
k
(t ÷ t
k
)
_ _ _ _¸
¸
¸
¸
¸
x=0
=
P
(÷1)
p÷1
(p ÷ 1)!
p
i=1
R
x
m
i
(t; s
m
i
)
_ _
,
(13:12)
where x9[o
1
; . . . ; o
N
]
T
, P is the set of distinct
partitions of {1; . . . ; N], each constituted by the
subsets {m
i
; i = 1; . . . ; p], x
m
i
is the [m
i
[dimen
sional vector whose components are those of x
having indices in m
i
. In (13.12), the almostperiodic
component extraction operator E
{a]
{·] extracts the
frequencies of the 2Nvariate fractionoftime joint
probability density function of the real and
imaginary parts of the timeseries x
k
(t) (k =
1; . . . ; N) according to
E
{a]
{g(x
(+)
1
1
(t ÷ t
1
); . . . ; x
(+)
N
N
(t ÷ t
N
))]
=
_
R
2N
g(x
(+)
1
1
; . . . ; x
(+)
N
N
)
f
{a]
x
1r
(t÷t
1
)x
1i
(t÷t
1
)···x
Nr
(t÷t
N
)x
Ni
(t÷t
N
)
(x
1r
; x
1i
; . . . ; x
Nr
; x
Ni
)
dx
1r
dx
1i
· · · dx
Nr
dx
Ni
, (13:13)
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 668
where x
kr
(t)9Re[x
k
(t)], x
ki
(t)9Im[x
k
(t)], x
kr
9
Re[x
k
], x
ki
9Im[x
k
], and
g(x
(+)
1
1
; . . . ; x
(+)
N
N
)9exp j
N
k=1
o
k
x
(+)
k
k
_ _
. (13.14)
In fact, by taking the Ndimensional Fourier
transform of the coefﬁcient of the Fourier series
expansion of the almostperiodic function (13.12)
C
b
x
(s)9¸C
x
(t; s)e
÷j2pbt
)
t
(13.15)
which is referred to as the Nthorder cyclic
temporal crosscumulant function (CTCCF), one
obtains the Nthorder cyclic spectral crosscumu
lant function (CSCCF) P
b
x
(f ). It can be written as
P
b
x
(f ) = P
b
x
(f
/
)d(f
T
1 ÷ b), (13.16)
where
P
b
x
(f
/
)9
_
R
N÷1
C
b
x
(s
/
)e
÷j2pf
/ T
s
/
ds
/
(13.17)
is the Nthorder cyclic crosspolyspectrum (CCP),
and
C
b
x
(s
/
)9C
b
x
(s)[
t
N
=0
(13.18)
is called the reduceddimension CTCCF (RD
CTCCF). The cyclic crosspolyspectrum is a well
behaved function under the mild conditions that
the timeseries x
N
(t) and x
k
(t ÷t
k
) (k =
1; . . . ; N ÷1) are asymptotically ([t
k
[ ÷ o) inde
pendent (in the FOT probability sense) so that
C
b
x
(s
/
) ÷ 0 as s
/
 ÷ o and, moreover, there
exists an 40 such that [C
b
x
(s
/
)[ = o(s
/

÷N÷1÷
) as
s
/
 ÷ o. Furthermore, except on a bsubmani
fold, the CCP P
b
x
(f
/
) is coincident with the RD
CSCMF S
b
x
(f
/
).
The CCP can also be expressed as
P
b
x
(f
/
) = lim
T÷o
1
T
cum{X
(+)
N
N;T
(t; (÷)
N
(a ÷f
/
T
1)),
X
(+)
k
k;T
(t; (÷)
k
f
k
),
k = 1; . . . ; N ÷ 1], (13:19)
where, in the computation of the cumulant in
(13.19), the stationary FOT expectation operation
can be adopted as T ÷ o. Eq. (13.19) reveals that
the CCP of N time series is the Nthorder cross
cumulant of their spectral components at frequen
cies f
k
whose sum is equal to b. In fact, Eq. (13.19)
can be rewritten as
P
b
x
(f
/
) = lim
B÷0
1
B
N÷1
cum{[(x
(+)
N
N
(t)e
÷j2p(a÷f
/
T
1)t
)
Q h
B
(t)]; [(x
(+)
k
k
(t)e
÷j2pf
k
t
) Q h
B
(t)],
k = 1; . . . ; N ÷ 1] (13:20)
which is the Nthorder temporal crosscumulant of
lowpass ﬁltered versions of the frequencyshifted
signals x
(+)
k
k
(t)e
÷j2pf
k
t
when the sum of the fre
quency shifts is equal to b and the bandwidth B
approaches zero.
As ﬁrst shown in [13.5] and then, in more detail,
in [13.9,13.13,13.15,13.17], the Nthorder temporal
cumulant function of a time series provides a
mathematical characterization of the notion of a
pure Nthorder sinewave. It is that part of the
sinewave present in the Nthorder lag product
waveform that remains after removal of all parts
that result from products of sinewaves in lower
order lag products obtained by factoring the Nth
order product.
13.2.2. Discretetime time series
In accordance with the deﬁnition for the
continuoustime case, N not necessarily distinct
discretetime complexvalued time series x
k
(n),
n c Z, exhibit joint Nthorder widesense cyclos
tationarity with cycle frequency ¯aa0 (¯a c [÷
1
2
;
1
2
[) if
at least one of the Nthorder CTCMF’s
¯
R
¯a
x
(m)9 lim
N÷o
1
2N ÷ 1
N
n=÷N
N
k=1
x
(+)
k
k
(n ÷ m
k
)e
÷j2p¯an
(13.21)
is not identically zero. In (13.21), x(n)9
[x
(+)
1
1
(n); . . . ; x
(+)
N
N
(n)]
T
and m9[m
1
; . . . ; m
N
]
T
.
The magnitude and phase of the CTCMF
(13.21) are the amplitude and phase of the
sinewave component with frequency ¯a contained
in the discretetime lag product whose temporal
expected value is the discretetime Nthorder
TCMF.
The Nfold discrete Fourier transform of the
CTCMF
¯
S
¯a
x
(m)9
mcZ
N
¯
R
¯a
x
(m)e
÷j2pm
T
m
, (13.22)
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 669
where m9[n
1
; . . . ; n
N
]
T
, is called the Nthorder
CSCMF and can be written as
¯
S
¯a
x
(m) =
¯
S
¯a
x
(m
/
)
÷o
r=÷o
d(¯a ÷ m
T
1 ÷ r), (13.23)
where the function
¯
S
¯a
x
(m
/
) is the Nthorder RD
CSCMF, which can be expressed as the (N ÷1)
fold discrete Fourier transform
¯
S
¯a
x
(m
/
) =
m
/
cZ
N÷1
¯
R
¯a
x
(m
/
)e
÷j2pm
/
T
m
/
(13.24)
of the Nthorder RDCTCMF
¯
R
¯a
x
(m
/
)9
¯
R
¯a
x
(m)[
m
N
=0
. (13.25)
Once the Nthorder discretetime TCCF
¯
C
x
(k; m)
is deﬁned analogously to the continuoustime case,
the Nthorder CTCCF is given by
¯
C
¯
b
x
(m)9 lim
N÷o
1
2N ÷ 1
N
n=÷N
¯
C
x
(n; m)e
÷j2p
¯
bn
.
(13.26)
Its discrete Fourier transform, referred to as the
Nthorder CSCCF, is given by
¯
P
¯
b
x
(m) =
¯
P
¯
b
x
(m
/
)
÷o
r=÷o
d(
¯
b ÷ m
T
1 ÷r), (13.27)
where the function
¯
P
¯
b
x
(m
/
) is referred to as the Nth
order CCP, which can be expressed as the (N ÷1)
fold discrete Fourier transform of the Nthorder
RDCTCCF
¯
C
¯
b
x
(m
/
)9
¯
C
¯
b
x
(m)[
m
N
=0
. (13.28)
Finally, it can be easily shown that the following
periodicity properties hold:
¯
R
¯a
x
(m) =
¯
R
¯a÷p
x
(m); p c Z, (13.29)
¯
S
¯a
x
(m) =
¯
S
¯a÷p
x
(m ÷ q); p c Z; q c Z
N
, (13.30)
¯
C
¯
b
x
(m) =
¯
C
¯
b÷p
x
(m); p c Z, (13.31)
¯
P
¯
b
x
(m) =
¯
P
¯
b÷p
x
(m ÷ q); p c Z; q c Z
N
. (13.32)
Analogous relations can be stated for the reduced
dimension statistics.
14. Applications to circuits, systems, and control
Applications of cyclostationarity to circuits,
systems, and control are in [14.1–14.31]. In circuit
theory, cyclostationarity has been exploited in
modelling noise [14.1,14.2,14.12,14.17,14.19,14.20,
14.24,14.26,14.28]. In system theory, cyclostation
ary or almost cyclostationary signals arise in
dealing with periodically or almostperiodically
time variant systems (see Section 3.6) [14.4–14.8,
14.13,14.14,14.23,14.25]. In control theory, cyclos
tationarity has been exploited in [14.3,14.10,14.16,
14.18,14.21,14.22].
On this subject, also see [9.4,9.59,10.5,10.24,
13.15,13.28,15.1,16.3].
Patents on applications of cyclostationarity to
system and circuit analysis and design are
[14.27,14.29,14.31].
15. Applications to acoustics and mechanics
Applications of cyclostationarity to acoustics
and mechanics are in [15.1–15.24]. Cyclostationar
ity has been exploited in acoustics and mechanics
for modelling road trafﬁc noise [15.2,15.3], for
analyzing music signals [15.6], and for describing
the vibration signals in mechanical systems. In
mechanical systems with moving parts, such as
engines, if some parameter such as speed, tem
perature, and load torque can be assumed to be
constant, then the dynamic physical processes
generate vibrations that can be modelled as
originating from periodic mechanisms such as
rotation and reciprocation of gears, belts, chains,
shafts, propellers, pistons, and so on [2.5]. Conse
quently, vibration signals exhibit periodic behavior
of one type or another with periods related to the
engine cycle. Often, the observed vibration signals
contain both an almostperiodic and an almost
cyclostationary component [15.1,15.5,15.8,15.10,
15.14,15.18,15.23]. Even if the almostcyclosta
tionary component has a power smaller than the
power of the almostperiodic component, it can be
useful; for example it can be successfully exploited
in early diagnosis of gear faults [15.4,15.8,15.9,
15.11,15.13,15.14,15.16,15.19–15.21,15.24].
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 670
Additional applications in the ﬁeld of acoustics
and mechanics can be found in [4.33,4.43,4.49,
14.21,21.2].
Patents on applications of cyclostationarity to
engine diagnosis are [15.12,15.17].
16. Applications to econometrics
In highfrequency ﬁnancial time series, such as
asset return, the repetitive patterns of openings
and closures of markets, the number of active
markets throughout the day, seasonally varying
preferences, and so forth, are sources of periodic
variations in ﬁnancialmarket volatility and other
statistical parameters. Autoregressive models with
periodically varying parameters provide appropri
ate descriptions of seasonally varying economic
time series [16.1–16.13]. Furthermore, neglecting
the periodic behavior gives rise to a loss in forecast
efﬁciency [2.15,16.8,16.10].
On this subject see also [12.40].
17. Applications to biology
Applications of cyclostationarity to biology
are in [17.1–17.18]. Applications of the concept
of spectral redundancy (spectral correlation
or cyclostationarity) have been proposed in
medical image signal processing [17.6,17.8], and
nondestructive evaluation [17.2]. Methods of
averaging were developed for estimating the
generalized spectrum that allow for the meaningful
characterization of phase information when
classic assumptions of stationarity do not hold.
This has led to many signiﬁcant performance
improvements in ultrasonic tissue characteriza
tion, particularly for liver and breast tissues
[17.5,17.10–17.18].
A patent on the application of cyclostationarity
to cholesterol detection is [17.4].
18. Level crossings
Level crossings of cyclostationary signals have
been characterized in [18.1–18.10].
19. Queueing
Exploitation of cyclostationarity in queueing
theory in computer networks is treated in
[19.1–19.4,19.6]. Queueing theory in car trafﬁc is
treated in [19.5].
On this subject, also see [14.8].
20. Cyclostationary random ﬁelds
A periodically correlated or cyclostationary
random ﬁeld is a secondorder random ﬁeld whose
mean and correlation have periodic structure
[20.6,20.7]. Speciﬁcally, a random ﬁeld x(n
1
; n
2
)
indexed on Z
2
is called strongly periodically
correlated with period (N
01
; N
02
) if and only if
there exists no smaller N
01
40 and N
02
40 for
which the mean and correlation satisfy
E{x(n
1
÷ N
01
; n
2
÷ N
02
)] = E{x(n
1
; n
2
)], (20.1)
E{x(n
1
÷ N
01
; n
2
÷ N
02
)x
+
(n
/
1
÷N
01
; n
/
2
÷N
02
)]
= E{x(n
1
; n
2
)x
+
(n
/
1
; n
/
2
)] (20:2)
for all n
1
, n
2
, n
/
1
, n
/
2
c Z. Cyclostationary random
ﬁelds are treated in [20.1–20.7].
21. Generalizations of cyclostationarity
21.1. General aspects
Generalizations of cyclostationary processes
and time series are treated in [21.1–21.17]. The
problem of statistical function estimation for
general nonstationary persistent signals is ad
dressed in [21.1,21.5] and limitations of previously
proposed approaches are exposed. In
[21.2–21.4,21.8], the class of the correlation auto
regressive processes is studied. Nonstationary
signals that are not ACS can arise from linear
timevariant, but not almostperiodically time
variant, transformations of ACS signals. Such
transformations occur, for example, in mobile
communications when the product of transmitted
signal bandwidth and observation interval is not
much smaller than the ratio between the pro
pagation speed in the medium and the relative
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 671
radial speed between transmitter and receiver
[8.35,21.11]. Two models for the output sig
nals of such transformations are the generalized
almostcyclostationary signals [21.7,21.9–21.11,
21.13,21.15–21.17], and the spectrally correlated
signals [21.6,21.12,21.14]. Moreover, communica
tions signals with parameters, such as the carrier
frequency and the baud rate, that vary slowly
with time cannot be modelled as ACS but,
rather, can be modelled as generalized almost
cyclostationary if the observation interval is large
enough [21.9].
21.2. Generalized almostcyclostationary signals
A continuoustime complexvalued time series
x(t) is said to be widesense generalized almost
cyclostationary (GACS) if the almostperiodic
component of its secondorder lag product admits
a (generalized) Fourier series expansion with both
coefﬁcients and frequencies depending on the lag
parameter t [21.9,21.10]:
E
{a]
{x(t ÷t)x
+
(t)] =
ncI
R
(n)
xx
+ (t)e
j2pa
n
(t)t
. (21.1)
In (21.1), I is a countable set, the frequencies a
n
(t)
are referred to as lagdependent cycle frequencies,
and the coefﬁcients, referred to as generalized
cyclic autocorrelation functions, are given by
R
(n)
xx
+ (t)9¸x(t ÷ t)x
+
(t)e
÷j2pa
n
(t)t
)
t
. (21.2)
For GACS signals, the cyclic autocorrelation
function can be expressed in terms of the general
ized cyclic autocorrelation functions by the follow
ing relationship:
R
a
xx
+ (t) =
ncI
R
(n)
xx
+ (t)d
a÷a
n
(t)
. (21.3)
Moreover, it results that
A
t
9{a c R : R
a
xx
+ (t)a0]
=
_
ncI
{a c R : a = a
n
(t)]. (21:4)
That is, for GACS signals, the support in the (a; t)
plane of the cyclic autocorrelation function R
a
xx
+ (t)
consists of a countable set of curves described by
the equations a = a
n
(t), n c I. For the GACS
signals, the set
A9
_
tcR
A
t
(21.5)
is not necessarily countable.
The ACS signals are obtained as the special case
of GACS signals for which the functions a
n
(t) are
constant with respect to t and are equal to the
cycle frequencies. In such a case the support of the
cyclic autocorrelation function in the (a; t) plane
consists of lines parallel to the t axis and the
generalized cyclic autocorrelation functions are
coincident with the cyclic autocorrelation func
tions. Moreover, the set A turns out to be
countable in this case (see (3.13)).
The higherorder characterization of the GACS
signals in the FOT probability framework is
provided in [21.9]. Linear ﬁltering is addressed in
[21.10,21.11] where the concept of expectation of
the impulseresponse function in the FOT prob
ability framework is also introduced. The problem
of sampling a GACS signal is considered in [21.13]
where it is shown that the discretetime signal
obtained by uniformly sampling a continuoustime
GACS signal is an ACS signal. The estimation of
the cyclic autocorrelation function for GACS
processes is addressed in [21.15,21.17] in the
stochastic process framework. In [21.16], a survey
of GACS signals is provided.
21.3. Spectrally correlated signals
A continuoustime complexvalued secondor
der harmonizable stochastic process x(t) is said to
be spectrally correlated (SC) if its Loe`ve bifre
quency spectrum can be expressed as [21.14]
S
xx
+ (f
1
; f
2
)9E{X(f
1
)X
+
(f
2
)]
=
ncI
S
(n)
xx
+ (f
1
)d(f
2
÷ C
n
(f
1
)), (21:6)
where I is a countable set, the curves
f
2
= C
n
(f
1
); n c I, describe the support of
S
xx
+ (f
1
; f
2
), and the functions S
(n)
xx
+ (f
1
), referred
to as the spectral correlation density functions,
describe the density of the Loe` ve bifrequency
spectrum on its support curves. The case of linear
support curves is considered in [21.6,21.12]. The
ACS processes are obtained as the special case of
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 672
SC processes for which the support curves are lines
with unity slope.
The bifrequency spectral correlation density
function
¯
S
xx
+ (f
1
; f
2
)9
ncI
S
(n)
xx
+ (f
1
)d
f
2
÷C
n
(f
1
)
(21.7)
is the density of the Loe` ve bifrequency spectrum
on its support curves f
2
= C
n
(f
1
). In [21.14] it is
shown that, if the location of the support curves is
unknown, the bifrequency spectral correlation
density function
¯
S
xx
+ (f
1
; f
2
) can be reliably
estimated by the timesmoothed crossperiodo
gram only if the slope of the support curves is not
too far from unity.
References
[2.1] H.L. Hurd, An investigation of periodically correlated
stochastic processes, Ph.D. Dissertation, Duke Uni
versity, Durham, NC, 1969.
[2.2] W.A. Gardner, Representation and estimation of
cyclostationary processes, Ph.D. Dissertation, Depart
ment of Electrical and Computer Engineering, Uni
versity of Massachusetts, reprinted as Signal and Image
Processing Lab Technical Report No. SIPL821,
Department of Electrical and Computer Engineering,
University of California, Davis, CA, 95616, 1982, 1972.
[2.3] K.S. Voychishin, Y.P. Dragan, Example of formation
of periodically correlated random processes, Radio
Eng. Electron. Phys. 18 (1973) 1426–1429 (English
translation of Radiotekh. Elektron. 18, 1957,1960).
[2.4] Y.P. Dragan, Structure and Representations of Models
of Stochastic Signals, Part 4, Naukova Dumka, Kiev,
1980.
[2.5] W.A. Gardner, Introduction to Random Processes with
Applications to Signals and Systems, Macmillan, New
York, 1985 (Chapter 12).
[2.6] W.A. Gardner, The spectral correlation theory of
cyclostationary time series, Signal Processing 11
(1986) 13–36 (Erratum: Signal Processing 11 405.)
Received EURASIP’s Best Paper of the Year award.
[2.7] A.M. Yaglom, Correlation Theory of Stationary and
Related Random Functions, vols. I and II, Springer,
New York, 1986.
[2.8] W.A. Gardner, Statistical Spectral Analysis: A Non
probabilistic Theory, PrenticeHall, Englewood Cliffs,
NJ, 1987.
[2.9] W.A. Brown, On the theory of cyclostationary signals,
Ph.D. Dissertation, Department of Electrical Engineer
ing and Computer Science, University of California,
Davis, CA, 1987.
[2.10] W.A. Gardner, Exploitation of spectral correlation in
cyclostationary signals, in: Fourth Annual ASSP
Workshop on Spectrum Estimation and Modeling,
Minneapolis, MN, 1988, pp. 1–6.
[2.11] W.A. Gardner, Introduction to Random Processes with
Applications to Signals and Systems, second ed.,
McGrawHill, New York, 1990 (Chapter 12).
[2.12] W.A. Gardner, W.A. Brown, Fractionoftime prob
ability for timeseries that exhibit cyclostationarity,
Signal Processing 23 (1991) 273–292.
[2.13] W.A. Gardner, C.K. Chen, The Random Processes
Tutor, McGrawHill, New York, 1991.
[2.14] W.A. Gardner, Exploitation of spectral redundancy in
cyclostationary signals, IEEE Signal Process. Mag. 8
(1991) 14–36.
[2.15] W.A. Gardner, An introduction to cyclostationary
signals, in: W.A. Gardner (Ed.), Cyclostationarity in
Communications and Signal Processing, IEEE Press,
New York, 1994, pp. 1–90.
[2.16] W.A. Gardner, C.M. Spooner, Cyclostationary signal
processing, in: C.T. Leondes (Ed.), Digital Signal
Processing Techniques and Applications, Academic
Press, New York, 1994.
[2.17] G.B. Giannakis, Cyclostationary signal analysis,
in: V.K. Madisetti, D.B. Williams (Eds.), The Digi
tal Signal Processing Handbook, CRC Press and
IEEE Press, Boca Raton, FL and New York, 1998
(Chapter 17).
[2.18] H.L. Hurd, A.G. Miamee, Periodically Correlated
Random Sequences: Spectral Theory and Practice,
Wiley, New York, 2006.
[3.1] L.I. Gudzenko, On periodic nonstationary processes,
Radio Eng. Electron. Phys. 4 (1959) 220–224 (in
Russian).
[3.2] V.L. Lebedev, On random processes having nonstatio
narity of periodic character, Nauchn. Dokl. Vysshch.
Shchk. Ser. Radiotekh. Elektron. 2 (1959) 32–34 (in
Russian).
[3.3] E.G. Gladyshev, Periodically correlated random se
quences, Sov. Math. Dokl. 2 (1961) 385–388.
[3.4] J. Kampe´ de Fe´ riet, Correlations and spectra for
nonstationary random functions, Math. Comput. 16
(77) (1962) 1–21.
[3.5] E.G. Gladyshev, Periodically and almost periodically
correlated random processes with continuous time para
meter, Theory Probab. Appl. (1963) 137–177 (in Russian).
[3.6] D.M. Willis, The statistics of a particular nonhomo
geneous Poisson process, Biometrika 51 (1964) 399–404.
[3.7] V.A. Markelov, Extrusions and phases of the periodi
cally nonstationary random process, Izv. VUZ Radio
tekhn. 9 (5) (1966).
[3.8] A.A. Kayatskas, Periodically correlated random processes,
Telecommun. Radio Eng. 23 (Part 2) (1968) 136–141.
[3.9] Y.P. Dragan, Periodically correlated random processes
and transformations with periodically varying para
meters, Otbor Peredacha Inform. 22 (1969) 27–33 (in
Russian).
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 673
[3.10] Y.P. Dragan, The spectral properties of periodically
correlated stochastic processes, Otbor Peredacha In
form. 30 (1971) 16–24 (in Russian).
[3.11] H. Ogura, Spectral representation of a periodic
nonstationary random process, IEEE Trans. Inform.
Theory IT17 (1971) 143–149.
[3.12] Y.P. Dragan, On foundations of the stochastic model of
rhythmic phenomena, Otbor Peredacha Inform. 31
(1972) 21–27 (in Russian).
[3.13] K.N. Gupta, Stationariness provided by ﬁltration to a
periodic nonstationary random process, J. Sound Vib.
23 (1972) 319–329.
[3.14] K.S. Voychishin, Y.P. Dragan, Some properties of the
stochastic model of systems with periodically varying
parameters, Algorithms Prod. Process. (Kiev) (1972)
11–20 (in Russian).
[3.15] K.S. Voychishin, Y.P. Dragan, The elimination
of rhythm from periodically correlated random pro
cesses, Otbor Peredacha Inform. 33 (1972) 12–16
(in Russian).
[3.16] R. Lugannani, Sample stability of periodically
correlated pulse trains, J. Franklin Inst. 296 (1973)
179–190.
[3.17] V.A. Melititskiy, V.O. Radzievskii, Statistical charac
teristics of a nonstationary Normal process and ways of
its modeling, Radio Electron. Commun. Syst. 16 (1973)
82–89.
[3.18] Y.P. Dragan, I.N. Yavorskii, A paradox of the
rhythmic model, Otbor Peredacha Inform. 41 (1974)
11–15 (in Russian).
[3.19] H.L. Hurd, Periodically correlated processes with
discontinuous correlation functions, Theory Probab.
Appl. 9 (1974) 804–807.
[3.20] H.L. Hurd, Stationarizing properties of random shifts,
SIAM J. Appl. Math. 26 (1) (1974) 203–212.
[3.21] Y.P. Dragan, General properties of the stochastic
model of rhythmic phenomena, Otbor Peredacha
Inform. 44 (1975) 3–14 (in Russian).
[3.22] Y.P. Dragan, The representation of a periodically
correlated random process by stationary components,
Otbor Peredacha Inform. 45 (1975) 7–20 (in Russian).
[3.23] W.A. Gardner, L.E. Franks, Characterization of
cyclostationary random signal processes, IEEE Trans.
Inform. Theory IT21 (1975) 4–14.
[3.24] Z.A. Piranashvili, L.O. Kavtaradze, Certain properties
of periodic nonstationary random processes, V.I. Lenin.
Sakharth. Politekh. Inst. Samecn. Shrom 3 (176), Math.
Mehk. (1975) 93–105 (in Russian).
[3.25] Y.P. Dragan, Harmonizability and spectral distri
bution of random processes with ﬁnite mean power,
Dokl. Arad. Nauk Ukrain 8 (1978) 679–684 (in
Russian).
[3.26] W.A. Gardner, Stationarizable random processes,
IEEE Trans. Inform. Theory IT24 (1978) 8–22.
[3.27] J. Rootenberg, S.A. Ghozati, Generation of a class of
nonstationary random processes, Internat. J. Systems
Sci. 9 (1978) 935–947.
[3.28] M.G. Bilyik, Some transformations of periodic inho
mogeneous random ﬁelds, Otbor Peredacha Inform. 57
(1979) 21–25 (in Russian).
[3.29] I. Honda, Sample periodicity of periodically correlated
processes, Keio Math. Rep. 5 (1980) 13–18.
[3.30] A.G. Miamee, H. Salehi, On the prediction of
periodically correlated stochastic processes, in: P.R.
Krishnaiah (Ed.), Multivariate Analysis—V, North
Holland Publishing Company, Amsterdam, 1980,
pp. 167–179.
[3.31] P. Kochel, Periodically stationary Markovian decision
models, Elektron. Informationsverarb. Kybernet. 16
(1980) 553–567 (in German).
[3.32] I. Honda, Spectral representation of periodically
correlated stochastic processes and approximate Four
ier series, Keio Math. Sem. Rep. 6 (1981) 11–16.
[3.33] A.I. Kapustinskas, Properties of periodically nonsta
tionary processes, Trudy Akad. Nauk Litov. SSR Ser. B
1 (122) (1981) 87–95 (in Russian).
[3.34] B.S. Rybakov, Correlation properties of the envelope and
phase of a periodically nonstationary Normal process,
Telecommun. Radio Eng. 35–36 (1981) 96–97 (English
translation of Elektrosvyaz and Radiotekhnika).
[3.35] I. Honda, On the spectral representation and related
properties of a periodically correlated stochastic pro
cess, Trans. IECE of Japan E 65 (1982) 723–729.
[3.36] H. Ogura, Y. Yoshida, Time series analysis of a
periodic stationary random process, Trans. Inst. Electr.
Comm. Eng. Japan 365A (1982) 22–29 (in Japanese).
[3.37] M. Pourahmadi, H. Salehi, On subordination and linear
transformation of harmonizable and periodically corre
lated processes, in: Probability Theory on Vector
Spaces, III, Springer (Lublin), Berlin, New York,
1983, pp. 195–213.
[3.38] A.I. Kamolor, Kolmogorov diameters of a class of
random processes, Dokl. Akad. Nauk USSR 8 (1984)
8–10 (in Russian).
[3.39] Y.P. Dragan, I.N. Yavorskii, Statistical analysis of
periodic random processes, Otbor Peredacha Inform. 71
(1985) 20–29 (in Russian).
[3.40] Y.P. Dragan, Periodic and periodically nonstationary
random processes, Otbor Peredacha Inform. 72 (1985)
3–17 (in Russian).
[3.41] I.N. Yavorskii, Statistical analysis of periodically
correlated random processes, Radiotekh. Elektron. 30
(1985) 1096–1104 (in Russian).
[3.42] B.S. Rybakov, Correlation coefﬁcients of the envelope
and phase of a signal for periodically nonstationary
Gaussian noise, Radiotekhnika 41 (1986) 3–7 (in
Russian).
[3.43] W.A. Gardner, Rice’s representation for cyclostation
ary processes, IEEE Trans. Commun. COM35 (1987)
74–78.
[3.44] W.A. Gardner, Common pitfalls in the application of
stationary process theory to timesampled and modu
lated signals, IEEE Trans. Commun. COM35 (1987)
529–534.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 674
[3.45] V.A. Melititskiy, A.I. Mosionzhik, A probabilistic
model of non Gaussian periodically nonstationary
radio signals, Sov. J. Commun. Technol. Electron. 32
(1987) 100–106.
[3.46] I.N. Yavorskii, Statistical analysis of periodically
correlated vector random processes, Otbor Peredacha
Inform. 76 (1987) 3–12.
[3.47] N.S. Akinshin, V.L. Rumyantsev, A.V. Mikhailov,
V.V. Melititskaya, Statistical characteristics of the
envelope of an additive mixture of a nonGaussian
periodically nonstationary radio signal and nonGaus
sian interference, Radioelectron. Commun. Systems 31
(1988) 89–92.
[3.48] A.Y. Dorogovtsev, L. Tkhuan, The existence of
periodic and stationary regimes of discrete dynamical
systems in a Banach space, Kibernetika (Kiev) (6)
(1988) 121–123, 136.
[3.49] S.M. Rytov, Y.A. Kratsov, V.I. Tatarskii, Principles of
Statistical Radiophysics. Correlation Theory of Ran
dom Processes, vol. 2, Springer, Berlin, 1988.
[3.50] R. Ballerini, W.P. McCormick, Extreme value theory
for processes with periodic variances, Stochastic Models
5 (1989) 45–61.
[3.51] H.L. Hurd, Representation of strongly harmonizable
periodically correlated processes and their covariances,
J. Multivariate Anal. 29 (1989) 53–67.
[3.52] I.N. Yavorskii, Statistical analysis of poly and nearly
periodicallycorrelated random processes, Otbor Per
edacha Inform. 79 (1989) 1–10 (in Russian).
[3.53] V.G. Alekseev, On the construction of spectral densities
of a periodically correlated random process, Problemy
Peredachi Inform. 26 (1990) 106–108.
[3.54] A.Y. Dorogovtsev, Stationary and periodic solutions of
a stochastic difference equation in a Banach space,
Teor. Veroyatnost. Mat. Statist. ( 42) (1990) 35–42.
[3.55] A.Y. Dorogovtsev, Necessary and sufﬁcient conditions
for existence of stationary and periodic solutions of a
stochastic difference equation in Hilbert space, Com
put. Math. Appl. 19 (1) (1990) 31–37.
[3.56] A.Y. Dorogovtsev, Stationary and periodic solutions of
stochastic difference and differential equations in
Banach space, in: New Trends in Probability and
Statistics, vol. 1, Bakuriani, 1990, VSP, Utrecht, 1991,
pp. 375–390.
[3.57] A.G. Miamee, Periodically correlated processes and
their stationary dilations, SIAM J. Appl. Math. 50 (4)
(1990) 1194–1199.
[3.58] D. Dehay, Contributions to the spectral analysis of
nonstationary processes, Dissertation for the Habilita
tion to Direct Research, Universite des Sciences et
Techniques de Lille Flandres Artois, 1991.
[3.59] H.L. Hurd, Correlation theory of almost periodically
correlated processes, J. Multivariate Anal. 37 (1991)
24–45.
[3.60] G.D. Z
˘
ivanovic´ , W.A. Gardner, Degrees of cyclosta
tionarity and their application to signal detection and
estimation, Signal Processing 22 (1991) 287–297.
[3.61] A. Makagon, A.G. Miamee, H. Salehi, Periodically
correlated processes and their spectrum, in: A.G.
Miamee (Ed.), Proceedings of the Workshop on
Nonstationary Stochastic Processes and their Applica
tions, World Scientiﬁc, Virginia, VA, 1991, pp. 147–164.
[3.62] H.L. Hurd, G. Kallianpur, Periodically correlated
processes and their relationship to L
2
[0; T]valued
stationary sequences, in: A.G. Miamee (Ed.), Proceed
ings of the Workshop on Nonstationary Stochastic
Processes and their Applications, World Scientiﬁc,
Virginia, VA, 1991, pp. 256–284.
[3.63] G.D. Z
˘
ivanovic´ , On the instantaneous frequency of
cyclostationary random signals, IEEE Trans. Signal
Process. 39 (7) (1991) 1604–1610.
[3.64] W.A. Gardner, A unifying view of coherence in signal
processing, Signal Processing 29 (1992) 113–140.
[3.65] H.L. Hurd, Almost periodically unitary stochastic
processes, Stochastic Process. Appl. 43 (1992) 99–113.
[3.66] D.G. Konstantinides, V.I. Pieterbarg, Extreme values
of the cyclostationary Gaussian random process, J.
Appl. Probab. 30 (1993) 82–97.
[3.67] V.P. Sathe, P.P. Vaidyanathan, Effects of multirate
systems on statistical properties of random signals,
IEEE Trans. Signal Process. 41 (1) (1993) 131–146.
[3.68] D. Dehay, Spectral analysis of the covariance of the
almost periodically correlated processes, Stochastic
Process. Appl. 50 (1994) 315–330.
[3.69] A. Makagon, A.G. Miamee, H. Salehi, Continuous
time periodically correlated processes: spectrum and
prediction, Stochastic Process. Appl. 49 (1994) 277–295.
[3.70] S. Cambanis, C. Houdre´ , On the continuous wavelet
transform of secondorder random processes, IEEE
Trans. Inform. Theory 41 (3) (1995) 628–642.
[3.71] S. Lambert, Extension of autocovariance coefﬁcients
sequence for periodically correlated random processes
by using the partial autocorrelation function, in:
Proceedings of VIII European Signal Processing Con
ference (EUSIPCO’96), Trieste, Italy, 1996.
[3.72] R.J. Swift, Almost periodic harmonizable processes,
Georgian Math. J. 3 (3) (1996) 275–292.
[3.73] L.E. Franks, Random processes, in: J. Gibson (Ed.),
The Communications Handbook, CRC Press and IEEE
Press, New York, 1997 (Chapter 5).
[3.74] H. Zhang, Maximum entropy modeling of periodically
correlated processes, IEEE Trans. Inform. Theory 43
(6) (1997) 2033–2035.
[3.75] G. De Nicolao, G. FerrariTrecate, On the Wold
decomposition of discretetime cyclostationary pro
cesses, IEEE Trans. Signal Process. 47 (7) (1999)
2041–2043.
[3.76] B. Lall, S.D. Joshi, R.K.P. Bhatt, Secondorder
statistical characterization of the ﬁlter bank and its
elements, IEEE Trans. Signal Process. 47 (6) (1999)
1745–1749.
[3.77] A. Makagon, Induced stationary process and structure
of locally square integrable periodically correlated
processes, Stud. Math. 136 (1) (1999) 71–86.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 675
[3.78] S. Akkarakaran, P.P. Vaidyanathan, Bifrequency and
bispectrum maps: a new look at multirate systems with
stochastic inputs, IEEE Trans. Signal Process. 48 (3)
(2000) 723–736.
[3.79] D. Alpay, A. Chevreuil, P. Loubaton, An extension
problem for discretetime periodically correlated sto
chastic processes, J. Time Series Anal. 22 (1) (2001)
1–12.
[3.80] A. Makagon, Characterization of the spectra of
periodically correlated processes, J. Multivariate Anal.
78 (1) (2001) 1–10.
[3.81] S. Touati, J.C. Pesquet, Statistical properties of the
wavelet decomposition of cyclostationary processes, in:
Proceedings of the IEEEEURASIP Workshop on
Nonlinear Signal and Image Processing, Baltimore,
MD, 2001.
[3.82] P. Borgnat, P. Flandrin, P.O. Amblard, Stochastic
discrete scale invariance, IEEE Signal Process. Lett. 9
(6) (2002) 181–184.
[3.83] A.Y. Dorogovtsev, .I.Discretesystems Periodicity in
distribution, Internat. J. Math. Math. Sci. 30 (2)
(2002) 65–127.
[3.84] A.G. Miamee, G.H. Shahkar, Shift operator for
periodically correlated processes, Indian J. Pure Appl.
Math. 33 (5) (2002) 705–712.
[3.85] H.J. Woerdeman, The Carathe´ odory–Toeplitz problem
with partial data, Linear Algebra Appl. 342 (2002)
149–161.
[3.86] R. Guidorzi, R. Diversi, Minimal representations
of MIMO timevarying systems and realization of
cyclostationary models, Automatica 39 (2003)
1903–1914.
[3.87] H.L. Hurd, T. Koski, The Wold isomorphism for
cyclostationary sequences, Signal Processing 84 (2004)
813–824.
[3.88] H.L. Hurd, T. Koski, Spectral theory of cyclostationary
arrays, in: A.C. Krinik, R.J. Swift (Eds.), Stochastic
Processes and Functional Analysis: Recent Advances. A
Volume in Honor of Professor M.M. Rao, Marcel
Dekker, New York, 2004.
[3.89] D.G. Konstantinides, V.I. Pieterbarg, S. Stamotovic,
Gnedenkotype limit theorems for cyclostationary w
2

processes, Lithuanian Math. J. 44 (2) (2004) 157–167.
[3.90] P. Borgnat, P. Amblard, P. Flandrin, Scale invariances
and Lamperti transformations for stochastic processes,
J. Phys. A: Math. Gen. 38 (2005) 2081–2101.
[3.91] S. LambertLacroix, Extension of autocovariance coef
ﬁcients sequence for periodically correlated processes, J.
Time Ser. Anal. 26 (3) (2005) 423–435.
[4.1] L.J. Herbst, Almost periodic variances, Ann. Math.
Stat. 34 (1963) 1549–1557.
[4.2] L.J. Herbst, The statistical Fourier analysis of var
iances, J. Roy. Statist. Soc. B 27 (1965) 159–165.
[4.3] A.I. Kapustinskas, Estimation of the parameters of a
periodically nonstationary autoregressive process, Tru
dy Akad. Nauk Litov. SSR Ser. B 4 (101) (1977)
115–121 (in Russian).
[4.4] A. Renger, Spectral analysis of periodically nonsta
tionary stochastic impulse processes, Z. Angew. Math.
Mech. 57 (1977) 681–692 (in German).
[4.5] A.I. Kapustinskas, Covariational estimation of the
parameters of a periodically nonstationary autoregres
sive process, Liet. Tsr Mokslu Akad. Darbai B Ser. 104
(1978) 113–119 (in Russian).
[4.6] Y.P. Dragan, V.P. Mezentsev, I.N. Yavorskii, Algo
rithm and program for calculating estimates of the
characteristics of periodically correlated random pro
cesses, Otbor Peredacha Inform. 58 (1979) 10–15 (in
Russian).
[4.7] S.M. Porotskii, M.Y. Tsymbalyuk, Extension of
stochastic approximation procedures to periodically
nonstationary random processes, Automat. Remote
Control 40 (Part 2) (1979) 606–609.
[4.8] W.A. Gardner, Cycloergodicity, Part I: Application to
spectral extraction, in: D.G. Lainiotis, N.S. Tzannes
(Eds.), Advances in Communications, D. Reidel, Hol
land, 1980, pp. 203–209.
[4.9] S.I. Yuran, Methods for the instrumental analysis of
periodically nonstationary random processes, Meas.
Tech. 23 (1980) 1069–1071.
[4.10] D. Osteyee, Testing Markov properties of timeseries,
in: Proceedings of TimeSeries Analysis, Houston, TX,
NorthHolland, Amsterdam, 1981, pp. 385–400.
[4.11] Y.P. Dragan, V.P. Mezentsev, I.N. Yavorskii, Symme
try of the covariance matrix of measurements of a
periodically correlated random process, Otbor Pereda
cha Inform. 66 (1982) 3–6 (in Russian).
[4.12] H.J. Newton, Using periodic autoregressions for multi
ple spectral estimation, Technometrics 24 (1982)
109–116.
[4.13] R.A. Boyles, W.A. Gardner, Cycloergodic properties
of discrete parameter nonstationary stochastic pro
cesses, IEEE Trans. Inform. Theory IT29 (1983)
105–114.
[4.14] I. Honda, A note on periodogram analysis of periodi
cally correlated stochastic processes, Keio Math. Sem.
Rep. 8 (1983) 1–7.
[4.15] I.N. Yavorskii, Properties of estimators of mathema
tical expectation and correlation function of periodi
callycorrelated random processes, Otbor Peredacha
Inform. 67 (1983) 22–28 (in Russian).
[4.16] Y.P. Dragan, V.P. Mezentsev, I.N. Yavorskii, Compu
tation of estimates for spectral characteristics of
periodically correlated random processes, Otbor Per
edacha Inform 69 (1984) 29–35 (in Russian).
[4.17] W.A. Gardner, Measurement of spectral correlation,
IEEE Trans. Acoust. Speech Signal Process. ASSP34
(1986) 1111–1123.
[4.18] I.N. Yavorskii, Component estimates of the probability
characteristics of periodically correlated random pro
cesses, Avtomatika 19 (1986) 44–48 (in Russian).
[4.19] I.N. Yavorskii, Interpolation of estimates of probability
characteristics of periodically correlated random pro
cesses, Avtometrika 1 (1987) 36–41 (in Russian).
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 676
[4.20] V.G. Alekseev, Estimating the spectral densities of a
Gaussian periodically correlated stochastic process,
Problems Inform. Transmission 24 (1988) 109–115.
[4.21] W.A. Brown, H.H. Loomis Jr., Digital implementations
of spectral correlation analyzers, in: Proceedings of the
Fourth Annual ASSP Workshop on Spectrum Estima
tion and Modeling, Minneapolis, MN, 1988, pp. 264–270
[4.22] V.P. Mezentsev, I.N. Yavorskii, Estimation of prob
ability characteristics of rhythmic signals as a problem
of linear ﬁltration, Radioelectron. Commun. Systems 31
(1988) 70–72.
[4.23] C.J. Tian, A limiting property of sample autocovar
iances of periodically correlated processes with applica
tion to period determination, J. Time Ser. Anal. 9
(1988) 411–417.
[4.24] H.L. Hurd, Nonparametric time series analysis for
periodically correlated processes, IEEE Trans. Inform.
Theory 35 (1989) 350–359.
[4.25] R.S. Roberts, Architectures for digital cyclic spectral
analysis, Ph.D. Dissertation, Department of Electrical
Engineering and Computer Science, University of
California, Davis, CA, 1989.
[4.26] I. Honda, On the ergodicity of Gaussian periodically
correlated stochastic processes, Trans. Inst. Electron.
Inform. Commun. Eng. E 73 (10) (1990) 1729–1737.
[4.27] C. Jacob, Ergodicity in periodic autoregressive models,
C. R. Acad. Sci. Ser. I 310 (1990) 431–434 (in French).
[4.28] D.E. Martin, Estimation of the minimal period of
periodically correlated processes, Ph.D. Dissertation,
University of Maryland, 1990.
[4.29] H. Sakai, Spectral analysis and lattice ﬁlter for periodic
autoregressive processes, Electron. Commun. Japan
(Part 3) 73 (1990) 9–15.
[4.30] V.G. Alekseev, Spectral density estimators of a
periodically correlated stochastic process, Problems
Inform. Transmission 26 (1991) 286–288.
[4.31] W.A. Gardner, Two alternative philosophies for
estimation of the parameters of timeseries, IEEE
Trans. Inform. Theory 37 (1991) 216–218.
[4.32] R.S. Roberts, W.A. Brown, H.H. Loomis Jr., Compu
tationally efﬁcient algorithms for cyclic spectral analy
sis, IEEE Signal Process. Mag. 8 (1991) 38–49.
[4.33] S. Yamaguchi, Y. Kato, A statistical study for
determining the minimum sample size for L
eq
estima
tion of periodic nonstationary random noise, Appl.
Acoust. 32 (1991) 35–48.
[4.34] J. Wilbur, J. Bono, Wigner/cycle spectrum analysis of
spread spectrum and diversity transmissions, IEEE J.
Ocean. Eng. 16 (1) (1991) 98–106.
[4.35] H.L. Hurd, J. Les´ kow, Estimation of the Fourier
coefﬁcient functions and their spectral densities for f
mixing almost periodically correlated processes, Statist.
Probab. Lett. 14 (1992) 299–306.
[4.36] H.L. Hurd, J. Les´ kow, Strongly consistent and asymp
totically normal estimation of the covariance for almost
periodically correlated processes, Statist. Decisions 10
(1992) 201–225.
[4.37] J. Les´ kow, A. Weron, Ergodic behavior and estimation
for periodically correlated processes, Statist. Probab.
Lett. 15 (1992) 299–304.
[4.38] W.A. Brown, H.H. Loomis Jr., Digital implementation
of spectral correlation analyzers, IEEE Trans. Signal
Process. 41 (2) (1993) 703–720.
[4.39] W.A. Gardner, R.S. Roberts, Onebit spectral correla
tion algorithms, IEEE Trans. Signal Process. 41 (1)
(1993) 423–427.
[4.40] S. Cambanis, C.H. Houdre´ , H.L. Hurd, J. Les´ kow,
Laws of large numbers for periodically and almost
periodically correlated processes, Stochastic Process.
Appl. 53 (1994) 37–54.
[4.41] D. Dehay, H.L. Hurd, Representation and estimation
for periodically and almost periodically correlated
random processes, in: W.A. Gardner (Ed.), Cyclosta
tionarity in Communications and Signal Processing,
IEEE Press, New York, 1994, pp. 295–326 (Chapter 6).
[4.42] M.J. Genossar, H. LevAri, T. Kailath, Consistent
estimation of the cyclic autocorrelation, IEEE Trans.
Signal Process. 42 (3) (1994) 595–603.
[4.43] N.L. Gerr, C. Allen, The generalized spectrum and
spectral coherence of harmonizable time series, Digital
Signal Process. 4 (1994) 222–238.
[4.44] J. Les´ kow, Asymptotic normality of the spectral density
estimator for almost periodically correlated stochastic
processes, Stochastic Process. Appl. 52 (1994) 351–360.
[4.45] R.S. Roberts, W.A. Brown, H.H. Loomis Jr., A review
of digital spectral correlation analysis: theory and
implementation, in: W.A. Gardner (Ed.), Cyclostatio
narity in Communications and Signal Processing, IEEE
Press, New York, 1994, pp. 455–479.
[4.46] T. Biedka, L. Mili, J.H. Reed, Robust estimation
of cyclic correlation in contaminated Gaussian noise,
in: Proceedings of 29th Asilomar Conference on
Signals, Systems and Computers, Paciﬁc Grove, CA,
1995
[4.47] D. Dehay, Asymptotic behavior of estimators of cyclic
functional parameters for some nonstationary pro
cesses, Statist. Decisions 13 (1995) 273–286.
[4.48] R.S. Roberts, J.H.H. Loomis, Parallel computation
structures for a class of cyclic spectral analysis
algorithms, J. VLSI Signal Process. 10 (1995) 25–40.
[4.49] B.M. Sadler, Acoustooptic cyclostationary signal
processing, Appl. Opt. 34 (23) (1995) 5091–5099.
[4.50] S.V. Schell, Asymptotic moments of estimated cyclic
correlation matrices, IEEE Trans. Signal Process. 43 (1)
(1995) 173–180.
[4.51] M.C. Sullivan, E.J. Wegman, Estimating spectral
correlations with simple nonlinear transformations,
IEEE Trans. Signal Process. 43 (6) (1995) 1525–1526.
[4.52] D. Dehay, J. Les´ kow, Functional limit theory for the
spectral covariance estimator, J. Appl. Probab. 33
(1996) 1077–1092.
[4.53] D. Dehay, V. Monsan, Random sampling estimation
for almostperiodically correlated processes, J. Time
Ser. Anal. 17 (5) (1996) 425–445.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 677
[4.54] A. Makagon, A.G. Miamee, Weak law of large
numbers for almost periodically correlated processes,
Proc. Amer. Math. Soc. 124 (6) (1996) 1899–1902.
[4.55] F. Dominique, J.H. Reed, Estimating spectral correla
tions using the least mean square algorithm, Electron.
Lett. 33 (3) (1997) 182–184.
[4.56] H. Li, Q. Cheng, Almost sure convergence analysis
of mixed time averages and kthorder cyclic sta
tistics, IEEE Trans. Inform. Theory 43 (4) (1997)
1265–1268.
[4.57] P.W. Wu, H. LevAri, Optimized estimation of
moments for nonstationary signals, IEEE Trans. on
Signal Process. 45 (5) (1997) 1210–1221.
[4.58] L.G. Hanin, B.M. Schreiber, Discrete spectrum of
nonstationary stochastic processes on groups, J. Theo
ret. Probab. 11 (4) (1998) 1111–1133.
[4.59] B.M. Sadler, A.V. Dandawate´ , Nonparametric estima
tion of the cyclic cross spectrum, IEEE Trans. Inform.
Theory 44 (1) (1998) 351–358.
[4.60] Z. Huang, W. Wang, Y. Zhou, W. Jiang, Asymptotic
analysis of estimated cyclic crosscorrelation function
between stationary and cyclostationary processes,
System Eng. Electron. 14 (3) (2003) 87–91.
[4.61] D. Dehay, V. Monsan, Discrete periodic sampling with
jitter and almost periodically correlated processes,
Statist. Inference Stochastic Process. (2005).
[5.1] W.R. Bennett, Statistics of regenerative digital trans
mission, Bell System Tech. J. 37 (1958) 1501–1542.
[5.2] L.E. Franks, Signal Theory, PrenticeHall, Englewood
Cliffs, NJ, 1969 (Chapter 8).
[5.3] A.A. Kayatskas, Quasiperiodically correlated random
processes, Telecommun. Radio Eng. (Part 2) 25 (1970)
145–146 (English translation of Radiotekhnika).
[5.4] P. van Der Wurf, On the spectral density of a
cyclostationary process, IEEE Trans. Commun. 22
(10) (1974) 1727–1730.
[5.5] Y.V. Berezin, I.V. Krasheninnikov, Distribution func
tions of the envelope and phase of a periodically non
stationary process frequently encountered in radio
physics, Geomagn. Aeron. 18 (1978) 309–312.
[5.6] O.P. Dragan, Y.P. Dragan, Y.V. Karavan, A pro
babilistic model of rhythmicity in the radiolysis of
solids, Otbor Peredacha Inform. 62 (1980) 26–31 (in
Russian).
[5.7] A. Papoulis, Random modulation: a review, IEEE
Trans. Acoust. Speech Signal Process. ASSP31 (1983)
96–105.
[5.8] A. Papoulis, Probability, Random Variables, and
Stochastic Processes, second ed., McGrawHill, New
York, 1984.
[5.9] W.A. Gardner, Spectral correlation of modulated
signals, Part I—Analog modulation, IEEE Trans.
Commun. COM35 (6) (1987) 584–594.
[5.10] W.A. Gardner, W.A. Brown, C.K. Chen, Spectral
correlation of modulated signals, Part II—Digital
modulation, IEEE Trans. Commun. COM35 (6)
(1987) 595–601.
[5.11] V.E. Gurevich, S.Y. Korshunov, Statistical character
istics of a narrowband cyclostationary process, Tele
commun. Radio Eng. (Part 2) 42 (1987) 87–91.
[5.12] C.K. Chen, Spectral correlation characterization of
modulated signals with application to signal detection
and source location, Ph.D. Dissertation, Department of
Electrical Engineering and Computer Science, Univer
sity of California, Davis, CA, 1989.
[5.13] D.C. Bukofzer, Characterization and applications of
multiplexed PAM/PPM processes, in: Proceedings of
the International Symposium on Information Theory
and Applications (ISITA’90), Hawaii, USA, 1990,
pp. 915–918.
[5.14] A.I. Kozlov, A.I. Mosionzhik, V.R. Rusinov, Statistical
characteristics of polarization parameters of non
Gaussian periodically non stationary radio signals,
Sov. J. Commun. Technol. Electron. 35 (1990) 130–134.
[5.15] J. Wilbur, R.J. McDonald, Nonlinear analysis of
cyclically correlated spectral spreading in modulated
signals, J. Acoust. Soc. Amer. 92 (1992) 219–230.
[5.16] M.R. Bell, R.A. Grubbs, JEM modeling and measure
ments for radar target identiﬁcation, IEEE Trans.
Aerospace 29 (1993) 73–87.
[5.17] J.M.H. Elmirghani, R.A. Cryan, F.M. Clayton, Spec
tral analysis of timing jitter effects on the cyclosta
tionary PPM format, Signal Processing 43 (3) (1995)
269–277.
[5.18] P. Gournay, P. Viravau, Theoretical spectral correla
tion of CPM modulations—Part I: Analytic results for
twostates CPFSK modulation (1REC), Ann. Te´ le´ com
mun. 53 (7–8) (1998) 267–278 (in French).
[5.19] P. Viravau, P. Gournay, Theoretical spectral correla
tion of CPM modulations—Part II: General calculation
method and analysis, Ann. Te´ le´ commun. 53 (7–8)
(1998) 279–288 (in French).
[5.20] M.Z. Win, On the power spectral density of digital
pulse streams generated by Mary cyclostationary
sequences in the presence of stationary timing jitter,
IEEE Trans. Commun. 46 (9) (1998) 1135–1145.
[5.21] D. Vuc˘ ic´ , M. Obradovic´ , Spectral correlation evalua
tion of MSK and offset QPSK modulation, Signal
Processing 78 (1999) 363–367.
[5.23] A. Napolitano, C.M. Spooner, Cyclic spectral analysis
of continuousphase modulated signals, IEEE Trans.
Signal Process. 49 (1) (2001) 30–44.
[6.1] A.S. Monin, Stationary and periodic time series in the
general circulation of the atmosphere, in: E.M. Rosen
blatt (Ed.), Proceedings of the Symposium on Time
Series Analysis, Wiley, New York, 1963, pp. 144–151.
[6.2] R.H. Jones, W.M. Brelsford, Time series with periodic
structure, Biometrika 54 (1967) 403–408.
[6.3] R.K. Bhuiya, Stochastic analysis of periodic hydrologic
process, J. Hydraulics Division, American Society of
Civil Engineers 97 (1971) 949–963.
[6.4] A. Renger, Behaviour of linear oscillatory systems
which are energized by stochastic impulse trains,
Maschinenbau Tech. 20 (1971) 596–600 (in German).
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 678
[6.5] K.S. Voychishin, Y.P. Dragan, A simple stochastic
model of the natural rhythmic processes, Otbor
Peredacha Inform. 2 (1971) 7–15 (in Russian).
[6.6] D.F. Dubois, B. Besserides, Nonlinear theory of
parametric instabilities in plasmas, Phys. Rev. A 14
(1976) 1869–1882.
[6.7] B. Arciplani, F. Carloni, M. Marseguerra, Neutron
counting statistics in a subcritical cyclostationary
multiplying system, Nucl. Instrum. Methods (Part I
(1979) 465–473.
[6.8] B. Arciplani, F. Carloni, M. Marseguerra, Neutron
counting statistics in a subcritical cyclostationary
multiplying system, Nucl. Instrum. Methods (Part II)
172 (1980) 531.
[6.9] K. Hasselmann, T.P. Barnett, Techniques of linear
prediction for systems with periodic statistics, J. Atmos.
Sci. 38 (1981) 2275–2283.
[6.10] W.K. Johnson, The dynamic pneumocardiogram: an
application of coherent signal processing to cardiovas
cular measurement, IEEE Trans. Biomed. Eng. BME
28 (1981) 471–475.
[6.11] Y.P. Dragan, I.N. Yavorskii, Rhythmics of sea waves
and underwater acoustic signals, Naukova Dumka,
Kiev, 1982 (in Russian).
[6.12] T.P. Barnett, Interaction of the monsoon and paciﬁc
trade wind system at interannual time scales. I. The
equatorial zone, Mon. Weather Rev. III (1983)
756–773.
[6.13] T.P. Barnett, H.D. Heinz, K. Hasselmann, Statistical
prediction of seasonal air temperature over Eurasia,
Tellus Ser. A (Sweden) 36A (1984) 132–146.
[6.14] Y.P. Dragan, V.A. Rozhkov, I.N. Yavorskii, Applica
tions of the theory of periodically correlated random
processes to the probabilistic analysis of oceanological
time series, in: V.A. Rozhkov (Ed.), In Probabilistic
Analysis and Modeling of Oceanological Processes,
Gidrometeoizdat, Leningrad, USSR, 1984, pp. 4–23 (in
Russian).
[6.15] W. Dunsmuir, Time series regression with periodically
correlated errors and missing data, in: Time Series
Analysis of Irregularly Observed Data, Springer, New
York, Berlin, 1984, pp. 78–107.
[6.16] C.M. Johnson, P. Lemke, T.P. Barnett, Linear predic
tion of sea ice anomalies, J. Geophys. Res. 90 (1985)
5665–5675.
[6.17] M.J. Ortiz, A. Ruiz de Elvira, A cyclostationary model
of sea surface temperature in the paciﬁc ocean, Tellus
Ser. A 37A (1985) 14–23.
[6.18] V.O. Alekseev, Extraction of the trend in a periodically
correlated time series, Atmos. Ocean. Phys. 23 (1987)
187–191.
[6.19] Y.P. Dragan, V.A. Rozhkov, I.N. Yavorskii, Methods
of Probabilistic Analysis of Oceanological Rhythmics,
Gidrometeoizdat, Leningrad, USSR, 1987 (in Russian).
[6.20] Y.P. Dragan, Principles of a linear theory of stochastic
test signals and their statistical analysis, Otbor Per
edacha Inform. 77 (1988) 2–7 (in Russian).
[6.21] K.Y. Kim, G.R. North, Seasonal cycle and second
moment statistics of a simple coupled climate system,
J. Geophys. Res. 97 (20) (1992) 437–448.
[6.22] P. Bloomﬁeld, H.L. Hurd, R. Lund, Periodic correla
tion in stratospheric ozone time series, J. Time Ser.
Anal. 15 (2) (1994) 127–150.
[6.23] R.B. Lund, H.L. Hurd, P. Bloomﬁeld, R.L. Smith,
Climatological time series with periodic correlation,
J. Climate 11 (1995) 2787–2809.
[6.24] K.Y. Kim, G.R. North, J. Huang, EOFs of one
dimensional cyclostationary time series: computations,
examples and stochastic modeling, J. Atmos. Sci. 53
(1996) 1007–1017.
[6.25] A.R. Kacimov, Y.V. Obnosov, N.D. Yakimov,
Groundwater ﬂow in a medium with parquettype
conductivity distribution, J. Hydrol. 226 (1999)
242–249.
[6.26] F. Gini, M. Greco, Texture modelling, estimation and
validation using measured sea clutter, IEE Proc.Radar,
Sonar Navig. 149 (3) (2002) 115–124.
[6.27] X. MenendezPidal, G.H. Abrego, System and method
for performing speech recognition in cyclostationary
noise environments, US Patent No. 6,785,648, August
31, 2004.
[7.1] W.A. Gardner, The structure of leastmeansquare
linear estimators for synchronous Mary signals, IEEE
Trans. Inform. Theory IT19 (1973) 240–243.
[7.2] W.A. Gardner, An equivalent linear model for marked
and ﬁltered doubly stochastic Poisson processes with
application to MMSE estimation for synchronous M
ary optical data signals, IEEE Trans. Commun.
Technol. COM24 (1976) 917–921.
[7.3] M.F. Mesiya, P.J. McLane, L.L. Campbell, Optimal
receiver ﬁlters for BPSK transmission over a bandlim
ited nonlinear channel, IEEE Trans. Commun. COM
26 (1978) 12–22.
[7.4] T.H.E. Ericson, Modulation by means of linear periodic
ﬁltering, IEEE Trans. Inform. Theory IT27 (1981)
322–327.
[7.5] J.C. Campbell, A.J. Gibbs, B.M. Smith, The cy
clostationary nature of crosstalk interference from
digital signals in multipair cable—Part I: Fundamen
tals, IEEE Trans. Commun. COM31 (5) (1983)
629–637.
[7.6] J.C. Campbell, A.J. Gibbs, B.M. Smith, The cyclosta
tionary nature of crosstalk interference from digital
signals in multipair cable—Part II: Applications and
further results, IEEE Trans. Commun. COM31 (5)
(1983) 638–649.
[7.7] F.K. Graef, Joint optimization of transmitter and
receiver for cyclostationary random signal processes,
in: Proceedings of the NATO Advanced Study Institute
on Nonlinear Stochastic Problems, Reidel, Algarve,
Portugal, 1983, pp. 581–592.
[7.8] J.P.A. Albuquerque, O. Shimbo, L.N. Ngugen, Mod
ulation transfer noise effects from a continuous digital
carrier to FDM/FM carriers in memoryless nonlinear
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 679
devices, IEEE Trans. Commun. COM32 (1984)
337–353.
[7.9] B.M. Smith, P.G. Potter, Design criteria for crosstalk
interference between digital signals in multipair cable,
IEEE Trans. Commun. COM34 (1986) 593–599.
[7.10] J.W.M. Bergmans, A.J.E.M. Janssen, Robust data
equalization, fractional tap spacing and the Zak trans
form, Philips J. Res. 23 (1987) 351–398.
[7.11] J.H. Reed, Timedependent adaptive ﬁlters for inter
ference rejection, Ph.D.Dissertation, Department of
Electrical Engineering and Computer Science, Univer
sity of California, Davis, CA, 1987.
[7.12] B.G. Agee, The baseband modulus restoral approach to
blind adaptive signal demodulation, in: Proceedings of
Digital Signal Processing Workshop, 1988.
[7.13] V. Joshi, D.D. Falconer, Reduced state sequence
estimation techniques for digital subscriber loop appli
cation, in: Proceedings of GLOBECOM ’88, vol. 2,
Hollywood, CA, 1988, pp. 799–803.
[7.14] J.H. Reed, A.A. Quilici, T.C. Hsia, A frequency domain
time dependent adaptive ﬂiter for interference rejection,
in: Proceedings of MILCOM 88, vol. 2, San Diego, CA,
1988, pp. 391–397.
[7.15] B.G. Agee, The property restoral approach to blind
adaptive signal extraction, Ph.D. Dissertation, Depart
ment of Electrical Engineering and Computer Science,
University of California, Davis, CA, 1989.
[7.16] B.G. Agee, The propertyrestoral approach to blind
adaptive signal extraction, in: Proceedings of CSIARO
Workshop on Advanced Topics in Communications,
1989.
[7.17] B.G. Agee, S. Venkataraman, Adaptive demodu
lation of PCM signals in the frequency domain, in:
Proceedings of the 23rd Asilomar Conference on
Signals, Systems and Computers, Paciﬁc Grove, CA,
1989.
[7.18] A. Fung, L.S. Lee, D.D. Falconer, A facility for near
end crosstalk measurements on ISDN subscriber loops,
in: Proceedings of GLOBECOM ’89, IEEE Global
Telecommunications Conference and Exhibition, vol. 3,
Dallas, TX, 1989, pp. 1592–1596.
[7.19] W.A. Gardner, W.A. Brown, Frequencyshift ﬁltering
theory for adaptive cochannel interference removal, in:
Conference Record, 23rd Asilomar Conference on
Signals, Systems and Computers, vol. 2, Paciﬁc Grove,
CA, 1989, pp. 562–567.
[7.20] J.H. Reed, C. Greene, T.C. Hsia, Demodulation of a
direct sequence spreadspectrum signal using an optimal
timedependent ﬁlter, in: Proceedings of IEEE Militar
Communications Conference (MILCOM’89), Boston,
MA, 1989, pp. 657–662.
[7.21] W.A. Gardner, S. Venkataraman, Performance of
optimum and adaptive frequencyshift ﬁlters for
cochannel interference and fading, in: Proceedings of
the 24th Annual Asilomar Conference on Signals,
Systems, and Computers, Paciﬁc Grove, CA, 1990,
pp. 242–247.
[7.22] J.P. Leduc, Image modelling for digital TV codecs, in:
Proceedings of the SPIE, Lausanne, Switzerland, 1990,
pp. 1160–1170.
[7.23] J.H. Reed, T.C. Hsia, The performance of time
dependent adaptive ﬁtters for interference rejection,
IEEE Trans. Acoust. Speech Signal Process. 38 (8)
(1990) 1373–1385.
[7.24] R. Mendoza, J.H. Reed, T.C. Hsia, B.G. Agee,
Interference rejection using timedependent constant
modulus algorithms (CMA) and the hybrid CMA/
spectral correlation discriminator, IEEE Trans. Signal
Process. 39 (9) (1991) 2108–2111.
[7.25] B.R. Petersen, D.D. Falconer, Minimum mean square
equalization in cyclostationary and stationary inter
ference—analysis and subscriber line calculations, IEEE
J. Sel. Areas Commun. 9 (6) (1991) 931–940.
[7.26] M. Abdulrahman, D.D. Falconer, Cyclostationary
crosstalk suppression by decision feedback equaliza
tion on digital subscriber loops, IEEE J. Sel. Areas
Commun. 12 (4) (1992) 640–649.
[7.27] H. Ding, M.M. Fahmy, Synchronised linear almost
periodically timevarying adaptive ﬁlters, IEE Proc.I
139 (24) (1992) 429–436.
[7.28] C. Pateros, G. Saulnier, Interference suppression and
multipath mitigation using an adaptive correlator direct
sequence spread spectrum receiver, in: Proceedings of
IEEE ICC, 1992, pp. 662–666.
[7.29] B.G. Agee, Solving the nearfar problem: exploitation
of spectral and spatial coherence in wireless personal
communication networks, in: Proceedings of Virginia
Tech. Third Symposium on Wireless Personal Commu
nications, vol. 15, 1993, pp. 1–12.
[7.30] H. Ding, M.M. Fahmy, Performance of outputmod
ulatorstructured linear almost periodically timevarying
adaptive ﬁlters, IEE Proc.I 140 (2) (1993) 114–120.
[7.31] W.A. Gardner, Cyclic Wiener ﬁltering: theory and
method, IEEE Trans. Commun. 41 (1) (1993) 151–163.
[7.32] J. Proakis, Reducedcomplexity simultaneous beam
forming and equalization for underwater acoustic
communications, in: Proceedings of Oceans’93, 1993.
[7.33] M. Stojanovic, J. Catipovic, J. Proakis, Adaptive
receivers for underwater acoustic communications: their
relation to beamforming and diversity combining, in:
Proceedings of COMCON, vol. 4, 1993.
[7.34] M. Stojanovic, J. Catipovic, J. Proakis, Adaptive
multichannel combining and equalization for under
water acoustic communications, J. Acoust. Soc. Amer.
94 (1993) 1621–1631.
[7.35] J.W.M. Bergmans, Data receiver comprising a control
loop with reduced sampling frequency, US Patent No.
5,311,558, May 10, 1994.
[7.36] I. Fijalkow, F. Lopez de Victoria, C. Johnson, Adaptive
fractionally spaced blind equalization, in: Proceedings
of Digital Signal Processing Workshop, 1994.
[7.37] G.D. Golden, Extended bandwidth transmitter for
crosstalk channels, US Patent No. 5,377,230, December
27, 1994.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 680
[7.38] U. Madhow, M. Honig, MMSE interference suppres
sion for directsequence spreadspectrum CDMA, IEEE
Trans. Commun. 42 (12) (1994) 3178–3188.
[7.39] P. Rapajic, B. Vucetic, Adaptive receiver structures for
asynchronous CDMA systems, IEEE J. Sel. Areas
Commun. 12 (4) (1994) 685–697.
[7.40] S. Roy, J. Yang, P.S. Kumar, Joint transmitter/receiver
optimization for multiuser communications, in: W.A.
Gardner (Ed.), Cyclostationarity in Communications
and Signal Processing, IEEE Press, New York, 1994,
pp. 329–361.
[7.41] G.K. Yeung, W.A. Brown, W.A. Gardner, A new
algorithm for blindadaptive frequencyshift ﬁltering,
in: 28th Annual Conference on Information Science and
Systems, Princeton, NJ, 1994.
[7.42] J.W.M. Bergmans, Transmission system with increased
sampling rate detection, US Patent No. 5,463,654,
October 31, 1995.
[7.43] W.A. Gardner, G.K. Yeung, W.A. Brown, Signal
reconstruction after spectral excision, in: Proceedings
of the 29th Annual Asilomar Conference on Signals,
Systems, and Computers, Paciﬁc Grove, CA, 1995.
[7.44] S. Miller, An adaptive direct sequence code division
multiple access receiver for multiuser interference
rejection, IEEE Trans. Commun. 43 (2) (1995)
1746–1755.
[7.45] J.J. Nicolas, J.S. Lim, Method and apparatus for
decoding broadcast digital HDTV in the presence of
quasicyclostationary interference, US Patent No.
5,453,797, September 26, 1995.
[7.46] J.H. Reed, N.M. Yuen, T.C. Hsia, An optimal receiver
using a timedependent adaptive ﬁlter, IEEE Trans.
Commun. 43 (2/3/4) (1995) 187–190.
[7.47] N. Uzun, R.A. Haddad, Cyclostationary modeling, and
optimization compensation of quantization errors in
subband codecs, IEEE Trans. Signal Process. 43 (9)
(1995) 2109–2119.
[7.48] S. Bensley, B. Aazhang, Subspacebased channel
estimation for code division multiple access systems,
IEEE Trans. Commun. 44 (8) (1996) 1009–1020.
[7.49] S. Ohno, H. Sakai, Optimization of ﬁlter banks using
cyclostationary spectral analysis, IEEE Trans. Signal
Process. 44 (11) (1996) 2718–2725.
[7.50] C. Pateros, G. Saulnier, An adaptive correlator receiver
for directsequence spreadspectrum communication,
IEEE Trans. Commun. 44 (1996) 1543–1552.
[7.51] E. Strom, S. Parkvall, S. Miller, B. Ottersten, Propaga
tion delay estimation in asynchronous directsequence
codedivision multiple access systems, IEEE Trans.
Commun. 44 (1) (1996) 84–93.
[7.52] B.G. Agee, P. Kelly, D. Gerlach, The backtalk airlink
for full exploitation of spectral and spatial diversity in
wireless communication systems, in: Proceedings of
Fourth Workshop on Smart Antennas in Wireless
Mobile Communications, 1997.
[7.53] S. Bensley, B. Aazhang, Maximumlikelihood synchro
nization of a single user for codedivision multiple
access communication systems, IEEE Trans. Commun.
46 (3) (1998) 392–399.
[7.54] J. Cui, D.D. Falconer, A.U.H. Sheikh, Blind adapta
tion of antenna arrays using a simple algorithm based
on small frequency offset, IEEE Trans. Commun. 46 (1)
(1998) 61–70.
[7.55] W.A. Gardner, S.V. Schell, GMSK signal processors
for improved communications capacity and quality, US
Patent No. 5,848,105, December 8, 1998.
[7.56] G. Gelli, L. Paura, A.M. Tulino, Cyclostationarity
based ﬁltering for narrowband interference suppression
in directsequence spreadspectrum systems, IEEE J.
Sel. Areas Commun. 16 (1998) 1747–1755.
[7.57] H. LevAri, Adaptive RLS ﬁltering under the cyclo
stationary regime, in: Proceedings of International
Conference on Acoustics, Speech, and Signal Processing
(ICASSP’98), 1998, pp. 2185–2188.
[7.58] M. Lops, G. Ricci, A.M. Tulino, Narrowbandinter
ference suppression in multiuser CDMA systems, IEEE
Trans. Commun. 46 (9) (1998) 1163–1175.
[7.59] U. Madhow, Blind adaptive interference suppression
for directsequence CDMA, Proc. IEEE (1998)
2049–2069.
[7.60] D. McLernon, Interrelationships between different
structures for periodic systems, in: Proceedings of IX
European Signal Processing Conference (EUSIP
CO’98), Rhodes, Greece, 1998, pp. 1885–1888.
[7.61] X. Wang, H.V. Poor, Blind multiuser detection: a
subspace approach, IEEE Trans. Inform. Theory 44 (2)
(1998) 30–44.
[7.62] A. Duverdier, B. Lacaze, D. Roviras, Introduction of
linear cyclostationary ﬁlters to model timevariant
channels, in: Proceedings of Global Telecom
munications Conference (GLOBECOM’99), 1999,
pp. 325–329.
[7.63] B. Francis, S. Dasgupta, Signal compression by
subband coding, Automatica 35 (1999) 1895–1908.
[7.64] A. Pandharipande, S. Dasgupta, Subband coding of
cyclostationary signals with static bit allocation, IEEE
Signal Process. Lett. 6 (11) (1999) 284–286.
[7.65] U. Madhow, M. Honig, On the average nearfar
resistance for MMSE detection of directsequence
CDMA signals with random spreading, IEEE Trans.
Inform. Theory 45 (9) (1999).
[7.66] N. Mangalvedhe, Development and analysis of adap
tive interference rejection techniques for direct sequence
code division multiple access systems, Ph.D. Disserta
tion, Bradley Department of Electrical Engineering,
Virginia Polytechnic Institute and State University,
1999.
[7.67] E. Strom, S. Miller, Properties of the singlebit single
user MMSE receiver for DSCDMA systems, IEEE
Trans. Commun. 47 (3) (1999) 416–425.
[7.68] A.M. Tulino, Interference suppression in CDMA
systems, Ph.D. Dissertation, Dipartimento di Ingegner
ia dell’Informazione, Seconda Universita` di Napoli,
Aversa, Italy, January 1999 (in Italian).
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 681
[7.69] J. Zhang, K.M. Wong, Z.Q. Luo, P.C. Ching, Blind
adaptive FRESH ﬁltering for signal extraction, IEEE
Trans. Signal Process. 47 (5) (1999) 1397–1402.
[7.70] B.G. Agee, Stackedcarrier discrete multiple tone
communication technology and combinations with code
nulling, interference cancellation, retrodirective com
munication adaptive antenna arrays, US Patent No.
6,128,276, October 3, 2000.
[7.71] A. Duverdier, B. Lacaze, On the use of periodic clock
changes to implement linear periodic timevarying
ﬁlters, IEEE Trans. Circuits Systems II: Analog Digital
Signal Process. 47 (11) (2000) 1152–1159.
[7.72] G. Gelli, L. Paura, A.R.P. Ragozzini, Blind widely
linear multiuser detection, IEEE Commun. Lett. 4 (6)
(2000) 187–189.
[7.73] M. Lops, A.M. Tulino, Simultaneous suppression of
multiaccess and narrowband interference in asynchro
nous CDMA networks, IEEE Trans. Veh. Technol. 49
(2000) 1705–1718.
[7.74] M. Martone, Blind adaptive detection of DS/CDMA
signals on timevarying multipath channels with anten
na array using higherorder statistics, IEEE Trans.
Commun. 48 (9) (2000) 1590–1600.
[7.75] S.J. Yu, F.B. Ueng, Implementation of cyclostation
ary signalbased adaptive arrays, Signal Processing 80
(2000) 2249–2254.
[7.76] S. Buzzi, M. Lops, A.M. Tulino, Partially blind
adaptive MMSE interference rejection in asynchronous
DS/CDMA networks over frequencyselective fading
channels, IEEE Trans. Commun. 49 (1) (2001) 94–108.
[7.77] S. Buzzi, M. Lops, A.M. Tulino, A new family of
MMSE multiuser receiver for interference sup
pression in DS/CDMA systems employing BPSK
modulation, IEEE Trans. Commun. 49 (1) (2001)
154–167.
[7.78] S. Buzzi, M. Lops, A.M. Tulino, Blind adaptive
multiuser detection for asynchronous dual rate DS/
CDMA systems, IEEE J. Sel. Areas Commun. 19 (2001)
233–244.
[7.79] W.A. Gardner, Suppression of cochannel interference
in GSM by predemodulation signal processing, in:
Proceedings of the 11th Virginia Tech. Symposium on
Wireless Personal Communications, 2001, pp. 217–228.
[7.80] W.A. Gardner, C.W. Reed, Making the most out of
spectral redundancy in GSM: Cheap CCI suppression,
in: Proceedings of the 36th Annual Asilomar Con
ference on Signals, Systems and Computers, Paciﬁc
Grove, CA, 2001.
[7.81] G. Gelli, F. Verde, Adaptive minimum variance
equalization with interference suppression capabilities,
IEEE Commun. Lett. 5 (12) (2001) 491–493.
[7.82] A. Sabharwal, U. Mitra, R. Moses, MMSE receivers
for multirate DSCDMA systems, IEEE Trans. Com
mun. 49 (12) (2001) 2184–2197.
[7.83] J. Thibault, P. Chevalier, F. Pipon, J.J. Monot, G.
Multedo, Method and device for space division multi
plexing of radio signals transmitted in cellular radio
communications, US Patent No. 6,240,098, May 29,
2001.
[7.84] A.M. Tulino, S. Verdu` , Asymptotic analysis of im
proved linear receivers for BPSKCDMA subject to
fading, IEEE J. Sel. Areas Commun. 19 (2001)
1544–1555.
[7.85] F. Verde, Equalization and interference suppression
techniques for highcapacity digital systems, Ph.D.
Dissertation, Dipartimento di Ingegneria Elettronica e
delle Telecomunicazioni, Universita` di Napoli Federico
II, Napoli, Italy, November 2001 (in Italian).
[7.86] B.G. Agee, M. Bromberg, D. Gerlach, D. Gibbons, J.T.
Golden, M. Ho, E. Hole, M. Jesse, R.L. Maxwell, R.G.
Mechaley Jr., R.R. Naish, D.J. Nix, D.J. Ryan, D.
Stephenson, High bandwidth efﬁcient communications,
US Patent No. 6,359,923, March 19, 2002.
[7.87] Z. Ding, P.H. Thomas, R. Wang, W. Tong, Cochannel
interference reduction in wireless communications
systems, US Patent No. 6,396,885, May 28, 2002.
[7.88] A. Gameiro, Simple receiver for systems with spectrally
overlapping narrowband and broadband signals, Wire
less Personal Commun. 23 (2002) 311–320.
[7.89] Z. Huang, Y. Zhou, W. Jiang, On cyclic correlation
matched ﬁltering, Acta Electron. Sinica 30 (12) (2002)
122–126 (in Chinese).
[7.90] B. Lacaze, D. Roviras, Effect of random permutations
applied to random sequences and related applications,
Signal Processing 82 (6) (2002) 821–831.
[7.91] G.D. Mandyam, Digitaltoanalog conversion of pulse
amplitude modulated systems using adaptive quantiza
tion, Wireless Personal Commun. 23 (2002) 253–281.
[7.92] B.G. Agee, Stacked carrier discrete multiple tone
communication systems, US Patent No. 6,512,737,
January 28, 2003.
[7.93] H. Arslan, K.J. Molnar, Methods of receiving co
channel signals by channel separation and successive
cancellation and related receivers, US Patent No.
6,574,235, June 3, 2003.
[7.94] G. Gelli, F. Verde, Blind FSRLPTV equalization and
interference rejection, IEEE Trans. Commun. 51 (2)
(2003) 145–150.
[7.95] G.D. Mandyam, Adaptive quantization in a pulse
amplitude modulated system, US Patent No. 6,535,564,
March 18, 2003.
[7.96] F. Trans, Means and method for increasing perfor
mances of interferencesuppression based receivers, US
Patent No. 6,553,085, April 22, 2003.
[7.97] J.H. Cho, Joint transmitter and receiver optimization in
additive cyclostationary noise, IEEE Trans. Inform.
Theory 50 (12) (2004) 3396–3405.
[7.98] G.H. Im, H.C. Won, Performance analysis of
cyclostationary interference suppression for multiuser
wired communication systems, J. Commun. Networks 6
(2) (2004) 93–105.
[7.99] T.L. Carroll, Lowinterference communications device
using chaotic signals, US Patent No. 6,854,058,
February 8, 2005.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 682
[7.100] F. Gourge, F. Roosen, Device enabling different
spreading factors while preserving a common rambling
code, in particular for transmission in a code division
multiple access cellular mobile radio system, US Patent
No. 6,868,077, March 15, 2005.
[7.101] O. Vourinen, T. Seppanen, J. Roning, J. Lilleberg, T.
Kolehmainen, Method for distinguishing signals from
one another, and ﬁlter, US Patent No. 6,847,689,
January 25, 2005.
[8.1] L.E. Franks, J. Bubrouski, Statistical properties of
timing jitter in a PAM timing recovery scheme, IEEE
Trans. Commun., COM22 (1974) 913–930.
[8.2] U. Mengali, E. Pezzani, Tracking properties of phase
locked loops in optical communication systems, IEEE
Trans. Commun. COM26 (1978) 1811–1818.
[8.3] L.E. Franks, Carrier and bit synchronization in data
communication—A tutorial review, IEEE Trans. Com
mun. COM28 (1980) 1107–1121.
[8.4] M. Moeneclaey, Comment on ‘Tracking performance
of the ﬁlter and square bit synchronizer’, IEEE Trans.
Commun. COM30 (1982) 407–410.
[8.5] M. Moeneclaey, Linear phaselocked loop theory for
cyclostationary input disturbances, IEEE Trans. Com
mun. COM30 (10) (1982) 2253–2259.
[8.6] C.Y. Yoon, W.C. Lindsay, Phaselocked loop perfor
mance in the presence of CW interference and additive
noise, IEEE Trans. Commun. COM30 (1982)
2305–2311.
[8.7] M. Moeneclaey, The optimum closedloop transfer
function of a phaselocked loop used for synchroniza
tion purposes, IEEE Trans. Commun. COM31 (1983)
549–553.
[8.8] C. Bilardi, S. Pupolin, Spectral analysis of the powers of
a PAM digital signal, Alta Frequenza 53 (2) (1984)
70–76.
[8.9] M. Moeneclaey, A fundamental lower bound on the
performance of practical joint carrier and bit synchro
nizers, IEEE Trans. Commun. COM32 (1984)
1007–1012.
[8.10] J.J. O’Reilly, Timing extraction for baseband digital
transmission, in: K.W. Cattermole, J.J. O’Reilly (Eds.),
Problems of Randomness in Communication Engineer
ing, Plymouth, London, 1984.
[8.11] S. Pupolin, C. Tomasi, Spectral analysis of line
regenerator time jitter, IEEE Trans. Commun. COM
32 (1984) 561–566.
[8.12] C.M. Monti, G.L. Pierobon, Block codes for linear
timing recovery in data transmission systems, IEEE
Trans. Commun. COM33 (1985) 527–534.
[8.13] W.A. Gardner, The role of spectral correlation in
design and performance analysis of synchronizers,
IEEE Trans. Commun. COM34 (11) (1986) 1089–1095.
[8.14] Z.H. Qiu, F.A. Cassara, Steadystate analysis of the
clock synchronizer for a narrowband communication
system, Int. J. Electron. 66 (1989) 551–570.
[8.15] N.M. Blachman, Beneﬁcial effects of spectral correla
tion on synchronization, in: W.A. Gardner (Ed.),
Cyclostationarity in Communications and Signal Pro
cessing, IEEE Press, New York, 1994, pp. 362–390.
[8.16] J. Riba, G. Vazquez, Bayesian recursive estimation of
frequency and timing exploiting the cyclostationarity
property, Signal Processing 40 (1994) 21–37.
[8.17] K.E. Scott, M. Kaube, K. Anvari, Timing and
automatic frequency control of digital receiver using
the cyclic properties of a nonlinear operation, US
Patent No. 5,282,228, January 25, 1994.
[8.18] K.E. Scott, E.B. Olasz, Simultaneous clock phase and
frequency offset estimation, IEEE Trans. Commun. 43
(7) (1995) 2263–2270.
[8.19] B.G. Agee, R.J. Kleinman, J.H. Reed, Soft synchroni
zation of direct sequence spreadspectrum signals, IEEE
Trans. Commun. 44 (11) (1996) 1527–1536.
[8.20] F. Gini, G.B. Giannakis, Frequency offset and symbol
timing recover in ﬂatfading channels: a cyclostationary
approach, IEEE Trans. Commun. 46 (1998) 400–411.
[8.21] P. Ciblat, Some estimation problems relative to non co
operative communications, Ph.D. Dissertation, Univ.
MarnelaValle´ e, MarnelaValle´ e, France, 2000 (in
French).
[8.20] E. Serpedin, A. Chevreuil, G.B. Giannakis, P. Louba
ton, Blind channel and carrier frequency offset estima
tion using periodic modulation precoders, IEEE Trans.
Signal Process. 8 (8) (2000) 2389–2405.
[8.23] H. Bo¨ lcskei, Blind estimation of symbol timing and
carrier frequency offset in wireless OFDM systems,
IEEE Trans. Commun. 49 (6) (2001) 988–999.
[8.24] A. Napolitano, M. Tanda, Blind parameter estimation
in multipleaccess systems, IEEE Trans. Commun. 49
(4) (2001) 688–698.
[8.25] P. Ciblat, P. Loubaton, E. Serpedin, G.B. Giannakis,
Performance analysis of blind carrier frequency offset
estimators for noncircular transmissions through fre
quencyselective channels, IEEE Trans. Signal Process.
50 (1) (2002) 130–140.
[8.26] P. Ciblat, P. Loubaton, E. Serpedin, G.B. Giannakis,
Asymptotic analysis of blind cyclic correlationbased
symbolrate estimators, IEEE Trans. Inform. Theory 48
(7) (2002) 1922–1934.
[8.27] I. Lakkis, D. O’Shea, M.K. Tayebi, B. Hatim,
Performance analysis of a class of nondataaided
frequency offset and symbol timing estimators for ﬂat
fading channels, IEEE Trans. Signal Process. 50 (9)
(2002) 2295–2305.
[8.28] Y. Wang, P. Ciblat, E. Serpedin, P. Loubaton,
Performance analysis of a class of nondataaided
frequency offset and symbol timing estimators for ﬂat
fading channels, IEEE Trans. Signal Process. 50 (9)
(2002) 2295–2305.
[8.29] P. Ciblat, E. Serpedin, On a blind fractionally sampling
based carrier frequency offset estimator for noncircular
transmissions, IEEE Signal Process. Lett. 10 (4) (2003)
89–92.
[8.30] P. Ciblat, L. Vandendorpe, Blind carrier frequency
offset estimation for noncircular constellationbased
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 683
transmissions, IEEE Trans. Signal Process. 51 (5)
(2003) 1378–1389.
[8.31] Y. Wang, E. Serpedin, P. Ciblat, An alternative blind
feedforward symbol timing estimator using two samples
per symbol, IEEE Trans. Commun. 51 (9) (2003)
1451–1455.
[8.32] Y. Wang, E. Serpedin, P. Ciblat, Optimal blind carrier
recovery for MPSK burst transmissions, IEEE Trans.
Commun. 51 (9) (2003) 1571–1581.
[8.33] P. Ciblat, E. Serpedin, A ﬁne blind frequency offset
estimator for OFDM/OQAM systems, IEEE Trans.
Signal Process. 52 (1) (2004) 291–296.
[8.34] I. Lakkis, D. O’Shea, M.K. Tayebi, Nondata aided
maximum likelihood based feedforward timing syn
chronization method, US Patent No. 6,768,780, July 27,
2004.
[8.35] A. Napolitano, M. Tanda, Dopplerchannel blind
identiﬁcation for noncircular transmissions in multi
pleaccess systems, IEEE Trans. Commun. 52 (12)
(2004) 2172–2191.
[8.36] A. Napolitano, S. Ricciardelli, M. Tanda, Blind
estimation of amplitudes, time delays and frequency
offsets in multiple access systems with circular transmis
sions, Signal Processing 85 (8) (2005) 1588–1601.
[9.1] K.L. Jordan, Discrete representations of random
signals, Ph.D. Dissertation, Department of Electrical
Engineering, Massachusetts Institute of Technology,
Cambridge, MA, July 1961.
[9.2] Y.P. Dragan, Expansion of random processes and their
noncommutative transformations, Otbor Peredacha
Inform. 22 (1969) 22–27 (in Russian).
[9.3] F.L. Rocha, Adaptive deconvolution of cyclostationary
signals. I, Rev. Telegr. Electron. 68 (1980) 1026–1030
(in Spanish).
[9.4] E.R. Ferrara Jr., B. Widrow, The timesequenced
adaptive ﬁlter, IEEE Trans. Acoust. Speech Signal
Process. ASSP29 (1981) 679–683.
[9.5] M. Hajivandi, Tracking performance of stochastic
gradient algorithms for nonstationary processes, Ph.D.
Dissertation, Department of Electrical and Com
puter Engineering, University of California, Davis,
CA, 1982.
[9.6] C.A. French, W.A. Gardner, Despreading spread
spectrum signals without the code, IEEE Trans.
Commun. COM (1986) 404–407.
[9.7] B.G. Agee, S.V. Schell, W.A. Gardner, Selfcoherence
restoral: a new approach to blind adaptation of antenna
arrays, in: Proceedings of the 21st Annual Asilomar
Conference on Signals, Systems, and Computers, Paciﬁc
Grove, CA, 1987, pp. 589–593.
[9.8] W.A. Gardner, Nonstationary learning characteristics
of the LMS algorithm, IEEE Trans. Circuits Systems
CAS34 (1987) 1199–1207.
[9.9] B.G. Agee, S.V. Schell, W.A. Gardner, The SCORE
approach to blind adaptive signal extraction: an
application of the theory of spectral correlation, in:
Proceedings of the IEEE/ASSP Fourth Workshop on
Spectral Estimation and Modeling, Minneapolis, MN,
1988, pp. 277–282.
[9.10] W.A. Gardner, Simpliﬁcation of MUSIC and ESPRIT
by exploitation of cyclostationarity, Proc. IEEE 76 (7)
(1988) 845–847.
[9.11] W.A. Gardner, C.K. Chen, Selective source location by
exploitation of spectral coherence, in: Proceedings of
the Fourth Annual ASSP Workshop on Spectrum
Estimation and Modeling, Minneapolis, MN, 1988,
pp. 271–276.
[9.12] W.A. Gardner, C.K. Chen, Interferencetolerant time
differenceofarrival estimation for modulated signals,
IEEE Trans. Acoustics Speech Signal Process. ASSP36
(1988) 1385–1395.
[9.13] V.A. Omel’chenko, A.V. Omel’chenko, Y.P. Dragan,
O.A. Kolesnikov, .I Recognising Gaussian periodically
correlated random signals, Radiotekhnika 85 (1988)
75–79 (in Russian).
[9.14] B. Sankar, N. Tugbay, The use of the WignerVille
distribution (WVD) for the classiﬁcation and the
analysis of modulation techniques, in: Proceedings of
ISCIS III, Cesme, Turkey, 1988, pp. 75–82.
[9.15] S.V. Schell, B.G. Agee, Application of the SCORE
algorithm and SCORE extensions to sorting in the
rankL spectral selfcoherence environment, in: Con
ference Record, 22nd Asilomar Conference on Signals,
Systems and Computers, Paciﬁc Grove, CA, 1988, pp.
274–278.
[9.16] J.S. Friedman, J.P. King, J.P. Pride III, Time difference
of arrival geolocation method, etc., US Patent No.
4,888,593, December 19, 1989.
[9.17] S.V. Schell, R.A. Calabretta, W.A. Gardner, B.G.
Agee, Cyclic MUSIC algorithms for signalselective
direction estimation, in: Proceedings of the Interna
tional Conference on Acoustics, Speech, and Signal
Processing (ICASSP’89), Glasgow, UK, 1989, pp.
2278–2281.
[9.18] S.V. Schell, W.A. Gardner, Signalselective direction
ﬁnding for fully correlated signals, in: Abstracts of Sixth
Multidimensional Signal Processing Workshop of the
IEEE/ASSP Society, Paciﬁc Grove, CA, 1989, pp.
139–140.
[9.19] C.M. Spooner, C.K. Chen, W.A. Gardner, Maximum
likelihood twosensor detection and TDOA estimation
for cyclostationary signals, in: Abstracts of Sixth
Multidimensional Signal Processing Workshop of the
IEEE/ASSP Society, Paciﬁc Grove, CA, 1989, pp.
119–120.
[9.20] G. Xu, T. Kailath, Array signal processing via
exploitation of spectral correlation—a combination of
temporal and spatial processing, in: Conference Record,
23rd Asilomar Conference on Signals, Systems and
Computers, Paciﬁc Grove, CA, 1989, pp. 945–949.
[9.21] B.C. Agee, S.V. Schell, W.A. Gardner, Spectral self
coherence restoral: a new approach to blind adaptive
signal extraction using antenna arrays, Proc. IEEE 78
(1990) 753–767.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 684
[9.22] B.G. Agee, D. Young, Blind capture and geolocation of
stationary waveforms using multiplatform temporal
and spectral selfcoherence restoral, in: Proceedings of
24th Asilomar Conference on Signals, Systems and
Computers, Paciﬁc Grove, CA, 1990.
[9.23] H. Ding, M.M. Fahmy, Inverse of linear periodically
timevarying ﬁltering, in: Proceedings of International
Conference on Acoustic, Speech, and Signal Processing
(ICASSP’90), vol. 3, Albuquerque, NM, 1990,
pp. 1217–1220.
[9.24] M. Hajivandi, W.A. Gardner, Measures of tracking
performance for the LMS algorithm, IEEE Trans.
Acoust. Speech Signal Process. 38 (1990) 1953–1958.
[9.25] V. Joshi, D.D. Falconer, Sequence estimation techni
ques for digital subscriber loop transmission with
crosstalk interference, IEEE Trans. Commun. 38
(1990) 1367–1374.
[9.26] S.V. Schell, Exploitation of spectral correlation for
signalselective direction ﬁnding, Ph.D. Dissertation,
Department of Electrical Engineering and Computer
Science, University of California, Davis, CA, 1990.
[9.27] S.V. Schell, W.A. Gardner, Signalselective highresolu
tion direction ﬁnding in multipath, in: Proceedings of
International Conference on Acoustic, Speech, and
Signal Processing (ICASSP’90), Albuquerque, NM,
1990, pp. 2667–2670.
[9.28] S.V. Schell, W.A. Gardner, Progress on signalselective
direction ﬁnding, in: Proceedings of the Fifth IEEE/
ASSP Workshop on Spectrum Estimation and Model
ing, Rochester, NY, 1990, pp. 144–148.
[9.29] S.V. Schell, W.A. Gardner, Detection of the number of
cyclostationary signals in unknown interference and
noise, in: Proceedings of the 24th Annual Asilomar
Conference on Signals, Systems, and Computers, Paciﬁc
Grove, CA, 1990, pp. 473–477.
[9.30] S.V. Schell, W.A. Gardner, The Cramer–Rao lower
bound for parameters of Gaussian cyclostationary
signals, in: Proceedings of International Symposium
on Information Theory and its Applications (ISI
TA’90), Honolulu, HI, 1990, pp. 255–258.
[9.31] G. Xu, T. Kailath, A simple and effective algorithm for
estimating time delay of communication signals, in:
Proceedings of International Symposium on Informa
tion Theory and its Applications (ISITA’90), 1990,
pp. 267–270.
[9.32] T. Biedka, B.G. Agee, Subinterval cyclic MUSIC—
robust DF with inaccurate knowledge of cycle fre
quency, in: Proceedings of 25th Asilomar Conference
on Signals, Systems and Computers, Paciﬁc Grove, CA,
1991.
[9.33] G. Gelli, A. Napolitano, L. Paura, Spectralcorrelation
based estimation of channel parameters by noncoherent
processing, in: Proceedings of China 1991 International
Conference on Circuits and Systems, Sheuzhen, China,
1991, pp. 354–357.
[9.34] G. Gelli, L. Izzo, L. Paura, G. Poggi, A cyclic SVD
based algorithm for multiple source localization, in:
Treizie` me Colloque sur le Traitement du Signal et des
Images (GRETSI’91), Juan LesPins, France, 1991.
[9.35] D.C. McLernon, Analysis of LMS algorithm with
inputs from cyclostationary random processes, Elec
tron. Lett. 27 (1991) 136–138.
[9.36] W.A. Gardner, C.K. Chen, Signalselective time
differenceofarrival estimation for passive location of
manmade signal sources in highly corruptive environ
ments. Part I: Theory and method IEEE Trans. Signal
Process. 40 (1992) 1168–1184.
[9.37] C.K. Chen, W.A. Gardner, Signalselective time
differenceofarrival estimation for passive location of
manmade signal sources in highly corruptive environ
ments. Part II: Algorithms and performance, IEEE
Trans. Signal Process. 40 (1992) 1185–1197.
[9.38] W.A. Gardner, C.M. Spooner, Comparison of auto
correlation and crosscorrelation methods for signal
selective TDOA estimation, IEEE Trans. Signal Pro
cess. 40 (10) (1992) 2606–2608.
[9.39] L. Izzo, L. Paura, G. Poggi, An interferencetolerant
algorithm for localization of cyclostationarysignal
sources, IEEE Trans. Signal Process. 40 (1992) 1682–1686.
[9.40] S.V. Schell, W.A. Gardner, The Crame´ r–Rao lower
bound for parameters of Gaussian cyclostationary
signals, IEEE Trans. Inform. Theory 38 (4) (1992)
1418–1422.
[9.41] G. Xu, T. Kailath, Directionofarrival estimation via
exploitation of cyclostationarity—a combination of
temporal and spatial ﬁltering, IEEE Trans. Signal
Process. 40 (7) (1992) 1775–1785.
[9.42] T. Biedka, Subspaceconstrained SCORE algorithms,
in: Proceedings of 27th Asilomar Conference on
Signals, Systems and Computers, Paciﬁc Grove, CA,
1993, pp. 716–720.
[9.43] L. Castedo, C. Tseng, L. Grifﬁths, A new cost function
for adaptive beamforming using cyclostationary signal
properties, in: Proceedings of IEEE International
Conference on Acoustics, Speech, and Signal Processing
(ICASSP’93), vol. 4, 1993, pp. 284–287.
[9.44] L. Castedo, C. Tseng, Behavior of adaptive beamfor
mers based on cyclostationary signal properties in
multipath environments, in: Proceedings of 27th Asilo
mar Conference on Signals, Systems and Computers,
vol. 1, Paciﬁc Grove, CA, 1993, pp. 653–657.
[9.45] W.A. Gardner, B.G. Agee, Selfcoherence restoring
signal extraction apparatus and method, US Patent No.
5,255,210, October 19, 1993.
[9.46] W.A. Gardner, S.V. Schell, Method and apparatus for
multiplexing communications signals through blind
adaptive spatial ﬁltering, US Patent No. 5,260,968,
November 9, 1993.
[9.47] S.V. Schell, W.A. Gardner, High resolution direction
ﬁnding, in: N.K. Bose, C.R. Rao (Eds.), Handbook of
Statistics, Elsevier, Amsterdam, NL, 1993.
[9.48] S.V. Schell, W.A. Gardner, Blind adaptive spatio
temporal ﬁltering for wideband cyclostationary signals,
IEEE Trans. Signal Process. 41 (1993) 1961–1964.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 685
[9.49] T. Biedka, M. Kahn, Methods for constraining a CMA
beamformer to extract a cyclostationary signal, in:
Proceedings of Second Workshop on Cyclostationary
Signals, Yountville, CA, 1994.
[9.50] L. Castedo, A. FueirasVidal, C. Tseng, L. Grifﬁths,
Linearly constrained adaptive beamforming using
cyclostationary signal properties, in: Proceedings of
IEEE International Conference on Acoustics, Speech,
and Signal Processing (ICASSP’94), vol. 4, 1994,
pp. 249–252.
[9.51] L.E. Franks, Polyperiodic linear ﬁltering, in: W.A.
Gardner (Ed.), Cyclostationarity in Communications
and Signal Processing, IEEE Press, New York, 1994,
pp. 240–266 (Chapter 4).
[9.52] W.A. Gardner, S.V. Schell, B.G. Agee, Selfcoherence
restoring signal extraction and estimation of signal
direction of arrival, US Patent No. 5,299,148, March
29, 1994.
[9.53] G. Gelli, Sensor arraybased parameter estimation of
cyclostationary signals, Ph.D. Dissertation, Diparti
mento di Ingegneria Elettronica, Universita` di Napoli
Federico II, Napoli, Italy, February 1994 (in Italian).
[9.54] N.L. Gerr, C. Allen, Timedelay estimation for harmo
nizable signals, Digital Signal Process. 4 (1994) 49–62.
[9.55] G.B. Giannakis, G. Zhou, Parameter estimation of
cyclostationary AM time series with application to
missing observations, IEEE Trans. Signal Process. 42
(9) (1994) 2408–2419.
[9.56] L. Izzo, A. Napolitano, L. Paura, Modiﬁed cyclic
methods for signal selective TDOA estimation, IEEE
Trans. Signal Process. 42 (11) (1994) 3294–3298.
[9.57] M. Kahn, M. Mow, W.A. Gardner, T. Biedka, A
recursive programmable canonical correlation analyzer,
in: Proceedings of Second Workshop on Cyclostation
ary Signals, Yountville, CA, 1994.
[9.58] S.V. Schell, An overview of sensor array processing for
cyclostationary signals, in: W.A. Gardner (Ed.), Cy
clostationarity in Communications and Signal Proces
sing, IEEE Press, New York, 1994, pp. 168–239
(Chapter 3).
[9.59] S.V. Schell, W.A. Gardner, Spatiotemporal ﬁltering
and equalization for cyclostationary signals, in: C.T.
Leondes (Ed.), Control and Dynamic Systems, vol. 64,
Academic Press, New York, 1994 (Chapter 1).
[9.60] S.V. Schell, Performance analysis of the cyclic MUSIC
method of direction estimation for cyclostationary
signals, IEEE Trans. Signal Process. 42 (11) (1994)
3043–3050.
[9.61] S. Shamsunder, G.B. Giannakis, Signal selective loca
lization of nonGaussian cyclostationary sources, IEEE
Trans. Signal Process. 42 (10) (1994) 2860–2864.
[9.62] Q. Wu, K.M. Wong, R. Ho, Fast algorithm for
adaptive beamforming of cyclic signals, IEE Proc.
Radar Sonar Navig. 141 (6) (1994) 312–318.
[9.63] T. Biedka, A method for reducing computations in
cyclostationarityexploiting beamforming, in: Proceed
ings of IEEE International Conference on Acoustics,
Speech, and Signal Processing (ICASSP’95), 1995,
pp. 1828–1831.
[9.64] L. Castedo, A.R. FigueirasVidal, An adaptive beam
forming technique based on cyclostationary signal
properties, IEEE Trans. Signal Process. 43 (7) (1995)
1637–1650.
[9.65] G. Gelli, Power and timing parameter estimation of
multiple cyclostationary signals from sensor array data,
Signal Processing 42 (1) (1995) 97–102.
[9.66] G.B. Giannakis, G. Zhou, Harmonics in multiplicative
and additive noise: parameter estimation using cyclic
statistics, IEEE Trans. Signal Process. 43 (9) (1995)
2217–2221.
[9.67] S. Shamsunder, G.B. Giannakis, B. Friedlander,
Estimating random amplitude polynomial phase sig
nals: a cyclostationary approach, IEEE Trans. Signal
Process. 43 (2) (1995) 492–505.
[9.68] S.V. Schell, W.A. Gardner, Programmable canonical
correlation analysis: a ﬂexible framework for blind
adaptive spatial ﬁltering, IEEE Trans. Signal Process.
43 (12) (1995) 2898–2908.
[9.69] G. Zhou, G.B. Giannakis, Retrieval of selfcoupled
harmonics, IEEE Trans. Signal Process. 43 (5) (1995)
1173–1185.
[9.70] G. Zhou, G.B. Giannakis, Harmonics in multiplicative
and additive noise: performance analysis of cyclic
estimators, IEEE Trans. Signal Process. 43 (6) (1995)
1445–1460.
[9.71] G. Zhou, Random amplitude and polynomial phase
modeling of nonstationary processes using higherorder
and cyclic statistics, Ph.D. Dissertation, University of
Virginia, Charlottesville, VA, January 1995.
[9.72] F. Mazzenga, F. Vatalaro, Parameter estimation in
CDMA multiuser detection using cyclostationary sta
tistics, Electron. Lett. 32 (3) (1996).
[9.73] Q. Wu, M. Wong, Blind adaptive beamforming for
cyclostationary signals, IEEE Trans. Signal Process. 44
(11) (1996) 2757–2767.
[9.74] G. Gelli, L. Izzo, Minimumredundancy linear arrays
for cyclostationaritybased source location, IEEE
Trans. Signal Process. 45 (10) (1997) 2605–2608.
[9.75] Y. Hongyi, B. Zhang, Fully blind estimation of time
delays and spatial signatures for cyclostationary signals,
Electron. Lett. 34 (25) (1998) 2378–2380.
[9.76] M. Ghogo, A. Swami, A.K. Nandi, Nonlinear least
squares estimation for harmonics in multiplicative and
additive noise, Signal Processing 79 (2) (1999) 43–60.
[9.77] M. Ghogo, A. Swami, B. Garel, Performance analysis
of cyclic statistics for the estimation of harmonics in
multiplicative and additive noise, IEEE Trans. Signal
Process. 47 (1999) 3235–3249.
[9.78] F. Gini, G.B. Giannakis, Hybrid FMpolynomial phase
signal modeling: parameter estimation and Crame´ r–
Rao bounds, IEEE Trans. Signal Process. 47 (2) (1999)
363–377.
[9.79] J.H. Lee, Y.T. Lee, Robust adaptive array beamform
ing for cyclostationary signals under cycle frequency
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 686
error, IEEE Trans. Antennas Propagation 47 (2) (1999)
233–241.
[9.80] M. Martone, Adaptive multistage beamforming using
cyclic higher order statistics, IEEE Trans. Signal
Process. 47 (10) (1999) 2867–2873.
[9.81] M.C. Sullivan, Wireless geolocation system, US Patent
No. 5, 999, 131, December 7, 1999.
[9.82] J.H. Lee, Y.T. Lee, W.H. Shih, Efﬁcient robust
adaptive beamforming for cyclostationary signals,
IEEE Trans. Signal Process. 48 (7) (2000) 1893–1901.
[9.83] F. Mazzenga, Blind adaptive parameter estimation for
CDMA systems using cyclostationary statistics, Eur.
Trans. Telecommun. Relat. Technol. 11 (5) (2000)
495–500.
[9.84] D.A. Streight, G.K. Lott, W.A. Brown, Maximum
likelihood estimates of the time and frequency differ
ences of arrival of weak cyclostationary digital com
munication signals, in: Proceedings of the 21st Century
Military Communications Conference (MILCOM
2000), 2000, 957–961.
[9.85] J. Xin, H. Tsuji, A. Sano, Higherorder cyclostationar
ity based direction estimation of coherent narrowband
signals, IEICE Trans. Fundam. E 83A (8) (2000)
1624–1633.
[9.86] T. Biedka, Analysis and development of blind ad
aptive beamforming algorithms, Ph.D. Dissertation,
Bradley Department of Electrical Engineering, Virginia
Polytechnic Institute and State University, Virginia,
2001.
[9.87] J.H. Lee, Y.T. Lee, A novel direction ﬁnding method
for cyclostationary signals, Signal Processing 81 (6)
(2001) 1317–1323.
[9.88] Y.T. Lee, J.H. Lee, Robust adaptive array beamform
ing with random error in cycle frequency, IEE Proc.
Radar, Sonar Navig. 148 (4) (2001) 193–199.
[9.89] J. Xin, A. Sano, Linear prediction approach to
direction estimation of cyclostationary signals in multi
path environments, IEEE Trans. Signal Process. 49 (4)
(2001) 710–720.
[9.90] Z.T. Huang, Y.Y. Zhou, W.L. Jiang, Q.Z. Lu, Joint
estimation of Doppler and timedifferenceofarrival
exploiting cyclostationarity property, IEE Proc. (4)
(2002) 161–165.
[9.91] Z. Huang, Y. Zhou, W. Jiang, Directionofarrival of
signal sources exploiting cyclostationarity property,
Acta Electron. Sinica 30 (3) (2002) 372–375 (in
Chinese).
[9.92] J.H. Lee, C.H. Tung, Estimating the bearings of near
ﬁeld cyclostationary signals, IEEE Trans. Signal Pro
cess. 50 (1) (2002) 110–118.
[9.93] M.R. Morelande, A.M. Zoubir, On the performance of
cyclic momentbased parameter estimators of amplitude
modulated polynomial phase signals, IEEE Trans.
Signal Process. 50 (3) (2002) 590–605.
[9.94] P. Charge´ , Y. Wang, J. Sillard, An extended cyclic
MUSIC algorithm, IEEE Trans. Signal Process. 51 (7)
(2003) 1695–1701.
[9.95] G. Gelli, L. Izzo, Cyclostationaritybased coherent
methods for widebandsignal source location, IEEE
Trans. Signal Process. 51 (10) (2003) 2471–2482.
[9.96] K.V. Krikorian, R.A. Rosen, Technique for robust
characterization of weak RF emitters and accurate time
difference of arrival estimation for passive ranging of
RF emitters, US Patent No. 6, 646, 602, November 11,
2003.
[9.97] J.H. Lee, Y.T. Lee, Robust technique for estimating
the bearings of cyclostationary signals, Signal Proces
sing 83 (5) (2003) 1035–1046.
[9.98] P. Bianchi, P. Loubaton, F. Sirven, Non dataaided
estimation of the modulation index of continuous phase
modulations, IEEE Trans. Signal Process. 52 (10)
(2004) 2847–2861.
[9.99] G. Gelli, D. Mattera, L. Paura, Blind wideband spatio
temporal ﬁltering based on higherorder cyclostationar
ity, IEEE Trans. Signal Process. 53 (4) (2005)
1282–1290.
[9.100] H. Yan, H.H. Fan, Wideband cyclic MUSIC algorithm,
Signal Processing 85 (2005) 643–649.
[10.1] R. Cerrato, B.A. Eisenstein, Deconvolution of cyclosta
tionary signals, IEEE Trans. Acoust. Speech Signal
Process., ASSP25 (1977) 466–476.
[10.2] R.D. Gitlin, S.B. Weinstein, Fractionallyspaced equal
ization: an improved digital transversal equalizer, Bell.
Syst. Tech. J. 60 (1981) 275–296.
[10.3] J.W.M. Bergmans, P.J. van Gerwen, K.J. Wouda, Data
signal transmission system using decision feedback
equalization, US Patent No. 4,870,657, September 26,
1989.
[10.4] T.L. Archer, W.A. Gardner, New methods for identify
ing the Volterra kernels of a nonlinear system, in:
Proceedings of the 24th Annual Asilomar Conference
on Signals, Systems, and Computers, Paciﬁc Grove,
CA, 1990, pp. 592–597.
[10.5] W.A. Gardner, Identiﬁcation of systems with cyclosta
tionary input and correlated input/output measurement
noise, IEEE Trans. Automat. Control 35 (4) (1990)
449–452.
[10.6] R.D. McCallister, D. Ronald, D.D. Shearer III,
Compensating for distortion in a communication
channel, US Patent No. 4,922,506, May 1, 1990.
[10.7] W.A. Gardner, A new method of channel identiﬁcation,
IEEE Trans. Commun. 39 (1991) 813–817.
[10.8] W.A. Gardner, T.L. Archer, Simpliﬁed methods for
identifying the Volterra kernels of nonlinear systems, in:
Proceedings of the 34th Midwest Symposium on
Circuits and Systems, Monterey, CA, 1991.
[10.9] A. DuelHallen, Equalizers for multiple input/multiple
output channels and PAM systems with cyclostationary
input sequence, IEEE J. Sel. Areas Commun. 10 (3)
(1992) 630–639.
[10.10] G.D. Golden, Use of cyclostationary signal to con
strain the frequency response of a fractionally spa
ced equalizer, US Patent No. 5,095,495, March 10,
1992.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 687
[10.11] G.D. Golden, Use of a fractionally spaced equalizer to
perform echo cancellation in a fullduplex modem, US
Patent No. 5,163,044, November 10, 1992.
[10.12] L. Izzo, A. Napolitano, M. Tanda, Spectralcorrelation
based methods for multipath channel identiﬁcation,
Eur. Trans. Telecommun. Relat. Technol. 3 (1) (1992)
341–348.
[10.13] W.A. Gardner, T.L. Archer, Exploitation of cyclosta
tionarity for identifying the Volterra kernels of non
linear systems, IEEE Trans. Inform. Theory 39 (3)
(1993) 535–542.
[10.14] G. Gelli, L. Izzo, A. Napolitano, L. Paura, Multipath
channel identiﬁcation by an improved Prony algorithm
based on spectral correlation measurements, Signal
Processing 31 (1) (1993) 17–29.
[10.15] Y. Chen, C.L. Nikias, Identiﬁability of a band limited
system from its cyclostationary output autocorrelation,
IEEE Trans. Signal Process. 42 (2) (1994) 483–485.
[10.16] Z. Ding, Blind channel identiﬁcation and equalization
using spectral correlation measurements, Part I: Fre
quencydomain analysis, in: W.A. Gardner (Ed.),
Cyclostationarity in Communications and Signal Pro
cessing, IEEE Press, New York, 1994, pp. 417–436.
[10.17] D. Hatzinakos, Nonminimum phase channel deconvo
lution using the complex cepstrum of the cyclic
autocorrelation, IEEE Trans. Signal Process. 42 (11)
(1994) 3026–3042.
[10.18] L. Izzo, A. Napolitano, L. Paura, Cyclostationarity
exploiting methods for multipathchannel identiﬁca
tion, in: W.A. Gardner (Ed.), Cyclostationarity in
Communications and Signal Processing, IEEE Press,
New York, 1994, pp. 91–416.
[10.19] Y. Li, Z. Ding, ARMA system identiﬁcation based on
secondorder cyclostationarity, IEEE Trans. Signal
Process. 42 (12) (1994) 3483–3494.
[10.20] A. Napolitano, Spectralcorrelation based methods for
multipath channel identiﬁcation, Ph.D. Dissertation,
Universita` di Napoli Federico II, Napoli, Italy,
Dipartimento di Ingegneria Elettronica, February
1994 (in Italian).
[10.21] L. Tong, G. Xu, T. Kailath, Blind identiﬁcation and
equalization based on second order statistics: a time
domain approach, IEEE Trans. Inform. Theory 40 (2)
(1994) 340–349.
[10.22] L. Tong, G. Xu, T. Kailath, Blind channel identiﬁcation
and equalization using spectral correlation measure
ments, Part II: A timedomain approach, in: W.A.
Gardner (Ed.), Cyclostationarity in Communications
and Signal Processing, IEEE Press, New York, 1994,
pp. 437–454.
[10.23] W.A. Gardner, L. Paura, Identiﬁcation of polyperiodic
nonlinear systems, Signal Processing 46 (1995) 75–83
[10.24] G.B. Giannakis, Polyspectral and cyclostationary ap
proaches for identiﬁcation of closedloop systems,
IEEE Trans. Automat. Control 40 (5) (1995) 882–885
[10.25] L. Tong, G. Xu, B. Hassibi, T. Kailath, Blind channel
identiﬁcation based on second order statistics: a
frequency domain approach, IEEE Trans. Inform.
Theory 41 (1) (1995) 329–334.
[10.26] J.K. Tugnait, On blind identiﬁability of multipath
channels using fractional sampling and secondorder
cyclostationary statistics, IEEE Trans. Inform. Theory
41 (1) (1995) 308–311.
[10.27] C. Andrieu, P. Duvat, A. Doucet, Bayesian deconvolu
tion of cyclostationary processes based on point
processes, in: Proceedings of VIII European Signal
Processing Conference (EUSIPCO’96), Trieste, Italy,
1996.
[10.28] H. Liu, G. Xu, L. Tong, T. Kailath, Recent develop
ments in blind channel equalization: from cyclostatio
narity to subspaces, Signal Processing 50 (1996) 83–99.
[10.29] J.K. Tugnait, Blind equalization and estimation of FIR
communication channels using fractional sampling,
IEEE Trans. Commun. 44 (3) (1996) 324–336.
[10.30] H.E. Wong, J.A. Chambers, Twostage interference
immune blind equaliser which exploits cyclostationary
statistics, Electron. Lett. 32 (19) (1996) 1763–1764.
[10.31] A. Chevreuil, P. Loubaton, Blind secondorder identi
ﬁcation of FIR channels: forced cyclostationarity and
structured subspace methods, IEEE Signal Process.
Lett. 4 (1997) 204–206.
[10.32] A. Chevreuil, Blind equalization in a cyclostationary
context, Ph.D. Dissertation, ENSTTelecom Paris,
Department of Signal Processing, Paris, France, 1997
(in French).
[10.33] G.B. Giannakis, S.D. Halford, Asymptotically optimal
blind fractionally spaced channel estimation and
performance analysis, IEEE Trans. Signal Process. 45
(7) (1997) 1815–1830.
[10.34] Y.C. Liang, A.R. Leyman, B.H. Soong, (Almost)
periodic FIR system identiﬁcation using thirdorder
cyclicstatistics, Electron. Lett. 33 (5) (1997) 356–357.
[10.35] S. Prakriya, D. Hatzinakos, Blind identiﬁcation of
linear subsystems of LTIZMNLLTI models with
cyclostationary inputs, IEEE Trans. Signal Process. 45
(8) (1997) 2023–2036.
[10.36] M. Tsatsanis, G.B. Giannakis, Transmitter induced
cyclostationarity for blind channel equalization, IEEE
Trans. Signal Process. 45 (7) (1997) 1785–1794.
[10.37] H.H. Zeng, L. Tong, Blind channel estimation using the
secondorder statistics: algorithms, IEEE Trans. Signal
Process. 45 (8) (1997) 1919–1930.
[10.38] G.B. Giannakis, E. Serpedin, Blind identiﬁcation of
ARMA channels with periodically modulated inputs,
IEEE Trans. Signal Process. 46 (11) (1998) 3099–3104.
[10.39] F. Guglielmi, C. Luschi, A. Spalvieri, Fractionality
spaced equalizing circuits and method, US Patent No.
5,751,768, May 12, 1998.
[10.40] Y.C. Liang, A.R. Leyman, (Almost) periodic moving
average system identiﬁcation using higher order cylic
statistics, IEEE Trans. Signal Process. 46 (3) (1998)
779–783.
[10.41] D. Mattera, L. Paura, Higherorder cyclostationarity
based methods for identifying Volterra systems by
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 688
input–output noisy measurements, Signal Processing 67
(1) (1998) 77–98.
[10.42] F. Mazzenga, Blind multipath channel identiﬁcation for
wideband communication systems based on cyclosta
tionary statistics, Eur. Trans. Telecommun. Relat.
Technol. 9 (1) (1998) 27–31.
[10.43] E. Serpedin, G.B. Giannakis, Blind channel identiﬁca
tion and equalization with modulation induced cyclos
tationarity, IEEE Trans. Signal Process. 46 (7) (1998)
1930–1944.
[10.44] A. Chevreuil, P. Loubaton, MIMO blind second
order equalization method and conjugate cyclostatio
narity, IEEE Trans. Signal Process. 47 (2) (1999)
572–578.
[10.45] R.W. Heat, G.B. Giannakis, Exploiting input cyclosta
tionarity for blind channel identiﬁcation in OFDM
systems, IEEE Trans. Signal Process. 47 (3) (1999)
848–856.
[10.46] D. Mattera, Identiﬁcation of polyperiodic Volterra
systems by means of input–output noisy measurements,
Signal Processing 75 (1) (1999) 41–50.
[10.47] A. Chevreuil, E. Serpedin, P. Loubaton, G.B. Gianna
kis, Blind channel identiﬁcation and equalization using
nonredundant periodic modulation precoders: perfor
mance analysis, IEEE Trans. Signal Process. 48 (6)
(2000) 1570–1586.
[10.48] F. Mazzenga, On the identiﬁability of a channel transfer
function from cyclic and conjugate cyclic statistics,
European Trans. Telecommun. Relat. Technol. 11 (3)
(2000) 293–296.
[10.49] H. Bo¨ lcskei, P. Duhamel, R. Hleiss, A subspacebased
approach to blind channel identiﬁcation in pulse
shaping OFDM/OQAM systems, IEEE Trans. Signal
Process. 49 (7) (2001) 1594–1598.
[10.50] H. Bo¨ lcskei, R.W. Health, A.J. Paulraj, Blind channel
identiﬁcation and equalization in OFDMbased multi
antenna systems, IEEE Trans. Signal Process. 50 (1)
(2002) 96–109.
[10.51] G. Gelli, F. Verde, Twostage interferenceresistant
adaptive periodically timevarying CMA blind equal
ization, IEEE Trans. Signal Process. 50 (3) (2002)
662–672.
[10.52] J.K. Tugnait, W. Luo, Linear prediction error method
for blind identiﬁcation of periodically timevarying
channels, IEEE Trans. Signal Process. 50 (12) (2002)
3070–3082.
[10.53] I. Bradaric, A.P. Petropulu, K.I. Diamantaras, Blind
MIMO FIR channel identiﬁcation based on second
order spectra correlations, IEEE Trans. Signal Process.
51 (6) (2003) 1668–1674.
[10.54] S. Houcke, A. Chevreuil, P. Loubaton, Blind equaliza
tioncase of unknown symbol period, IEEE Trans.
Signal Process. 51 (3) (2003) 781–793.
[10.55] J. Antoni, P. Wagstaff, J.C. Henrio, H
a
—A consistent
estimator for frequency response function with input
and output noise, IEEE Trans. Instrumentat. Meas. 53
(2) (2004) 457–465.
[10.56] S. Celebi, Methods and devices for simplifying blind
channel estimation of the channel impulse response for a
DMT signal, US Patent No. 6,782,042, August 24, 2004.
[10.57] S.V. Narasimhan, M. Hazarataiah, P.V.S. Giridhar,
Channel blind identiﬁcation based on cyclostationarity
and group delay, Signal Processing 85 (7) (2005)
1275–1286.
[11.1] M.G. Bilyik, Periodic nonstationary processes, Otbor
Peredacha Inform. 50 (1977) 28–34 (in Russian).
[11.2] D. Bukofzer, Coherent and noncoherent detection of
signals in cyclostationary noise and of cyclostationary
signals in stationary noise, Ph.D. Dissertation, Depart
ment of Electrical and Computer Engineering, Uni
versity of California, Davis, 1979.
[11.3] Y.P. Dragan, V.P. Mezentsev, I.N. Yavorskii, The
problem of veriﬁcation of a stochastic rhythmicity
model, Otbor Peredacha Inform. 59 (1980) 3–12 (in
Russian).
[11.4] B.S. Rybakov, Optimum detection of a class of random
Gaussian pulsed signals in the presence of noise with a
priori parametric uncertainty, Radio Eng. Electron.
Phys. 25 (1980) 62–66 (English translation of Radio
tekhnika Electron. 25, 1198–1202).
[11.5] W.A. Gardner, Structural characterization of locally
optimum detectors in terms of locally optimum
estimators and correlators, IEEE Trans. Inform.
Theory IT28 (1982) 924–932.
[11.6] H.L. Hurd, Clipping degradation in a system used to
detect timeperiodic variance ﬂuctuations, J. Acoust.
Soc. Amer. 72 (1982) 1827–1830.
[11.7] W.A. Gardner, Nonparametric signal detection and
identiﬁcation of modulation type for signals hidden in
noise: an application of the new theory of cyclic spectral
analysis, in: Proceedings of Ninth Symposium on Signal
Processing and its Applications, 1983, pp. 119–124.
[11.8] D. Bukofzer, Optimum and suboptimum detector
performance for signals in cyclostationary noise, IEEE
J. Ocean. Eng. OE12 (1987) 97–115.
[11.9] W.A. Gardner, Signal interception: a unifying theore
tical framework for feature detection, in: Proceedings of
the 11th GRETSI Symposium on Signal and Image
Processing, Nice, France, 1987, pp. 69–72.
[11.10] W.A. Gardner, C.M. Spooner, Cyclic spectral analysis
for signal detection and modulation recognition, in:
Proceedings of MILCOM 88, San Diego, CA, 1988, pp.
419–424.
[11.11] W.A. Gardner, Signal interception: a unifying theore
tical framework for feature detection, IEEE Trans.
Commun. COM36 (1988) 897–906, awarded S.O. Rice
Prize Paper from the IEEE Communications Society.
[11.12] L. Izzo, L. Paura, M. Tanda, Optimum and subopti
mum detection of weak signals in cyclostationary non
Gaussian noise, Eur. Trans. Telecommun. Relat.
Technol. 1 (3) (1990) 233–238.
[11.13] N.L. Gerr, H.L. Hurd, Graphical methods for deter
mining the presence of periodic correlation, J. Time Ser.
Anal. 12 (1991) 337–350.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 689
[11.14] A.V. Vecchia, R. Ballerini, Testing for periodic auto
correlations in seasonal time series data, Biometrika 78
(1) (1991) 53–63.
[11.15] W.A. Gardner, C.M. Spooner, Signal interception:
performance advantages of cyclic feature detectors,
IEEE Trans. Commun. 40 (1) (1992) 149–159.
[11.16] L. Izzo, L. Paura, M. Tanda, Signal interception in non
Gaussian noise, IEEE Trans. Commun. 40 (1992)
1030–1037.
[11.17] A. Napolitano, L. Paura, M. Tanda, Signal detection in
cyclostationary generalized Gaussian noise with un
known parameters, Eur. Trans. Telecommun. Relat.
Technol. 3 (1) (1992) 39–43.
[11.18] M. Tanda, Cyclostationaritybased algorithms for
signal interception in nonGaussian noise, Ph.D. Dis
sertation, Dipartimento di Ingegneria Elettronica, Uni
versita` di Napoli Federico II, Napoli, Italy, February
1992 (in Italian).
[11.19] W.A. Gardner, C.M. Spooner, Detection and source
location of weak cyclostationary signals: simpliﬁcations
of the maximumlikelihood receiver, IEEE Trans.
Commun. 41 (6) (1993) 905–916.
[11.20] A.V. Dandawate´ , G.B. Giannakis, Statistical test for
presence of cyclostationarity, IEEE Trans. Signal
Process. 42 (9) (1994) 2355–2369.
[11.21] C.M. Spooner, W.A. Gardner, Robust feature detection
for cyclostationary signals, IEEE Trans. Commun. 42
(5) (1994) 2165–2173.
[11.22] R.W. Gardner, Detection of a digitally modulated
signal using binary line enhancement, US Patent No.
5,455,846, October 3, 1995.
[11.23] P. Gournay, P. Nicolas, Cyclic spectral analysis and
timefrequency analysis for automatic transmission
classiﬁcation, in: Proceedings of Quinzieme Colloque
GRETSI, JuanlesPins, F, 1995 (in French).
[11.24] L. Izzo, M. Tanda, Delayandmultiply detectors for
signal interception in nonGaussian noise, in: Proceed
ings of 15th GRTESI Symposium on Signal and Image
Processing, JuanLesPins, France, 1995.
[11.25] G. Gelli, L. Izzo, L. Paura, Cyclostationaritybased
signal detection and source location in nonGaussian
noise, IEEE Trans. Commun. 44 (3) (1996) 368–376.
[11.26] P. Rostaing, Detection of modulated signals by
exploiting their cyclostationarity properties: appli
cation to sonar signals, Ph.D. Dissertation, University
of NiceSophia Antipolis, Nice, France, 1996 (in
French).
[11.27] G.K. Yeung, W.A. Gardner, Searchefﬁcient methods
of detection of cyclostationary signals, IEEE Trans.
Signal Process. 44 (5) (1996) 1214–1223.
[11.28] D. Boiteau, C. Le Martret, A general maximum
likelihood framework for modulation classiﬁcation, in:
Proceedings of International Conference on Acoustics,
Speech, and Signal Processing (ICASSP’98), vol. 4,
1998, pp. 2165–2168.
[11.29] P. Marchand, Detection and classiﬁcation of digital
modulations by higher order cyclic statistics, Ph.D.
Dissertation, INPGNat. Poly. Inst. Grenoble, Greno
ble, France, 1998 (in French).
[11.30] P. Marchand, J.L. Lacoume, C. Le Martret, Multiple
hypothesis modulation classiﬁcation based on cyclic
cumulants of different orders, in: Proceedings of Inter
national Conference on Acoustics, Speech, and Signal
Processing (ICASSP’98), vol. 4, 1998, pp. 2157–2160.
[11.31] O. Besson, P. Stoica, Simple test for distinguishing
constant from time varying amplitude in harmonic
retrieval problem, IEEE Trans. Signal Process. 47 (4)
(1999) 1137–1141.
[11.32] P. Rostaing, E. Thierry, T. Pitarque, Asymptotic
performance analysis of cyclic detectors, IEEE Trans.
Commun. 47 (1) (1999) 10–13.
[11.33] N.S. Subotic, B.J. Thelen, D.A. Carrara, Cyclostation
ary signal models for the detection and characterization
of vibrating objects in SAR data, in: Proceedings of the
32nd Asilomar Conference on Signals, Systems, and
Computers, Paciﬁc Grove, CA, 1998, pp. 1304–1308.
[11.34] A. Ferre´ ol, P. Chevalier, On the behavior of current
second and higher order blind source separation
methods for cyclostationary sources, IEEE Trans.
Signal Process. 48 (6) (2000) 1712–1725.
[11.35] K. AbedMeraim, Y. Xiang, J.H. Manton, Y. Hua,
Blind source separation using secondorder cyclosta
tionary statistics, IEEE Trans. Signal Process. 49 (4)
(2001) 694–701.
[11.36] A. Ferreol, Process of cyclic detection in diversity of
polarization of digital cyclostationary radioelectric
signals, US Patent No. 6,430,239, August 6, 2002.
[11.37] A. Ferre´ ol, P. Chevalier, L. Albera, Secondorder blind
separation of ﬁrst and secondorder cyclostationary
sources—application to AM, and deterministic sources,
IEEE Trans. Signal Process. 52 (4) (2004) 845–861.
[11.38] J.A. Sills, Q.R. Black, Signal recognizer for commu
nications signals, US Patent No. 6,690,746, February
10, 2004.
[11.39] J. Antoni, F. Guillet, M. El Badaoui, F. Bonnardot,
Blind separation of convolved cyclostationary pro
cesses, Signal Processing 85 (2005) 51–66.
[12.1] W.M. Brelsford, Probability predictions and time series
with periodic structure, Ph.D. Dissertation, Johns
Hopkins University, Baltimore, MD, 1967.
[12.2] C.W.J. Granger, Some new time series models: non
linear, bilinear and nonstationary, Statistician 27 (3–4)
(1978) 237–254.
[12.3] M. Pagano, On periodic and multiple autoregressions,
Ann. Statist. 6 (1978) 1310–1317.
[12.4] B.M. Troutman, Some results in periodic autoregres
sion, Biometrika 66 (1979) 219–228.
[12.5] J.D. Salas, R.A. Smith, Correlation properties of
periodic AR(p) models, in: Proceedings of the Third
International Symposium on Stochastic Hydraulics,
Tokyo, 1980,107–115
[12.6] G.C. Tiao, M.R. Grupe, Hidden periodic autoregres
sivemoving average models in time series data,
Biometrika 67 (1980) 365–373.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 690
[12.7] S. Shell, K.K. Biowas, A.K. Sihka, Modelling and
predictions of stochastic processes involving periodi
cities, Appl. Math. Modeling 5 (1981) 241–245.
[12.8] H. Sakai, Circular lattice ﬁltering using Pagano’s
method, IEEE Trans. Acoust. Speech Signal Process.
ASSP (1982) 279–287.
[12.9] J.D. Salas, D.C. Boes, R.A. Smith, Estimation of
ARMA models with seasonal parameters, Water
Resour. Res. 18 (1982) 1006–1010.
[12.10] H. Sakai, Covariance matrices characterization by a set
of scalar partial autocorrelation coefﬁcients, Ann.
Statist. 11 (1983) 337–340.
[12.11] A.V. Vecchia, Aggregation and estimation for periodic
autoregressive moving average models, Ph.D. Disserta
tion, Colorado State University, Fort Collins, CO,
1983.
[12.12] J. Andel, Periodic autoregression with exogenous
variables and equal variances, in: W. Grossman, et al.
(Eds.), Probability Theory and Mathematical Statistics
with Applications, Reidel, Dordrecht, NL, Boston,
MA, 1985, pp. 237–245.
[12.13] J. Andel, On periodic autoregression with unknown
mean, Appl. Math. 30 (1985) 126–139.
[12.14] T. Cipra, Periodic moving average process, Appl. Math.
30 (1985) 218–229.
[12.15] R.M. Thompstone, K.W. Hipel, A.I. McLeod, Group
ing of periodic autoregressive models, in: O.D. Ander
son, J.K. Ord, E.A. Robinson (Eds.), Time Series
Analysis, vol. 6, Elsevier, Amsterdam, Holland, 1985,
pp. 35–49.
[12.16] A.V. Vecchia, Periodic autoregressivemoving average
(PARMA) modeling with applications to water re
sources, Water Resources Bull. 21 (1985) 721–730.
[12.17] A.V. Vecchia, Maximum likelihood estimation for
periodic autoregressive moving average models, Tech
nometrics 27 (1985) 375–384.
[12.18] J. Andel, A. Rubio, On interpolation on periodic
autoregressive processes, Appl. Math. 31 (1986)
480–485.
[12.19] S. Bittanti, The periodic prediction problem for
cyclostationary processes—an introduction, in: Pro
ceedings of the NATO Advanced Research Workshop,
Groningen, 1986, 239–249.
[12.20] B. Fernandez, J.D. Salas, Periodic gamma autoregres
sive processes for operational hydrology, Water Re
sour. Res. 22 (1986) 1385–1396.
[12.21] J.T.B. Obeysekera, J.D. Salas, Modeling of aggregated
hydrologic time series, J. Hydrol. 86 (1986) 197–219.
[12.22] T. Cipra, P. Tusty, Estimation in multiple autoregres
sivemoving average models using periodicity, J. Time
Ser. Anal. 8 (1987) 293–300.
[12.23] A.I. McLeod, D.J. Noakes, K.W. Hipel, R.M. Thomp
stone, Combining hydrologic forecasts, J. Water
Resour. Plan. Manage. 113 (1987) 29–41.
[12.24] P. Bartolini, J.D. Salas, J.T.B. Obeysekera, Multi
variate periodic ARMA(1,1) processes, Water Resour.
Res. 24 (1988) 1237–1246.
[12.25] W.K. Li, Y.V. Hui, An algorithm for the exact
likelihood of periodic autoregressive moving average
models, Commun. Statist. Simulat. Comput. 17 (1988)
1483–1494.
[12.26] J. Andel, Periodic autoregression with exogenous
variables and periodic variances, Cesk. Akod. Ved.
Aplikace Motemotiky 34 (1989) 387–395.
[12.27] H. Sakai, F. Sakaguchi, Simultaneous conﬁdence bands
for the spectral estimate of twochannel autoregressive
processes, J. Time Ser. Anal. 11 (1990) 49–56.
[12.28] T.A. Ula, Periodic covariance stationarity of multi
variate periodic autoregressive moving average pro
cesses, Water Resour. Res. 26 (1990) 855–861.
[12.29] S. Bittanti, P. Bolzern, G. De Nicolao, L. Piroddi, D.
Purassanta, A minimum prediction error algorithm for
estimation of periodic ARMA models, in: Proceedings
of the European Control Conference, Grenoble,
France, 1991
[12.30] H. Sakai, On the spectral density matrix of a periodic
ARMA process, J. Time Ser. Anal. 12 (1991) 73–82.
[12.31] P.L. Anderson, A.V. Vecchia, Asymptotic results for
periodic autoregressive movingaverage processes, J.
Time Ser. Anal. 14 (1993) 1–18.
[12.32] M. Bentarzi, M. Hallin, On the invertibility of periodic
movingaverage models, J. Time Ser. Anal. 15 (1993)
263–268.
[12.33] A.I. McLeod, Parsimony, model adequacy and periodic
correlation in time series forecasting, Int. Statist. Rev.
61 (1993) 387–393.
[12.34] A.G. Miamee, Explicit formulas for the best linear
predictor and predictor error matrix of a periodically
correlated process, SIAM J. Math. Anal. 24 (1993)
703–711.
[12.35] T.A. Ula, Forecasting of multivariate periodic auto
regressive movingaverage processes, J. Time Ser. Anal.
14 (6) (1993).
[12.36] S. Bittanti, P. Bolzern, G. De Nicolao, L. Piroddi,
Representation, in: W.A. Gardner (Ed.), predic
tion, IEEE Press, New York, 1994, pp. 267–294
(Chapter 5).
[12.37] A.G. Miamee, On recent developments in prediction
theory for cyclostationary processes, in: W.A. Gardner
(Ed.), Cyclostationarity in Communications and Signal
Processing, IEEE Press, New York, 1994, pp. 480–493.
[12.38] G.J. Adams, G.C. Goodwin, Parameter estimation for
periodic ARMA models, J. Time Ser. Anal. 16 (1995)
127–146.
[12.39] G.N. Boshnakov, Recursive computation of the para
meters of periodic autoregressive moving average
processes, J. Time Ser. Anal. 17 (1995) 333–349.
[12.40] M. Bentarzi, M. Hallin, Locally optimal tests against
periodical autocorrelation: parametric and nonpara
metric approaches, Econometric Theory 12 (1996)
88–112.
[12.41] A. Dandawate´ , G.B. Giannakis, Modeling periodic
moving average processes using cyclic statistics, IEEE
Trans. Signal Process. 44 (3) (1996) 673–684.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 691
[12.42] P.L. Anderson, M.M. Meerschaert, Periodic moving
averages of random variables with regularly varying
tails, Ann. Statist. 25 (1997) 771–785.
[12.43] R. Lund, I. Basawa, Modeling and inference for
periodically correlated time series, in: S. Gosh (Ed.),
Asymptotics, Nonparametrics and Time Series, 1997
[12.44] M. Bentarzi, M. Hallin, J. Appl. Probab. 35 (1) (1998)
46–54.
[12.45] P.L. Anderson, M.M. Meerschaert, A. Vecchia, In
novations algorithm for periodically stationary time
series, Stochastic Process. Appl. 83 (1999) 149–169.
[12.46] A. Makagon, Theoretical prediction of periodically
correlated sequences, Probab. Math. Statist. 19 (2)
(1999) 287–322.
[12.47] S. LambertLacroix, On periodic autoregressive pro
cesses estimation, IEEE Trans. Signal Process. 48 (6)
(2000) 1800–1803.
[12.48] R. Lund, I.V. Basawa, Recursive prediction and like
lihood evaluation for periodic ARMA models, J. Time
Ser. Anal. 21 (1) (2000) 75–93.
[12.49] I.V. Basawa, R. Lund, Large sample properties of
parameter estimates for periodic ARMA models, J.
Time Ser. Anal. 22 (6) (2001) 651–664.
[12.50] H.L. Hurd, A. Makagon, A.G. Miamee, On AR(1)
models with periodic and almost periodic coefﬁcients,
Stochastic Process. Appl. 100 (2002) 167–185.
[12.51] A. Bibi, C. Franq, Consistent and asymptotically
Normal estimators for cyclically timedependent linear
models, Ann. Inst. Statist. Math. 55 (1) (2003) 41–68.
[12.52] Q. Shao, R. Lund, Computation and characterization
of autocorrelations and partial autocorrelations in
periodic ARMA models, J. Time Ser. Anal. 25 (3)
(2004) 359–372.
[12.53] P.L. Anderson, M.M. Meerschaert, Parameter estima
tion for periodically stationary time series, J. Time Ser.
Anal. 26 (4) (2005) 489–518.
[13.1] V.O. Alekseev, Symmetry properties of highorder
spectral densities of stationary and periodicnonsta
tionary stochastic processes, Problems Inform. Trans
mission 23 (1987) 210–215.
[13.2] D.E. Reed, M.A. Wickert, Nonstationary moments of a
random binary pulse train, IEEE Trans. Inform.
Theory 35 (1989) 700–703.
[13.3] A.V.T. Cartaxo, A.A.D. Albuquerque, A general
property of the norder moment generating function
of strictsense cyclostationary processes, in: E. Arikan
(Ed.), Proceedings of the 1990 Bilkent International
Conference on New Trends in Communication, Control
and Signal Processing, vol. 2, Ankara, Turkey, Elsevier
Science, Amsterdam, Netherlands, 1990, pp. 1548–1552
[13.4] W.A. Gardner, Spectral characterization of nth order
cyclostationarity, in: Proceedings of the Fifth IEEE/
ASSP Workshop on Spectrum Estimation and Model
ing, Rochester, NY, 1990, pp. 251–255
[13.5] W.A. Gardner, C.M. Spooner, Higher order cyclosta
tionarity, cyclic cumulants, and cyclic polyspectra, in:
Proceedings of the 1990 International Symposium on
Information Theory and its Applications (ISITA’90),
Honolulu, HI, 1990, pp. 355–358
[13.6] C.M. Spooner, W.A. Gardner, An overview of higher
order cyclostationarity, in: A.G. Miamee (Ed.), Pro
ceeding of the Workshop on Nonstationary Stochastic
Processes and their Applications, World Scientiﬁc,
Virginia, VA, 1991, pp. 110–125.
[13.7] C.M. Spooner, W.A. Gardner, Estimation of cyclic
polyspectra, in: Proceedings of 25th Annual Asilomar
Conference on Signals, Systems, and Computers, Paciﬁc
Grove, CA, 1991
[13.8] G.D. Z
˘
ivanovic´ , Some aspects of the higherorder
cyclostationary theory, in: Proceedings of the Interna
tional Conference on Acoustics, Speech, and Signal
Processing (ICASSP’91), 1991
[13.9] C.M. Spooner, Theory and application of higherorder
cyclostationarity, Ph.D. Dissertation, Department of
Electrical and Computer Engineering, University of
California, Davis, CA, 1992.
[13.10] A.V. Dandawate´ , Exploiting cyclostationarity and
higherorder statistics in signal processing, Ph.D.
Dissertation, University of Virginia, Charlottesville,
VA, January 1993
[13.11] A. Dandawate´ , G.B. Giannakis, Nonparametric cyclic
polyspectral analysis of AM signals and processes with
missing observations, IEEE Trans. Inform. Theory 39
(6) (1993) 1864–1876.
[13.12] A.V. Dandawate´ , G.B. Giannakis, Nonparametric
polyspectral estimators for kthorder (almost) cyclosta
tionary processes, IEEE Trans. Inform. Theory 40 (1)
(1994) 67–84.
[13.13] W.A. Gardner, C.M. Spooner, The cumulant theory of
cyclostationary timeseries, Part I: Foundation, IEEE
Trans. Signal Process. 42 (12) (1994) 3387–3408.
[13.14] C.M. Spooner, W.A. Gardner, The cumulant theory of
cyclostationary timeseries, Part II: Development and
applications, IEEE Trans. Signal Process. 42 (12) (1994)
3409–3429.
[13.15] W.A. Gardner, C.M. Spooner, Cyclostationary signal
processing, in: C.T. Leondes (Ed.), Control and
Dynamic Systems, vol. 65, Academic Press, New York,
1994 (Chapter 1).
[13.16] L. Izzo, A. Napolitano, L. Paura, MIMO Volterra system
input/output relations for cyclic higherorder statistics, in:
Proceedings of VII European Signal Processing Confer
ence (EUSIPCO’94), Edinburgh, UK, 1994
[13.17] C.M. Spooner, Higherorder statistics for nonlinear
processing of cyclostationary signals, in: W.A. Gardner
(Ed.), Cyclostationarity in Communications and Signal
Processing, IEEE Press, New York, 1994, pp. 91–167
(Chapter 2).
[13.18] A.V. Dandawate´ , G.B. Giannakis, Asymptotic theory
of mixed time averages and Kthorder cyclic moment
and cumulant statistics, IEEE Trans. Inform. Theory 41
(1) (1995) 216–232.
[13.19] P. Marchand, P.O. Amblard, J.L. Lacoume,
Higher than secondorder statistics for complexvalued
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 692
cyclostationary signals, in: Proceedings of 15th GRET
SI Symposium, JuanLesPins, France, 1995, pp. 69–72
[13.20] A. Napolitano, Cyclic higherorder statistics: input/
output relations for discrete and continuoustime
MIMO linear almostperiodically timevariant systems,
Signal Processing 42 (2) (1995) 147–166, received
EURASIP’s Best Paper of the Year award
[13.21] S.V. Schell, Higherorder cyclostationarity properties of
coded communication signals, in: Proceedings of the
29th Annual Asilomar Conference on Signals, Systems,
and Computers, Paciﬁc Grove, CA, 1995
[13.22] C.M. Spooner, Classiﬁcation of cochannel com
munication signals using cyclic cumulants, in: Proceed
ings of the 29th Annual Asilomar Conference on
Signals, Systems, and Computers, Paciﬁc Grove, CA,
1995
[13.23] L. Izzo, A. Napolitano, Higherorder cyclostationarity
properties of sampled timeseries, Signal Processing 54
(1996) 303–307.
[13.24] G. Zhou, G.B. Giannakis, Polyspectral analysis of
mixed processes and coupled harmonics, IEEE Trans.
Inform. Theory 42 (3) (1996) 943–958.
[13.25] L. Izzo, A. Napolitano, Higherorder statistics for
Rice’s representation of cyclostationary signals, Signal
Processing 56 (1997) 279–292.
[13.26] J.L. Lacoume, P.O. Amblard, P. Common, Higher
Order Statistics for Signal Processing, Masson, Paris,
1997 (in French).
[13.27] L. Izzo, A. Napolitano, Multirate processing of time
series exhibiting higherorder cyclostationarity, IEEE
Trans. Signal Process. 46 (2) (1998) 429–439.
[13.28] F. Flagiello, L. Izzo, A. Napolitano, A computationally
efﬁcient and interference tolerant nonparametric algo
rithm for LTI system identiﬁcation based on higher
order cyclic statistics, IEEE Trans. Signal Process. 48
(4) (2000) 1040–1052.
[13.29] A. Napolitano, C.M. Spooner, Medianbased cyclic
polyspectrum estimation, IEEE Trans. Signal Process.
48 (2000) 1462–1466.
[13.30] C.M. Spooner, W.A. Brown, G.K. Yeung, Automatic
radiofrequency environment analysis, in: Proceedings
of the 34th Annual Asilomar Conference on Signals,
Systems, and Computers, Paciﬁc Grove, CA, 2000
[13.31] C.M. Spooner, On the utility of sixthorder cyclic
cumulants for RF signal classiﬁcation, in: Proceedings
of the 35th Annual Asilomar Conference on Signals,
Systems, and Computers, Paciﬁc Grove, CA, 2001
[14.1] A.N. Bruyevich, Fluctuations in autooscillators for
periodically nonstationary shot noise, Radio Eng. 23
(1968) 91–96.
[14.2] V.I. Voloshin, Frequency ﬂuctuations in isochronous
selfoscillators at periodically nonstationary random
actions, Radiophys. Quantum Electron.
[14.3] E.N. Rozenvasser, Periodically Nonstationary Control
Systems, Izdat. Nauka, Moskow, 1973 (in Russian).
[14.4] Y.P. Dragan, I.N. Yavorskii, Hydroacoustic commu
nication channels with surface scattering represented in
terms of linear timevarying ﬁlters, Otbor Peredacha
Inform. 39 (1974) 23–31 (in Russian).
[14.5] J. Rootenberg, S.A. Ghozati, Stability properties of
periodic ﬁlters, Int. J. Systems Sci. 8 (1977) 953–959.
[14.6] T. Strom, S. Signell, Analysis of periodically switched
linear circuits, IEEE Trans. Circuits Systems CAS
(1977) 531–541.
[14.7] A.M. Silaiev, A.V. Yakimov, Periodic nonstationarity
of noise in harmonic systems, Radiotekhnika Elektron.
24 (1979) 1806–1811 (in Russian).
[14.8] V.A. Zaytsev, A.V. Pechinkin, Optimal control of
servicing in a multichannel system, Eng. Cybernet. 17
(1979) 36–42
[14.9] L. Pelkowitz, Frequency domain analysis of wrap
around error in fast convolution algorithms, IEEE
Trans. Acoust. Speech Signal Process. ASSP (1981)
413–422.
[14.10] S. Bittanti, G. Guardabassi, Optimal cyclostationary
control: a parameteroptimization frequencydomain
approach, in: H. Akashi (Ed.), Proceedings of the
Eighth Triennial World Congress of the International
Federation of Automatic Control, vol. 2, Kyoto, Japan,
1982, pp. 857–862
[14.11] Y. Isokawa, An identiﬁcation problem in almost and
asymptotically almost periodically correlated processes,
J. Appl. Probab. 19 (1982) 456–462.
[14.12] K. Vokurka, Application of the group pulse processes
in the theory of Barkhausen noise, Czech. J. Phys. B 32
(1982) 1384–1398.
[14.13] V.A. Semenov, A.I. Smirnov, On the stability of linear
stochastic systems with periodically nonstationary
parametric excitation, Mech. Solids 18 (1983) 14–20.
[14.14] E.R. Ferrara Jr., Frequencydomain implementations
of periodically timevarying ﬁlters, IEEE Trans.
Acoust. Speech Signal Process. ASSP (1985) 883–892.
[14.15] G.V. Obrezkov, Y.A. Sizyakov, Approximate analysis
of pulsed tracking systems when there is ﬂuctuating
interference, Radioelectron. Commun. Systems 28
(1985) 63–65 (English translation of Izv. VUZ Radio
electron.).
[14.16] D. Williamson, K. Kadiman, Cyclostationarity in the
digital regulation of continuous time systems, in: C.I.
Byrnes, A. Lindquist (Eds.), Frequency Domain and
State Space Methods for Linear Systems, NorthHol
land, Amsterdam, 1985, pp. 297–309.
[14.17] Y.P. Dragan, B.I. Yavorskii, A model of the test signal
noise in channels with digital signal processing, Radio
electron. Commun. Systems 29 (1986) 17–21.
[14.18] K. Kadiman, D. Williamson, Discrete minimax linear
quadratic regulation of continuoustime systems, Auto
matica 23 (1987) 741–747.
[14.19] V.G. Leonov, L.Y. Mogilevskaya, Y.L. Khotuntsev,
A.V. Frolov, Correlation relations between the com
plex amplitudes of the noise components in bipolar
and ﬁeldeffect transistors in the presence of strong
signals, Sov. J. Commun. Technol. Electron. 32 (1987)
126–130.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 693
[14.20] J. Goette, C. Gobet, Exact noise analysis of SC circuits
and an approximate computer implementation, IEEE
Trans. Circuits Systems 36 (1989) 508–521.
[14.21] S. Bittanti, D.B. Hernandez, G. Zerbi, The simple
pendulum and the periodic LQG control problem, J.
Franklin Inst. 328 (1991) 299–315.
[14.22] D. Williamson, Digital Control and Implementation,
PrenticeHall, Inc., Englewood Cliffs, NJ, 1991.
[14.23] S. Bittanti, G. De Nicolao, Spectral factorization of
linear periodic systems with application to the optimal
prediction of periodic ARMA models, Automatica
(1992).
[14.24] M. Okumura, H. Tanimoto, T. Itakura, T. Sugawara,
Numerical noise analysis for nonlinear circuits with a
periodic large signal excitation including cyclostation
ary noise sources, IEEE Trans. Circuits Systems (1993)
581–590.
[14.25] H.L. Hurd, C.H. Jones, Dynamical systems with
cyclostationary orbits, in: R. Katz (Ed.), The Chaos
Paradigm, AIP Press, New York, 1994.
[14.26] J. Roychowdhury, D. Long, P. Feldmann, Cyclosta
tionary noise analysis of large RF circuits with multi
tone excitations, IEEE J. Solid (3) (1998) 324–336.
[14.27] J. Roychowdhury, P. Feldmann, D.E. Long, Method
of making an integrated circuit including noise model
ing and prediction, US Patent No. 6, 072, 947, June 6,
2000.
[14.28] F. Bonani, S. Donati, G. Ghione, Noise source
modeling for cyclostationary noise analysis in large
signal device operation, IEEE Trans. Electron. Devices
49 (9) (2002) 1640–1647.
[14.29] K.K. Gullapalli, Method and apparatus for analyzing
small signal response and noise in nonlinear circuits, US
Patent No. 6,536,026, March 18, 2003.
[14.30] P. Shiktorov, E. Starikov, V. Gruz˘ inskis, L. Reggiani,
L. Varani, J.C. Vaissie` re, S.P.T. Gonza´ les, Monte Carlo
simulation in electronic noise in semiconductor materi
als and devices operating under cyclostationary condi
tions, J. Comput. Electron. 2 (2003) 455–458.
[14.31] J.S. Roychowdhury, Apparatus and method for re
ducedorder modeling of timevarying systems and
computer storage medium containing the same, US
Patent No. 6,687,658, February 3, 2004.
[15.1] S. Braun, B. Seth, Analysis of repetitive mechanism
signature, J. Sound Vib. 70 (1980) 513–526.
[15.2] S. Yamaguchi, Y. Kato, A practical method of
predicting noise produced by road trafﬁc controlled
by trafﬁc signals, J. Acoust. Soc. Amer. 86 (1989)
2206–2214.
[15.3] Y. Kato, S. Yamaguchi, A prediction method for
probability distribution of road trafﬁc noise at an
intersection, Acoust. Australia 18 (1990) 46–50.
[15.4] D. Koenig, J. Boehme, Application of cyclostationarity
and timefrequency analysis to engine car diagnostics,
in: Proceedings of International Conference on Acous
tics, Speech, and Signal Processing (ICASSP), Adelaide,
Australia, 1994, 149–152.
[15.5] A.C. McCormick, A.K. Nandi, Cyclostationarity in
rotating machinery vibrations, Mech. Systems Signal
Process. 12 (2) (1998) 225–242.
[15.6] T.R. Black, K.D. Donohue, Pitch determination of
music signals using the generalized spectrum, in:
Proceedings of IEEE SouthEast Conference ’00,
Nashville, TN, 2000, pp. 104–109.
[15.7] T.R. Black, K.D. Donohue, Frequency correlation
analysis for periodic echoes, in: Proceedings of IEEE
SouthEast Conference ’00, Nashville, TN, 2000,
pp. 131–138.
[15.8] C. Capdessus, M. Sidahmed, J.L. Lacoume, Cyclosta
tionary processes: application in gear faults early
diagnosis, Mech. Systems Signal Process. 14 (3) (2000)
371–385.
[15.9] G. Dalpiaz, A. Rivola, R. Rubini, Effectiveness and
sensitivity of vibration processing techniques for local
fault detection in gears, Mech. Systems Signal Process.
14 (3) (2000) 387–412.
[15.10] I. Antoniadis, G. Glossiotis, Cyclostationary analysis of
rollingelement bearing vibration signals, J. Sound Vib.
248 (5) (2001) 829–845.
[15.11] L. Bouillaut, M. Sidahmed, Cyclostationary approach
and bilinear approach: comparison, applications to
early diagnosis for helicopter gearbox and classiﬁcation
method based on HOCS, Mech. Systems Signal Process.
15 (5) (2001) 923–943.
[15.12] G.F.P. DusserreTelmon, D. Flores, F. Prieux, Damage
detection of motor pieces, European Patent No.
1111364, June 27, 2001.
[15.13] R.B. Randall, J. Antoni, S. Chobsaard, The relation
ship between spectral correlation and envelope analysis
in the diagnostics of bearing faults and other cyclosta
tionary machine signals, Mech. Systems Signal Process.
15 (5) (2001) 945–962.
[15.14] J. Antoni, J. Daniere, F. Guillet, Effective vibration
analysis of IC engines using cyclostationarity, Part I—A
methodology for condition monitoring, J. Sound Vib.
257 (5) (2002) 815–837.
[15.15] J. Antoni, J. Daniere, F. Guillet, R.B. Randall,
Effective vibration analysis of IC engines using cyclos
tationarity, Part II —New results on the reconstruction
of the cylinder pressure, J. Sound Vib. 257 (5) (2002)
839–856.
[15.16] J. Antoni, R.B. Randall, Differential diagnosis of gear
and bearing faults, ASME J. Vib. Acoust. 124 (2) (2002)
165–171.
[15.17] G.F.P. DusserreTelmon, D. Flores, F. Prieux, Process
for the detection of damage to components of an
engine, US Patent No. 6, 389, 887, May 21, 2002.
[15.18] A. Raad, J. Antoni, M. Sidahmed, Thirdorder cyclic
characterization of vibration signals in rotating
machinery, in: Proceedings of XI European Signal
Processing Conference (EUSIPCO’02), Tolouse,
France, 2002.
[15.19] J. Antoni, R.B. Randall, A stochastic model for
simulation and diagnostics of rolling element bearings
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 694
with localized faults, ASME J. Vib. Acoust. 125 (3)
(2003) 282–289.
[15.20] L. Li, L. Qu, Cyclic statistics in rolling bearing
diagnosis, J. Sound Vib. 267 (2003) 253–265.
[15.21] A. Raad, Contributions to the higherorder cyclic
statistics: applications to gear faults, Ph.D. Disserta
tion, Universite´ de Technologie Compie` gne, Com
pie` gne, France, November 2003 (in French).
[15.22] S. Tardu, Characterization of unsteady time periodical
turbulent ﬂows, Comptes Rendus Mecanique 331
(2003) 767–774 (in French).
[15.23] J. Antoni, F. Bonnardot, A. Raad, M. El Badaoui,
Cyclostationary modelling of rotating machine vibra
tion signal, Mech. Systems Signal Process. 18 (6) (2004)
1285–1314.
[15.24] Z.K. Zhu, Z.H. Feng, F.R. Kong, Cyclostationarity
analysis for gearbox condition monitoring: approaches
and effectiveness, Mech. Systems Signal Process. 19
(2005) 467–482.
[16.1] E. Parzen, M. Pagano, An approach to modeling
seasonally stationary timeseries, J. Econometrics 9
(1979) 137–153.
[16.2] D.R. Osborn, J.P. Smith, The performance of periodic
autoregressive models in forecasting seasonal UK
consumption, J. Bus. Econom. Statist. 7 (1989)
117–127 (see also J. Appl. Econometrics 3 (1989) 255
266).
[16.3] R.M. Todd, Periodic linearquadratic models of sea
sonality, J. Econom. Dyn. Control 14 (1990) 763–796.
[16.4] D.R. Osborn, The implications of periodically varying
coefﬁcients for seasonal timeseries processes, J. Econo
metrics 48 (1991) 373–384.
[16.5] P.H. Franses, R. Paap, Model selection in periodic
autoregressions, Oxford Bull. Econom. Statist. 56
(1994) 421–439.
[16.6] H.P. Boswijk, P.H. Franses, Testing for periodic
integration, Econom. Lett. 48 (1995) 241–248.
[16.7] P.H. Franses, The effects of seasonally adjusting a
periodic autoregressive process, Comput. Statist. Data
Anal. 19 (1995) 683–704.
[16.8] T. Bollerslev, E. Ghysels, Periodic autoregressive
conditional heteroscedasticity, Amer. Statist. Assoc. J.
Bus. Econom. Statist. 14 (2) (1996) 139–151.
[16.9] D. Dehay, J. Les´ kow, Testing stationarity for stock
market data, Econom. Lett. 50 (1996) 205–212.
[16.10] E. Ghysels, A. Hall, H.S. Lee, On periodic structures
and testing for seasonal unit roots, J. Amer. Statist.
Assoc. 91 (1996) 1551–1559.
[16.11] P.H. Franses, M. McAleer, Cointegration analysis of
seasonal time series, J. Econom. Survey 12 (1998)
651–678.
[16.12] E. Ghysels, D.R. Osborne, The Econometric Analysis
of Seasonal Time Series, Themes in Modern Econo
metrics, Cambridge University Press, Cambridge, UK,
2001.
[16.13] E. BroszkiewiczSuwaj, A. Makagon, R. Weron, A.
Wy"oman´ ska, On detecting and modeling periodic
correlation in ﬁnancial data, Physica A Statist. Mech.
Appl. 336 (1–2) (2004) 196–205.
[17.1] C.J. Finelli, J.M. Jenkins, A cyclostationary least mean
squares algorithm for discrimination of ventricular
tachycardia from sinus rhythm, in: Proceedings of
the Annual International Conference of the IEEE
Engineering in Medicine and Biology Society, vol. 13,
1991.
[17.2] K.D. Donohue, J. Bressler, T. Varghese, N.M. Bilgu
tay, Spectral correlation in ultrasonic pulseecho signal
processing, IEEE Trans. Ultrason. Ferroelectrics Fre
quency Control 40 (3) (1993) 330–337.
[17.3] K.D. Donohue, T. Varghese, N.M. Bilgutay, Spectral
redundancy in characterizing scatterer structures from
ultrasonic echoes, in: Review of Progress in Quantita
tive NonDestructive Evaluation, vol. 13, Brunswick,
Maine, 1993, pp. 951–958.
[17.4] M. Savic, Detection of cholestrol deposits in arteries,
US Patent No. 5,327,893, July 12, 1994.
[17.5] T. Varghese, K.D. Donohue, Characterization of tissue
microstructure with spectral crosscorrelation, Ultra
sound Imaging 15 (1994) 238–254.
[17.6] T. Varghese, K.D. Donohue, Meanscatter spacing
estimates with spectral correlation, J. Acoust. Soc.
Amer. 96 (6) (1994) 3504–3515.
[17.7] K.D. Donohue, H.Y. Cheah, Spectral correlation ﬁlters
for ﬂaw detection, in: Proceedings of 1995 Ultrasonics
Symposium, Seattle, WA, 1995, pp. 725–728.
[17.8] T. Varghese, K.D. Donohue, Estimating mean scatterer
spacing with frequencysmoothed spectral autocorrela
tion function, IEEE Trans. Ultrason. Ferroelectrics
Frequency Control 42 (3) (1995) 451–463.
[17.9] T. Varghese, K.D. Donohue, J.P. Chatterjee, Specular
echo imaging with spectral correlation, in: Proceedings
of 1995 Ultrasonics Symposium, Seattle, WA, 1995, pp.
1315–1318.
[17.10] T. Varghese, K.D. Donohue, Spectral redundancy in
tissue characterization, in: Twentieth International
Symposium on Ultrasonic Imaging and Tissue Char
acterization, 1995.
[17.11] K.D. Donohue, T. Varghese, F. Forsberg, E.J. Halpern,
The analysis and classiﬁcation of smallscale tissue
structures using the generalized spectrum, in: Proceed
ings of the International Congress on Acoustics and
Acoustical Society of America, Seattle WA, 1998,
pp. 2685–2686.
[17.12] K.D. Donohue, F. Forsberg, C.W. Piccoli, B.B. Gold
berg, Analysis and classiﬁcation of tissue with scatterer
structure templates, IEEE Trans. Ultrason. Ferro
electrics Frequency Control 46 (2) (1999) 300–310.
[17.13] K.D. Donohue, L. Haung, V. Genis, F. Foresberg,
Duct size estimation in breast tissue, in: Proceedings
of 1999 Ultrasonics Symposium, Reno, NV, 1999,
pp. 1353–1356.
[17.14] L. Huang, K.D. Donohue, V. Genis, F. Forsberg, Duct
detection and wall spacing estimation in breast tissue,
Ultrason. Imaging 22 (3) (2000) 137–152.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 695
[17.15] K.D. Donohue, L. Huang, T. Burks, F. Forsberg, C.W.
Piccoli, Tissue classiﬁcation with generalized spectrum
parameters, Ultrasound Med. Biol. 27 (11) (2001)
1505–1514.
[17.16] K.D. Donohue, L. Huang, Ultrasonic scatterer struc
ture classiﬁcation with the generalized spectrum, in:
Proceedings of the IEEE International Conference on
Acoustics, Speech and Signal Processing, Salt Lake
City, UT, 2001.
[17.17] K.D. Donohue, L. Huang, G. Georgiou, F.S. Cohen,
C.W. Piccoli, F. Forsberg, Malignant and benign breast
tissue classiﬁcation performance using a scatterer
structure preclassiﬁer, IEEE Trans. Ultrason. Ferro
electrics Frequency Control 50 (6) (2003) 724–729.
[17.18] S. Gefen, O.J. Tretiak, C. Piccoli, K.D. Donohue, A.P.
Petropulu, P.M. Shankar, V.A. Dumane, L. Huang,
M.A. Kutay, V. Genis, F. Forsberg, J.M. Reid, B.B.
Goldberg, ROC analysis of classiﬁers based on ultra
sonic tissue characterization features, IEEE Trans.
Med. Imaging 22 (2) (2003) 170–177.
[18.1] V.A. Markelov, On extrusions and relative time of
staying of the periodically nonstationary random
process, Izv. VUZ Radiotekhnika 9 (4) (1996) (in
Russian).
[18.2] V.A. Markelov, Axis crossings and relative time of
existence of a periodically nonstationary random
process, Sov. Radiophys. 9 (1966) 440–443 (in Russian).
[18.3] Y.P. Dragan, Properties of counts of periodically
correlated random processes, Otbor Peredacha Inform.
33 (1972) 9–12.
[18.4] M.G. Bilyik, Upcrossings of periodically nonstationary
processes, Otbor Peredacha Inform. 36 (1973) 28–31 (in
Russian).
[18.5] M.G. Bilyik, On the theory of upcrossings of
periodically nonstationary processes, Otbor Peredacha
Inform. 38 (1974) 15–21 (in Russian).
[18.6] M.G. Bilyik, Models and level crossings of periodically
nonstationary processes, Otbor Peredacha Inform. 39
(1974) 3–7 (in Russian).
[18.7] M.G. Bilyik, Some properties of periodically nonsta
tionary processes and inhomogeneous ﬁelds, Otbor
Peredacha Inform. 48 (1976) 27–33 (in Russian).
[18.8] M.G. Bilyik, Statistical properties of level crossings of a
periodic nonstationary process and a periodic inhomo
geneous ﬁeld, Otbor Peredacha Inform. 52 (1977) 3–8
(in Russian).
[18.9] M.G. Bilyik, High crossings of periodically nonsta
tionary Gaussian process, Otbor Peredacha Inform. 59
(1980) 24–31 (in Russian).
[18.10] D.E.K. Martin, Detection of periodic autocorrelation
in time series data via zero crossing, J. Time Ser. Anal.
20 (4) (1999) 435–452.
[19.1] R.R. Anderson, G.T. Foschini, B. Gopinath, A queuing
model for a hybrid data multiplexer, Bell System Tech.
J. 58 (1979) 279–300.
[19.2] L. Clare, I. Rubin, Queueing analysis of TDMA with
limited and unlimited buffer capacity, in: Proceedings of
IEEE INFOCOM’83, San Diego, CA, 1983,
pp. 229–238.
[19.3] M. Kaplan, A singleserver queue with cyclostationary
arrivals and arithmetic service, Oper. Res. 31 (1983)
184–205.
[19.4] M.H. Ackroyd, Stationary and cyclostationary ﬁnite
buffer behaviour computation via Levinson’s method,
AT (1984) 2159–2170.
[19.5] T. Nakatsuka, Periodic property of streetcar congestion
at the ﬁrst station, J. Oper. Res. Soc. Japan 29 (1986)
1–20.
[19.6] D. Habibi, D.J.H. Lewis, Fourier analysis for modelling
some cyclic behaviour of networks, Comput. Commun.
19 (1996) 426–434.
[20.1] Y.P. Dragan, I.N. Yavorskii, The periodic correla
tionrandom ﬁeld as a model for bidimensional ocean
waves, Otbor Peredacha Inform. 51 (1982) 15–21 (in
Russian).
[20.2] V.G. Alekseev, On spectral density estimates of
Gaussian periodically correlated random ﬁelds, Probab.
Math. Statist. 11 (2) (1991) 157–167.
[20.3] H.L. Hurd, Spectral correlation of randomly jittered
periodic functions of two variables, in: Proceedings of
the 29th Annual Asilomar Conference on Signals,
Systems, and Computers, Paciﬁc Grove, CA, 1995.
[20.4] W. Chen, G.B. Giannakis, N. Nandhakumar, Spatio
temporal approach for timevarying image motion
estimation, IEEE Trans. Image Process. 10 (1996)
1448–1461.
[20.5] H. Li, Q. Cheng, B. Yuan, Strong laws of large numbers
for two dimensional processes, in: Proceedings of
Fourth International Conference on Signal Processing
(ICSP’98), 1998.
[20.6] D. Dehay, H. Hurd, Spectral estimation for strongly
periodically correlated random ﬁelds deﬁned on R
2
,
Math. Methods Statist. 11 (2) (2002) 135–151.
[20.7] H. Hurd, G. Kallianpur, J. Farshidi, Correlation and
spectral theory for periodically correlated random ﬁelds
indexed on Z
2
, J. Multivariate Anal. 90 (2) (2004)
359–383.
[21.1] W.A. Gardner, Correlation estimation and timeseries
modeling for nonstationary processes, Signal Processing
15 (1988) 31–41.
[21.2] J.C. Hardin, A.G. Miamee, Correlation autoregressive
processes with application to helicopter noise, J. Sound
Vib. 142 (1990) 191–202.
[21.3] A.G. Miamee, J.C. Hardin, On a class of nonstationary
stochastic processes, Sankhya: Indian J. Statist. Ser. A
(Part 2) 52 (1990) 145–156.
[21.4] J.C. Hardin, A. Makagon, A.G. Miamee, Correlation
autoregressive sequences: a summary, in: A.G. Miamee
(Ed.), Proceeding of the Workshop on Nonstationary
Stochastic Processes and their Applications, World
Scientiﬁc, Virginia, VA, 1991, pp. 165–175.
[21.5] W.A. Gardner, On the spectral coherence of nonsta
tionary processes, IEEE Trans. Signal Process. 39
(1991) 424–430.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 696
[21.6] J. Allen, S. Hobbs, Detecting target motion by
frequencyplane smoothing, in: Proceedings of the
26th Asilomar Conference on Systems and Computers,
Paciﬁc Grove, CA, 1992, pp. 1042–1047.
[21.7] L. Izzo, A. Napolitano, Timefrequency representations
of generalized almostcyclostationary processes, in:
Proceedings of 16th GRETSI Symposium on Signal
and Image Processing, 1997.
[21.8] A. Makagon, A.G. Miamee, On the spectrum of
correlation autoregressive sequences, Stochastic Pro
cess. Appl. 69 (1997) 179–193.
[21.9] L. Izzo, A. Napolitano, The higherorder theory of
generalized almostcyclostationary timeseries, IEEE
Trans. Signal Process. 46 (11) (1998) 2975–2989.
[21.10] L. Izzo, A. Napolitano, Linear timevariant transfor
mations of generalized almostcyclostationary signals,
Part I: Theory and method, IEEE Trans. Signal
Process. 50 (12) (2002) 2947–2961.
[21.11] L. Izzo, A. Napolitano, Linear timevariant transfor
mations of generalized almostcyclostationary signals,
Part II: Development and applications, IEEE Trans.
Signal Process. 50 (12) (2002) 2962–2975.
[21.12] K.S. Lii, M. Rosenblatt, Spectral analysis for
harmonizable processes, Ann. Statist. 30 (1) (2002)
258–297.
[21.13] L. Izzo, A. Napolitano, Sampling of generalized almost
cyclostationary signals, IEEE Trans. Signal Process. 51
(6) (2003) 1546–1556.
[21.14] A. Napolitano, Uncertainty in measurements on
spectrally correlated stochastic processes, IEEE Trans.
Inform. Theory 49 (9) (2003) 2172–2191.
[21.15] A. Napolitano, Meansquare consistency of statistical
function estimators for generalized almostcyclostation
ary processes, in: Proceedings of XII European Signal
Processing Conference (EUSIPCO 2004), Vienna,
Austria, 2004.
[21.16] L. Izzo, A. Napolitano, Generalized almostcyclosta
tionary signals, in: P.W. Hawkes (Ed.), Advances in
Imaging and Electron Physics, vol. 135, Elsevier,
Amsterdam, 2005, pp. 103–223 (Chapter 3).
[21.17] A. Napolitano, Asymptotic Normality of statistical
function estimators for generalized almostcyclostation
ary processes, in: Proceedings of XIII European Signal
Processing Conference (EUSIPCO 2005), Antalya,
Turkey, 2005.
[22.1] O.I. Koronkevich, Sci. Ann. L’vovskii Inst. 44 (5) (1957).
[22.2] R.L. Stratonovich, Topics in the Theory of Random
Noise, vols. I and II, Gordon and Breach, New York,
1967 (Revised English edition translated by R.A.
Silverman).
[22.3] K.S. Voychishin, Y.P. Dragan, et al., Coll. Algorithmi
zation of Industrial Processes, Cybernetics Institute,
AN USSR, 1972 (in Russian).
[22.4] V. Zalud, V.N. Kulesov, LowNoise Semiconductor
Circuits, SNTL, Prague, 1980, Section 1.18 (in Czech).
[22.5] B.N. Gartshtein, V.V. Smelyakov, Assessing character
istics of periodic random signals, Radiotekhnika 60
(1982) 116–123 (in Russian).
[22.6] P. Kochel, A dynamic multilocation supply model with
redistribution between the stores, Math. Oper. Stat. Ser.
Optim. 13 (1982) 267–286 (in German).
[22.7] G.G. Kuznetsov, P.A. Chernenko, Mathematical model
and device for diagnostics of periodically nonstationary
random processes, Elektron. Model. 6 (1984) 45–48 (in
Russian).
[22.8] A.J.E.M. Janssen, The Zak transform: a signal trans
form for sampled timecontinuous signals, Philips J.
Res. 43 (1988) 23–69.
Further references
[23.1] N. Wiener, Generalized harmonic analysis, Acta Math.
55 (1930) 117–258.
[23.2] A.S. Besicovitch, Almost Periodic Functions, Cam
bridge University Press, London, 1932 (also New York:
Dover, 1954).
[23.3] H. Bohr, Almost Periodic Functions, Springer, Berlin,
1933 (also Chelsea, New York, 1974).
[23.4] N. Wiener, The Fourier Integral and Certain of its
Applications, Cambridge University Press, London,
1933 (also New York: Dover, 1958).
[23.5] M. Loe` ve, Probability Theory, Van Nostrand, Prince
ton, NJ, 1963.
[23.6] L. Amerio, Prouse, AlmostPeriodic Functions and
Functional Equations, Van Nostrand, New York, 1971.
[23.7] J. Bass, Cours de Mathe´ matiques, vol. III, Masson &
Cie, Paris, 1971.
[23.8] R.A. Meyer, C.S. Burrus, A uniﬁed analysis of multi
rate and periodically timevarying digital ﬁlters, IEEE
Trans. Circuits Systems CAS (1975) 162–168.
[23.9] E. Pfaffelhuber, Generalized harmonic analysis for
distributions, IEEE Trans. Inform. Theory IT21
(1975) 605–611.
[23.10] A.H. Zemanian, Distribution Theory and Transform
Analysis, Dover, New York, 1987.
[23.11] C. Corduneanu, Almost Periodic Functions, Chelsea,
New York, 1989.
[23.12] T. Chen, L. Qiu, Linear periodically timevarying
discretetime systems: aliasing and LTI approximations,
Systems Control Lett. 30 (1997) 225–235.
[23.13] A.S. Mehr, T. Chen, On aliascomponent matrices of
discretetime periodically timevarying systems, IEEE
Signal Process. Lett. 8 (2001) 114–116.
[23.14] A.S. Mehr, T. Chen, Representation of linear periodi
cally timevarying multirate systems, IEEE Trans.
Signal Process. 50 (9) (2002) 2221–2229.
[23.15] J. Leskow, A. Napolitano, Foundations of the non
stochastic approach for time series analysis, Signal
Process., submitted.
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 697
ARTICLE IN PRESS
640 W.A. Gardner et al. / Signal Processing 86 (2006) 639–697
4.
5.
6. 7.
8. 9. 10.
11. 12. 13.
14. 15. 16. 17. 18. 19. 20. 21.
Time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1. Continuoustime time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2. Discretetime time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4. Link between the stochastic and fractionoftime approaches . . . . . . . . . . . 3.5. Complex processes and time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6. Linear ﬁltering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6.1. Structure of linear almostperiodically timevariant systems . . . . . . 3.6.2. Input/output relations in terms of cyclic statistics . . . . . . . . . . . . . 3.7. Product modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.8. Supports of cyclic spectra of band limited signals . . . . . . . . . . . . . . . . . . . 3.9. Sampling and aliasing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.10. Representations by stationary components . . . . . . . . . . . . . . . . . . . . . . . . 3.10.1. Continuoustime processes and time series . . . . . . . . . . . . . . . . . . 3.10.2. Discretetime processes and time series. . . . . . . . . . . . . . . . . . . . . Ergodic properties and measurement of characteristics . . . . . . . . . . . . . . . . . . . . 4.1. Estimation of the cyclic autocorrelation function and the cyclic spectrum . . 4.2. Two alternative approaches to the analysis of measurements on time series . Manufactured signals: modelling and analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1. General aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2. Examples of communication signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1. Double sideband amplitudemodulated signal . . . . . . . . . . . . . . . 5.2.2. Pulseamplitudemodulated signal . . . . . . . . . . . . . . . . . . . . . . . . 5.2.3. Directsequence spreadspectrum signal . . . . . . . . . . . . . . . . . . . . Natural signals: modelling and analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Communications systems: analysis and design . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1. General aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2. Cyclic Wiener ﬁltering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Synchronization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.1. Spectral line generation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Signal parameter and waveform estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . Channel identiﬁcation and equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.1. General aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.2. LTIsystem identiﬁcation with noisymeasurements . . . . . . . . . . . . . . . . . . 10.3. Blind LTIsystem identiﬁcation and equalization . . . . . . . . . . . . . . . . . . . . 10.4. Nonlinearsystem identiﬁcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Signal detection and classiﬁcation, and source separation . . . . . . . . . . . . . . . . . . Periodic AR and ARMA modelling and prediction. . . . . . . . . . . . . . . . . . . . . . . Higherorder statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.2. Higherorder cyclic statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.2.1. Continuoustime time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.2.2. Discretetime time series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Applications to circuits, systems, and control . . . . . . . . . . . . . . . . . . . . . . . . . . . Applications to acoustics and mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Applications to econometrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Applications to biology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Level crossings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Queueing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cyclostationary random ﬁelds. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Generalizations of cyclostationarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.1. General aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
646 646 648 649 649 650 650 650 651 652 653 654 654 655 656 656 657 658 658 659 659 659 660 661 661 661 661 663 663 663 664 664 664 665 665 665 666 666 666 667 667 669 670 670 671 671 671 671 671 671 671
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 641
21.2. Generalized almostcyclostationary signals 21.3. Spectrally correlated signals . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Further references . . . . . . . . . . . . . . . . . . . . . . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
672 672 673 697
1. Introduction Many processes encountered in nature arise from periodic phenomena. These processes, although not periodic functions of time, give rise to random data whose statistical characteristics vary periodically with time and are called cyclostationary processes [2.5]. For example, in telecommunications, telemetry, radar, and sonar applications, periodicity is due to modulation, sampling, multiplexing, and coding operations. In mechanics it is due, for example, to gear rotation. In radio astronomy, periodicity results from revolution and rotation of planets and on pulsation of stars. In econometrics, it is due to seasonality; and in atmospheric science it is due to rotation and revolution of the earth. The relevance of the theory of cyclostationarity to all these ﬁelds of study and more was ﬁrst proposed in [2.5]. Widesense cyclostationary stochastic processes have autocorrelation functions that vary periodically with time. This function, under mild regularity conditions, can be expanded in a Fourier series whose coefﬁcients, referred to as cyclic autocorrelation functions, depend on the lag parameter; the frequencies, called cycle frequencies, are all multiples of the reciprocal of the period of cyclostationarity [2.5]. Cyclostationary processes have also been referred to as periodically correlated processes [2.18,3.5]. More generally, if the frequencies of the (generalized) Fourier series expansion of the autocorrelation function are not commensurate, that is, if the autocorrelation function is an almostperiodic function of time, then the process is said to be almostcyclostationary [3.26] or, equivalently, almostperiodically correlated [3.5]. The almostperiodicity property of the autocorrelation function is manifested in the frequency domain as correlation among the spectral components of the process that are separated by amounts equal to the cycle frequencies. In contrast to this, widesense stationary processes have
autocorrelation functions that are independent of time, depending on only the lag parameter, and all distinct spectral components are uncorrelated. As an alternative, the presence of periodicity in the underlying datagenerating mechanism of a phenomenon can be described without modelling the available data as a sample path of a stochastic process but, rather, by modelling it as a single function of time [4.31]. Within this nonstochastic framework, a timeseries is said to exhibit secondorder cyclostationarity (in the wide sense), as ﬁrst deﬁned in [2.8], if there exists a stable quadratic timeinvariant transformation of the timeseries that gives rise to ﬁnitestrength additive sinewave components. In this paper, a concise survey of the literature (in all languages in which a substantial amount of research has been published) on cyclostationarity is presented and includes an extensive bibliography and list of issued patents. Citations are classiﬁed into 22 categories and listed, for each category, in chronological order. In Section 2, general treatments and tutorials on the theory of cyclostationarity are cited. General properties of processes and timeseries are presented in Section 3. In Section 4, the problem of estimating statistical functions is addressed. Models for manufactured and natural signals are considered in Sections 5 and 6, respectively. Communications systems and related problems are treated in Sections 7–11. Speciﬁcally, the analysis and design of communications systems is addressed in Section 7, the problem of synchronization is addressed in Section 8, the estimation of signal parameters and waveforms is addressed in Section 9, the identiﬁcation and equalization of channels is addressed in Section 10, and the signal detection and classiﬁcation problems and the problem of source separation are addressed in Section 11. Periodic autoregressive (AR) and autoregressive movingaverage (ARMA) modelling and prediction are treated in Section 12. In Section 13, theory and
ARTICLE IN PRESS
642 W.A. Gardner et al. / Signal Processing 86 (2006) 639–697
applications of higherorder cyclostationarity are presented. We address applications to circuits, systems, and control in Section 14, to acoustics and mechanics in Section 15, to econometrics in Section 16, and to biology in Section 17. Applications to the problems of level crossing and queueing are addressed in Sections 18 and 19, respectively. Cyclostationary random ﬁelds are treated in Section 20. In Section 21, some classes of nonstationary signals that extend the class of almostcyclostationary signals are considered. Finally, some miscellaneous references are listed [22.1–22.8]. Further references only indirectly related to cyclostationarity are [23.1–23.15]. To assist readers in ‘‘going to the source’’, seminal contributions—if known— are identiﬁed within the literature published in English. In some cases, identiﬁed sources may have been preceded in the literature of another language, most likely Russian. For the most part, the subject of this survey developed independently in the literature published in English.
[13.4,13.5], and treated in depth in [13.9,13.13, 13.14,13.17]. An analogous treatment for stochastic processes is given in [13.10].
3. General properties and structure of stochastic processes and timeseries 3.1. Introduction General properties of cyclostationary processes (see [3.1–3.91]) are derived in terms of the Fourier series expansion of the autocorrelation function. The frequencies, called cycle frequencies, are multiples of the reciprocal of the period of cyclostationarity and the coefﬁcients, referred to as cyclic autocorrelation functions, are continuous functions of the lag parameter. Cyclostationary processes are characterized in the frequency domain by the cyclic spectra, which are the Fourier transforms of the cyclic autocorrelation functions. The cyclic spectrum at a given cycle frequency represents the density of correlation between two spectral components of the process that are separated by an amount equal to the cycle frequency. Almostcyclostationary processes have autocorrelation functions that can be expressed in a (generalized) Fourier series whose frequencies are possibly incommensurate. The ﬁrst contributions to the analysis of the general properties of cyclostationary stochastic processes are in the Russian literature [3.1–3.3, 3.5,3.7–3.9]. The problem of spectral analysis is mainly treated in [3.10,3.11,3.32,3.35,3.51,3.53, 3.61,3.68,3.69,3.80,3.88]. The stationarizing effects of random shifts are examined in [3.20,3.26,3.57] and the important impact of random shifts on cycloergodic properties is exposed in [2.15]. The problem of ﬁltering is considered in [3.9,3.13,3.67]. The mathematical link between the stochastic and nonstochastic approaches, which was ﬁrst established in [2.5,2.12], is rigorously treated in [3.87]. Wavelet analysis of cyclostationary processes is addressed in [3.70,3.81]. For furtherrelated references, see the general treatments in [2.1,2.2,2.4–2.8,2.11–2.13,2.15,2.18], and also [4.13,4.31,4.41,13.2,13.3,13.6,16.3,18.3, 21.9,21.14].
2. General treatments General treatments on cyclostationarity are in [2.1–2.18]. The ﬁrst extensive treatments of the theory of cyclostationary processes can be found in the pioneering works of Hurd [2.1] and Gardner [2.2]. In [4.13,2.5,2.11], the theory of secondorder cyclostationary processes is developed mainly with reference to continuoustime stochastic processes, but discretetime is the focus in [4.13]. Discretetime processes are treated more generally in [2.17] in a manner largely analogous to that in [2.5]. The statistical characterization of cyclostationary timeseries in the nonstochastic framework is introduced and treated in depth [2.6,2.8,2.12,2.14] with reference to continuoustime signals and in [2.15] for both continuous and discretetime signals. The case of complex signals is introduced and treated in depth in [2.8,2.9]. Finally, a rigorous treatment of periodically correlated processes within the framework of harmonizable processes is given in [2.18]. The theory of higherorder cyclostationarity in the nonstochastic framework is introduced in
ARTICLE IN PRESS
W.A. Gardner et al. / Signal Processing 86 (2006) 639–697 643
3.2. Stochastic processes 3.2.1. Continuoustime processes Let us consider a continuoustime realvalued stochastic process fxðt; oÞ; t 2 R; o 2 Og, with abbreviated notation xðtÞ when this does not create ambiguity, deﬁned on a probability space ðO; F; PÞ, where O is the sample space, equipped with the sﬁeld F, and P is a probability measure deﬁned on the elements of F. The process xðtÞ is said to be Nthorder cyclostationary in the strict sense [2.2,2.5] if its Nthorder distribution function F xðtþt1 ÞÁÁÁxðtþtNÀ1 ÞxðtÞ ðx1 ; . . . ; xNÀ1 ; xN Þ 9Pfxðt þ t1 Þpx1 ; . . . ; xðt þ tNÀ1 Þ pxNÀ1 ; xðtÞpxN g is periodic in t with some period, say T 0 : F xðtþt1 þT 0 ÞÁÁÁxðtþtNÀ1 þT 0 ÞxðtþT 0 Þ ðx1 ; . . . ; xNÀ1 ; xN Þ ¼ F xðtþt1 ÞÁÁÁxðtþtNÀ1 ÞxðtÞ ðx1 ; . . . ; xNÀ1 ; xN Þ 8t 2 R 8ðt1 ; . . . ; tNÀ1 Þ 2 RNÀ1 ð3:2Þ 8ðx1 ; . . . ; xN Þ 2 RN . ð3:1Þ
frequencies. Widesense cyclostationary processes have also been called periodically correlated processes (see e.g., [2.1,3.3,3.59]). As ﬁrst shown in [3.23], xðtÞ and its frequencyshifted version xðtÞej2pnt=T 0 are correlated. The widesense stationary processes are the special case of cyclostationary processes for which Rn=T 0 ðtÞc0 only for x n ¼ 0. It can be shown that if xðtÞ is cyclostationary with period T 0 , then the stochastic process xðt þ yÞ, where y is a random variable that is uniformly distributed in ½0; T 0 Þ and is statistically independent of xðtÞ, is wide sense stationary [2.5,3.20,3.26,3.57]. A more general class of stochastic processes is obtained if the autocorrelation function Rx ðt; tÞ is almost periodic in t for each t. A function zðtÞ is said to be almost periodic if it is the limit of a uniformly convergent sequence of trigonometric polynomials in t [23.2,23.3,23.4, Paragraphs 24–25,23.6,23.7, Part 5, 23.11]: X zðtÞ ¼ za ej2pat , (3.8)
a2A
The process xðtÞ is said to be secondorder cyclostationary in the wide sense [2.2,2.5] if its mean EfxðtÞg and autocorrelation function Rx ðt; tÞ9Efxðt þ tÞxðtÞg are periodic with some period, say T 0 : Efxðt þ T 0 Þg ¼ EfxðtÞg, Rx ðt þ T 0 ; tÞ ¼ Rx ðt; tÞ (3.4) (3.5) (3.3)
where A is a countable set, the frequencies a 2 A are possibly incommensurate, and the coefﬁcients za are given by za 9hzðtÞeÀj2pat it Z 1 T=2 ¼ lim zðtÞeÀj2pat dt. T!1 T ÀT=2
ð3:9Þ
for all t and t. Therefore, by assuming that the Fourier series expansion of Rx ðt; tÞ is convergent to Rx ðt; tÞ, we can write Rx ðt; tÞ ¼
þ1 X n¼À1
Rn=T 0 ðtÞej2pðn=T 0 Þt , x
(3.6)
where the Fourier coefﬁcients Z T 0 =2 1 n=T 0 Rx ðtÞ9 Rx ðt; tÞeÀj2pðn=T 0 Þt dt T 0 ÀT 0 =2
(3.7)
are referred to as cyclic autocorrelation functions and the frequencies fn=T 0 gn2Z are called cycle
Such functions are called almostperiodic in the sense of Bohr [23.3, Paragraphs 84–92] or, equivalently, uniformly almost periodic in t in the sense of Besicovitch [23.2, Chapter 1]. The process xðtÞ is said to be Nthorder almostcyclostationary in the strict sense [2.2,2.5] if its Nthorder distribution function (3.1) is almostperiodic in t. With reference to the autocorrelation properties, a continuoustime realvalued stochastic process xðtÞ is said to be almostcyclostationary (ACS) in the wide sense [2.2,2.5,3.26] if its autocorrelation function Rx ðt; tÞ is an almost periodic function of t (with frequencies not depending on t). Thus, it is the limit of a uniformly convergent sequence of trigonometric polynomials in t: X Ra ðtÞej2pat , (3.10) Rx ðt; tÞ ¼ x
a2A
xðtÞ is said to be asymptotically almost cyclostationary [3. f À aÞg dt. for every t). at least in the sense of distributions (generalized functions) [23. Gardner et al. and Z tþZ=2 X Z ðt.26]. and Z a (3.5]. where the Fourier transform is assumed to exist. As ﬁrst shown in [3. tÞeÀj2pat dt ÀT=2 (3.A. for an ACS process.1. The density of spectral correlation on this support is described by the cyclic spectra Sa ðf Þ.5. tÞ ¼ 0. with probability 1. Let At be the set At 9fa 2 R : Ra ðtÞa0g. (3) The timeaveraged autocorrelation function R0 ðtÞ9hRx ðt. Widesense almostcyclostationary processes have also been called almostperiodically correlated processes (see e. Let us now consider the secondorder (widesense) characterization of ACS processes in the frequency domain. The support in theÈ 1 . tÞ. tÞit is continuous for t ¼ 0 x (and. f 2 Þ planeÉ the ðf of spectral correlation function E X ðf 1 ÞX Ã ðf 2 Þ consists of parallel lines with unity slope. f Þ Df !0 T!1 T ÀT=2 ÂX Ã ðt. Let Z X ðf Þ9 xðtÞeÀj2pft dt (3.59]).10). it is shown that the ACS processes are characterized by the following conditions: (1) The set [ A9 At t2R In the special case where limjtj!1 rx ðt. then the process xðtÞ and its frequencyshifted version xðtÞej2pat are correlated when a 2 A. 1=Df ð3:20Þ where the function X Ra ðtÞej2pat x a2A is not necessarily continuous in t and the term rx ðt.16) is the cyclic autocorrelation function at cycle frequency a. hence. if xðtÞ is an ACS process. [2. tÞ is uniformly continuous in t and t.13) (3.11) hrx ðt. (3. correlation exists between spectral components that are separated by amounts equal to the cycle frequencies.14) More generally. which can be x expressed as [2. tÞ does not contain any ﬁnitestrength addi where the order of the two limits cannot be reversed. x a2A (3.. f Þ9 xðsÞeÀj2pfs ds. (2) The autocorrelation function Rx ðt.18) where dðÁÞ is the Dirac delta.17) R be the stochastic process obtained from Fourier transformation of the ACS process xðtÞ. tÞeÀj2pat it ¼ 0 8a 2 R. superscript Ã is complex conjugation. x (3. a stochastic process xðtÞ is said to exhibit cyclostationarity at cycle frequency a if Ra ðtÞc0 [2. / Signal Processing 86 (2006) 639–697 where Ra ðtÞ9 lim x T!1 tive sinewave component: 1 T Z T=2 Rx ðt. that is sup Efjxðt þ tÞ À xðtÞj2 g ! 0 t2R as t ! 0.10]. (3. in the sense of distributions we obtain X EfX ðf 1 ÞX Ã ðf 2 Þg ¼ Sa ðf 1 Þdðf 2 À f 1 þ aÞ. By using (3. a 2 A. the autocorrelation x function can be expressed as X Rx ðt.26].19) Sx ðf Þ9 Ra ðtÞeÀj2pf t dt x R is countable.12) In [3.3.g.15) x a2A is referred to as the cyclic spectrum at cycle frequency a. tÞ ¼ Ra ðtÞej2pat þ rx ðt. (3. In such a case.59]. Therefore. The widesense cyclostationary processes are obtained as a special case of the ACS processes when A fn=T 0 gn2Z for some T 0 .21) tÀZ=2 .ARTICLE IN PRESS 644 W.3. (4) The process is meansquare continuous.5] Z 1 T=2 a Sx ðf Þ ¼ lim lim EfDfX 1=Df ðt. (3.
Moreover.8. and wðf Þ is the indeﬁnite integral of X ðf Þ or the integrated Fourier transform of xðtÞ [23. with abbreviated notation xðnÞ when this does not create ambiguity. for widesense stationary processes the autocorrelation function does not depend on t.4].2.5] it was dubbed the Cyclic Wiener–Khinchin Relation. f þ a=2Þ ÀT=2 A covariance (or autocorrelation) function Efxðt1 ÞxÃ ðt2 Þg is said to be harmonizable if it can be expressed as Z Efxðt1 ÞxÃ ðt2 Þg ¼ ej2pðf 1 t1 Àf 2 t2 Þ dgðf 1 .32) . They are linked to the asymmetric deﬁnitions (3. 1Þ : Rx ðmÞc0g 2 2 (3.18. It represents the timeaveraged statistical correlation (with zero lag) of two spectral components at frequencies f and f À a. mÞ9Efxðn þ mÞxðnÞg (3.31) is the cyclic autocorrelation function at cycle frequency e and a a e a ee A9fe 2 ½À1. If a stochastic process is harmonizable.1. as the bandwidth approaches zero. e. dgðf 1 . it is shown that a necessary condition for a stochastic process to be harmonizable is that it be secondorder continuous. 1=Df x respectively. the cyclic spectrum reduces to the power spectrum or spectral density function S0 ðf Þ x and (3.2. then it follows that [2.23) Df EfX 1=Df ðt. (3. mÞeÀj2pe 2N þ 1 n¼ÀN N!1 (3.22) ACS. Consequently. equivalently. In [23. Efxðt þ tÞxðtÞg ¼ R0 ðtÞ x (3.59] X Sa ðf 1 Þdðf 2 À f 1 þ aÞ df 1 df 2 . Thus. x (3. The stochastic process xðnÞ is said to be secondorder almostcyclostationary in the wide sense [2. Therefore. when (3. mÞ ¼ (3. ð3:28Þ where gðf 1 . f 2 Þ ¼ x a2A (3.11]).5. the cyclic spectrum Sa ðf Þ is also called x the spectral correlation density function. f 2 Þ is a covariance of bounded variation on R Â R and the integral is a Fourier– Stieltjes transform [23. n 2 Z.3.26) Finally. Gardner et al.2. f À a=2Þg dt ¼ Sa ðf þ a=2Þ.5.29) with probability one. no correlation exists between distinct spectral components. Moreover. it is shown that a stochastic process is harmonizable if and only if its covariance is harmonizable. Discretetime processes Let us consider a discretetime realvalued stochastic process fxðn. [2. For a ¼ 0. f 2 Þ of bounded variation on R Â R such that Z xðtÞ ¼ ej2pft dwðf Þ (3.20) was discovered in [2.25) R 3.ARTICLE IN PRESS W. we have dwðf Þ ¼ X ðf Þ df and dgðf 1 .2.11) and (3. it can be expressed as X e an e ea Rx ðn. a secondorder stochastic process xðtÞ is said to be harmonizable if there exists a secondorder stochastic process wðf Þ with covariance function Efwðf 1 ÞwÃ ðf 2 Þg ¼ gðf 1 .4.2. o 2 Og.24) R2 ÂX Ã ðt. / Signal Processing 86 (2006) 639–697 645 Therefore. in the sense of distributions [23.5].10].2. if the stochastic process is harmonizable and is an almostperiodic function of the discretetime parameter n. e e2A a where a ee Rx ðmÞ9 lim N X 1 an e Rx ðn.A.30) Rx ðmÞej2pe . f 2 Þ ¼ Efdwðf 1 Þ dwÃ ðf 2 Þg ¼ EfX ðf 1 ÞX Ã ðf 2 Þg df 1 df 2 . f 2 Þ. note that symmetric deﬁnitions of cyclic autocorrelation function and cyclic spectrum have been widely used (see.g.5].2. In contrast. EfX ðf 1 ÞX Ã ðf 2 Þg ¼ S0 ðf 1 Þdðf 2 À f 1 Þ..13] if its autocorrelation function e Rx ðn.20) by the relationships Z 1 T=2 lim Efxðt þ t=2Þxðt À t=2Þg T!1 T ÀT=2 ÂeÀj2pat dt ¼ Ra ðtÞeÀjpat x and 1 lim lim Df !0 T!1 T Z T=2 ð3:27Þ and.19) reduces to the Wiener–Khinchin Relation. oÞ.19) and (3.
38) Let xðtÞ.36) where xap ðtÞ is an almostperiodic function and xr ðtÞ a residual term not containing ﬁnitestrength additive sinewave components.12]. By using (3. Gardner et al. a (possibly time varying) probability distribution function can be constructed and this leads to an expectation operation and all the associated familiar probabilistic concepts and parameters. etc. starting from this single time series. the autocorrelation function e Rx ðn. such as stationarity. Such a model can be unnecessarily abstract when there is only a single time series at hand.33) n2Z be the stochastic process obtained from Fourier transformation (in a generalized sense) of the ACS process xðnÞ.1.41]. a e f In general. probabilistic parameters are deﬁned through inﬁnitetime averages of a single time series (and functions of this time series) rather than through expected values or ensemble averages of a stochastic process. the stochastic process. nonstationarity. (3. (3. cyclostationarity.8. Note that the cyclic autocorrea ee a lation function Rx ðmÞ is periodic in e with period 1. independence. treated more in depth in [2.A.ARTICLE IN PRESS 646 W. mÞ is periodic in n with period N 0 and the process xðnÞ is said to be cyclostationary in the wide sense. Speciﬁcally. t 2 R. that is. Time series 3.12]. the set A (or A1 ) contains possibly incommensurate cycle frequencies e. In the FOT probability approach. In [3. This ap proach turns out to be more appropriate when an ensemble of realizations does not exist and would have to be artiﬁcially introduced just to create a mathematical model.5. Let X e X ðnÞ9 xðnÞeÀj2pnn (3. cumulants.3. variance. Moreover. in the sense of distributions it can be shown that [3. 3.35) is the cyclic spectrum at cycle frequency e. Continuoustime time series A statistical analysis framework for time series that is an alternative to the classical stochasticprocess framework is the fractionoftime (FOT) probability framework ﬁrst introduced for ACS time series in [2.4. . Starting from such a decomposition.12.30).3.30) can be equivalently a e ee extended to the set A1 9fe 2 ½0. In the FOT probability framework. see also [2. the sum in (3.37) The almostperiodic component extraction operator Efag fÁg is deﬁned to be the operator that extracts all the ﬁnitestrength additive sinewave components of its argument. The extension of the Wold isomorphism to cyclostationary sequences was ﬁrst introduced in [2. Efag fxðtÞg9xap ðtÞ. For comprehensive treatments of the FOT probability framework see [2. mean. The a e a e a cyclic spectrum Sx ðnÞ is periodic in both n and e with period 1. / Signal Processing 86 (2006) 639–697 is a countable set. .2.8. . (3. In the special a e case where A1 f0. see [23.87]. 1=N 0 . 1Þ : Rx ðmÞc0g.2. let us consider the decomposition xðtÞ9xap ðtÞ þ xr ðtÞ. that is. be a realvalued continuoustime time series (a single function of time) and let us . moments.2. a Sx ðn1 Þ ¼ ‘2Z e eA a2 where X e a ea ee Rx ðmÞeÀj2pnm Sx ðnÞ9 m2Z Ã ð3:34Þ (3. the expectation operator is deﬁned. Thus.3. If N 0 ¼ 1 then xðnÞ is widesense stationary.15]. The timevariant FOT probability framework is based on the decomposition of functions of a time series into their (possibly zero) almostperiodic components and residual terms.15] and. for more mathematical rigor on foundations and existence proofs.5].59] e e EfX ðn1 ÞX ðn2 Þg X X e ea dðn2 À n1 þ e À ‘Þ. signals are modelled as single functions of time (time series) rather than sample paths of stochastic processes. hxr ðtÞeÀj2pat it 0 8a 2 R.2. it is shown that discretetime cyclostationary processes are harmonizable.15]. that is. ðN 0 À 1Þ=N 0 g for some integer N 0 . and ﬁnally— with mathematical rigor—in [3. for any ﬁniteaveragepower time series xðtÞ. .
Finally. Speciﬁcally. the time series is said to be widesense almost cyclostationary. when an ensemble of realizations or sample paths exists. Moreover. ð3:45Þ x a2A where A is a countable set and Ra ðtÞ9hxðt þ tÞxðtÞeÀj2pat it x (3. where ( 1. the secondorder derivative. 1.12] it is shown that the function of x F fag ðxÞ9Efag f1fxðtÞpxg g xðtÞ (3.44) which reveals that Efag fÁg is the expectation operator with respect to the distribution function F fag ðxÞ for the time series xðtÞ. F fag ðÀ1Þ ¼ 0. Moreover. The classiﬁcation of the kind of nonstationarity of a time series is made on the basis of the elements contained in the set A. is referred to as the fundamental theorem of temporal expectation. In [2. ﬁrst introduced in [2. All the other signals are the random signals. is a valid secondorder joint probability density function [2. it can be shown that the function Rx ðt. Analogously. Efag fxðtÞg ¼ xðtÞ. for any wellbehaved function gðÁÞ. / Signal Processing 86 (2006) 639–697 647 assume that the set G1 of frequencies (for every x) of the almostperiodic component of the function of t 1fxðtÞpxg is countable. denoted by f xðtþtÞxðtÞ ðx1 . as usual.ARTICLE IN PRESS W. In general. (3. whereas the adjective random is used in reference to a single function of time. it has values in ½0.42) That is. x2 Þ. t : xðtÞ4x is the indicator function of the set ft 2 R : xðtÞpxg.40) let us assume that the set G2 of frequencies (for every x1 .41) xðtÞ R is a valid secondorder joint cumulative distribution function for every ﬁxed t and t. is non decreasing. the function of x1 and x2 F fag xðtþtÞxðtÞ ðx1 . by analogy with the corresponding fundamental theorem of expectation from probability theory.12]. tÞ9Efag fxðt þ tÞxðtÞg (3.39) 1fxðtÞpxg 9 0. and xðtÞ F fag ðþ1Þ ¼ 1. x2 and t) of the almostperiodic component of the function of t 1fxðtþtÞpx1 g 1fxðtÞpx2 g is countable. t : xðtÞpx. fag of F fag xðtþtÞxðtÞ ðx1 . we have xðtÞ xap ðtÞ and. For an almostperiodic signal xðtÞ. In fact.41). the almostperiodic functions are the deterministic signals in the FOT probability framework. (3. Gardner et al. tÞ ¼ x1 x2 f fag xðtþtÞxðtÞ ðx1 . x2 Þ dx1 dx2 R2 X ¼ Ra ðtÞej2pat . the set A can contain incommensurate cycle frequencies a and. a secondorder characterization for the realvalued time series xðtÞ can be obtained by using the almostperiodic component extraction operator as the expectation operator. x2 Þ 9Efag f1fxðtþtÞpx1 g 1fxðtÞpx2 g g ð3:43Þ for any t is a valid cumulative distribution function except for the rightcontinuity property (in the discontinuity points). Furthermore. is a valid autocorrelation function and can be characterized by Z Rx ðt. the adjective stochastic is adopted. That is.46) is the (nonstochastic) cyclic autocorrelation function at cycle frequency a ða 2 RÞ.A. except for the rightcontinuity property (in the discontinuity points) with respect to x1 and x2 . it follows that Z fag E fgðxðtÞÞg ¼ gðxÞf fag ðxÞ dx (3. its derivative with resxðtÞ pect to x. denoted by f fag ðxÞ. is a valid probability xðtÞ density function and. then the time series xðtÞ is said to be widesense stationary. Note that the term ‘‘random’’ here is not intended to be synonymous with ‘‘stochastic’’. x2 Þ. note that the periodic component with period T 0 of an almostperiodic time series (or lag product time series) zðtÞ can be extracted by exploiting the synchronized averaging identity . If the set A contains only the element a ¼ 0.8]. in such a case. hence. In the special case where A fk=T 0 gk2Z the time series xðtÞ is said to be widesense cyclostationary. The xðtÞ result (3. with respect to x1 and x2 . Then.
f Þ is deﬁned according to (3. In (3. which was ﬁrst dubbed the Wiener Relation in [2.5. n 2 Z. as in the stochastic framework. That is.9).3. Discretetime time series The characterization of discretetime time series (sequences) is similar to that of continuoustime time series. 1=Df ð3:48Þ is a valid secondorder joint probability density function [2. In the special case ð3:50Þ is a valid secondorder joint cumulative distribution function for every ﬁxed n and m. a Obviously.52) can be a e a equivalently extended to the set A1 9fe 2 ½0. which is deﬁned analogously to its continuoustime counterpart. the function of x1 and x2 fe ag F xðnþmÞxðnÞ ðx1 . / Signal Processing 86 (2006) 639–697 introduced in [2.8] Z a S x ðf Þ ¼ Ra ðtÞeÀj2pf t dt (3.51) where X 1=Df ðt. 3. xðnþmÞxðnÞ 1 2 xðnþmÞxðnÞ 1 2 where the Fourier coefﬁcients zk=T 0 are deﬁned according to (3.50). The secondorder derivative with respect to x1 and x2 a ag of F feg ðx .8]: þ1 X k¼À1 zk=T 0 ej2pðk=T 0 Þt N X 1 zðt À nT 0 Þ. Moreover. in general. Thus.ARTICLE IN PRESS 648 W. f À aÞ dt. We consider here only the widesense secondorder characterization. Let xðnÞ. x Þ.19). 2N þ 1 n¼ÀN ¼ lim N!1 ð3:47Þ the rightcontinuity property (in the discontinuity ag points) with respect to x1 and x2 . x Þ. The Cyclic Wiener Relation introduced in [2. 1Þ : Rx ðmÞc0g 2 2 R2 ð3:52Þ links the cyclic autocorrelation function to the cyclic spectrum. be a realvalued discretetime e time series.2. ¼ e2A a e where a e a ee A9fe 2 ½À1. x2 and m) of the almostperiodic component of the function of n 1fxðnþmÞpx1 g 1fxðnÞpx2 g is countable. which is frequently called the Wiener– Khinchin Relation. x2 Þ ag 9Efe f1fxðnþmÞpx1 g 1fxðnÞpx2 g g (3. f ÞX Ã ðt. as in the continuoustime case. Gardner et al. mÞ ¼ x1 x2 f xðnþmÞxðnÞ ðx1 . the a ee cyclic autocorrelation function Rx ðmÞ is periodic in e with period 1. except for .12].21) and the order of the two limits cannot be reversed.49) x R is a valid autocorrelation function and can be characterized by Z fe ag e Rx ðn. e can contain incommensurate cycle frequencies set A e and. Let us assume that the set G2 of frequencies (for every x1 . Then.53) is a countable set and a ee Rx ðmÞ9 lim N X 1 an xðn þ mÞxðnÞeÀj2pe 2N þ 1 n¼ÀN N!1 (3. it can be shown that the function a e Rx ðn.2. x2 Þ dx1 dx2 X e an ea Rx ðmÞej2pe .5] to distinguish it from the Khinchin Relation (3. ACS time series are characterized in the spectral domain by the (nonstochastic) cyclic spectrum or spectral correlation density function at cycle frequency a: Z 1 T=2 S a ðf Þ9 lim lim Df x Df !0 T!1 T ÀT=2 ÂX 1=Df ðt.A. This relation generalizes that for a ¼ 0. denoted by f fe ðx . in such a case. the classiﬁcation of the kind of nonstationarity of discretetime time series is made on the basis of e the elements contained in set A. the time series is said to be a widesense almost cyclostationary. the sum in (3. Furthermore. mÞ9Efeg fxðn þ mÞxðnÞg (3. Efe fÁg denotes the discretetime almostperiodic component extraction operator. 1Þ : e a e Rx ðmÞc0g.54) is the (nonstochastic) discretetime cyclic autocorrelation function at cycle frequency e.
5] X e a ee ea Rx ðmÞeÀj2pnm S x ðnÞ ¼ m2Z (3. The Fourier coefﬁcients Ra Ã ðtÞ9hxðt þ tÞxÃ ðtÞeÀj2pat it .4. t 2 R. 3.13. in place of the almostperiodic component extraction operator Efag fÁg.59) Dn!0 N!1 1 2N þ 1 N X n¼ÀN Dn (3.2.31. results derived in the FOT probability approach for time series can be interpreted in the classical stochastic process framework by substituting. . Link between the stochastic and fractionoftime approaches In the timevariant nonstochastic framework (for ACS time series). hence.87. 23. xx Rb ðtÞ9hxðt þ tÞxðtÞeÀj2pbt it xx (3.64) .56) Analogous equivalences hold in the discretetime case. tÞ9Efag fxðt þ tÞxÃ ðtÞg X ¼ Ra Ã ðtÞej2pat . . the statistical expectation operator EfÁg. x x Sa ðf Þ ¼ S a ðf Þ.2. xx a2AxxÃ and bÁc denotes the closest odd integer.8. . Rx ðt.2.13.ARTICLE IN PRESS W.14.13. Complex processes and time series The case of complexvalued processes and time series. x x (3. . is extensively treated in [2.15. Therefore. the time series xðnÞ is said to be widee sense cyclostationary.4. The link between the two approaches is treated in depth in [2. tÞ ð3:58Þ and. most results are presented in the FOT probability framework.57) links the cyclic autocorrelation function to the cyclic spectrum. The Cyclic Wiener relation [2. respectively. If the set A contains only the element e ¼ 0. the stochastic and FOT autocorrelation functions are identical for almost all sample paths xðtÞ.2. / Signal Processing 86 (2006) 639–697 649 e where A1 f0. nÞX Ã b1=Dnc ðn. In fact. 1=N 0 .9]. tÞ9Efag fxðt þ tÞxðtÞg X Rb ðtÞej2pbt ¼ xx b2Axx ð3:62Þ which are called the autocorrelation function and conjugate autocorrelation function.2. then the time series xðnÞ is said to be a widesense stationary.25].3. and the expectation operator is the almostperiodic component extraction operator.5. so too are their cyclic components and frequencydomain counterparts.2. nÞ9 nþM X k¼nÀM xðkÞeÀj2pnk (3. ðN 0 À 1Þ=N 0 g for some integer N 0 . ﬁrst treated in depth in [2. In the following.63) (3.15]. tÞ9Efxðt þ tÞxðtÞg ¼ Efag fxðt þ tÞxðtÞg9Rx ðt.8.5. by assuming that fxðt.A. in the case of processes exhibiting suitable ergodicity properties.60) À eÞ. Let xðtÞ be a zeromean complexvalued continuoustime time series. a ð3:55Þ ÂX b1=Dnc ðn. 3.12.20. o 2 Og is a stochastic process satisfying appropriate ergodicity proper ð3:61Þ Rxx ðt. n where X 2Mþ1 ðn.11. oÞ. Its widesense characterization can be made in terms of the two secondorder moments RxxÃ ðt. Discretetime ACS time series are characterized in the spectral domain by the (nonstochastic) cyclic spectrum or spectral correlation density function at cycle frequency e: a a ee S x ðnÞ9 lim lim ties (see Section 4). the almostperiodic functions play the same role as that played by the deterministic functions in the stochasticprocess framework. Gardner et al.43.6. Ra ðtÞ ¼ Ra ðtÞ.3.13.13. Several results with reference to higherorder statistics are reported in Section 13.9.
2.6.70) into (3. Gardner et al.72) From (3. and input–output relation Z yðtÞ ¼ hðt. is also considered in followon work in [7.68) xx xx R is said to be linear almostperiodically timevariant (LAPTV) if the impulseresponse function admits the Fourier series expansion X hðt. 3.2. Properties of linear periodically timevarying systems are analyzed in [23. for example.64) are simultaneously represented by Ra ðÃÞ ðtÞ9 xðt þ tÞxðÃÞ ðtÞeÀj2pat t .23.23. If G contains only the element s ¼ 0. In the following.71b) It should be noted that.65) xx xx R More general timevariant linear ﬁltering is addressed in [21. xx (3. i ¼ 1.11].A. a 2 AxxðÃÞ .8.2. then the system is linear timeinvariant (LTI). (3. (3.31].6.71a) it follows that a LAPTV systems performs a linear timeinvariant ﬁltering of frequencyshifted version of the input signal. t 2 R. uÞxðuÞ du (3.20] as well as the early Russian work in [3. impulseresponse function hðt. tÞ ¼ Efag fxðt þ t=2ÞxÃ ðt À t=2Þg and RxxÃ ðt. (3.13. respectively.69) we see that the output yðtÞ can be expressed in the two equivalent forms [2.51]: X yðtÞ ¼ hs ðtÞ ½xðtÞej2pst (3. By substituting (3. In the special case for which G fk=T 0 gk2Z for some period T 0 .69) R S b ðf Þ9 xx Z R Rb ðf ÞeÀj2pf t dt xx (3. an analogous notation is adopted in these references and others for the (conjugate) cyclic autocorrelation function and the (conjugate) cyclic spectrum.66) are simultaneously represented by Z S a ðÃÞ ðf Þ9 Ra ðÃÞ ðtÞeÀj2pf t dt. Equivalently. Rxx ðt. from (3.21.1.66) are called the cyclic spectrum and the conjugate cyclic spectrum.8.65) and (3. output yðtÞ.6.2. For this reason LAPTV ﬁltering is also referred to as frequencyshift (FRESH) ﬁltering [7. gs ðtÞ9hs ðtÞeÀj2pst .5] for cyclostationary processes and generalized in [2. in [2. be two possiblycomplex ACS continuoustime time series with . the system is said to be linear periodically timevariant (LPTV). Moreover. Structure of linear almostperiodically timevariant systems A linear timevariant system with input xðtÞ.12. when it does not create ambiguity. both functions deﬁned in (3.9] to ACS signals. Linear ﬁltering The linear almostperiodically timevariant ﬁltering of ACS signals introduced in [2.9. Input/output relations in terms of cyclic statistics Let xi ðtÞ.60.14]. tÞ ¼ Efag fxðt þ t=2Þxðt À t=2Þg.51.14.2. a different conjugation notation in the subscripts is adopted to denote the autocorrelation and the conjugate autocorrelation function. (3.9.9].71a) s2G ¼ where X s2G ½gs ðtÞ xðtÞej2pst . both functions deﬁned in (3. uÞ ¼ hs ðt À uÞej2psu .70) s2G where G is a countable set. respectively.ARTICLE IN PRESS 650 W. 3. 3.10.71b) it follows that a LAPTV systems performs a frequency shift of linear timeinvariant ﬁltered versions of the input.6.9. Speciﬁcally.13. uÞ. 13. / Signal Processing 86 (2006) 639–697 are referred to as the cyclic autocorrelation function at cycle frequency a and the conjugate cyclic autocorrelation function at conjugate cycle frequency b.67) Analogously.8.8. (3. Let us denote by the superscript ðÃÞ an optional complex conjugation.23.63) and (3. The Fourier transforms of the cyclic autocorrelation function and conjugate cyclic autocorrelation function Z S a Ã ðf Þ9 Ra Ã ðtÞeÀj2pf t dt.2] and followed by [2.13.9].
then we obtain the input–output relations for LAPTV systems in terms of cyclic autocorrelation functions and cyclic spectra: X X bÀs þs Rb Ã ðtÞ ¼ ½RxxÃ 1 2 ðtÞej2ps1 t yy s1 2G s2 2G rb ðtÞ.81). If the set A12 contains at least one nonzero element. where t denotes convolution with respect to t.75) is a countable set. cc (3. ð3:78Þ 3.A. and ðÃÞ is conjugation. 12 t ð3:83Þ S xxÃ 1 bÀs þs2 Sb Ã ðf Þ ¼ yy X X s1 2G s2 2G ðf À s1 Þ ð3:84Þ The (conjugate) crosscorrelation of the outputs Z yi ðtÞ ¼ hi ðt. Gardner et al. in general. Note that. then their joint probability density function factors into the product of the . tÞ ¼ Rax ðÃÞ ðtÞej2pax t . y1 ¼ y2 ¼ y. only those s1 2 G 1 and s2 2 G 2 such that b À s1 À ðÀÞs2 2 A12 give nonzero contribution. Rb y1 y2 ðÃÞ ðtÞ9hy1 ðt þ tÞyðÃÞ ðtÞeÀj2pbt it 2 X X bÀs ÀðÀÞs 1 2 ¼ ½R ðÃÞ ðtÞej2ps1 t s1 2G 1 s2 2G2 x1 x2 RccðÃÞ ðt.76) si 2G i and. s2 hsi ðtÞeÀj2pf t dt ð3:81Þ where Ra ðÃÞ x1 x2 ðtÞ9hx1 ðt þ tÞxðÃÞ ðtÞeÀj2pat it 2 (3. and Z rg 1 s2 ðÃÞ ðtÞ9 hs1 ðt þ sÞhðÃÞ ðsÞeÀj2pgs ds. Let us consider now two linear LAPTV systems whose impulseresponse functions admit the Fourier series expansions X hi ðt.80) and (3. tÞ9Efag fy1 ðt þ tÞyðÃÞ ðtÞg 2 X X X ½Ra ¼ a2A12 s1 2G 1 s2 2G2 hs1 ðt þ sÞhÃ2 ðsÞeÀj2pbs ds. Eqs. see Sections 7 and 10. (3. (3.74) H si ðf Þ9 (3.ARTICLE IN PRESS W.81) can be specialized to several cases of interest. in the sums in (3.85) x1 x2 ðÃÞ ðtÞej2ps1 t For further examples and applications. uÞ ¼ hsi ðt À uÞej2psi u . (3.86) xx ax 2AxxðÃÞ rs1 s21ðÃÞ t aþs þðÀÞs2 ðtÞej2pðaþs1 þðÀÞs2 Þt . tÞ9Efag fx1 ðt þ tÞxðÃÞ ðtÞg 2 X a ¼ R ðÃÞ ðtÞej2pat . / Signal Processing 86 (2006) 639–697 651 secondorder (conjugate) crosscorrelation function Rx ðÃÞ 1 x2 Sb y1 y2 ðÃÞ ðf Þ9 ðt.87) t rb1 s2 ðÃÞ ðtÞ.75)).77) R ÂH s1 ðf ÞH Ã2 ðf À bÞ. For example. i ¼ 1. s where rb ðtÞ9 12 Z R is given by Ry ðÃÞ 1 y2 ðt. a2A12 x1 x2 Rb ðÃÞ ðtÞeÀj2pf t dt y1 y2 X X bÀs ÀðÀÞs 1 2 ¼ S ðÃÞ ðf À s1 Þ R s1 2G1 s2 2G2 x1 x2 Z ð3:73Þ where Z ÂH s1 ðf ÞH ðÃÞ ððÀÞðb À f ÞÞ. ðÀÞ is an optional minus sign that is linked to ðÃÞ.80) and (3. set A12 depends on whether ðÃÞ is conjugation or not and can be different from the sets A11 and A22 (both deﬁned according to (3.79) s2 s R Thus. 2. s (3. h1 ¼ h2 ¼ h. s ð3:80Þ If xðtÞ and cðtÞ are statistically independent in the FOT probability sense. tÞ ¼ X ac 2AccðÃÞ RacðÃÞ ðtÞej2pac t . if x1 ¼ x2 ¼ x. uÞxi ðuÞ du. then the time series x1 ðtÞ and x2 ðtÞ are said to be jointly ACS. Product modulation Let xðtÞ and cðtÞ be two ACS signals with (conjugate) autocorrelation functions X RxxðÃÞ ðt. 2 (3.82) is the (conjugate) cyclic crosscorrelation between x1 and x2 at cycle frequency a and A12 9fa 2 R : Ra x1 x2 ðÃÞ R ðtÞc0g (3.7. i ¼ 1. (3.
we obtain S a ðÃÞ ðf Þ ¼ S a ðÃÞ ðf ÞHðf ÞH ðÃÞ ððÀÞða À f ÞÞ. jf jpB. (3.89) (3. (3. f Þ 2 R Â R : Hðf Þ ÂH ðÃÞ ððÀÞða À f ÞÞa0g.9. we have (see Fig. xðtÞ xðtÞ hLPF ðtÞ.91) where. S0 Ã ðf Þ 0 xx for f eðÀB.5. jf j4B: 2B Accounting for (3.A.97). f Þ 2 R Â R : Sa ðÃÞ ðf Þa0g yy yy fða.96) it follows that the support in the ða.2. then RccðÃÞ ðt.90) yy xx ccðÃÞ ax 2AxxðÃÞ Let xðtÞ be a strictly bandlimited lowpass signal with monolateral bandwidth B. tÞRxxðÃÞ ðt.92) g2G where hLPF ðtÞ is the impulseresponse function of the ideal lowpass ﬁlter with harmonicresponse function: ( 1. f Þ plane of the (conjugate) cyclic spectrum of yðtÞ is such that supp½Sa ðÃÞ ðf Þ9fða. (3. Observe that. Cyclicspectrum support of a lowpass signal.95) −B B f 3. . ðÀÞðaÀgÞ e ð3:93Þ (3. tÞ ¼ cðt þ tÞcðÃÞ ðtÞ X X cg1 cðÃÞ eðÀÞj2pg2 t ¼ g2 g1 2G g2 2G 2B Âej2pðg1 þðÀÞg2 Þt and Ra ðÃÞ ðtÞ ¼ cc S a ðÃÞ ðf Þ ¼ cc X g2G j2pðaÀgÞt cg cðÃÞ .88) From (3. (3. BÞ. / Signal Processing 86 (2006) 639–697 marginal probability densities [2. 1.94) X g2G cg cðÃÞ ðÀÞðaÀgÞ dðf þ g À aÞ. and y1 ¼ y2 ¼ y.96) −2B Fig.81) to the case x1 ¼ x2 ¼ x. only those (conjugate) cycle frequencies ax such that a À ax 2 AccðÃÞ give nonzero contribution.12] and the (conjugate) autocorrelation functions of the product waveform yðtÞ ¼ cðtÞxðtÞ also factor. if cðtÞ is an almostperiodic function X cðtÞ ¼ cg ej2pgt . xx ccðÃÞ (3. 1) supp½Sa ðÃÞ ðf Þ fða.99) 0. ð3:97Þ Therefore.2. tÞ ¼ RccðÃÞ ðt. tÞ. Then. Gardner et al.98) S a ðÃÞ ðf Þ ¼ yy X ax 2AxxðÃÞ Z R S ax ðÃÞ ðlÞSaÀax ðf À lÞ dl.8]: X Ra ðÃÞ ðtÞ ¼ Rax ðÃÞ ðtÞRaÀax ðtÞ. the (conjugate) cyclic autocorrelation function and the (conjugate) cyclic spectrum of yðtÞ are [2. f Þ 2 R Â R : H LPF ðf Þ xx ÂH LPF ðf À aÞa0g. f H LPF ðf Þ ¼ rect 9 (3.8. (3.8. Supports of cyclic spectra of band limited signals By specializing (3.2. in the sums.ARTICLE IN PRESS 652 W. RyyðÃÞ ðt. ð3:100Þ where the fact that rectðf Þ is real and even is used. yy xx (3. h1 ¼ h2 ¼ h (LTI). that is.
Cyclicspectrum support of a highpass signal.101) where hHPF ðtÞ is the impulseresponse function of the ideal highpass ﬁlter with harmonicresponse function f H HPF ðf Þ ¼ 1 À rect . Gardner et al. S0 Ã ðf Þ 0 for f 2 ðÀb. f Þ 2 R Â R : H HPF ðf Þ ÂH HPF ðf À aÞa0g.107) The (conjugate) autocorrelation function of the discretetime signal xðnÞ can be shown to be the sampled version of the (conjugate) autocorrelation function of the continuoustime signal xa ðtÞ: ag Efe fxðn þ mÞxðÃÞ ðnÞg Fig.8]. (3.6. S 0 Ã ðf Þ 0 for f eðÀB. (3. let xðtÞ be a strictly bandlimited bandpass signal. BÞ. we have (see Fig.97).A.102) 2b From (3.9. respectively.97). ð3:108Þ . Therefore. that is.2.t¼mT s . 2. ð3:103Þ Finally. Cyclicspectrum support of a bandpass signal. ð3:106Þ It is noted that supports for symmetric deﬁnitions of cyclic spectra (3. that is. we have (see Fig. (3. the continuoustime signal xa ðtÞ: xðnÞ9xa ðtÞjt¼nT s .105) for (3. xðtÞ xðtÞ hBPF ðtÞ. 3. Sampling and aliasing where H LPF ðf Þ and H HPF ðf Þ are given by (3. f Þ 2 R Â R : H LPF ðf ÞH LPF ðf À aÞa0g \ fða. f Þ 2 R Â R : H BPF ðf ÞH BPF ðf À aÞa0g ¼ fða.99) and (3.28) are reported in [2. xx xðtÞ xðtÞ hHPF ðtÞ. ÀbÞ[ xx ðb.104) −B −b 2B 2b b −2b B f −2B Fig. (3. accounting 2b −b b −2b f Let xðnÞ be the sequence obtained by uniformly sampling. ¼ Efag fxa ðt þ tÞxa ðtÞgjt¼nT s . (3.ARTICLE IN PRESS W. / Signal Processing 86 (2006) 639–697 653 Let xðtÞ be a strictly bandlimited highpass signal. 3) supp½Sa ðÃÞ ðf Þ xx fða. 2) supp½S a ðÃÞ ðf Þ fða. f Þ 2 R Â R : H HPF ðf Þ xx ÂH HPF ðf À aÞa0g. Then. where hBPF ðtÞ is the impulseresponse function of the ideal bandpass ﬁlter with harmonicresponse function H BPF ðf Þ ¼ H LPF ðf ÞH HPF ðf Þ. bÞ. 3.102). with period T s ¼ 1=f s . where 0oboB. Then.
only the replica with p ¼ 0 gives nonzero contribution in the base support region in (3. Representations by stationary components 3.A. (3. T 0 ¼ KT s þ . then the discretetime signal is almostcyclostationary [13. / Signal Processing 86 (2006) 639–697 However.a¼ef .22.112) In such a case. the (conjugate) cyclic autocorrelation functions and the (conjugate) cyclic spectra of xðnÞ can be expressed in terms of the (conjugate) cyclic autocorrelation functions and the (conjugate) cyclic spectra of xa ðtÞ by the relations [2.and/or time. ð3:113Þ Sa xa xa ðf Þ 8 < T Se ðnÞj ea . if T 0 is the period of cyclostationarity.e a¼a=f s ¼ : 0 ðÃÞ jajp f2s .23. Continuoustime processes and time series A continuoustime widesense cyclostationary signal (process or timeseries) xðtÞ can be expressed in terms of singularly and jointly widesense stationary signals with nonoverlapping spectral bands [2. the (conjugate) cyclic autocorrelation functions of xðnÞ are not sampled versions of the (conjugate) cyclic autocorrelation functions of xa ðtÞ because of the presence of aliasing in the cyclefrequency domain. ja À f jpBg fða. it follows that supp½S a ðÃÞ ðf Þ xx fða. jajp2Bg.23]. with period T s .116) . the (conjugate) cyclic autocorrelation functions and the cyclic spectra of the continuoustime signal xa ðtÞ are amplitude. From (3. xx e a¼a=f s a R ðÃÞ ðtÞjt¼mT s ¼ xa xa > :0 jajp fs . the support of each replica in (3. Gardner et al.2. If.77. In [3. f Þ 2 R Â R : jf jpB.10. f Þ 2 R Â R : jf jpB.ARTICLE IN PRESS 654 W. Consequently.5. Speciﬁcally. for the (conjugate) cyclic spectra.44]. consequently.109) and (3. K integer.13.a¼af s signal xðnÞ [13.20.110) Let xðtÞ be a strictly bandlimited lowpass signal with monolateral bandwidth B. is a discretetime cyclostationary signal with cyclostationarity period K. f Þ 2 R Â R : jf À qf s jpB.109) The discretetime signal obtained by sampling. (3. otherwise: ð3:114Þ e . jf jp f2s .8. and we deﬁne e xk ðtÞ9½xðtÞeÀj2pðk=T 0 Þt h0 ðtÞ.3.5. a cyclostationary continuoustime signal with cyclostationarity period T 0 ¼ KT s .23]: X aÀpf a s ee RxxðÃÞ ðmÞ ¼ R ðÃÞ ðtÞjt¼mT p2Z xa xa s . (3. 2 otherwise. 3.or frequencyscaled versions of the (conjugate) cyclic autocorrelation functions and cyclic spectra of the discretetime where h0 ðtÞ9ð1=T 0 Þ sincðt=T 0 Þ is the impulseresponse function of an ideal lowpass ﬁlter with monolateral bandwidth 1=ð2 T 0 Þ.13.3. ja À pf s jp2Bg and. with 0ooT s and incommensurate with T s .110) is contained in the set fða. however.20]: 8 > ee < Ra ðÃÞ ðmÞj . aliasing in both the spectralfrequency and cyclefrequency domains occurs. That is. a sufﬁcient condition for assuring that the replicas in (3. ð3:111Þ Thus. common pitfalls arising in the application of the stationary signal theory to time sampled cyclostationary signals are examined. s xxðÃÞ n¼f =f s .4. Thus.41].10.100). a s s T s p2Z q2Z xa xðÃÞ a (3.110) do not overlap is f s X4B.3.1. (3.110).115) a 1 X X aÀpf s ee S xxðÃÞ ðnÞ ¼ S ðf À qf s Þjf ¼nf . then xðtÞ can be expressed by the harmonic series representation: xðtÞ ¼ þ1 X k¼À1 e xk ðtÞej2pðk=T 0 Þt .
.10. . The signal xðnÞ is widesense cyclostationary with period N 0 if and only if the set of subsampled xi ðnÞ are jointly widesense stationary [3.65]. . i ¼ 0. It follows that a continuoustime widesense cyclostationary scalar signal xðtÞ is equivalent to the inﬁnitedimensional vectorvalued widesense stationary signal e e e e e ½.121) Therefore.23]. The subsampled (or decimated) timeseries: xi ðnÞ9xðnN 0 þ iÞ. (3. . In this case.120) which is independent of t.A.2. N 0 À 1. .12. Another representation of a cyclostationary signal by stationary components is the translation series representation in terms of any complete orthonormal set of basis functions [3. That is. ð3:117Þ 3. . a discretetime widesense cyclostationary scalar signal xðnÞ is equivalent to the N 0 dimensional vectorvalued widesense stationary signal e e ½e0 ðnÞ. .1.2. . A harmonic series representation can also be obtained for an ACS process xðtÞ.2. / Signal Processing 86 (2006) 639–697 655 where Efag fek ðt þ tÞeÃ ðtÞg x xh Z 1=ð2 T 0 Þ ðhÀkÞ=T 0 SxxÃ f ¼ À1=ð2 T 0 Þ k j2pf t þ df e T0 with stationary components has been demonstrated to exist for ACS processes. . xN 0 À1 ðnÞ. if N 0 is the period of cyclostationarity.3. . 0 0 k e Xk f À ¼ T0 0 otherwise. . x0 ðtÞ.123) where the frequencies lk are possibly incommene surate and the processes xk ðtÞ are singularly and jointly widesense stationary but not necessarily band limited. Given the set of time series xi ðnÞ. .17. x1 ðtÞ. . Let xðnÞ be a discretetime realvalued widesense cyclostationary timeseries with period N 0 . i ¼ 0. No translation series representation where dg is the Kronecker delta (dg ¼ 1 for g ¼ 0 and dg ¼ 0 for ga0). . . . One example uses ` the Karhunen–Loeve expansion of the cyclostationary signal xðtÞ on the intervals t 2 ½nT 0 . jf À k=T jp1=ð2 T Þ.12. .4. N 0 À 1 (3.119) xi ð‘ÞdnÀiÀ‘N 0 .ARTICLE IN PRESS W. . the original signal xðnÞ can be reconstructed by using the synthesis formula: xðnÞ ¼ N 0 À1 X i¼0 X ‘2Z e xk ðtÞej2plk t . (3. provided that it belongs to the subclass of the almostperiodically unitary processes [3. since only spectral components of xðtÞ with frequencies separated by an integer multiple of 1=T 0 can be correlated. . . x1 ðnÞ. . . then xðnÞ can be expressed by the harmonic series representation: xðnÞ ¼ N 0 À1 X k¼0 e xk ðnÞej2pðk=N 0 Þn . In fact..122) constitute what is called the polyphase decomposition of xðnÞ. where T 0 is the period of cyclostationarity. then the only pairs of e e spectral components in xk ðtÞ and xh ðtÞ that can be correlated are those with the same frequency f. due to the .30].41. xðtÞ ¼ þ1 X k¼À1 where h0 ðnÞ9ð1=N 0 Þ sincðn=N 0 Þ is the ideal lowpass ﬁlter with monolateral bandwidth 1=ð2N 0 Þ.3]. xÀk ðtÞ.5. xk ðtÞ. there is no spectral crosscorrelation at e e distinct frequencies in xk ðtÞ and xh ðtÞ. ðn þ 1ÞT 0 Þ for all integers n. xÀ1 ðtÞ. Thus. Gardner et al. . Discretetime processes and time series A discretetime widesense cyclostationary signal (process or timeseries) xðnÞ can be expressed in terms of a ﬁnite number of singularly and jointly e widesense stationary signals xk ðnÞ with non overlapping bands [2. (3. . . This result reﬂects the e fact that the Fourier transforms X k ðf Þ of the e signals xk ðtÞ have nonoverlapping support of width 1=T 0 ( X ðf Þ. (3. x A further decomposition of a discretetime cyclostationary signal can be obtained in terms of subsampled components. . and we deﬁne e xk ðnÞ9½xðnÞeÀj2pðk=N 0 Þn h0 ðnÞ. ð3:118Þ therefore.
The problem of cyclic leakage in the estimate of a cyclic statistic at cycle frequency a arising from cyclic statistics at cycle frequencies different from a. In the frequency domain.8.3. Consistency for estimators of the cyclic spectrum has been addressed in [2. / Signal Processing 86 (2006) 639–697 cyclostationarity of xðnÞ: E fxi ðn þ mÞxk ðnÞg ¼ Efag fxððn þ mÞN 0 þ iÞxðnN 0 þ kÞg ¼ Efag fxðmN 0 þ iÞxðkÞg.4.5.2) is a consistent estimator of the cyclic spectrum Sa ðf Þ.and fourthorder cumulants.4.11)). Such a property is generally expressed in terms of mixing conditions or summability of second. Df ! 0.4.30. ﬁrst observed in [2. Under such mixing conditions.13. t0 . TÞ9 xðt þ tÞxðtÞeÀj2pat dt (4.A. f ÞDf x Z f þDf =2 1 1 X T ðt0 . x Df !0 T!1 ð4:5Þ where X T ðt.55.9.36.ARTICLE IN PRESS 656 W.4.61].18].4.2.13. 4.4.4.19) x and (3. Gardner et al. the cyclic periodogram I a ðt. The ﬁrst treatment for ACS processes is in [4. f ÞDf À Sa ðf Þ x is an asymptotically (T ! 1. the cyclic correlogram Z 1 t0 þT=2 Ra ðt. In [2. f À aÞ.1. with T Df ! 1) zeromean complex normal random variable for each f and t0 .3. . is addressed in [2. Ergodic properties and measurement of characteristics 4.17].13].13.4. 4.35.4. f ÞDf x ¼ lim lim Sa 1=Df ðt.2. ð3:124Þ fag cyclic periodogram Sa T ðt0 .17].4. Consistency for estimators of the cyclic autocorrelation function for cyclostationary and/or ACS processes has been addressed in [2.4. is an asymptotically unbiased but not consistent estimator of the cyclic spectrum Sa ðf Þ (see (3.8].43.1) x T t0 ÀT=2 is a consistent estimator of the cyclic autocorrelation function Ra ðtÞ (see (3. lÞ is deﬁned according to (3. f ÞX Ã ðt. Moreover. Estimation of the cyclic autocorrelation function and the cyclic spectrum Ergodic properties and measurements of characteristics are treated in [4. f À aÞ ds 1=Df ð4:4Þ is asymptotically equivalent to the frequency smoothed cyclic periodogram in the sense that Df !0 T!1 lim lim S a T ðt.20)). l À aÞ dl 9 T Df f ÀDf =2 T ð4:3Þ 4. the frequencysmoothed where the order of the two limits on each side cannot be reversed.17].1.1.12]. T T (4. under the abovementioned mixing conditions. Note that both the timesmoothed and frequencysmoothed cyclic periodograms exhibit spectral frequency resolution on the order of Df and a cycle frequency resolution on the order of 1=T [4.44.42. However. f ÞT .4. TÞ À Ra ðtÞ x is an asymptotically (T ! 1) zeromean complex normal random variable for each t and t0 .20.23.4.24.8. x pﬃﬃﬃﬃ a T ½Rx ðt.1. Moreover.59. which is independent of n. x pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ a T Df ½S xT ðt. The ﬁrst detailed study of the variance of estimators of the cyclic spectrum is given in [2. In particular.14. it is shown that the timesmoothed cyclic periodogram: Z 1 tþT=2 a Sx1=Df ðt.8] and is based on the FOT framework. it is shown that a strong stationary spectrally overlapping noise component added to a cyclostationary signal degrades the performance (bias and variance) of the estimators of cyclic statistics at nonzero cycle frequencies because of the leakage from the zero cycle frequency. Consistent estimates of secondorder statistical functions of an ACS stochastic process can be obtained provided that the stochastic process has ﬁnite or ‘‘effectively ﬁnite’’ memory.4. f ÞT 9 Df T tÀT=2 ÂX 1=Df ðs.15. t0 . f ÞX Ã ðs. lÞX Ã ðt0 .4. f Þ9 x 1 X T ðt.21).1–4.
51].4.14.10. the kind of convergence of the estimators to be considered as the datarecord length approaches inﬁnity is the convergence of the function sequence of the ﬁnitetime averages (indexed by the datarecord length).1.20.13.12.).15].39.2. the convergence in the FOT probability framework must be considered ‘‘pointwise’’.4.2. Given a . 13.9.12.59.4.3.17] and also in [4.2.32.21. Two alternative approaches to the analysis of measurements on time series In the FOT probability approach.5.51].9].9.50.13.14. passing such frequencyshifted versions through two lowpass ﬁlters hDf ðtÞ with bandwidth Df and unity passband height. Spectral correlation analyzer.22.41].9.1].3. Cyclostationary feature measurements in the nonstochastic approach are treated in considerable depth in [4.4.26] or ‘‘almost sure sense’’ [4. For measurements of cyclic higherorder statistics see [4. see the general treatments [2.7. Problems arising from the presence of jitter in measurements are addressed in [4.11.11. Estimators of the FOT probabilistic parameters are obtained by considering ﬁnitetime averages of the same quantities involved in the inﬁnitetime averages. Reliable spectral estimates with reduced computational requirements can be obtained by using nonlinear transformations of the data [4. Measurements on cyclostationary random ﬁelds are treated in [20.3.13.25. and then correlating the output signals (see Fig.11.29]. the inﬁnitetime averages exist).A.s. The problem of measurement of statistical functions for more general classes of nonstationary signals is considered in [21.s. without the necessity of requiring ergodicity properties as in the stochastic process framework.1. were ﬁrst treated in depth in [4.38.11.21. in the FOT probability framework. Therefore.17].56.12.8. in the ‘‘temporal meansquare sense’’ [2. Thus. probabilistic parameters are deﬁned through inﬁnitetime averages of functions of a single time series (such as products of time.8.18.20.31].8. for example.4. A survey of estimation problems is given in [4.f )T Fig.and frequencyshifted versions of the time series) rather than through expected values or ensemble averages of a stochastic process.48].5.15.3.2.18.3.9.45.2. A spectral correlation analyzer can be realized by frequency shifting the signal xðtÞ by two amounts differing by a.4. Strictsense ergodic properties of ACS processes. let us consider the convergence of time series in the temporal meansquare sense (t.31].23. in this order [4. 4.2.2. 12. / Signal Processing 86 (2006) 639–697 657 e−j2πft × x (t) × e−j2π( f− )t h∆f (t) × h∆f (t) (⋅)* 〈⋅〉T ∆f Sx1/∆ f (t.m. their asymptotic estimators converge by deﬁnition to the true values. For further references.21.5]. in the ‘‘stochastic meansquare sense’’ [3.ARTICLE IN PRESS W.13. unlike the stochastic process framework where convergence must be deﬁned.13.72.2. 4).57. referred to as cycloergodicity in the strict sense.13. Computationally efﬁcient digital implementations of cyclic spectrum analyzers are developed and analyzed in [4.2.18] and also see [3.56] or ‘‘in distribution’’.21. see also [2.49.15]. By following the guidelines in [2. 4. and taking the limit as the correlation time T ! 1 and the bandwidth Df ! 0.2. 4. assuming the abovementioned limits exist (that is.2. or in the ‘‘sense of generalized functions (distributions)’’ [23.4. Gardner et al.2.13]. which are exactly the inﬁnitetime averages. Therefore.4.35.53.4.61].41.13.39. The spectral correlation density function S a ðf Þ is x obtained by normalizing the output by Df .35.59.13.
t þ T=2g.e.. In such a case. u 2 ½t À T=2.10] for nonstochastic time series. Thus. The analytical cyclic spectral analysis of mathematical models of analog and digitally modulated signals was ﬁrst carried out in [2.ARTICLE IN PRESS 658 W.1.23]..8.7.11).19. On this general subject.10) is just the almostperiodic component contained in the time series zðtÞ.23].7) ÀT=2 and we assume that T!1 lim zb ðtÞT ¼ zb ðt:m:s:s:Þ 8b 2 R.21]. Manufactured signals: modelling and analysis 5. scanning. biasfzb ðtÞT g9Efag fzb ðtÞT g À zb ’ hzb ðtÞT it À zb . the series P 2 b2B jzb j is summable [2.9] and.12) it follows that hjzb ðtÞT À zb j2 it ’ varfzb ðtÞT g þ jbiasfzb ðtÞT gj2 ð4:13Þ and this approximation become exact as T ! 1. It is worthwhile to emphasize that. such as modulation.8.9). Moreover.22].17.5.2.5. Therefore. unlike the stochastic process framework where the variance accounts for ﬂuctuations of the estimates over the ensemble of sample paths.18.2.13]. probabilistic functions are deﬁned in terms of the almostperiodic component extraction operation.5. and also see [2.1–5. (4.2.5] for stochastic processes and in [5. The effects of timing jitter on the cyclostationarity properties of communications signals are addressed in [2. see the general treatments [2. which plays the same role as that played by the statistical expectation operation in the stochastic process framework [2. / Signal Processing 86 (2006) 639–697 time series zðtÞ (such as a lag product of another time series).8) that is. ð4:11Þ .8. estimates obtained by using different time segments asymptotically do not depend on the central point of the segment. and (4. accounting for (4. Therefore. The function zb ðtÞT is an estimator of zb based on the observation fzðuÞ.10) The magnitude and phase of zb are the amplitude and phase of the ﬁnitestrength additive complex sinewave with frequency b contained in the time series zðtÞ. In fact. in the FOT probability framework the variance accounts for the ﬂuctuations of the estimates in the time parameter t.5. the central point of the ﬁnitelength time series segment adopted for the estimation. It can be shown that.12. (4.15. T!1 lim hjzb ðtÞT À zb j2 it ¼ 0 8b 2 R. varfzb ðtÞT g9Efag fjzb ðtÞT À Efag fzb ðtÞT gj2 g ’ hjzb ðtÞT À hzb ðtÞT it j2 it . ð4:12Þ 5.9) where the approximation becomes exact equality in the limit as T ! 1.A.5. multiplexing and coding operations [5.11. viz.9. Gardner et al. the assumption that the estimator asymptotically approaches the true value (the inﬁnitetime average) in the meansquare sense is equivalent to the statement that the estimator is meansquare consistent in the FOT probability sense. the righthand side in (4.13]. 6. (4. from (4. sampling. lim T!1 b2B b2B (4. then the set B9fb 2 R : zb a0g is countable. The effects of multiplexing are considered in [5.6) zb ðtÞT 9 T tÀT=2 zb 9 lim 1 T Z þT=2 T!1 zðuÞeÀj2pbu du (4.15].2. it follows that X X zb ðtÞT ej2pbt ¼ zb ej2pbt ðt:m:s:s:Þ.11.5. if the time series zðtÞ has ﬁniteaveragepower (i. General aspects Cyclostationarity in manmade communications signals is due to signal processing operations used in the construction and/or subsequent processing of the signal.8). Continuousphase frequencymodulated signals are treated in [5.9. in the FOT probability framework. (4. hjzðtÞj2 it o1). we deﬁne Z 1 tþT=2 zðuÞeÀj2pbu du.5.2.
1) with stationary modulating signal sðtÞ having triangular autocorrelation function.ARTICLE IN PRESS W. Then a third example of a more sophisticated communication signal is considered. as a function of a and t.A. 5(a) the magnitude of the cyclic autocorrelation function Ra ðtÞ. for the DSBAM signal x (5. The (conjugate) cyclic autocorrelation function and (conjugate) cyclic spectrum of xðtÞ .5) S a ðf Þ x ¼ : Ç f 0 ÞeÆj2f0 . xðtÞ is cyclostationary with period 1=ð2 f 0 Þ. as a function of a and f.7. and (b) magnitude of the cyclic x spectrum Sa ðf Þ.1) The cyclic autocorrelation function and cyclic spectrum of xðtÞ are [5. 5(b) the magnitude of the cyclic spectrum S a ðf Þ. s ð5:3Þ and t. (5. s S a ðf Þ ¼ 1 fS a ðf À f 0 Þ þ S a ðf þ f 0 Þ x s s 4 þ S aþ2f 0 ðf þ f 0 ÞeÀj2f0 s þ S aÀ2f 0 ðf À f 0 Þej2f0 g. / Signal Processing 86 (2006) 639–697 659 5.2. In Fig. (5. The derivations of the cyclic statistics can be accomplished by using the results of Sections 3.6 and 3. Gardner et al. otherwise: (5. respectively. two fundamental examples of cyclostationary communication signals are considered and their widesense cyclic statistics are described. 5. < a 1 0 Æj2pf 0 t Æj2f0 Rx ðtÞ ¼ 4 Rs ðtÞe e .4) 81 0 0 > 4 fSs ðf À f 0 Þ þ S s ðf þ f 0 Þg.2.21]. a ¼ Æ2f 0 . as a function of a and f. 5. is an ACS sequence whose cyclostationarity is possibly induced by framing.2. ak 2 C.2. 5.1.6) k2Z ð5:2Þ Fig. (a) Magnitude of the cyclic autocorrelation function Ra ðtÞ. Examples of communication signals In this section. as a function of a x where qðtÞ is a complexvalued square integrable pulse and fak gk2Z . If sðtÞ is a widesense stationary signal then Ra ðtÞ ¼ R0 ðtÞda and s s 81 0 a ¼ 0. are x reported for the DSBAM signal (5. (5. < 1 0 S ðf >4 s a ¼ 0. Pulseamplitudemodulated signal Let xðtÞ be the complexvalued pulseamplitude modulated (PAM) signal: X xðtÞ9 ak qðt À kT 0 Þ. > 2 Rs ðtÞ cosð2pf 0 tÞ. 0 Thus.1). multiplexing. or coding [13. > : 0 otherwise. and in Fig.9]: Ra ðtÞ ¼ 1 Ra ðtÞ cos ð2pf 0 tÞ x 2 s þ 1 fRaþ2f 0 ðtÞeÀj2pf 0 t eÀj2f0 s 4 þ RaÀ2f 0 ðtÞej2pf 0 t ej2f0 g. Double sideband amplitudemodulated signal Let xðtÞ be the (realvalued) double sideband amplitudemodulated (DSBAM) signal: xðtÞ9sðtÞ cosð2pf 0 t þ f0 Þ. a ¼ Æ2f 0 .
10) are the (conjugate) cyclic autocorrelation function and the (conjugate) cyclic spectrum. (5.14) In Fig.6). as a function of a and f.e¼aT 0 0a T 0 aa ðÃÞ ÂQðf ÞQ ððÀÞða À f ÞÞ. .17) k2Z S a Ã ðf Þ xx e0 RaaÃ ð0Þ ¼ dðaT 0 mod 1Þ Qðf ÞQÃ ðf À aÞ.2. where a ee RaaðÃÞ ðmÞ9 lim N X 1 ak akþm aðÃÞ eÀj2pe . the cyclic autocorrelation function and the cyclic spectrum of xðtÞ become Ra Ã ðtÞ ¼ xx e0 RaaÃ ð0Þ dðaT 0 mod 1Þ ra Ã ðtÞ. / Signal Processing 86 (2006) 639–697 are [5. a (5. as a function of a and t.11) R and ra ðÃÞ ðtÞ9qðtÞ ½qðÃÞ ðÀtÞej2pat qq Z ¼ qðt þ tÞqðÃÞ ðtÞeÀj2pat dt. ð5:12Þ If the sequence fak gk2Z is widesense stationary and white. k 2N þ 1 k¼ÀN N!1 (5. Thus. (5.10]: Ra ðÃÞ ðtÞ ¼ xx a 1 X ee ½R ðÃÞ ðmÞe¼aT a 0 T 0 m2Z aa Âra ðÃÞ ðt À mT 0 Þ. In this case. ð5:7Þ ð5:8Þ respectively.3. 6(a) the magnitude of the cyclic autocorrelation function Ra ðtÞ. In this case. 6. and rectangular pulse qðtÞ9rectððt À T 0 =2Þ=T 0 Þ.ARTICLE IN PRESS 660 W. 6(b) the magnitude of the cyclic spectrum S a ðf Þ. k 2 Z. for the PAM signal x (5. qq T0 (5. and (b) magnitude of the cyclic x spectrum Sa ðf Þ. Gardner et al. qq S a ðÃÞ ðf Þ ¼ xx a 1 ee ½S ðÃÞ ðnÞn¼fT .13) where mod denotes the modulo operation.12) reduces to t ra ðtÞ ¼ eÀjpaðT 0 ÀtÞ rect qq 2T 0 jtj Â 1À ð5:16Þ T 0 sincðaðT 0 À jtjÞÞ.9) X e a ee ea RaaðÃÞ ðmÞeÀj2pnm S aaðÃÞ ðnÞ9 m2Z (5. as a function of a and f. R Fig. of the sequence fak gk2Z . xðtÞ is cyclostationary with period T 0 .9. Directsequence spreadspectrum signal Let xðtÞ be the directsequence spreadspectrum (DSSS) baseband PAM signal X xðtÞ9 ak qðt À kT 0 Þ. respectively.15) respectively.5. T0 (5. are x reported for the PAM signal (5. as a function of a x and t. xðtÞ exhibits cyclostationarity with cycle frequencies a ¼ k=T 0 .6) with stationary modulating sequence fak gk2Z .A. Z Qðf Þ9 qðtÞeÀj2pft dt (5. T0 5. (a) Magnitude of the cyclic autocorrelation function Ra ðtÞ. and in Fig. then a ee e0 RaaÃ ðmÞ ¼ RaaÃ ð0Þdðe mod 1Þ dm . that is.
Cyclic Wiener ﬁltering À mT 0 Þ ga ðÃÞ ðtÞ.1–7. leading to the linearconjugatelinear (LCL) structure [2. and qðtÞ9 N c À1 X n¼0 cn pðt À nT c Þ (5.18).22) with CðnÞ9 N c À1 X n¼0 cn eÀj2pnn .3. General aspects Cyclostationarity properties of modulated signals can be suitably exploited in the analysis and design of communications systems (see [7.1].8.14.14. A patent on a speech recognition technique exploiting cyclostationarity is [6.98]. The problems of synchronization. / Signal Processing 86 (2006) 639–697 661 where fak gk2Z .5. c ca ð5:20Þ where ga ðÃÞ ðtÞ9 cc N c À1 N c À1 X X n1 ¼0 n2 ¼0 cn1 cðÃÞ eÀj2pan2 T c n2 ð5:21Þ Âdðt À ðn1 À n2 ÞT c Þ and a ee GccðÃÞ ðnÞ9CðnÞCðÃÞ ððÀÞðe À nÞÞ a (5.1.e¼aT 0 0a T 0 aa ðÃÞ ÂPðf ÞP ððÀÞða À f ÞÞ a ee ÂGccðÃÞ ðnÞjn¼fT .7. cyclostationarity is due to rotation and revolution of the earth [6.21–6.2. (5.16.47. Applications in hydrology are consid The problem of optimum linear ﬁltering consists of designing the linear transformation of the data xðtÞ that minimizes the meansquared error of the ﬁlter output relative to a desired signal. for periodic ARMA modelling and the prediction problem in hydrology.24]: a 1 X ee Ra ðÃÞ ðtÞ ¼ ½RaaðÃÞ ðmÞe xx a¼aT 0 T0 m2Z a ÂrppðÃÞ ðt ered in [6. channel identiﬁcation and equalization. fc0 .2.11.11.5.26].6. . and T c is the chip period such that T 0 ¼ N cT c.6.18) is the spreading waveform. signal parameter and waveform estimation.7.6.5.e¼aT .20.7.5. 12. say dðtÞ.10.8. cN c À1 g is the N c length spreading sequence (code) with cn 2 C.24].9.21. the optimum ﬁlter is obtained by processing both xðtÞ and xÃ ðtÞ.6.A. On the analysis and design of communications systems.13.13]. The cyclostationary nature of interference in communications systems is characterized in [7.19.47. The problem of optimum ﬁltering (cyclic Wiener ﬁltering) of ACS signals is addressed in Section 7. see [12.27]. .7. and also see [3.2.18.12. Communications systems: analysis and design 7.23) 6. The (conjugate) cyclic autocorrelation function and (conjugate) cyclic spectrum of xðtÞ are [2. 2.7.8. Gardner et al. see the general treatments [2. If dðtÞ and xðtÞ are jointly ACS signals. 10. cc ð5:19Þ S a ðÃÞ ðf Þ ¼ xx a 1 ee ½S ðÃÞ ðnÞn¼fT .7.5.2.12.3.1. 7.6.25.12.8. ak 2 C.6. for the modelling of ocean waves as a twodimensional cyclostationary random ﬁeld see [20.7. 7.6.1–6.25.8].6.9. In the case of complex data. In climatology and atmospheric science.8.101]) since the signals involved are typically ACS (see Section 5).2.23]. .27].6. Natural signals: modelling and analysis Cyclostationarity occurs in data arising from a variety of natural (not manmade) phenomena due to the presence of periodic mechanisms in the phenomena [6.ARTICLE IN PRESS W.22. FRESH ﬁltering consists of periodically or almostperiodically timevariant ﬁltering of xðtÞ and xÃ ðtÞ and .16. and signal detection and classiﬁcation are treated in Sections 8–11. In (5.6.6. is an ACS sequence. T 0 is the symbol period. the optimum ﬁltering is referred to (as ﬁrst suggested in [7.7. .17.31]) as cyclic Wiener ﬁltering (and also frequencyshift (FRESH) ﬁltering).9].23.9.2. 6.6.
5) ¼ S g ðf Þ 8g 2 F dx . the sums are extended to all frequency shifts s 2 G and Z 2 Gc such that g À s 2 AxxÃ and Z À g 2 Axx .6) s2G hc ðt.14.8b). adaptive demodulators.76.79–7. in terms of cyclic autocorrelations and cyclic spectra in [2. uÞxÃ ðuÞ du R R ﬁlter design [7.7. the sums are extended to all frequency shifts s 2 G and Z 2 G c such that g À s 2 Axx and Z À g 2 AxxÃ .8]. for each x xx g 2 F dxÃ .7.24.98].75.56.7.5]).7.57.69. xðtÞ ¼ dðtÞ þ nðtÞ where nðtÞ is the interference) [7. Eqs.31].31.21. for each g 2 F dx .A. 7. dx then it follows that the optimum ﬁlters are LAPTV: X hðt.19. uÞ ¼ hs ðt À uÞej2psu . Z Applications of FRESH ﬁltering to interference suppression have been considered in [7.5.7.2. (7.6). b Efag f½dðt þ tÞ À dðt þ tÞxÃ ðtÞg ¼ 0 8t 2 R 8t 2 R.7) By substituting (7.19. cycle frequencies of the signal of interest and interference. in (7.23.30. dx RbÃ xÃ ðtÞ x RÀb ðtÞÃ xx ð7:8bÞ be the LCL estimate of the desired signal dðtÞ obtained from the data xðtÞ. Let b dðtÞ9yðtÞ þ yc ðtÞ Z Z ¼ hðt. and (7.2. uÞxðuÞ du þ hc ðt. respectively.7.3a) and (7.7.7. b Efag f½dðt þ tÞ À dðt þ tÞxðtÞg ¼ 0 8t 2 R 8t 2 R. / Signal Processing 86 (2006) 639–697 adding the results.73.7.9. where the frequency shifts are chosen in accordance with the cycle frequencies of xðtÞ and dðtÞ (e.g. Adaptive FRESH ﬁltering is addressed in [7. in (7.46. .81]. Gardner et al.27.7.7.7.23] and. dx X s2G ð7:8aÞ ej2pst ½RgÀs ðtÞ hs ðtÞ xx X Z2Gc ZÀg ej2pZt ½RxxÃ ðtÞÃ hc ðtÞ Z þ ð7:1Þ ¼ Rg ðtÞ 8g 2 F dx .11. To minimize the meansquared error b Efag fjdðtÞ À dðtÞj2 g (7. (7. Z (7.81.8a). dx (7. Cyclic Wiener ﬁltering or FRESH ﬁltering can be recognized to be linked to FSE’s.1). RAKE ﬁlters.20. that is.7.7. (7. and LMMSE despreaders for recovery of baseband symbol data from PAM.7. and.7.7. ð7:9aÞ SgÀs ðf À sÞH s ðf À sÞ xx X Z2Gc þ S ZÀg ðZ À f ÞÃ H c ðf À ZÞ Z xxÃ ð7:9bÞ E fdðt þ tÞx ðtÞg ¼ fag ðÃÞ X g2F dxðÃÞ Rg ðÃÞ ðtÞej2pgt . ð7:3bÞ If xðtÞ and dðtÞ are singularly and jointly ACS. 7.3.7.21.85.7. uÞ ¼ X Z2Gc hc ðt À uÞej2pZu . we obtain the system of simultaneous where H s ðf Þ and H c ðf Þ are the Fourier transforms Z of hs ðtÞ and hc ðtÞ. The problem of optimum LPTV and LAPTV ﬁltering was ﬁrst addressed in [2.8a) and (7.7.74.ARTICLE IN PRESS 662 W.3b).51]: X ej2pst ½RgÀs ðtÞ hs ðtÞ xxÃ s2G þ X Z2Gc ZÀg ej2pZt ½Rxx ðtÞÃ hc ðtÞ Z ¼ Rg Ã ðtÞ 8g 2 F dxÃ ..58.7) into (7.4) Efag fxðt þ tÞxðÃÞ ðtÞg ¼ xx a2AxxðÃÞ where the fact that ¼ and Ra Ã x ðtÞ ¼ RÀaÃ ðtÞÃ is used and.8b) can be reexpressed in the frequency domain: X gÀs SxxÃ ðf À sÞH s ðf À sÞ s2G ð7:3aÞ þ X Z2Gc ZÀg S xx ðZ À f ÞÃ H c ðf À ZÞ Z ¼ S g Ã ðf Þ dx X s2G 8g 2 F dxÃ .77.7. X Ra ðÃÞ ðtÞej2pat .68.7. (7.94.2) a necessary and sufﬁcient condition is that the error signal be orthogonal to the data (orthogonality condition [2. 7.
101]. on blind and nonblind adaptive demodulation of PAM signals and LMMSE blind despreading of shortcode DSSS/CDMA signals using FSE’s and RAKE ﬁltering structures.8. if the desired and interfering signals have different cyclic parameters such as carrier frequency or baud rate.100].15].7.42.7. Cyclostationarity properties are exploited for synchronization in [8. consisting of desired signal plus interfering signal.8. / Signal Processing 86 (2006) 639–697 663 GMSK/GFSK. In particular.1. 9. Spectral line generation Let xðtÞ be a realvalued secondorder widesense ACS time series. if xðtÞ is the binary PAM signal deﬁned in (5.6) with qðtÞ real and duration limited to an interval strictly less than T 0 .8.34]. More generally. at a cycle frequency exhibited by the desired signal but not by the interference.8. Patents of inventions for the analysis and design of communications systems exploiting cyclostationarity are [7.8.7. etc.28.2) For communications signals.4 and 13). (see Section 5). In fact. Phaselock loops are analyzed in [8.7.96. Therefore.15–7.28–8.7. This signal selectivity by exploitation of cyclostationarity was ﬁrst suggested in [4.7. tÞ not containing any ﬁnitestrength additive sinewave component (see (3.95. pulse rate.7.7. as well as multicarrier or direct frequency diversity spread spectrum systems for adaptive transmission and combining.36].45.99–7.17. quadratic or higherorder nonlinear timeinvariant transformations of an ACS signal give rise to time series containing ﬁnitestrength additive sinewave components whose frequencies are the second or higherorder cycle frequencies of the original signal.A.1.3) 8.37.36].48. symbol rate. On direct frequencydiverse multicarrier transceivers. (8.2. 7.55. All synchronization schemes can be recognized to exploit the second. Patents of inventions exploiting the FRESH ﬁltering are [7.29. frame rate. and DSSS signals.7.34.44.87.7.39.11.8.17. ACS signals enable spectral lines to be generated by passage through a stable nonlinear timeinvariant transformation (see Sections 10.7. by deﬁnition. see [7. Therefore.7.27. then it .12.1–9.1–8.7.32–7.44) and (3. then they exhibit cyclostationarity at different cycle frequencies and.8.33.7].86.53.52]. and for separation of baseband symbol streams in overlapped signal environments.7.3.8]. Blind or nondataaided synchronization algorithms are described in [8. 1g.38.93]. Parameters that can be estimated in the presence of interference and/or high noise include carrier frequency ð8:1Þ where h‘x ðt.1].61.1) leads to a signal suitable for synchronization purposes. 7. Gardner et al.70.7.5.ARTICLE IN PRESS W.8. Synchronization 8. The spectral analysis of timing waveforms with regenerated spectral lines is treated in [8.26. sampling frequency. See also [2. That is. the synchronization signal x2 ðtÞ is periodic with period T 0 .46)): xðt þ tÞxðtÞ ¼ Efag fxðt þ tÞxðtÞg þ ‘x ðt.8.7.65–7. the secondorder lag product xðt þ tÞxðtÞ can be decomposed into the sum of its almostperiodic component and a residual term ‘x ðt. Signal parameter and waveform estimation Cyclostationarity properties can be exploited to design signal selective algorithms for signal parameter and waveform estimation [9. follows that X q2 ðt À kT 0 Þ x2 ðtÞ ¼ k2Z (8. tÞeÀj2pat it 0 8a 2 R.7. the extraction of the almostperiodic component in (8.35.13]. 0Þ ¼ 0.13–8. the cycle frequencies a 2 A are related to parameters such as sinewave carrier frequency.35.51.7.59.83.7.8.or higherorder cyclostationarity features of signals [8. According to the results of Section 3.36. For example.7.7.92. Patents on synchronization techniques exploiting cyclostationarity are [8. parameters of the desired signal can be extracted by estimating cyclic statistics of the received data.7. ¼ x a2A and ‘x ðt. tÞ X Ra ðtÞej2pat þ ‘x ðt.67]. and ak 2 fÀ1. consequently.5–8.20–8.7. see [7.7. tÞ.7.8.50.
and also see [3.9.9. 9.56]. On this subject see the general treatments [2.33. bandwidths.9. Patents of inventions on signal parameter and waveform estimation exploiting cyclostationarity are [9.56.38. Array processing problems.9.36–9.9.9.11.77).2.ARTICLE IN PRESS 664 W.80.57–9. have been considered.10. and also see [7.10.9.16.9.9.67. nðtÞ.2) on the basis of the observed noisy signals vðtÞ and zðtÞ vðtÞ ¼ xðtÞ þ nðtÞ.5.10.15.7) reveal the ability of cyclostationaritybased algorithms to be signal selective.9.12. / Signal Processing 86 (2006) 639–697 and phase.32.74.11]. 9.34.9.5) Therefore.7. and blind adaptation algorithms. Patents of inventions on blind system identiﬁcation exploiting cyclostationarity are [10.11].1.39–9.9.75.94.5.10.65.9.9.60.23. equivalently.19.7.39.95.19.11.47–9. (3.9. there are difﬁculties in application of some of this work—difﬁculties associated with lack of ergodic properties of stochasticprocess models adopted for polynomialphase signals and signals with coupled harmonics.72.7.15.9.97.81) specializes to Sa Ã ðf Þ ¼ Sa Ã ðf ÞHðf Þ.93].A.9. 10. time.9. periodically and almostperiodically timevariant systems.14.9. pulse rate and phase. y1 ¼ y. But. (10.9. (10.3.3) (10.9.20–9.82. under the assumption that nðtÞ (10.54.96].4).99. By assuming x1 ¼ x2 ¼ x.8.9.9. and so on. Eqs. yx xx (10. direction of arrival.30.9. vv xx nn Sa Ã ðf Þ ¼ S a Ã ðf Þ þ Sa Ã ðf Þ zv yx mn ¼ S a Ã ðf ÞHðf Þ þ S a Ã ðf Þ xx mn ð10:7Þ provided that xðtÞ is uncorrelated with both nðtÞ and mðtÞ.9.8. provided that there is at least one cycle frequency of the desired signal that is not shared by the disturbance.79. In fact. are treated in [7. both timedifferenceofarrival and frequencydifferenceof arrival are estimated.1) R of an LTI system with input/output relation yðtÞ ¼ hðtÞ xðtÞ (10.9. signal power level. have been developed for LTI systems.90].86–9.50.24. from the model (10.5.8. the harmonicresponse function: Z Hðf Þ9 hðtÞeÀj2pft dt (10.9.20.9.21.83].84.24.2.3) and (10.9.10.46. including spatial ﬁltering and direction ﬁnding.71. On this subject see the general treatments [2. Harmonics in additive and multiplicative noise are considered in [9. and timeinvariant.6) and (10.9.81. we obtain Sa Ã ðf Þ ¼ S a Ã ðf Þ þ S a Ã ðf Þ. 9.2.4) 10. where xðtÞ. In [9.73.9.2.3. LTIsystem identiﬁcation with noisymeasurements Cyclostationaritybased techniques can be exploited in channel identiﬁcation and equalization problems in order to separate the desired and disturbance contributions in noisy input/output measurements. techniques for noisy input/output measurement. Gardner et al.9.45.8.78.7.52.9.70.59. 7.57].9.10. y2 ¼ x.91. modulation indices.6) .9.6.100].66.9.13. Linear and nonlinear systems.2.31. 9.9.2.18.9. zðtÞ ¼ yðtÞ þ mðtÞ.2.68.2.13.17.88.55.7. General aspects Cyclostationaritybased techniques have been exploited for channel identiﬁcation and equalization [10.28.11].77].12.9.2.9. h1 ¼ h (LTI).9. The polynomialphase signal parameter estimation is addressed in [9.26–9.1–10. and choosing ðÃÞ to be conjugation.9. Channel identiﬁcation and equalization 10.7.76.and frequencydifference of arrival.10.7. 9. and mðtÞ are zeromean time series.9.9.9.36. and h2 ¼ d in (3.69.25.2.6].9. Also.1.24. Signal selective timedifferenceofarrival estimation algorithms are considered in [9.1.2.21.11.9.22.53. Let us consider the problem of estimating the impulseresponse function hðtÞ or.44.9.2. 10.9.9.
36. and source separation Signal detection techniques designed for cyclostationary signals take account of the periodicity or almost periodicity of the signal autocorrelation function [11.33.5. cyclostationarity properties of the output signal are suitable to be exploited for recovering both phase and magnitude of the system harmonicresponse function as ﬁrst observed in [10.25].19.2).8.13].24.7.13. S a Ã ðf Þ 0). and for nonGaussian noise.9. The detection problem for additive Gaussian noise is addressed in [11.55]. 10.10. Thus.11. identiﬁcation and equalization techniques for Volterra systems excited by ACS signals make use of higherorder cyclostationarity properties.11. .21.9.13. Nonlinearsystem identiﬁcation Let yðtÞ the output signal of a Volterra system excited by the input signal xðtÞ: þ1 XZ kn ðt1 . relative to stationary input signals. the harmonicmn response function for the system is given by Hðf Þ ¼ Sa Ã ðf Þ yx Sa Ã ðf Þ xx ¼ S a Ã ðf Þ zv .35.e.16.7. S a Ã ðf Þ vv (10.10. . there are potentially substantial advantages to using cyclostationary input signals. Blind LTIsystem identiﬁcation and equalization By specializing (3. this identiﬁcation method is highly tolerant to disturbances in practice. . Sa Ã ðf Þ 0) and nðtÞ and mðtÞ do nn not exhibit joint cyclostationarity with cycle frequency a (i.24. in [11.11. The problem of signal detection in cyclostationary noise is treated in [11.15. 11.8) provides a system identiﬁcation formula that is intrinsically immune to the effects of noise and interference as ﬁrst observed in [2.10.7].10.27.10.11.10. Thus. / Signal Processing 86 (2006) 639–697 665 does not exhibit cyclostationarity at cycle frequency a (i.18. referred to as the temporal moment function.11.9.11.5. tn Þxðt þ t1 Þ Á Á Á yðtÞ ¼ n¼1 Rn That is. for a ¼ 0.22.17]. yy xx (10.3)): X xðt þ t1 Þ Á Á Á xðt þ tn Þ ¼ Ra ðsÞej2pat þ ‘x ðt.10.10. The identiﬁcation of LTI systems based on noisy input/output measurements is considered in [2.18.10.10. 10.10. (10. 10.11.10.28–10.1–11.10.11.1. .5].10..19.84) to the case of the LTI system (10.41.16.4.12.12) Thus.20.10.47.42–10.9) xðt þ tn Þ dt1 Á Á Á dtn .50–10.2. 11. ð10:11Þ Accounting for the results of Sections 3.45. Signal detection and classiﬁcation. Gardner et al.14. x a2Ax (10. respectively.11.3 and 13. provided that a sufﬁciently long integration time is used for the cyclic spectral estimates.10. Singlecycle and multicycle detectors exploit one or multiple cycle frequencies.14.10.10.12.10.11.2.10. Hðf Þ can be expressed in terms of the cyclic spectra of the noisy input and output signals. which.10.9.9) and (10. the input lagproduct waveform can be decomposed into the sum of an almostperiodic component. unlike those in terms of autocorrelation functions and power spectra.48.32].3.21.8.8.10. reduces to the input/output relationship in terms of power spectra: S 0 Ã ðf Þ ¼ S0 Ã ðf ÞjHðf Þj2 .23.9–11.12. Tests for the presence of cyclostationarity are proposed in [11. Therefore.4. for purposes of Volterra system modelling and identiﬁcation as ﬁrst observed in [10.31] by . 2.46.10. and a residual term not containing any ﬁnitestrength additive sinewave component (see (13.53]. 11. Both timeinvariant and almostperiodically timevariant nonlinear systems are treated in [10.11.8. Cyclostationarity properties are exploited for blind identiﬁcation (without measurements of the system input) of linear systems and for blind equalization techniques in [9.59. 11.25. sÞ.31.e.16].15.10.ARTICLE IN PRESS W.11.36–10.39].11.10.10) From (10.12. preserve phase information of the harmonicresponse function Hðf Þ..8) 10.10) it follows that input/output relationships for LTI systems in terms of cyclic statistics.38. In fact.11.11.11. we get S a Ã ðf Þ ¼ Sa Ã ðf ÞHðf ÞH Ã ðf À aÞ yy xx (10.26.10.26.37.A. 10.20.5.
6.40.60.39]. Introduction As ﬁrst deﬁned in [13.6.12.12.2.52].13]. [16.37]: PðnÞ X k¼0 13.34.12.22.5.19.11.11.12.1) where xðnÞ and yðnÞ are the input and output signals.12.27.12–12.12.11.13] by exploiting the properties of the support of the spectral correlation function.36.9.22].45.30. even if they exhibit the same secondorder cyclostationarity ak ðnÞyðn À kÞ ¼ QðnÞ X m¼0 bm ðnÞxðn À mÞ.2.2.11.11.13. The problem of ﬁtting an AR model to data yðnÞ is equivalent to the problem of solving for a linear predictor for yðnÞ. If the bandwidth of Hðf Þ is smaller than the smallest nonzero secondorder cycle frequency of the input signal xðtÞ. see also [13. Periodic AR and ARMA systems are treated in [12.16. Motivations to study HOCS properties of signals include the following: (1) Signals not exhibiting secondorder cyclostationarity can exhibit HOCS [13. Patents on detection and signal recognition exploiting cyclostationarity are [11.7] for applications to econometrics.11.sspitech. 12.29] for the parameter estimation problem. see the general treatments [2.17.16. a signal xðtÞ is said to exhibit higherorder cyclostationarity (HOCS) if there exists a homogeneous nonlinear transformation of xðtÞ of order greater than two such that the output of this transformation contains ﬁnitestrength additive sinewave components.46. 12.12.69.ARTICLE IN PRESS 666 W.38].13. and [21. 11. automatic. 12. Higherorder statistics 12.5.16] for modelling of atmospheric and hydrologic signals.4.74.13. (12.43.36.34–12.1–12.35. however. periodic ARMA systems are a special case of discretetime periodically timevarying systems.12.1. On this subject see the general treatments [2.12.12. let us consider the input/output relationship for LTI systems in terms of cyclic spectra (10. Thus.27. The modelling and prediction problem is treated in [12.37.13]. [4.7.14. 4.23.6.4. [6. (3) The exploitation of HOCS can be useful for signal classiﬁcation [13.12. See also the URL http://www.1.12.3.32.72.16. and the coefﬁcients ak ðnÞ and bm ðnÞ and the orders PðnÞ and QðnÞ are periodic functions with the same period N 0 . and also see [3.21.22].51].2.11.12. respectively.4].31.3] for the prediction problem. When they are excited by an almostcyclostationary input signal. When they are excited by a stationary or cyclostationary input signal xðnÞ. Gardner et al.22.9.28–12.4] for application to modelling helicopter noise.41.19. On the subject of periodic AR and ARMA modelling and prediction.36. communicationsignal classiﬁcation software systems.47–12. 14. (2) Narrowband ﬁltering can destroy secondorder (widesense) cyclostationarity.9).15. In fact.9.9.13.30.5. and also see [3. 9.2.11.28–11.23.12. they give rise to a cyclostationary output signal yðnÞ.5.31].12.2. different communication signals.30]. / Signal Processing 86 (2006) 639–697 exploiting the asymptotic properties of the cyclic correlogram and in [11.8.49. . The problem of cyclostationary source separation is considered in [11. The parameter estimation problem is addressed in [12.11.11.3.38–12.12.12. com for information on general purpose. Modulation classiﬁcation techniques are proposed in [11. The signal yðtÞ.4.11.2. then the output signal yðtÞ does not exhibit secondorder widesense cyclostationarity. 13.13]).30.7.12. Speciﬁcally.6.1–12.53].13].37.12. 12. Periodic AR and ARMA modelling and prediction Periodic autoregressive (AR) and autoregressive moving average (ARMA) (discretetime) systems are characterized by input/output relationships described by difference equations with periodically timevarying coefﬁcients and system orders [12. 16.26.8.23–12.12. can exhibit HOCS (see also [3.50. where xðnÞ is the modelﬁttingerror time series. they give rise to an almostcyclostationary output signal.A.
. uKÀ1 T .1–13.9.13. (13. . in addition. for some s. The function S a ðf 0 Þ.40.15.13. . 10.4.2. .20]. .41. [10. sÞ9Efag fLx ðt. as outlined below. [4. sÞ9 N Y k¼1 xk k ðt þ tk Þ ðÃÞ (13.1) where s9½t1 .34.67.10. . rex ferred to as the reduceddimension CSCMF (RDCSCMF).13.18.8] for the spectral analysis of PAM signals.ARTICLE IN PRESS W.4) x RN where f 9½f 1 .9.6) x RNÀ1 . .9–21. Continuoustime time series Let us consider the column vector ðÃÞ ðÃÞ xðtÞ9½x1 1 ðtÞ.9. . in [13.2) contains a ﬁnitestrength additive sinewave component with frequency a.31].10.34] for applications to source separation.5] and developed in [13. f N T . .46] for applications to nonlinear system identiﬁcation and equalization.21. can be expressed as Z 0T 0 a 0 Sx ðf Þ ¼ Ra ðs0 ÞeÀj2pf s ds0 . xN N ðtÞT whose components are N not necessarily distinct complexvalued continuoustime timeseries and ðÃÞk represents optional complex conjugation of the kth signal xk ðtÞ.10.15.13.93. On the subject of higherorder cyclostationarity and its applications see [13.81] for the wavelet decomposition.9.28–11.30] for applications to signal classiﬁcation. The Ndimensional Fourier transform of the CTCMF Z T a Sx ðf Þ9 Ra ðsÞeÀj2pf s ds. [11.24.71. 13. N time series exhibit widesense joint Nthorder cyclostationarity with cycle frequency aa0 if. . whose amplitude and phase are the magnitude and phase of Ra ðsÞ.9. xN N ðt þ tN Þ (for the given conjugation conﬁguration).21. [11.13.10. 1T . x respectively. [9. [7. . . . [21.5) where 1 is the vector ½1. . . . (13. Gardner et al. can exhibit different cyclic features of higher order. .13. .13] for the higherorder characterization of generalized almostcyclostationary signals.10.61. tN T is not identically zero. Rx ðsÞ9 k¼1 t (13. Thus. . [8.15. (4) Exploitation of HOCS can be useful for many estimation problems.23] for applications in acoustics and mechanics.23. For treatments within the stochastic framework.26] or extend to higherorder statistics the link between the two frameworks discussed in Section 3. .14.17]. The Nthoder temporal crossmoment function (TCMF) is deﬁned by Rx ðt. different behaviors can be obtained for different conjugation conﬁgurations [13. .11.1.80.74] for blind adaptive detection.9. see [3. is called the Nthorder cyclic spectral crossmoment function (CSCMF) and can be written as Sa ðf Þ ¼ S a ðf 0 Þdðf T 1 À aÞ.56] for estimation issues.13. 9. uK T into the reduceddimension version u0 9½u1 . .55.69–9.A.23] for the cyclic higherorder properties of CPM signals. [5. for discrete and continuous time. .9.13. the lag product waveform Lx ðt. The N timeseries exhibit joint Nthorder widesense cyclostationarity with cycle frequency aa0 if at least one of the Nthorder cyclic temporal crossmoment functions (CTCMFs) * + N Y ðÃÞ a Àj2pat k xk ðt þ tk Þe .4.18.13.20] for applications to signal detection.2.85. 13.13. Moreover. .12. / Signal Processing 86 (2006) 639–697 667 properties. cyclic higherorder (joint) statistics for both continuoustime and discretetime time series are presented in the FOT probability framework as ﬁrst introduced for continuous time in [13. [11. Higherorder cyclic statistics In this section. and. 13.9. x x (13. and prime denotes the operator that transforms a vector u9½u1 .99] for application to signal parameter and waveform estimation.10. [15. . . sÞg X Ra ðsÞej2pat ¼ x a2Ax ð13:3Þ where Ax is the countable set (not depending on s) of the Nthorder cycle frequencies of the time ðÃÞ ðÃÞ series x1 1 ðt þ t1 Þ. .17].13. .14. see [13.
. ð13:13Þ . . . . .9) tÀT=2 and ðÀÞk denotes an optional minus sign that is linked to the optional conjugation ðÃÞk . . Eq. . Moreover.. xN N ðt þ tN ÞÞg Z ðÃÞ ðÃÞ ¼ gðx1 1 . . xNi Þ Â dx1r dx1i Á Á Á dxNr dxNi . . Ra ðsÞc0 or. i ¼ 1. . ðÀÞN ða À f T 1ÞÞ ðÃÞ 0 Â where NÀ1 Y k¼1 X k. but rather it oscillates.1) is.. it is shown that the RDCSCMF S b ðf 0 Þ x can contain impulsive terms if the vector f with P f N ¼ b À NÀ1 f k lies on the bsubmanifold. f k Þ9 tþT=2 xk ðsÞeÀj2pf k s ds (13. relation (13. the Nthorder generalization of the Cyclic Wiener Relation.8) and Nthorder cyclic temporal moment (13. by using hB ðtÞ9B rectðBtÞ jtjp1 2 (13.7) is the reduceddimension CTCMF (RDCTCMF). sÞ9cumfxk k ðt þ tk Þ.ARTICLE IN PRESS 668 W. . . . equivalently. . Let us note that in general the function Ra ðs0 Þ is x not absolutely integrable because it does not in general decay as ks0 k ! 1. Ng with p41 such that each sum ami ¼ P k2mi f k is a jmi jthorder cycle frequency of xðtÞ. the almostperiodic component extraction operator Efag fÁg extracts the frequencies of the 2Nvariate fractionoftime joint probability density function of the real and imaginary parts of the timeseries xk ðtÞ ðk ¼ 1.4) between the Nthorder cyclic spectral moment function (13. P i¼1 ðÃÞ with B91=T and rectðtÞ ¼ 1 if and rectðtÞ ¼ 0 if jtj41. Thus. .T ðu.Tk ðu. Gardner et al. k¼1 if there exists at least one partition fm1 .8) can be rewritten as 2 Z 1 tþZ=2 1 ðÃÞ a 0 S x ðf Þ ¼ lim lim ½ðxN N ðuÞ B!0 Z!1 Z tÀZ=2 BNÀ1 ÂeÀj2pðaÀf Â NÀ1 Y k¼1 0T ð13:12Þ where x9½o1 . smi Þ . . Such a property is the generalization to the order N42 of the spectral correlation property of signals that exhibit secondorder cyclostationarity. where jmi j is the number of elements in mi . . .A. pg. . xN N Þ ðÃÞ ðÃÞ 1Þu Þ hB ðuÞ ½ðxk k ðuÞeÀj2pf k u Þ hB ðuÞ du ð13:11Þ ðÃÞ which is the Nthorder temporal crossmoment of lowpassﬁltered versions of the frequencyshifted ðÃÞ signals xk k ðtÞeÀj2pf k t when the sum of the frequency shifts is equal to a and the bandwidth B approaches zero. S a ðf 0 Þ can contain impulses and.10) [13. .4]. consex quently. . .13]. x In [13. . is nonzero. S a ðf 0 Þc0) if and x x only if the Nthorder temporal crossmoment of their spectral components at frequencies f k .T ðt. whose sum is equal to a. (13. In fact. (13. . mp g of f1. x1i .12).5). . . . . In the spectralfrequency domain. Sa ðf Þ can contain products of impulses. The RDCSCMF can also be expressed as Z 1 tþZ=2 1 a 0 S x ðf Þ ¼ lim lim T!1 Z!1 Z tÀZ=2 T N ÂX N. xNr . at ﬁrst pointed out in R2N Âf fag ðtþt1 Þx1i ðtþt1 ÞÁÁÁxNr ðtþtN ÞxNi ðtþtN Þ x1r ðx1r . In (13. P is the set of distinct partitions of f1. ðÀÞk f k Þ du. Ng. i. . . . . . . . NÞ according to Efag fgðx1 1 ðt þ t1 Þ. Ng @N loge ¼ ðÀjÞN @o Á Á Á @oN ( 1 " #) N X ðÃÞk Efag exp j ok xk ðt þ tk Þ k¼1 x¼0 " # p X Y pÀ1 ¼ ðÀ1Þ ðp À 1Þ! Rxmi ðt. each constituted by the subsets fmi .8) reveals that N timeseries exhibit joint Nthorder widesense cyclostationarity with cycle frequency a (i. . . oN T . / Signal Processing 86 (2006) 639–697 where Ra ðs0 Þ9Ra ðsÞjtN ¼0 x x (13. k ¼ 1.e. (13. ðÃÞ ð13:8Þ Z X k. xmi is the jmi jdimensional vector whose components are those of x having indices in mi . Eq. a wellbehaved function can be introduced starting from the Nthorder temporal crosscumulant function (TCCF): Cx ðt. . . .e.
sÞeÀj2pbt it x (13. T!1 T ðÃÞ X k. . / Signal Processing 86 (2006) 639–697 669 where xkr ðtÞ9Re½xk ðtÞ. k ¼ 1. . ðÀÞN ða À f T 1ÞÞ. The Nfold discrete Fourier transform of the CTCMF X e a T ee ea Sx ðmÞ9 Rx ðmÞeÀj2pm m . . . (13. . xN Þ9 exp j ok xk . . . the CCP Pb ðf 0 Þ is coincident with the RDx CSCMF S b ðf 0 Þ. 13. . and C b ðs0 Þ9Cb ðsÞjtN ¼0 x x (13.17) is the Nthorder cyclic crosspolyspectrum (CCP). .13.A. k ¼ 1. Discretetime time series In accordance with the deﬁnition for the continuoustime case. Eq.13.18) is called the reduceddimension CTCCF (RDCTCCF). there x exists an 40 such that jC b ðs0 Þj ¼ oðks0 kÀNþ1À Þ as x ks0 k ! 1. . N À 1g.ARTICLE IN PRESS W. xki 9Im½xk . . The cyclic crosspolyspectrum is a wellbehaved function under the mild conditions that the timeseries xN ðtÞ and xk ðt þ tk Þ ðk ¼ 1. Gardner et al. xðnÞ9 ðÃÞ ðÃÞ ½x1 1 ðnÞ. moreover. and " # N X ðÃÞk ðÃÞN ðÃÞ1 gðx1 .13.17]. ðÀÞk f k Þ. . in [13. It can be written as x Pb ðf Þ ¼ Pb ðf 0 Þdðf T 1 À bÞ. 1½Þ if a a 2 2 at least one of the Nthorder CTCMF’s a ee Rx ðmÞ9 lim N N X Y ðÃÞ 1 a xk k ðn þ mk ÞeÀj2pen N!1 2N þ 1 n¼ÀN k¼1 In fact.19) (13. . xkr 9 Re½xk . N not necessarily distinct discretetime complexvalued time series xk ðnÞ. .22) . . The magnitude and phase of the CTCMF (13. the stationary FOT expectation operation can be adopted as T ! 1. .16) C b ðs0 ÞeÀj2pf x ds0 (13.21) are the amplitude and phase of the sinewave component with frequency e contained a in the discretetime lag product whose temporal expected value is the discretetime Nthorder TCMF. N À 1g ð13:20Þ which is the Nthorder temporal crosscumulant of lowpass ﬁltered versions of the frequencyshifted ðÃÞ signals xk k ðtÞeÀj2pf k t when the sum of the frequency shifts is equal to b and the bandwidth B approaches zero. .Tk ðt. in the computation of the cumulant in (13. in more detail. . exhibit joint Nthorder widesense cyclostationarity with cycle frequency ea0 ðe 2 ½À1. x The CCP can also be expressed as Pb ðf 0 Þ ¼ lim x 1 0 ðÃÞN cumfX N. m2ZN where. n 2 Z. by taking the Ndimensional Fourier transform of the coefﬁcient of the Fourier series expansion of the almostperiodic function (13. . Furthermore. In fact.5] and then.21). . Eq.T ðt. except on a bsubmanifold.9. x x where Pb ðf 0 Þ9 x Z RNÀ1 0 T s0 (13. . .21) is not identically zero. ½ðxk k ðtÞeÀj2pf k t Þ hB ðtÞ. the Nthorder temporal cumulant function of a time series provides a mathematical characterization of the notion of a pure Nthorder sinewave.2. mN T . ð13:19Þ (13.2. . one obtains the Nthorder cyclic spectral crosscumulant function (CSCCF) Pb ðf Þ.14) k¼1 can be rewritten as Pb ðf 0 Þ ¼ lim x 1 NÀ1 B!0 B cumf½ðxN N ðtÞeÀj2pðaÀf ðÃÞ ðÃÞ 0T 1Þt Þ hB ðtÞ. N À 1Þ are asymptotically (jtk j ! 1) independent (in the FOT probability sense) so that C b ðs0 Þ ! 0 as ks0 k ! 1 and. xki ðtÞ9Im½xk ðtÞ. In (13.19).12) Cb ðsÞ9hCx ðt. It is that part of the sinewave present in the Nthorder lag product waveform that remains after removal of all parts that result from products of sinewaves in lower order lag products obtained by factoring the Nthorder product. (13. .15. As ﬁrst shown in [13.15) which is referred to as the Nthorder cyclic temporal crosscumulant function (CTCCF). (13. xN N ðnÞT and m9½m1 .19) reveals that the CCP of N time series is the Nthorder crosscumulant of their spectral components at frequencies f k whose sum is equal to b.13.
14.14. nN T .14.27.21. Analogous relations can be stated for the reduceddimension statistics.25) e Once the Nthorder discretetime TCCF Cx ðk.14. and so on [2.14.8.2.3].17. (13. Cyclostationarity has been exploited in acoustics and mechanics for modelling road trafﬁc noise [15. . Applications to acoustics and mechanics Applications of cyclostationarity to acoustics and mechanics are in [15. Consequently. pistons. mÞ is deﬁned analogously to the continuoustime case.2.31].6]. it can be easily shown that the following periodicity properties hold: a aþp ee ee Rx ðmÞ ¼ Rx ðmÞ.14. In circuit theory. cyclostationarity has been exploited in modelling noise [14. . such as engines. In mechanical systems with moving parts. (13.25].59. for analyzing music signals [15. p 2 Z. is called the Nthorder CSCMF and can be written as þ1 X a a ee ee dðe À mT 1 À rÞ. propellers.4–14.10.16. if some parameter such as speed. is given by þ1 X e e eb eb Px ðmÞ ¼ Px ðm0 Þ dðe À mT 1 À rÞ.13.30) (13.19. and control are in [14.15.4.1. referred to as the Nthorder CSCCF.1. which can be expressed as the ðN À 1Þfold discrete Fourier transform X 0T 0 a a ee ee S x ðm0 Þ ¼ Rx ðm0 ÞeÀj2pm m (13.28) Finally. Even if the almostcyclostationary component has a power smaller than the power of the almostperiodic component.24].10.24.14. Applications to circuits.14.26) Its discrete Fourier transform.5].15. which can be expressed as the ðN À 1Þfold discrete Fourier transform of the Nthorder RDCTCCF e e eb eb C x ðm0 Þ9Cx ðmÞjmN ¼0 .14.13.31].10.15. cyclostationary or almost cyclostationary signals arise in dealing with periodically or almostperiodically time variant systems (see Section 3. In control theory.15.14. 14.5.A.6) [14. Often.18. 15.20.9. systems. .28.23) a ee where the function S x ðm 0 Þ is the Nthorder RDCSCMF. Gardner et al.26.14.1–15.24]. e e eb e bþp Cx ðmÞ ¼ Cx ðmÞ.1. / Signal Processing 86 (2006) 639–697 where m9½n1 .ARTICLE IN PRESS 670 W.15. systems.8.14.15.16.8. p 2 Z. q 2 ZN .14.10. mÞeÀj2pbn . cyclostationarity has been exploited in [14. 14.4. (13. 14.15. N!1 2N þ 1 n¼ÀN (13.1–14.14.19–15. and control Applications of cyclostationarity to circuits. (13. vibration signals exhibit periodic behavior of one type or another with periods related to the engine cycle. the Nthorder CTCCF is given by e eb Cx ðmÞ9 lim N X 1 e e Cx ðn.14. . p 2 Z. (13. In system theory. Patents on applications of cyclostationarity to system and circuit analysis and design are [14.23]. 15.28].12.5.14.23. then the dynamic physical processes generate vibrations that can be modelled as originating from periodic mechanisms such as rotation and reciprocation of gears.14.21. for example it can be successfully exploited in early diagnosis of gear faults [15. e e eb e bþp Px ðmÞ ¼ Px ðm þ qÞ. and for describing the vibration signals in mechanical systems.29) (13. On this subject.3.14.27) e eb where the function Px ðm0 Þ is referred to as the Nthorder CCP. the observed vibration signals contain both an almostperiodic and an almostcyclostationary component [15.29.15. also see [9.11. chains.18.14.32) a aþp ee ee Sx ðmÞ ¼ Sx ðm þ qÞ.15.31) (13.15.13. shafts.3].15. it can be useful. p 2 Z. . q 2 ZN .24) m0 2ZNÀ1 of the Nthorder RDCTCMF a a ee ee Rx ðm0 Þ9Rx ðmÞjmN ¼0 .22]. belts.24.15.15.15.9. and load torque can be assumed to be constant. 13. temperature.16. b r¼À1 15. a Sx ðmÞ ¼ Sx ðm0 Þ r¼À1 14.15.
1. Methods of averaging were developed for estimating the generalized spectrum that allow for the meaningful characterization of phase information when classic assumptions of stationarity do not hold.4.A.1–20. Speciﬁcally.6.8].8]. n02 þ N 02 Þg ¼ Efxðn1 .40]. N 02 Þ if and only if there exists no smaller N 01 40 and N 02 40 for which the mean and correlation satisfy Efxðn1 þ N 01 . transformations of ACS signals.1–19.21. the repetitive patterns of openings and closures of markets. Cyclostationary random for all ﬁelds are treated in [20. in mobile communications when the product of transmittedsignal bandwidth and observation interval is not much smaller than the ratio between the propagation speed in the medium and the relative 18.1–18.2]. Applications of the concept of spectral redundancy (spectral correlation or cyclostationarity) have been proposed in medical image signal processing [17. are sources of periodic variations in ﬁnancialmarket volatility and other statistical parameters.1–21. n2 .1–17. Gardner et al.10].5] and limitations of previously proposed approaches are exposed.49.1) Efxðn1 þ N 01 . n1 . n02 17.1. This has led to many signiﬁcant performance improvements in ultrasonic tissue characterization. but not almostperiodically timevariant. 21. (20. Furthermore.2].15.12.17.19.13].20. neglecting the periodic behavior gives rise to a loss in forecast efﬁciency [2. General aspects Generalizations of cyclostationary processes and time series are treated in [21. Nonstationary signals that are not ACS can arise from linear timevariant. n2 þ N 02 ÞxÃ ðn01 þ N 01 .43. 16. Patents on applications of cyclostationarity to engine diagnosis are [15. In [21.7].18].18]. The problem of statistical function estimation for general nonstationary persistent signals is addressed in [21. Level crossings Level crossings of cyclostationary signals have been characterized in [18. On this subject.6]. for example.6. Queueing theory in car trafﬁc is treated in [19. a random ﬁeld xðn1 . 19. particularly for liver and breast tissues [17.15.17]. n2 ÞxÃ ðn01 .2–21.4.10–17. 14.1–16. n2 þ N 02 Þg ¼ Efxðn1 . and nondestructive evaluation [17. also see [14.21. seasonally varying preferences. On this subject see also [12. A patent on the application of cyclostationarity to cholesterol detection is [17. n2 Þ indexed on Z2 is called strongly periodically correlated with period ðN 01 .8].8. the class of the correlation autoregressive processes is studied. Applications to biology Applications of cyclostationarity to biology are in [17.5].ARTICLE IN PRESS W. 20. Autoregressive models with periodically varying parameters provide appropriate descriptions of seasonally varying economic time series [16.7]. Cyclostationary random ﬁelds A periodically correlated or cyclostationary random ﬁeld is a secondorder random ﬁeld whose mean and correlation have periodic structure [20.17.16.10]. the number of active markets throughout the day.21. Applications to econometrics In highfrequency ﬁnancial time series. Generalizations of cyclostationarity 21.4. Queueing Exploitation of cyclostationarity in queueing theory in computer networks is treated in [19.21. Such transformations occur. such as asset return.17].33. n02 Þg ð20:2Þ 2 Z.16. and so forth. n2 Þg. n01 .4]. / Signal Processing 86 (2006) 639–697 671 Additional applications in the ﬁeld of acoustics and mechanics can be found in [4.5.4. .
14].12. Moreover. The estimation of the cyclic autocorrelation function for GACS processes is addressed in [21.10. n 2 I.35. rather. and the spectrally correlated signals [21. the support in the ða.2) is not necessarily countable.7.2. and the coefﬁcients. f 2 Þ9EfX ðf 1 ÞX Ã ðf 2 Þg X ðnÞ ¼ SxxÃ ðf 1 Þdðf 2 À Cn ðf 1 ÞÞ. the set [ At A9 t2R (21. the set A turns out to be countable in this case (see (3.21. referred xx to as the spectral correlation density functions.10]: X ðnÞ Efag fxðt þ tÞxÃ ðtÞg ¼ RxxÃ ðtÞej2pan ðtÞt .5) In (21.21.15–21.21.1) n2I signals.12]. that vary slowly with time cannot be modelled as ACS but. such as the carrier frequency and the baud rate. n2I For GACS signals. are given by RðnÞÃ ðtÞ9hxðt þ tÞxÃ ðtÞeÀj2pan ðtÞt it . the curves f 2 ¼ Cn ðf 1 Þ.21. (21.17] in the stochastic process framework. ¼ n2I ð21:4Þ That is.ARTICLE IN PRESS 672 W. I is a countable set. The problem of sampling a GACS signal is considered in [21.9]. / Signal Processing 86 (2006) 639–697 radial speed between transmitter and receiver [8.11. Gardner et al. Spectrally correlated signals A continuoustime complexvalued secondorder harmonizable stochastic process xðtÞ is said to `ve be spectrally correlated (SC) if its Loe bifrequency spectrum can be expressed as [21. referred to as generalized cyclic autocorrelation functions.6. Linear ﬁltering is addressed in [21. a survey of GACS signals is provided. In [21. The ACS signals are obtained as the special case of GACS signals for which the functions an ðtÞ are constant with respect to t and are equal to the cycle frequencies. it results that At 9fa 2 R : Ra Ã ðtÞa0g xx [ fa 2 R : a ¼ an ðtÞg.11] where the concept of expectation of the impulseresponse function in the FOT probability framework is also introduced. tÞ plane consists of lines parallel to the t axis and the generalized cyclic autocorrelation functions are coincident with the cyclic autocorrelation functions. Generalized almostcyclostationary signals A continuoustime complexvalued time series xðtÞ is said to be widesense generalized almostcyclostationary (GACS) if the almostperiodic component of its secondorder lag product admits a (generalized) Fourier series expansion with both coefﬁcients and frequencies depending on the lag parameter t [21. describe the support of SxxÃ ðf 1 . for GACS signals.13)). communications signals with parameters. xx (21.21.13] where it is shown that the discretetime signal obtained by uniformly sampling a continuoustime GACS signal is an ACS signal.9].21. The higherorder characterization of the GACS signals in the FOT probability framework is provided in [21.1).9–21.21.9. The case of linear support curves is considered in [21. Moreover.6. In such a case the support of the cyclic autocorrelation function in the ða.14] SxxÃ ðf 1 .16].15.3) xx n2I ð21:6Þ Moreover. n 2 I.3. the cyclic autocorrelation function can be expressed in terms of the generalized cyclic autocorrelation functions by the following relationship: X ðnÞ Ra Ã ðtÞ ¼ RxxÃ ðtÞdaÀan ðtÞ . can be modelled as generalized almostcyclostationary if the observation interval is large enough [21. (21. the frequencies an ðtÞ are referred to as lagdependent cycle frequencies. 21. and the functions S ðnÞÃ ðf 1 Þ.13. For the GACS where I is a countable set. The ACS processes are obtained as the special case of . 21.A.21.17].21. Two models for the output signals of such transformations are the generalized almostcyclostationary signals [21. f 2 Þ. 21. tÞ plane of the cyclic autocorrelation function Ra Ã ðtÞ xx consists of a countable set of curves described by the equations a ¼ an ðtÞ.11]. ` describe the density of the Loeve bifrequency spectrum on its support curves.
6] W. Minneapolis. Gardner. 1–90. Department of Electrical and Computer Engineering. second ed. [2. Dragan. Periodically correlated random processes and transformations with periodically varying parameters. Periodically correlated random sequences. Otbor Peredacha Inform. [2.). Nauchn. Exploitation of spectral redundancy in cyclostationary signals. Radiotekh. 1972. Signal Processing 23 (1991) 273–292. On the theory of cyclostationary signals. Phys.A. [2.1960). Gardner. On periodic nonstationary processes. University of California. pp. Digital Signal Processing Techniques and Applications.A. Dissertation. Williams (Eds. [3.1] H. Gardner et al.K. 23 (Part 2) (1968) 136–141. 22 (1969) 27–33 (in Russian).M. 18 (1973) 1426–1429 (English translation of Radiotekh. Theory Probab. Kayatskas. Kiev.A. Wiley. [2. Department of Electrical Engineering and Computer Science. University of Massachusetts. Cyclostationarity in Communications and Signal Processing. Miamee. Lebedev. New York. CA. Spooner. in: C.D. Correlation Theory of Stationary and Related Random Functions. Macmillan.T.A.2] V. Gardner.3] K. Fractionoftime probability for timeseries that exhibit cyclostationarity. The bifrequency spectral correlation density function X ðnÞ ¯ SxxÃ ðf 1 .7] V. Gardner. Boca Raton. Radio Eng. [2. [3. Cyclostationary signal analysis. f 2 Þ9 S xxÃ ðf 1 Þdf 2 ÀCn ðf 1 Þ (21. 9 (5) (1966).). (1963) 137–177 (in Russian).2] W.5] E.7] A. 18. Electron. 8 (1991) 14–36. D. f 2 Þ can be reliably estimated by the timesmoothed crossperiodogram only if the slope of the support curves is not too far from unity. [2. 1969.12] W. 1990 (Chapter 12).14] it is shown that. Dissertation. Electron. 1985 (Chapter 12). [2.S. 1987. Introduction to Random Processes with Applications to Signals and Systems. [2. 1982.A. Englewood Cliffs.M. New York. Periodically correlated random processes.8] A. [2. The spectral correlation theory of cyclostationary time series. Correlations and spectra for nonstationary random functions. Giannakis. NC. Structure and Representations of Models of Stochastic Signals. 1–6. A. Math. Izv. 1994. Math. Gardner.5] W. Sov. Ph. FL and New York. Gardner. Representation and estimation of cyclostationary processes. Mag. Gladyshev. References [2.1] L.8] W. Academic Press. VUZ Radiotekhn. 1980.A. [2. Yaglom. Periodically and almost periodically correlated random processes with continuous time parameter. Brown. Gardner. vols.4] J. Willis. Naukova Dumka.3] E. Gardner.10] W. In [21. 1987.G. 16 (77) (1962) 1–21. 1994. 1957. New York.15] W.D.L. Dragan.I.A. Phys. An investigation of periodically correlated stochastic processes. Hurd. CRC Press and IEEE Press. Ser.G. in: W. Brown.D. Hurd. Markelov. Example of formation of periodically correlated random processes.A. [3. Gardner. Periodically Correlated Random Sequences: Spectral Theory and Practice. Gardner (Ed. PrenticeHall. 1998 (Chapter 17). Ph.P. [3. [3. Exploitation of spectral correlation in cyclostationary signals. 95616.A. Chen.9] W.A. pp.M. 2 (1961) 385–388. 2 (1959) 32–34 (in Russian). Comput. Elektron. New York. Cyclostationary signal processing. IEEE Signal Process. The Digital Signal Processing Handbook.A.9] Y. Dissertation.L. An introduction to cyclostationary signals. [3. Gardner. Part 4. Duke University. Gudzenko. W. 4 (1959) 220–224 (in Russian). IEEE Press. [2. Telecommun.17] G. Shchk. [2. 2006.4] Y. [3. Dokl.A.16] W. . SIPL821. Introduction to Random Processes with Applications to Signals and Systems. C. Biometrika 51 (1964) 399–404..A.L. [2. if the location of the support curves is unknown. Dokl. University of California.B. Department of Electrical and Computer Engineering. The Random Processes Tutor. New York.P.11] W.A. Leondes (Ed.6] D. 1991. Voychishin.A. 1988. The statistics of a particular nonhomogeneous Poisson process. Davis. Signal Processing 11 (1986) 13–36 (Erratum: Signal Processing 11 405. [2. McGrawHill. Vysshch. Extrusions and phases of the periodically nonstationary random process. Appl.A. Gardner.18] H. [2.7) n2I ` is the density of the Loeve bifrequency spectrum on its support curves f 2 ¼ Cn ðf 1 Þ. / Signal Processing 86 (2006) 639–697 673 SC processes for which the support curves are lines with unity slope. [3. the bifrequency spectral correlation ¯ density function SxxÃ ðf 1 . ´ ´ [3.ARTICLE IN PRESS W.G. Elektron. in: V.13] W.) Received EURASIP’s Best Paper of the Year award. New York. Ph. Springer. Radio Eng.A.P. Dragan. I and II. C. NJ. reprinted as Signal and Image Processing Lab Technical Report No. in: Fourth Annual ASSP Workshop on Spectrum Estimation and Modeling. 1986. Durham. Gladyshev. [2. Davis. Radio Eng. Kampe de Feriet. CA. Y. On random processes having nonstationarity of periodic character. McGrawHill. New York. Madisetti.14] W. [2. Statistical Spectral Analysis: A Nonprobabilistic Theory.).K. MN.B.
[3. Math. [3. [3. Math. 57 (1979) 21–25 (in Russian). Systems Sci. Theory IT24 (1978) 8–22. 195–213. [3. Ghozati.I.18] Y.N. A paradox of the rhythmic model.37] M. (Kiev) (1972) 11–20 (in Russian).A.A. Otbor Peredacha Inform.G. Nauk Ukrain 8 (1978) 679–684 (in Russian). 30 (1971) 16–24 (in Russian). Dokl. Theory IT17 (1971) 143–149. Ogura. Characterization of cyclostationary random signal processes. Dragan. Elektron. Electr.20] H. IEEE Trans. Statistical characteristics of a nonstationary Normal process and ways of its modeling. 296 (1973) 179–190. Generation of a class of nonstationary random processes. Yoshida. Nauk USSR 8 (1984) 8–10 (in Russian). Otbor Peredacha Inform. Dragan.ARTICLE IN PRESS 674 W. [3. [3. Rybakov. IEEE Trans. Yavorskii. Correlation coefﬁcients of the envelope and phase of a signal for periodically nonstationary Gaussian noise.44] W. Salehi. Comm.P. SSR Ser. Radzievskii. Gupta. Dragan. Theory IT21 (1975) 4–14.13] K. On the spectral representation and related properties of a periodically correlated stochastic process. Bilyik. Rybakov. Stationariness provided by ﬁltration to a periodic nonstationary random process. J. 9 (1978) 935–947. 72 (1985) 3–17 (in Russian). III. [3.34] B. Spectral representation of a periodic nonstationary random process.S. Dokl. J. Dragan. Salehi.32] I.S. Honda. 30 (1985) 1096–1104 (in Russian).O. [3.). [3. Inform. Gardner. Dragan. 31 (1972) 21–27 (in Russian). Piranashvili. 1980.31] P.S.43] W. Periodically correlated processes with discontinuous correlation functions.12] Y.L.A.P.40] Y. I. Gardner et al. Algorithms Prod. B 1 (122) (1981) 87–95 (in Russian). Y. 35–36 (1981) 96–97 (English translation of Elektrosvyaz and Radiotekhnika). Sample periodicity of periodically correlated processes. Common pitfalls in the application of stationary process theory to timesampled and modulated signals. Otbor Peredacha Inform. Appl. pp.N. V. 41 (1974) 11–15 (in Russian).15] K. [3. Kamolor. 33 (1972) 12–16 (in Russian). Nauk Litov. On foundations of the stochastic model of rhythmic phenomena. Eng.P. Melititskiy. Y.28] M. [3. [3. L. Lugannani.A. Kochel.41] I. H. Rootenberg.E. Properties of periodically nonstationary processes. Dragan. Otbor Peredacha Inform. Commun. Some transformations of periodic inhomogeneous random ﬁelds. S.I. Keio Math. Informationsverarb. Sound Vib. 1983.19] H. SIAM J. Telecommun. Inform. 26 (1) (1974) 203–212. Correlation properties of the envelope and phase of a periodically nonstationary Normal process. COM35 (1987) 74–78. Dragan. Dragan. Statistical analysis of periodically correlated random processes. [3. [3.39] Y. Hurd.14] K. [3. L.25] Y. Sample stability of periodically correlated pulse trains.27] J. Ogura. Radio Eng.24] Z. V. 167–179. [3. Arad. Process. Some properties of the stochastic model of systems with periodically varying parameters. On the prediction of periodically correlated stochastic processes. Franks. Springer (Lublin). Commun. Stationarizing properties of random shifts. Miamee.26] W.P.33] A. Krishnaiah (Ed. [3. [3. 9 (1974) 804–807. Otbor Peredacha Inform.S. Inst.P. [3. NorthHolland Publishing Company. Statistical analysis of periodic random processes.L. Pourahmadi. Politekh. Appl. Spectral representation of periodically correlated stochastic processes and approximate Fourier series. Certain properties of periodic nonstationary random processes. Commun. [3. Y. Gardner. IEEE Trans. IEEE Trans. Berlin.36] H. Sakharth. 44 (1975) 3–14 (in Russian).P.42] B. 16 (1980) 553–567 (in German). Honda. Trans. Inst.A. 45 (1975) 7–20 (in Russian). [3. pp. Franklin Inst. in: P. Rep. Samecn. [3. Theory Probab. 23 (1972) 319–329. [3. Radiotekhnika 41 (1986) 3–7 (in Russian). in: Probability Theory on Vector Spaces. The elimination of rhythm from periodically correlated random processes. [3.P. Radio Electron. 16 (1973) 82–89. Time series analysis of a periodic stationary random process.P. Rice’s representation for cyclostationary processes. H.17] V. Otbor Peredacha Inform.11] H. Shrom 3 (176). Gardner. / Signal Processing 86 (2006) 639–697 [3. On subordination and linear transformation of harmonizable and periodically correlated processes. Elektron. Periodically stationary Markovian decision models. I. Sem. New York. [3.I. Otbor Peredacha Inform. Periodic and periodically nonstationary random processes. Syst. [3. Keio Math. Otbor Peredacha Inform.R. Lenin.A. (1975) 93–105 (in Russian). IEEE Trans. Inform. Harmonizability and spectral distribution of random processes with ﬁnite mean power. Voychishin. The representation of a periodically correlated random process by stationary components. Stationarizable random processes.A.23] W.21] Y.22] Y. Kavtaradze. [3.16] R.35] I. 6 (1981) 11–16. Gardner. [3. Japan 365A (1982) 22–29 (in Japanese). Trans. Trudy Akad. General properties of the stochastic model of rhythmic phenomena.10] Y.G. [3. The spectral properties of periodically correlated stochastic processes. . 71 (1985) 20–29 (in Russian). Kybernet. Akad. [3.P. Kapustinskas. [3. Yavorskii. [3. Rep.30] A. Dragan.P. Radiotekh. COM35 (1987) 529–534. IECE of Japan E 65 (1982) 723–729. [3. Dragan. Internat.38] A.29] I.O. Multivariate Analysis—V. J. Honda. 5 (1980) 13–18. Otbor Peredacha Inform. Yavorskii. [3. Amsterdam.N. Mehk.A. Hurd. Voychishin.N. Kolmogorov diameters of a class of random processes.
Hurd.G.A.G. [3. Gibson (Ed. Dorogovtsev. Kallianpur. Principles of Statistical Radiophysics. Stochastic Process.N. Theory 41 (3) (1995) 628–642. Dorogovtsev. IEEE Trans. Stochastic Process. Appl.Y. Zhang.46] I. ˘ ´ [3. [3.I. Miamee (Ed. [3. Extreme value theory for processes with periodic variances. Gardner. Alekseev.63] G.A. Random processes.56] A.54] A. 32 (1987) 100–106. ( 42) (1990) 35–42. J. [3. G. Virginia. Salehi. The Communications Handbook. Multivariate Anal. 1990. Akinshin. 50 (1994) 315–330. Gardner. H. [3. in: A. Correlation Theory of Random Processes. Appl. [3.M. Hurd. Appl. Statist. Signal Process. Vaidyanathan.P. Degrees of cyclostationarity and their application to signal detection and estimation.53] V. Multivariate Anal. Comput. FerrariTrecate. Periodically correlated processes and their relationship to L2 ½0. [3.G.K. V.G. V.47] N. Bakuriani. 50 (4) (1990) 1194–1199. Periodically correlated processes and their spectrum. 30 (1993) 82–97. Radioelectron. Signal Process. Periodically correlated processes and their stationary dilations. Veroyatnost.J. J. Yavorskii. Tvalued stationary sequences. Swift.L. 47 (7) (1999) 2041–2043.Y. V. Salehi. Commun. The existence of periodic and stationary regimes of discrete dynamical systems in a Banach space.A. S. CRC Press and IEEE Press. [3. Secondorder statistical characterization of the ﬁlter bank and its elements. Theory 43 (6) (1997) 2033–2035.51] H.L.73] L. Makagon.67] V. Dorogovtsev.49] S. pp.I.ARTICLE IN PRESS W. Gardner et al. [3.66] D. 1991. 1997 (Chapter 5). Otbor Peredacha Inform. A probabilistic model of non Gaussian periodically nonstationary radio signals.L. IEEE Trans.55] A. 1991. Commun. [3. [3. P.65] H. 1. Math. Dissertation for the Habilitation to Direct Research. Lall. Contributions to the spectral analysis of nonstationary processes. Melititskaya. SIAM J.60] G.S. Almost periodic harmonizable processes. [3. Stationary and periodic solutions of a stochastic difference equation in a Banach space.D.76] B. 47 (6) (1999) 1745–1749. J. Dehay.58] D. Inform. Georgian Math. [3. Math. C.A. Dorogovtsev. 2. Appl. 1996.74] H.I. 1988. Statistical analysis of periodically correlated vector random processes. VA. Mikhailov. Lambert. Tatarskii. Mat. Mosionzhik. Rumyantsev. World Scientiﬁc. J. 29 (1989) 53–67.L.59] H. Maximum entropy modeling of periodically correlated processes. [3.D.48] A. [3. Miamee (Ed.77] A.and nearly periodicallycorrelated random processes.).E. Sov. [3. Technol.71] S. J. Electron. Universite des Sciences et Techniques de Lille Flandres Artois.Y. Yavorskii. Italy.G. Probab. World Scientiﬁc. Signal Processing 29 (1992) 113–140. Utrecht. IEEE Trans. Melititskiy.V. De Nicolao. 39 (7) (1991) 1604–1610. H.57] A. [3. Kibernetika (Kiev) (6) (1988) 121–123. Statistical characteristics of the envelope of an additive mixture of a nonGaussian periodically nonstationary radio signal and nonGaussian interference. Sathe. [3. 1991. Statistical analysis of poly. [3. in: A. V. Systems 31 (1988) 89–92. Berlin. 256–284.45] V.64] W. Appl. Signal Process.75] G. Franks. / Signal Processing 86 (2006) 639–697 [3.72] R. Stud. in: Proceedings of VIII European Signal Processing Conference (EUSIPCO’96). Extension of autocovariance coefﬁcients sequence for periodically correlated random processes by using the partial autocorrelation function. Dehay. ´ [3. 3 (3) (1996) 275–292. Teor. 76 (1987) 3–12. IEEE Trans.Y. 43 (1992) 99–113. Miamee.L. Makagon.70] S. Proceedings of the Workshop on Nonstationary Stochastic Processes and their Applications.68] D. IEEE Trans. Springer.52] I. [3. VA. pp. 675 [3. vol.G. . On the instantaneous frequency of cyclostationary random signals. Continuous time periodically correlated processes: spectrum and prediction. IEEE Trans. [3. ˘ ´ [3.50] R.). in: New Trends in Probability and Statistics.P. Otbor Peredacha Inform. Rytov. Almost periodically unitary stochastic processes. Stochastic Models 5 (1989) 45–61. 41 (1) (1993) 131–146. Hurd. Ballerini. Signal Process. Signal Processing 22 (1991) 287–297. A. Pieterbarg. pp. Houdre. Problemy Peredachi Inform. 136 (1) (1999) 71–86. 49 (1994) 277–295. 136. Representation of strongly harmonizable periodically correlated processes and their covariances. 26 (1990) 106–108. Zivanovic. Extreme values of the cyclostationary Gaussian random process. 37 (1991) 24–45. Hurd.N. 1991. Miamee.P.G. in: J. A. Necessary and sufﬁcient conditions for existence of stationary and periodic solutions of a stochastic difference equation in Hilbert space.61] A.P. On the construction of spectral densities of a periodically correlated random process.69] A. On the Wold decomposition of discretetime cyclostationary processes. Math. [3. Stochastic Process. Induced stationary process and structure of locally square integrable periodically correlated processes. Makagon. L. New York. Tkhuan. Proceedings of the Workshop on Nonstationary Stochastic Processes and their Applications. Trieste. Effects of multirate systems on statistical properties of random signals. Bhatt. [3. R. G. Cambanis. 147–164. [3.62] H. Inform.A. 19 (1) (1990) 31–37.D. Y. Spectral analysis of the covariance of the almost periodically correlated processes. 79 (1989) 1–10 (in Russian). Miamee. VSP. vol. A.V. Kratsov. W. Correlation theory of almost periodically correlated processes. [3. [3. A unifying view of coherence in signal processing. Konstantinides. [3. Zivanovic. Joshi.). 375–390. On the continuous wavelet transform of secondorder random processes. Virginia. Stationary and periodic solutions of stochastic difference and differential equations in Banach space. W. [3. Appl. McCormick. A.
The Caratheodory–Toeplitz problem with partial data. 66 (1982) 3–6 (in Russian). An extension problem for discretetime periodically correlated stochastic processes. Cycloergodic properties of discrete parameter nonstationary stochastic processes. Otbor Peredacha Inform.C. IEEE Signal Process. Technometrics 24 (1982) 109–116.90] P. Koski. Shahkar. J. J. Almost periodic variances. [4.6] Y. V. Dragan.2] L. 34 (1963) 1549–1557. Gardner. Acoust. 342 (2002) 149–161. Vaidyanathan. Ann.N. [3.11] Y. Swift (Eds. Reidel.S.7] S. Dragan.J. [4. T. Measurement of spectral correlation.A. I. Using periodic autoregressions for multiple spectral estimation. Akkarakaran. V. [4.O. Tech. Extension of autocovariance coefﬁcients sequence for periodically correlated processes. 2001. 8 (1983) 1–7.3] A. Avtometrika 1 (1987) 36–41 (in Russian).). 9 (6) (2002) 181–184. Flandrin. [3. P. ´ [3. Remote Control 40 (Part 2) (1979) 606–609.P. Statistical properties of the wavelet decomposition of cyclostationary processes. in: D.I. Mezentsev. [4. Gardner. Automatica 39 (2003) 1903–1914. Borgnat. in: Proceedings of the IEEEEURASIP Workshop on Nonlinear Signal and Image Processing. [4.I.19] I. J.91] S. Math. Estimation of the parameters of a periodically nonstationary autoregressive process. Lainiotis. Marcel Dekker. 26 (3) (2005) 423–435. Amblard. [3. M. [3. Diversi. Avtomatika 19 (1986) 44–48 (in Russian). Math. 33 (5) (2002) 705–712. 385–400. 203–209. Dorogovtsev. A Volume in Honor of Professor M. pp. Mezentsev. Roy. Otbor Peredacha Inform 69 (1984) 29–35 (in Russian).P.81] S. 67 (1983) 22–28 (in Russian).14] I.A. .N. Angew. Liet. Math. Gen. Part I: Application to spectral extraction.N. N.17] W. Amsterdam. Signal Processing 84 (2004) 813–824. 1981. Gardner. Koski. Stamotovic.16] Y. in: Proceedings of TimeSeries Analysis. Yavorskii. MD. IEEE Trans. G.A. Extension of stochastic approximation procedures to periodically nonstationary random processes. Pieterbarg. Woerdeman. Pure Appl.18] I. Borgnat. Advances in Communications. Characterization of the spectra of periodically correlated processes. Shift operator for periodically correlated processes. Symmetry of the covariance matrix of measurements of a periodically correlated random process.N. Kapustinskas.M. Otbor Peredacha Inform. Osteyee. A.87] H. Z. [4.83] A. J. Methods for the instrumental analysis of periodically nonstationary random processes. Anal. in: A. Yuran. B 4 (101) (1977) 115–121 (in Russian).M. Nauk Litov. 1980. J. Minimal representations of MIMO timevarying systems and realization of cyclostationary models. NorthHolland. P.Y. Spectral theory of cyclostationary arrays. Mezentsev. Tsymbalyuk.80] A.88] H. Sem.G. 78 (1) (2001) 1–10. Herbst.12] H. P.1] L. Algorithm and program for calculating estimates of the characteristics of periodically correlated random processes. Yavorskii. [3.P. A note on periodogram analysis of periodically correlated stochastic processes. Mech. Speech Signal Process. V. [4.P. J. The statistical Fourier analysis of variances. Holland. [3.13] R. Spectral analysis of periodically nonstationary stochastic impulse processes. R. 44 (2) (2004) 157–167.A. Automat. Computation of estimates for spectral characteristics of periodically correlated random processes. 57 (1977) 681–692 (in German).85] H. A: Math. TX. 30 (2) (2002) 65–127. [3. Renger.J. Meas. Covariational estimation of the parameters of a periodically nonstationary autoregressive process. Sci. W.79] D. Internat. Amblard. The Wold isomorphism for cyclostationary sequences. Hurd. SSR Ser.N.8] W. Math. B 27 (1965) 159–165. Houston.Y. Gardner et al.84] A. Otbor Peredacha Inform.P. 23 (1980) 1069–1071.G. Theory IT29 (1983) 105–114.10] D. 22 (1) (2001) 1–12.9] S.ARTICLE IN PRESS 676 W. P. Component estimates of the probability characteristics of periodically correlated random processes. / Signal Processing 86 (2006) 639–697 [4. Alpay. Scale invariances and Lamperti transformations for stochastic processes.5] A. P. 48 (3) (2000) 723–736. Makagon. Gnedenkotype limit theorems for cyclostationary w2 processes. [3. [3. R. P. Baltimore. 58 (1979) 10–15 (in Russian). [4. Time Series Anal. Multivariate Anal.N. [4. Darbai B Ser. Bifrequency and bispectrum maps: a new look at multirate systems with stochastic inputs. Indian J. [4.82] P. Stat. [3.89] D. Herbst.I. Stochastic Processes and Functional Analysis: Recent Advances. Chevreuil. pp.P. Boyles. Kapustinskas. Konstantinides.P. Keio Math. T. Porotskii.I. Lithuanian Math. ASSP34 (1986) 1111–1123. Inform.A. Yavorskii. J. Rep.J. Time Ser. [4. Yavorskii. .Discretesystems Periodicity in distribution. Miamee. Signal Process.H. I. Newton.78] S. Hurd. New York. [3. Rao. 104 (1978) 113–119 (in Russian). Math. [4. 2004. Cycloergodicity. Pesquet. J. [4. Guidorzi. Touati. Soc. Loubaton. Linear Algebra Appl.L. I. Flandrin.J.86] R. Yavorskii. D. [4. Properties of estimators of mathematical expectation and correlation function of periodicallycorrelated random processes.C. Tsr Mokslu Akad. IEEE Trans.4] A.J. Honda. LambertLacroix. [4. Trudy Akad. [3.I. [4.). Phys. V. Interpolation of estimates of probability characteristics of periodically correlated random processes. Dragan. [3. 38 (2005) 2081–2101. IEEE Trans. Stochastic discrete scale invariance.G. [4. Lett. Krinik. [4. S. Yavorskii.15] I. Testing Markov properties of timeseries. Tzannes (Eds.L. Statist.
Bono. [4.38] W. Systems 31 (1988) 70–72. Transmission 24 (1988) 109–115. in: W. Transmission 26 (1991) 286–288.S. Les´ kow.45] R. Gardner (Ed. [4. 1995 [4. Loomis Jr. Eng.H. 1989. Appl. Strongly consistent and asymptotically normal estimation of the covariance for almost periodically correlated processes. Biedka. Dehay. [4.40] S.26] I. Anal. [4. IEEE Press. Appl.33] S.A.A.29] H.27] C. Wilbur. Hurd. 41 (2) (1993) 703–720. Radioelectron. H.A.46] T. C. Loomis. Weron.ARTICLE IN PRESS W. IEEE Trans. New York. J. Inform. H.42] M.G.23] C. CA. Opt. Stochastic Process. Japan (Part 3) 73 (1990) 9–15. W.32] R.28] D. [4. [4.21] W.30] V. Asymptotic moments of estimated cyclic correlation matrices. Digital Signal Process. Ph.A. Sakai. Jacob. Houdre. Ergodic behavior and estimation for periodically correlated processes. Mezentsev.47] D. Inform. Brown. Roberts. Brown. 1994. Asymptotic normality of the spectral density estimator for almost periodically correlated stochastic processes. Tian. 10 (1995) 25–40. Hurd. E. LevAri. Schell.41] D. Alekseev. Estimation of the minimal period of periodically correlated processes. [4. Probab.L. Signal Process.20] V. [4. [4.H.52] D. E 73 (10) (1990) 1729–1737.P. Trans.51] M. Kato. [4. J. Sci. Gardner. [4. in: Proceedings of 29th Asilomar Conference on Signals. [4. Statist. Representation and estimation for periodically and almost periodically correlated random processes. J. 42 (3) (1994) 595–603. J. Sadler. I. Sullivan.L. Systems and Computers.36] H. Davis. Cyclostationarity in Communications and Signal Processing. Martin.). 53 (1994) 37–54. 52 (1994) 351–360.53] D. R.31] W. Loomis Jr.G.A. IEEE Trans. Problems Inform. 17 (5) (1996) 425–445. J.H.50] S. 1988. MN.A. [4.D. 677 [4. Theory 37 (1991) 216–218...J. Ocean. [4. Inst. Brown. Genossar. 43 (6) (1995) 1525–1526.H. Signal Process. Minneapolis.V. Kailath.A. Electron.25] R. C. 9 (1988) 411–417. Spectral density estimators of a periodically correlated stochastic process. Appl.C.. 32 (1991) 35–48. Loomis Jr. Eng. Les´ kow. Mili. Statist.H. in: W.44] J. University of California. Acoust. VLSI Signal Process. Les´ kow. Roberts. Lett. 15 (1992) 299–304. [4. Commun. On the ergodicity of Gaussian periodically correlated stochastic processes.22] V. 295–326 (Chapter 6). 14 (1992) 299–306. Decisions 10 (1992) 201–225.L. Parallel computation structures for a class of cyclic spectral analysis algorithms. H.24] H. Dissertation. Yavorskii. [4. [4. W. Statist. Probab. [4.N. J. Laws of large numbers for periodically and almost periodically correlated processes. Allen. Gardner et al. 41 (1) (1993) 423–427. Hurd.S. [4. / Signal Processing 86 (2006) 639–697 [4. pp. 34 (23) (1995) 5091–5099. Estimating the spectral densities of a Gaussian periodically correlated stochastic process.). Yamaguchi. Two alternative philosophies for estimation of the parameters of timeseries. J. Monsan. Statist.34] J. Digital implementations of spectral correlation analyzers. H.J. Roberts. Probab. [4. Wigner/cycle spectrum analysis of spread spectrum and diversity transmissions.43] N. J.L. H. H. [4. L. A limiting property of sample autocovariances of periodically correlated processes with application to period determination. [4. 4 (1994) 222–238. Estimation of probability characteristics of rhythmic signals as a problem of linear ﬁltration. Estimation of the Fourier coefﬁcient functions and their spectral densities for fmixing almost periodically correlated processes. T.35] H. Gardner. Digital implementation of spectral correlation analyzers. Signal Process. V. C.H. Dehay. Brown.E. Acad. H. J.D. Reed. Cambanis. Robust estimation of cyclic correlation in contaminated Gaussian noise. A statistical study for determining the minimum sample size for Leq estimation of periodic nonstationary random noise. Onebit spectral correlation algorithms. Ser. R. J. CA. The generalized spectrum and spectral coherence of harmonizable time series. 43 (1) (1995) 173–180.49] B. Appl. . IEEE Trans. Architectures for digital cyclic spectral analysis.L. Honda. A.S. [4. [4. 264–270 [4. Estimating spectral correlations with simple nonlinear transformations. [4. I 310 (1990) 431–434 (in French). Ergodicity in periodic autoregressive models. Roberts. Cyclostationarity in Communications and Signal Processing.H. Decisions 13 (1995) 273–286. Anal.J. Functional limit theory for the spectral covariance estimator. Mag. University of Maryland. Department of Electrical Engineering and Computer Science.H. Acoustooptic cyclostationary signal processing. Dissertation. Commun. Alekseev. Roberts. IEEE Trans. Appl. Signal Process. IEEE Signal Process. IEEE Trans.S. Time Ser. Consistent estimation of the cyclic autocorrelation. ´ [4. IEEE J.48] R. New York. Lett.39] W. [4. Gerr. Spectral analysis and lattice ﬁlter for periodic autoregressive processes. Signal Process.A. IEEE Trans. 1994. Loomis Jr. Gardner (Ed. Wegman. Ph. Hurd. 455–479. A review of digital spectral correlation analysis: theory and implementation. Time Ser. Dehay. 33 (1996) 1077–1092. Random sampling estimation for almostperiodically correlated processes. Les´ kow. IEEE Press. Asymptotic behavior of estimators of cyclic functional parameters for some nonstationary processes. Computationally efﬁcient algorithms for cyclic spectral analysis. IEEE Trans.M. Les´ kow.A. 1990. in: Proceedings of the Fourth Annual ASSP Workshop on Spectrum Estimation and Modeling. Paciﬁc Grove. J. Y. Commun. pp. Les´ kow. 8 (1991) 38–49. Theory 35 (1989) 350–359. Dehay.. [4. Hurd. Problems Inform. 16 (1) (1991) 98–106. [4. Inform. pp. Nonparametric time series analysis for periodically correlated processes. Stochastic Process.S.L. Electron. [4.37] J.
Y.V. American Society of Civil Engineers 97 (1971) 949–963. A probabilistic model of rhythmicity in the radiolysis of solids. On the power spectral density of digital pulse streams generated by Mary cyclostationary sequences in the presence of stationary timing jitter. Vucic. Time series with periodic structure. Dragan. Win. Probability. Stochastic analysis of periodic hydrologic process. Signal Theory.I. [5. McGrawHill. Zhou. Mosionzhik. LevAri. IEEE Trans. Inform. IEEE Trans. Quasiperiodically correlated random processes.A. Technol. Nonlinear analysis of cyclically correlated spectral spreading in modulated signals. IEEE Trans. Sadler.8] A. Hydraulics Division. . Jones. IEEE Trans. Bukofzer. / Signal Processing 86 (2006) 639–697 [5. pp. A. 35 (1990) 130–134.6] O. [4. COM35 (6) (1987) 595–601.A. Spooner. Dissertation. Davis. [5. Electron.17] J. NJ. Schreiber. Spectral analysis of timing jitter effects on the cyclostationary PPM format. 1963. Statistical characteristics of polarization parameters of nonGaussian periodically non stationary radio signals. C. Y.A.3] R. PrenticeHall. 14 (3) (2003) 87–91. 1984.V. Characterization and applications of multiplexed PAM/PPM processes.P. IEEE Trans. pp. Aeron. R. McDonald.R. ´ [4. Soc.58] L. Clayton.).2] L. Commun.61] D.Z. [6. J. S. Renger. A. Chen. [5. Papoulis. Dandawate. [5. Karavan. Dragan.E. Radio Eng. Gournay.M. 92 (1992) 219–230.4] A.. [5.J. Englewood Cliffs. Ann. 20 (1971) 596–600 (in German). Estimating spectral correlations using the least mean square algorithm. [5. Kozlov.10] W. Spectral correlation evalua˘ ´ tion of MSK and offset QPSK modulation. Part I—Analog modulation. [4. Q. [5. Speech Signal Process. Hanin. second ed. V. Distribution functions of the envelope and phase of a periodically non stationary process frequently encountered in radio physics. Kayatskas.A. Cryan. C. 49 (1) (2001) 30–44. Electron. Rusinov.20] M. Amer.I. [6. on Signal Process. Wilbur. Bennett. IEEE Trans. 62 (1980) 26–31 (in Russian). and Stochastic Processes. Theory 43 (4) (1997) 1265–1268. M. Spectral correlation of modulated signals. W.W. Telecommun. ASSP31 (1983) 96–105.M. Telecommun. Theory 44 (1) (1998) 351–358. [4. Wiley. Franks. [6. Commun. [4. Weak law of large numbers for almost periodically correlated processes. F. R. 53 (7–8) (1998) 279–288 (in French). Commun. [5.A. Y.R. Obradovic. in: E. 1990. Telecommun. Dehay. IEEE Trans. Monsan. Korshunov. J. Bell System Tech. IEEE Trans.55] F. Makagon. Telecommun.M. 33 (3) (1997) 182–184.H. University of California. Hawaii. Berezin. Signal Process.59] B. Statistics of regenerative digital transmission. Napolitano.ARTICLE IN PRESS 678 W. [4. Radio Eng. Commun. [4. Huang. Li. Gurevich. New York. [4.12] C. 11 (4) (1998) 1111–1133.A. A. Statist. Chen.K.G. J. Brelsford.54] A.H. System Eng. 22 (10) (1974) 1727–1730. Random Variables.V. Biometrika 54 (1967) 403–408. 1969 (Chapter 8).M. Aerospace 29 (1993) 73–87.M. Cheng. Statistical characteristics of a narrowband cyclostationary process. Probab.G. Monin. 144–151.60] Z. Proceedings of the Symposium on Time Series Analysis.7] A. JEM modeling and measurements for radar target identiﬁcation. Lett. van Der Wurf.S. ´ [5. Amer. P. Spectral correlation of modulated signals. Bell. Spectral correlation characterization of modulated signals with application to signal detection and source location. Stationary and periodic time series in the general circulation of the atmosphere. Theoret. Random modulation: a review.57] P. Signal Processing 43 (3) (1995) 269–277.H.M. Maschinenbau Tech.21] D. Ann. W. Inform.14] A. [5. Acoust. P. IEEE Trans.15] J. J. Discrete periodic sampling with jitter and almost periodically correlated processes. Optimized estimation of moments for nonstationary signals. Gardner. Miamee. Ph. Bhuiya. Signal Processing 78 (1999) 363–367. Cyclic spectral analysis of continuousphase modulated signals. Papoulis. [5.1] A. Brown. Gardner et al. I.16] M. H. IEEE Trans. Acoust. Rosenblatt (Ed. 53 (7–8) (1998) 267–278 (in French). Almost sure convergence analysis of mixed time averages and kthorder cyclic statistics.2] R. Discrete spectrum of nonstationary stochastic processes on groups. CA. [5. Theoretical spectral correlation of CPM modulations—Part II: General calculation ´ ´ method and analysis. Electron. COM35 (6) (1987) 584–594. R.9] W. W. [5. Viravau. [5. Y.K. Math.18] P. Nonparametric estimation of the cyclic cross spectrum. [5. Gardner. 1989.56] H. Jiang.5] Y. Inference Stochastic Process.V. [5. Part II—Digital modulation. 124 (6) (1996) 1899–1902.1] W. Theoretical spectral correlation of CPM modulations—Part I: Analytic results for ´ ´ twostates CPFSK modulation (1REC).13] D. 45 (5) (1997) 1210–1221. 37 (1958) 1501–1542. 915–918.A.M.23] A.D. On the spectral density of a cyclostationary process. Dominique.19] P. (Part 2) 42 (1987) 87–91. Wu. Geomagn.4] P. in: Proceedings of the International Symposium on Information Theory and Applications (ISITA’90). Asymptotic analysis of estimated cyclic crosscorrelation function between stationary and cyclostationary processes.K. Elmirghani. Soc. Otbor Peredacha Inform. J. (2005). [5. [5. [6. New York. Reed. Wang. V. USA. 46 (9) (1998) 1135–1145. [5. Commun.P. (Part 2) 25 (1970) 145–146 (English translation of Radiotekhnika). 18 (1978) 309–312. J. [5. Krasheninnikov. Gournay. Department of Electrical Engineering and Computer Science.C. Grubbs.E. [5.11] V. Proc. W.R. Viravau. B. Behaviour of linear oscillatory systems which are energized by stochastic impulse trains.3] A. Sov.
Kacimov. Gidrometeoizdat.17] M.B.A. Arciplani. Instrum. Shimbo. 23 (1987) 187–191.J. Voychishin. G.J. Gardner.5] J. I. Campbell.648. Rozhkov. estimation and validation using measured sea clutter. Y. Anal. The equatorial zone.M. J. Mesiya. J. Methods (Part I (1979) 465–473.R. Ortiz. Periodic correlation in stratospheric ozone time series. System and method for performing speech recognition in cyclostationary noise environments. Hasselmann. Inform.R. Lund. Yavorskii. T.4] T.P. G.A. B.H. Abrego. A (Sweden) 36A (1984) 132–146. The dynamic pneumocardiogram: an application of coherent signal processing to cardiovascular measurement.18] V.P. Johnson. A simple stochastic model of the natural rhythmic processes.N. IEE Proc. US Patent No. Ericson. Instrum.J.7] B. A cyclostationary model of sea surface temperature in the paciﬁc ocean. Ngugen. L. Lemke. Mon. Bloomﬁeld. Sci. [6. Marseguerra. Neutron counting statistics in a subcritical cyclostationary multiplying system. [7. in: V. 226 (1999) 242–249.D. pp.P.11] Y.23] R. [6. Gardner. Linear prediction of sea ice anomalies.L.P. H. Extraction of the trend in a periodically correlated time series. Carloni. Obnosov. Naukova Dumka. MenendezPidal.P. Time Ser. Commun.A. P.13] T. I. Nonlinear theory of parametric instabilities in plasmas. J.26] F.22] P.K.E. III (1983) 756–773.R. Biomed. [6. McLane.6] D.F. USSR. Algarve. A 14 (1976) 1869–1882. Atmos. [7. Yakimov. J.M. R. Besserides.J. 6. IEEE Trans. IEEE Trans. An equivalent linear model for marked and ﬁltered doubly stochastic Poisson processes with application to MMSE estimation for synchronous Mary optical data signals.L. Dragan. IEEE Trans. Techniques of linear prediction for systems with periodic statistics.L. Hurd.P. Graef.D. Ruiz de Elvira. examples and stochastic modeling. Statistical prediction of seasonal air temperature over Eurasia. Leningrad. Methods of Probabilistic Analysis of Oceanological Rhythmics. 1983. Modulation by means of linear periodic ﬁltering. O. Phys. EOFs of onedimensional cyclostationary time series: computations. BME28 (1981) 471–475.8] B. Kiev. [6. F. Eng. Heinz. Nucl. Neutron counting statistics in a subcritical cyclostationary multiplying system. 1987 (in Russian).8] J. Applications of the theory of periodically correlated random processes to the probabilistic analysis of oceanological time series. New York. Gini. M. [6. K. Sonar Navig.Y. August 31.20] Y.1] W. 90 (1985) 5665–5675. Geophys.P.N. Leningrad. Rhythmics of sea waves and underwater acoustic signals. J. Barnett. COM24 (1976) 917–921. Methods (Part II) 172 (1980) 531.A. Marseguerra. IEEE Trans. Otbor Peredacha Inform. [6. Dragan. Huang. B. Optimal receiver ﬁlters for BPSK transmission over a bandlimited nonlinear channel. Smith. P. [6.16] C. Campbell. Alekseev. Hasselmann.N. North. Barnett. COM31 (5) (1983) 638–649. pp.21] K. 77 (1988) 2–7 (in Russian). / Signal Processing 86 (2006) 639–697 [6. Climatological time series with periodic correlation. A. 38 (1981) 2275–2283.V. Campbell. 581–592. [6.3] M. A.C. Climate 11 (1995) 2787–2809. COM26 (1978) 12–22. [6. Dragan. 1984. Interaction of the monsoon and paciﬁc trade wind system at interannual time scales. Dubois. Phys.10] W.O. F. Commun. Atmos. Johnson. J. The structure of leastmeansquare linear estimators for synchronous Mary signals. Res.N. N. Tellus Ser. Springer. 4–23 (in Russian). Principles of a linear theory of stochastic test signals and their statistical analysis.25] A. Commun.15] W.L. 2 (1971) 7–15 (in Russian). [7. 1984.Y. 53 (1996) 1007–1017. Albuquerque. in: Time Series Analysis of Irregularly Observed Data.27] X. Reidel.14] Y. A 37A (1985) 14–23. IEEE Trans. Sci. Dragan.S. Arciplani. V.ARTICLE IN PRESS W. Kim. M. 2004. [7. G. Hydrol. Seasonal cycle and secondmoment statistics of a simple coupled climate system. Yavorskii.7] F. [6. Barnett.C. Yavorskii. IEEE Trans. Lund.A. Gibbs. B. [6.19] Y. [6. Gibbs. Time series regression with periodically correlated errors and missing data.P. H. [6.).12] T.785. P. Portugal. Carloni.9] K. 97 (20) (1992) 437–448.Radar. Rozhkov (Ed. [7. Smith.5] K. 15 (2) (1994) 127–150. 78–107. Bloomﬁeld. Rev. North. Smith. I. Ocean. Joint optimization of transmitter and receiver for cyclostationary random signal processes.24] K.A. 679 [6. [6. [6. Gardner et al. [6. Atmos. Dragan.H. Theory IT27 (1981) 322–327. Hurd. Kim. Dunsmuir. [6. Inform. Technol. L.2] W. Rozhkov. Texture modelling. Greco. Nucl. Tellus Ser. 1982 (in Russian). Modulation transfer noise effects from a continuous digital carrier to FDM/FM carriers in memoryless nonlinear . Groundwater ﬂow in a medium with parquettype conductivity distribution. [7. Geophys.K.A. pp.M. In Probabilistic Analysis and Modeling of Oceanological Processes. Barnett. USSR. [6. Commun. Gidrometeoizdat. [6.P.6] J. The cyclostationary nature of crosstalk interference from digital signals in multipair cable—Part II: Applications and further results.F. Weather Rev. [6. Berlin. J. I. J. in: Proceedings of the NATO Advanced Study Institute on Nonlinear Stochastic Problems. [7. M. [6. Y. IEEE Trans. The cyclostationary nature of crosstalk interference from digital signals in multipair cable—Part I: Fundamentals. R. Otbor Peredacha Inform. COM31 (5) (1983) 629–637.P. A. Theory IT19 (1973) 240–243. 149 (3) (2002) 115–124. Res. [7. H. V. T.
J. Hsia. Systems. 1989. The performance of timedependent adaptive ﬁtters for interference rejection.C. [7. Agee. Falconer. Bergmans. Agee. Lopez de Victoria.21] .I 139 (24) (1992) 429–436.G. [7. 15.D. Dallas. IEEE Trans. 799–803. CA.H.26] M. J.20] [7. The propertyrestoral approach to blind adaptive signal extraction. B. The property restoral approach to blind adaptive signal extraction.S. IEEE J. V. [7. in: Conference Record.G. 1989. Cyclostationary crosstalk suppression by decision feedback equalization on digital subscriber loops. CA.A. pp. Mendoza.16] [7. Proakis. The baseband modulus restoral approach to blind adaptive signal demodulation. 657–662. vol. May 10. Joshi. in: Proceedings of Digital Signal Processing Workshop. vol. in: Proceedings of GLOBECOM ’89. A.M. IEE Proc. Lausanne. CA. pp. Catipovic.G. Design criteria for crosstalk interference between digital signals in multipair cable. IEEE Trans. 23 (1987) 351–398. pp. Falconer. S. pp. Abdulrahman. [7. Demodulation of a direct sequence spreadspectrum signal using an optimal timedependent ﬁlter.10] [7. University of California.24] R. [7.G. Interference suppression and multipath mitigation using an adaptive correlator direct sequence spread spectrum receiver.E. Venkataraman. D.M. Data receiver comprising a control loop with reduced sampling frequency. A frequency domain time dependent adaptive ﬂiter for interference rejection. [7. Systems and Computers. Gardner et al.P. in: Proceedings of GLOBECOM ’88. Golden. J. Adaptive multichannel combining and equalization for underwater acoustic communications. Venkataraman. 1989. Paciﬁc Grove. 9 (6) (1991) 931–940. Ph. J. IEEE J. Hsia. M. Acoust. Gardner. 2. Proakis. vol. 1990. 1989.H.27] H. in: Proceedings of IEEE ICC. Reed.C. B. Third Symposium on Wireless Personal Communications.D. Performance of optimum and adaptive frequencyshift ﬁlters for cochannel interference and fading. Fijalkow. Fahmy. vol. San Diego. 1993. Ding. Potter. vol. C. Res.ARTICLE IN PRESS 680 W. J. IEEE Trans. CA. Bergmans.A. L. 1992. Hsia. Soc. Department of Electrical Engineering and Computer Science.18] [7. Fahmy. Minimum mean square equalization in cyclostationary and stationary interference—analysis and subscriber line calculations. Saulnier. T. T. pp. 94 (1993) 1621–1631. [7. MA.34] M. B.558. Agee.22] J.H. Lee. A facility for near end crosstalk measurements on ISDN subscriber loops. 5. Falconer.35] J.311. [7. J.12] [7. pp. D.25] B.D. Adaptive fractionally spaced blind equalization. Reed. 1989. [7. IEEE Global Telecommunications Conference and Exhibition. in: Proceedings of IEEE Militar Communications Conference (MILCOM’89).J. T. COM32 (1984) 337–353. Reducedcomplexity simultaneous beamforming and equalization for underwater acoustic communications.I 140 (2) (1993) 114–120.G. Ding. Sel. [7. TX. G. Davis. Robust data equalization.D. in: Proceedings of the 24th Annual Asilomar Conference on Signals. 23rd Asilomar Conference on Signals.C. Hollywood.M.G. J. [7. Janssen. Johnson. B. 562–567. Petersen. 1994. Areas Commun.19] [7.G. Amer. / Signal Processing 86 (2006) 639–697 devices. Paciﬁc Grove. in: Proceedings of Virginia Tech. vol. Reed. 1993. Reed. [7. 391–397. Leduc.A. fractional tap spacing and the Zak transform. pp. 1987. Stojanovic. CA. W. Systems and Computers. 1592–1596. 1988. Philips J. US Patent No. Switzerland. pp. B. Falconer. Commun. Image modelling for digital TV codecs. D. Timedependent adaptive ﬁlters for interference rejection.11] [7. Extended bandwidth transmitter for crosstalk channels.32] J.33] M.Dissertation. Stojanovic. in: Proceedings of COMCON. US Patent No. in: Proceedings of the 23rd Asilomar Conference on Signals. 1989.A.M. 1988. C. Hsia. Cyclic Wiener ﬁltering: theory and method. Sel. pp. and Computers. 242–247. CA. Adaptive demodulation of PCM signals in the frequency domain. IEEE Trans. Pateros. Interference rejection using timedependent constant modulus algorithms (CMA) and the hybrid CMA/ spectral correlation discriminator. 2. Areas Commun. COM34 (1986) 593–599. Proakis. Davis.D. 41 (1) (1993) 151–163. [7. Fung. Agee. Reed. in: Proceedings of Digital Signal Processing Workshop.37] G. in: Proceedings of Oceans’93. University of California.30] H. 39 (9) (1991) 2108–2111.H. [7. 1–12. 1160–1170. W. Adaptive receivers for underwater acoustic communications: their relation to beamforming and diversity combining. Agee. Speech Signal Process. A. Catipovic.29] B. W. 1994.A.13] [7. IEE Proc.377.R. Frequencyshift ﬁltering theory for adaptive cochannel interference removal.M. 38 (8) (1990) 1373–1385. 662–666. Reduced state sequence estimation techniques for digital subscriber loop application. in: Proceedings of MILCOM 88.C. Boston. Synchronised linear almost periodically timevarying adaptive ﬁlters.M. J. Acoust. 4.14] [7.D. [7. B. Paciﬁc Grove.W. Commun. Solving the nearfar problem: exploitation of spectral and spatial coherence in wireless personal communication networks.D.15] [7. Smith. in: Proceedings of CSIARO Workshop on Advanced Topics in Communications. T.A. 1990. Ph. A. Commun. 5. 12 (4) (1992) 640–649. CA. Agee. D. J. Brown.17] [7. F.H. 1993. 1994. IEEE Trans.230. Quilici. Greene.23] J.31] W. Performance of outputmodulatorstructured linear almost periodically timevarying adaptive ﬁlters. Gardner. M. Signal Process. 2. [7. J. December 27.36] I. Gardner. Department of Electrical Engineering and Computer Science. P. S. in: Proceedings of the SPIE. 1988.9] [7. Dissertation. [7.28] C. 3.W.
IEEE Trans. A. Ph. Optimization of ﬁlter banks using cyclostationary spectral analysis. LevAri.62] [7. Brown. NJ.67] [7. Interference suppression in CDMA systems. Areas Commun. IEEE Trans. [7. Signal compression by subband coding. in: Proceedings of the 29th Annual Asilomar Conference on Signals. S. Inform. Madhow. Cui. Yang. IEEE Signal Process.V. US Patent No.57] [7. Theory 45 (9) (1999). MMSE interference suppression for directsequence spreadspectrum CDMA.M. Gardner. Ricci. Aazhang. A new algorithm for blindadaptive frequencyshift ﬁltering. pp. Development and analysis of adaptive interference rejection techniques for direct sequence code division multiple access systems.52] B.51] E. Kumar. B.41] G. IEEE Trans. A. Ottersten. Cyclostationary modeling. D. 5. [7. G. Dissertation. Cyclostationarity in Communications and Signal Processing. Miller. Sel. B. P. Sheikh.42] J.A. H. G. 46 (9) (1998) 1163–1175. B. Honig. [7. W. IEEE Trans.58] [7.D. W. Blind adaptive interference suppression for directsequence CDMA.A. H.D.ARTICLE IN PRESS W. Virginia Polytechnic Institute and State University.U. Brown. Duverdier. [7. S.68] access communication systems. Nicolas. in: Proceedings of Global Telecommunications Conference (GLOBECOM’99). Gardner.45] J. S. Princeton. 1999. pp. 2185–2188.453.56] [7. Commun. Areas Commun. Dasgupta. Pandharipande. US Patent No. G. 43 (2) (1995) 1746–1755. in: Proceedings of International Conference on Acoustics.43] W. D. pp. Maximumlikelihood synchronization of a single user for codedivision multiple681 [7. Yeung. Bergmans. Aazhang.D. 6 (11) (1999) 284–286. Bradley Department of Electrical Engineering. Speech. Propagation delay estimation in asynchronous directsequence codedivision multiple access systems. Seconda Universita di Napoli.M. P.A.59] [7. An adaptive direct sequence code division multiple access receiver for multiuser interference rejection. Narrowbandinterference suppression in multiuser CDMA systems. U. Commun. Yuen.46] J.49] S. September 26. Commun. Adaptive receiver structures for asynchronous CDMA systems. January 1999 (in Italian). S. [7. S. Paura. Subspacebased channel estimation for code division multiple access systems. Vucetic. pp. IEEE J. [7. M. IEEE Trans.654. 1998. [7.V. Gerlach.K.H. Aversa. T. and Computers. J. R. Automatica 35 (1999) 1895–1908.S.M. 46 (3) (1998) 392–399. Systems. [7. Tulino. G.53] S. 16 (1998) 1747–1755. 1998. Madhow. [7. Lops. Falconer. Commun. W. Blind multiuser detection: a subspace approach. 43 (2/3/4) (1995) 187–190. Gardner. A. Commun. 1998. US Patent No. Roviras. X. Sel.848. M. B. IEEE Trans. Signal Process. N. Haddad. Properties of the singlebit singleuser MMSE receiver for DSCDMA systems. Pateros.M. [7. Signal reconstruction after spectral excision. and optimization compensation of quantization errors in subband codecs.J. [7.A. 1995. 329–361. 12 (4) (1994) 685–697. Uzun. Kelly.797. S.61] [7. Mangalvedhe. Paciﬁc Grove.63] [7. Rhodes. Dasgupta. B. Francis. Italy. 5. D. Tulino. Proc. 44 (1) (1996) 84–93.65] [7.H. IEEE Trans.39] P. in: Proceedings of Fourth Workshop on Smart Antennas in Wireless Mobile Communications.50] C. Yeung. [7. Transmission system with increased sampling rate detection. N. Introduction of linear cyclostationary ﬁlters to model timevariant channels. and Signal Processing (ICASSP’98). B. Dissertation. On the average nearfar resistance for MMSE detection of directsequence CDMA signals with random spreading.K. Saulnier.S.54] [7. An adaptive correlator receiver for directsequence spreadspectrum communication. Reed. 42 (12) (1994) 3178–3188. Commun. Poor. McLernon. Miller. Joint transmitter/receiver optimization for multiuser communications.C. 5. [7.64] [7. 1885–1888.47] N. 43 (9) (1995) 2109–2119. CA. Agee. Gardner (Ed. L. Cyclostationaritybased ﬁltering for narrowband interference suppression in directsequence spreadspectrum systems. E. Greece.105. 325–329. IEEE Trans. in: W. IEEE Trans. Honig. Interrelationships between different structures for periodic systems. [7. IEEE Press. D. Lim. IEEE Trans. Bensley.55] [7. Gelli. IEEE Trans. 1994. Commun. Strom. Gardner et al.G.). 44 (11) (1996) 2718–2725.A. [7. Theory 44 (2) (1998) 30–44. Bensley. Commun. December 8. A. Ph.463.M. W. IEEE (1998) 2049–2069. Dipartimento di Ingegner` ia dell’Informazione. H. in: 28th Annual Conference on Information Science and Systems. Madhow. Commun. 44 (8) (1996) 1009–1020. A. Hsia. October 31. 1995. Roy. The backtalk airlink for full exploitation of spectral and spatial diversity in wireless communication systems. J. U. 44 (1996) 1543–1552. Rapajic. Lacaze. IEEE J. Lett. IEEE Trans. / Signal Processing 86 (2006) 639–697 [7. Signal Process.A. Miller. M. in: Proceedings of IX European Signal Processing Conference (EUSIPCO’98). An optimal receiver using a timedependent adaptive ﬁlter. Sakai. IEEE Trans.66] [7.38] U. Parkvall. Blind adaptation of antenna arrays using a simple algorithm based on small frequency offset. IEEE Trans.40] S. Method and apparatus for decoding broadcast digital HDTV in the presence of quasicyclostationary interference. Ohno. Strom. Tulino. J.W. GMSK signal processors for improved communications capacity and quality. 1997. Adaptive RLS ﬁltering under the cyclostationary regime. 1999.A. 47 (3) (1999) 416–425. New York.44] S.48] S. 1995. Subband coding of cyclostationary signals with static bit allocation. . Wang. Commun.A. A. Inform.60] [7. 1994. 46 (1) (1998) 61–70. Schell.
Naish.574. M.94] [7. 2003.737. Jesse. 6. [7. K. Verde.396. IEEE Trans. 19 (2001) 1544–1555. IEEE Trans. Acta Electron. [7. Mitra. IEEE Trans. 6.058. Blind adaptive FRESH ﬁltering for signal extraction.240.553.79] W. Gardner. A.885. 51 (2) (2003) 145–150. M. R. Making the most out of spectral redundancy in GSM: Cheap CCI suppression.L. G.085. Inform. Zhang. Sel.G. May 28. Ragozzini.L.88] [7. J. Tulino. March 19. Gardner. Thomas. Stephenson. Napoli. F. D. W. Mandyam. F. Trans. Symposium on Wireless Personal Communications. Z.91] [7. Lops. Tong. 5 (12) (2001) 491–493.T. Asymptotic analysis of im` proved linear receivers for BPSKCDMA subject to fading. D. March 18. P. Ph. Monot. Bromberg. Hole. February 8. C. IEEE Commun. F. 6. P. Dissertation. Multedo.A.512.80] W. [7.R. Commun.J. [7.C. US Patent No. [7. Lops. Pipon. Signal Processing 80 (2000) 2249–2254.75] S. A.B. Sinica 30 (12) (2002) 122–126 (in Chinese).99] .89] [7.Q. Martone. Y. Chevalier. Thibault. Stackedcarrier discrete multiple tone communication technology and combinations with code nulling. Arslan.J. Cho. US Patent No.M. US Patent No. L. M. Lett. U. 49 (2000) 1705–1718.98] [7. Signal Process. On cyclic correlation matched ﬁltering.95] [7. Ding. W. 2003. Lacaze. Agee. Lett. [7.P. Sel. 4 (6) (2000) 187–189. 6. Buzzi. Veh.W. M.G. US Patent No. Partially blind adaptive MMSE interference rejection in asynchronous DS/CDMA networks over frequencyselective fading channels. 2002.G. A. November 2001 (in Italian). Gelli. Lops. US Patent No. Mandyam. Commun. in: Proceedings of the 36th Annual Asilomar Conference on Signals. R.93] [7. Gameiro. Roviras. 2001. A. J. B. 2001. Tulino. 217–228. Circuits Systems II: Analog Digital Signal Process. T. G. Verdu. 49 (1) (2001) 94–108. [7. Suppression of cochannel interference in GSM by predemodulation signal processing. Ryan. R. Gelli. [7.M. pp. Systems and Computers.H. 48 (9) (2000) 1590–1600. [7. Blind adaptive multiuser detection for asynchronous dual rate DS/ CDMA systems. Tulino. D. K. Performance analysis of cyclostationary interference suppression for multiuser wired communication systems. Lowinterference communications device using chaotic signals. Nix. Gerlach. Commun.90] [7. G. Agee. Gibbons. 6. Italy. Won. IEEE J. Paura. A.564. F.86] [7. Dipartimento di Ingegneria Elettronica e ` delle Telecomunicazioni.82] A.D. US Patent No. Simple receiver for systems with spectrally overlapping narrowband and broadband signals. IEEE Trans. April 22.D. P.70] B. / Signal Processing 86 (2006) 639–697 communications. IEEE Trans. IEEE Trans. G. Signal Processing 82 (6) (2002) 821–831. Jiang. S. R. F. Golden.J. Ueng. H. Yu. Blind adaptive detection of DS/CDMA signals on timevarying multipath channels with antenna array using higherorder statistics. Wong. Theory 50 (12) (2004) 3396–3405. 2000. [7. Sabharwal. Gelli.R. Commun. Universita di Napoli Federico II. Gardner et al.535. Luo. MMSE receivers for multirate DSCDMA systems.M.M.G. Huang. US Patent No.77] S. retrodirective communication adaptive antenna arrays.A. Stacked carrier discrete multiple tone communication systems. M. US Patent No. Effect of random permutations applied to random sequences and related applications. Buzzi. Networks 6 (2) (2004) 93–105. 2005.276. Commun. Digitaltoanalog conversion of pulse amplitude modulated systems using adaptive quantization. interference cancellation. G. October 3. [7. Lacaze. Carroll.97] [7. R.923. On the use of periodic clock changes to implement linear periodic timevarying ﬁlters. Blind widely linear multiuser detection. Wang. J. Maxwell. E. Moses. Adaptive minimum variance equalization with interference suppression capabilities. in: Proceedings of the 11th Virginia Tech. Paciﬁc Grove.85] [7. IEEE Trans. 49 (1) (2001) 154–167.71] A. Technol.235. Z.72] G. Commun. High bandwidth efﬁcient communications. IEEE Trans. US Patent No. Areas Commun. Tulino. 2001. J. B.. January 28. 49 (12) (2001) 2184–2197.78] S.H. IEEE Trans. 19 (2001) 233–244. 6. Joint transmitter and receiver optimization in additive cyclostationary noise. Reed.83] J.M. 23 (2002) 311–320. Verde. 23 (2002) 253–281.69] J. Lops.H.96] [7. D.76] S.D.M. A new family of MMSE multiuser receiver for interference suppression in DS/CDMA systems employing BPSK modulation. [7. B. [7. Verde.359.854. Tulino. Equalization and interference suppression techniques for highcapacity digital systems. A. Zhou. Simultaneous suppression of multiaccess and narrowband interference in asynchronous CDMA networks. 6. Blind FSRLPTV equalization and interference rejection. Ching.ARTICLE IN PRESS 682 W. June 3. Adaptive quantization in a pulseamplitude modulated system. 2002.J. Means and method for increasing performances of interferencesuppression based receivers. Buzzi. [7. May 29.A. IEEE J. H. Ho. Agee. Implementation of cyclostationary signalbased adaptive arrays. Mechaley Jr. F. IEEE Commun.87] [7.098. CA. 6. Methods of receiving cochannel signals by channel separation and successive cancellation and related receivers. 47 (5) (1999) 1397–1402. D.128. 2003.J. Wireless Personal Commun.84] [7. 6. Z. Im. Duverdier. Wireless Personal Commun. M. Method and device for space division multiplexing of radio signals transmitted in cellular radio [7. [7. Molnar. 2003.C. Areas Commun. A.74] M. 47 (11) (2000) 1152–1159. D. Cochannel interference reduction in wireless communications systems. B.81] G.92] [7.73] M.
6. Plymouth. P. Pupolin.28] [8. Dissertation. Serpedin. Ph. IEEE Trans. M. Comment on ‘Tracking performance of the ﬁlter and square bit synchronizer’.H. Tomasi. Linear phaselocked loop theory for cyclostationary input disturbances. Commun. G. 10 (4) (2003) 89–92. Serpedin. Serpedin. L. [7. Signal Process. O’Reilly. E. COM30 (10) (1982) 2253–2259. Bubrouski. Commun. 683 [8.5] M.J.K.25] [8. 66 (1989) 551–570. 2000 (in French). The optimum closedloop transfer function of a phaselocked loop used for synchronization purposes.20] [8. Commun. 50 (1) (2002) 130–140. Spectral analysis of the powers of a PAM digital signal. Inform. P.J. IEEE Trans. G. 6. Commun. E. P. Simultaneous clock phase and frequency offset estimation. Performance analysis of a class of nondataaided frequency offset and symbol timing estimators for ﬂatfading channels. [8. Kleinman. US Patent No.3] L. Lett. Vourinen. Pezzani. P.Y. Pierobon. IEEE Trans. T. London. O’Reilly (Eds.B. Blind channel and carrier frequency offset estimation using periodic modulation precoders. Commun.21] [8.6] C. Kaube. Ciblat. [8.. E. in particular for transmission in a code division multiple access cellular mobile radio system. [8.29] [8. 1984.A. Commun. Kolehmainen. IEEE Trans. Commun. Some estimation problems relative to non cooperative communications. Loubaton.4] M. New York. M.17] [8. IEEE Trans. Moeneclaey. MarnelaVallee. 50 (9) (2002) 2295–2305. 46 (1998) 400–411. On a blind fractionally samplingbased carrier frequency offset estimator for noncircular transmissions. Roosen.B.M. [8. Qiu. The role of spectral correlation in design and performance analysis of synchronizers. Agee. Signal Process. Phaselocked loop performance in the presence of CW interference and additive noise. Wang. [8. Anvari. Gardner (Ed. J. M. Univ. COM31 (1983) 549–553.13] W. P. Ciblat. COM32 (1984) 561–566.E. January 25. IEEE Trans.7] M. / Signal Processing 86 (2006) 639–697 [7. J. Carrier and bit synchronization in data communication—A tutorial review. W. Giannakis. in: K. 5. [8. F. E. Tayebi. Mengali.847. Blind parameter estimation in multipleaccess systems. Y. IEEE Trans.D. [8. COM28 (1980) 1107–1121. COM22 (1974) 913–930. E. Commun. Commun.26] [8.689. Statistical properties of timing jitter in a PAM timing recovery scheme. Seppanen.9] M. [8. T. US Patent No.). Commun.1] L. K. COM34 (11) (1986) 1089–1095. Steadystate analysis of the clock synchronizer for a narrowband communication system. G.19] [8.14] Z. Signal Processing 40 (1994) 21–37.E. Ciblat. A. R.G. Signal Process. Blind carrier frequency offset estimation for noncircular constellationbased . IEEE Trans. K.2] U. Loubaton. Scott. Cattermole. Roning. H.11] S. Franks. Moeneclaey. P. Beneﬁcial effects of spectral correlation on synchronization. Asymptotic analysis of blind cyclic correlationbased symbolrate estimators. Napolitano. IEEE Trans. IEEE Trans. Reed. and ﬁlter. J. ´ ´ MarnelaVallee. Lilleberg.W. E. Cassara. Timing and automatic frequency control of digital receiver using the cyclic properties of a nonlinear operation. P. Monti. Hatim.E. P. Ciblat.100] F. Franks. Bolcskei. 43 (7) (1995) 2263–2270. S. IEEE Press. IEEE Trans. 44 (11) (1996) 1527–1536. [8.101] O. Yoon.). B. [8. in: W. Gardner et al.12] C.282.868. Tracking properties of phaselocked loops in optical communication systems. Chevreuil. March 15. 2005. Serpedin. K.M.C. IEEE Trans.J. [8. Giannakis. F. Method for distinguishing signals from one another. COM30 (1982) 2305–2311. Moeneclaey.30] Cyclostationarity in Communications and Signal Processing. Device enabling different spreading factors while preserving a common rambling code.18] [8. F. A fundamental lower bound on the performance of practical joint carrier and bit synchronizers. IEEE Trans. COM32 (1984) 1007–1012. Gourge.24] [8. Alta Frequenza 53 (2) (1984) 70–76. IEEE Trans. Electron. A. COM30 (1982) 407–410. G. 8 (8) (2000) 2389–2405. E. G. B. Commun. IEEE Trans. C. O’Shea. 1994. Blind estimation of symbol timing and ¨ carrier frequency offset in wireless OFDM systems. IEEE Trans. Lindsay.ARTICLE IN PRESS W. 2005.B. IEEE Signal Process.8] C. Int. Ciblat.10] J. Blachman. Bayesian recursive estimation of frequency and timing exploiting the cyclostationarity property. 49 (4) (2001) 688–698. D. Vazquez.A. Block codes for linear timing recovery in data transmission systems. Gini. Loubaton. I. Gardner.E. Pupolin. Vandendorpe. Giannakis. Tanda.A. Commun. IEEE Trans. Performance analysis of a class of nondataaided frequency offset and symbol timing estimators for ﬂatfading channels. IEEE Trans. J.27] [8. France.B. 1994.228. Serpedin.20] [8. Commun.16] [8. [8. P. Signal Process. J. J.B.077. US Patent No. Riba. IEEE Trans. Timing extraction for baseband digital transmission. COM26 (1978) 1811–1818. Commun. IEEE Trans. pp. Commun. Theory 48 (7) (2002) 1922–1934.23] [8. Olasz. COM33 (1985) 527–534. Commun.H.A. [8. January 25. Spectral analysis of line regenerator time jitter. Bilardi. [8. Giannakis. IEEE Trans. Problems of Randomness in Communication Engineering. Performance analysis of blind carrier frequency offset estimators for noncircular transmissions through frequencyselective channels. Moeneclaey.15] N. 50 (9) (2002) 2295–2305. G. J. 362–390. Loubaton. Scott. Soft synchronization of direct sequence spreadspectrum signals. Lakkis. [8. 49 (6) (2001) 988–999. P. Frequency offset and symbol timing recover in ﬂatfading channels: a cyclostationary approach. Ciblat.L.
. J.A. Nondata aided maximum likelihood based feedforward timing synchronization method. 2004.A.K.18] [9. E. W. Time difference of arrival geolocation method. W. / Signal Processing 86 (2006) 639–697 transmissions. Blind estimation of amplitudes. Schell. Commun. .P.9] [9. 1982. W. J. Gardner. 6. Omel’chenko. IEEE Trans. 51 (9) (2003) 1451–1455. Gardner.13] [8. The use of the WignerVille distribution (WVD) for the classiﬁcation and the analysis of modulation techniques. Y. Ph. Serpedin.S. pp. R. Ph.15] [9.34] [9. Kolesnikov. IEEE Trans.780.K. Widrow. pp.A.A. B. Array signal processing via exploitation of spectral correlation—a combination of temporal and spatial processing. Napolitano. Gardner.ARTICLE IN PRESS 684 W. July 27. Despreading spreadspectrum signals without the code. 23rd Asilomar Conference on Signals. Xu. CA. King.G.D. CA. Schell. C. and Computers. Rocha. CA. W. Tracking performance of stochastic gradient algorithms for nonstationary processes. Proc. Nonstationary learning characteristics of the LMS algorithm. ASSP29 (1981) 679–683. University of California. 2278–2281.V. Gardner. December 19.V. Hajivandi. Ciblat. 589–593. 1989. W.A.A. 1988. W. Y.11] [8. 52 (1) (2004) 291–296. Commun. Pride III.19] [9. Napolitano. Tugbay. IEEE Trans. B. Gardner. Spooner.V. Davis. A. IEEE Trans. Ciblat.A. Department of Electrical Engineering. Selfcoherence restoral: a new approach to blind adaptation of antenna arrays. 1989. Proc. 52 (12) (2004) 2172–2191. O’Shea. MN. N.888. The timesequenced adaptive ﬁlter. 119–120. ASSP36 (1988) 1385–1395.V. Department of Electrical and Computer Engineering. Expansion of random processes and their noncommutative transformations. A ﬁne blind frequency offset estimator for OFDM/OQAM systems. Gardner et al. Commun. S. S. C. IEEE Trans. B.A. Gardner.A. MN.8] [9. Wang. Selective source location by exploitation of spectral coherence. pp. 4. July 1961. Gardner. Adaptive deconvolution of cyclostationary signals. Y. S. Wang.21] . W. Minneapolis. O. Massachusetts Institute of Technology. Friedman. US Patent No. 75–82. Speech Signal Process. French. Signal Process. in: Proceedings of the IEEE/ASSP Fourth Workshop on Spectral Estimation and Modeling.C. etc. A. Gardner.A.M. V.35] [9. Serpedin.I Recognising Gaussian periodicallycorrelated random signals. 277–282. M. W. Agee. Agee. in: Proceedings of ISCIS III. Discrete representations of random signals. B. Paciﬁc Grove. pp.4] [9. Systems and Computers. CA. Turkey.20] [9. Otbor Peredacha Inform. MA. UK. Tanda. P. B. Dragan. in: Conference Record. 274–278.V. in: Conference Record. in: Proceedings of the International Conference on Acoustics. Cesme. S. Y.P.33] [9. D. pp. Calabretta.1] [9. and Signal Processing (ICASSP’89). Signal Processing 85 (8) (2005) 1588–1601. 22nd Asilomar Conference on Signals. IEEE 76 (7) (1988) 845–847. [8. Signal Process. Cambridge. pp.K. Commun.5] [9. 1988.7] [9.10] [9. Gardner. Ferrara Jr. 271–276. T. Tanda. IEEE 78 (1990) 753–767. Spectral selfcoherence restoral: a new approach to blind adaptive signal extraction using antenna arrays. Paciﬁc Grove.12] [8. Agee. A. CA. Serpedin. J. Paciﬁc Grove. Simpliﬁcation of MUSIC and ESPRIT by exploitation of cyclostationarity. M. Telegr. IEEE Trans. Acoust. W. in: Proceedings of the 21st Annual Asilomar Conference on Signals.A. Systems and Computers.A. 139–140. Dopplerchannel blind identiﬁcation for noncircular transmissions in multipleaccess systems. E. IEEE Trans. Chen.3] [9.2] [9. Ciblat. Dissertation. The SCORE approach to blind adaptive signal extraction: an application of the theory of spectral correlation. pp. C. Gardner. S. F.A. Kailath.. Schell. Sankar. Gardner.R. in: Proceedings of the Fourth Annual ASSP Workshop on Spectrum Estimation and Modeling. E.V. S. M. Glasgow. 1989. 68 (1980) 1026–1030 (in Spanish). Circuits Systems CAS34 (1987) 1199–1207. Omel’chenko. Schell.6] [9. 945–949. Interferencetolerant timedifferenceofarrival estimation for modulated signals.36] [9. P. P. Chen. Cyclic MUSIC algorithms for signalselective direction estimation. 51 (9) (2003) 1571–1581. G. pp. Dragan. 22 (1969) 22–27 (in Russian). 1989. COM (1986) 404–407.K. 51 (5) (2003) 1378–1389. C. Acoustics Speech Signal Process. I. Speech.D. US Patent No. M.G. Chen. Systems. 1988.17] [9. Schell.V.593. pp.16] [9. Tayebi. 1988. Agee. Lakkis. Application of the SCORE algorithm and SCORE extensions to sorting in the rankL spectral selfcoherence environment.768.G. Optimal blind carrier recovery for MPSK burst transmissions. 1989. Dissertation. Agee.L. Rev. IEEE Trans. Electron. Ricciardelli. time delays and frequency offsets in multiple access systems with circular transmissions. E. Signalselective direction ﬁnding for fully correlated signals. IEEE Trans.14] [8. An alternative blind feedforward symbol timing estimator using two samples per symbol. in: Abstracts of Sixth Multidimensional Signal Processing Workshop of the IEEE/ASSP Society. CA. 1987. in: Abstracts of Sixth Multidimensional Signal Processing Workshop of the IEEE/ASSP Society. W. W. K.L. S. Jordan.32] [8. B. Paciﬁc Grove.G. Paciﬁc Grove.31] [9. Radiotekhnika 85 (1988) 75–79 (in Russian). I. B.A. Minneapolis. Maximum likelihood twosensor detection and TDOA estimation for cyclostationary signals.P.A. Schell.P.A. C.
IEEE Trans. B. Blind adaptive spatiotemporal ﬁltering for wideband cyclostationary signals. W. vol. C. 1. in: Proceedings of the Fifth IEEE/ ASSP Workshop on Spectrum Estimation and Modeling. Signalselective timedifferenceofarrival estimation for passive location of manmade signal sources in highly corruptive environments. Part II: Algorithms and performance. 1990. / Signal Processing 86 (2006) 639–697 [9. Xu. 38 (1990) 1953–1958. Gardner. Schell.27] S. Schell. in: Proceedings of 27th Asilomar Conference on Signals. 40 (7) (1992) 1775–1785. 40 (10) (1992) 2606–2608. Chen. L.A.35] [9. Chen. Gardner. 5. [9. G. Joshi.210. Juan. L. NM. T. and Signal Processing (ICASSP’90). Agee. Gardner. [9. IEEE Trans. L. in: Proceedings of 27th Asilomar Conference on Signals. 1991. C. Schell. 27 (1991) 136–138.V. Schell.V. Schell. 4. in: 685 [9.D.31] G. 144–148. [9. pp. Kailath.29] S. 40 (1992) 1185–1197. Paciﬁc Grove. in: Proceedings of International Conference on Acoustic. W. [9. Signal Process. Measures of tracking performance for the LMS algorithm. Grifﬁths. in: Proceedings of International Symposium on Information Theory and its Applications (ISITA’90). 716–720. Commun. C.V. 473–477. Tseng. in: Proceedings of 24th Asilomar Conference on Signals. pp.M.A. D.A.255. Albuquerque. The Cramer–Rao lower bound for parameters of Gaussian cyclostationary signals. Paura. in: N. [9.A.34] G. 1993.44] [9. vol. Sequence estimation techniques for digital subscriber loop transmission with crosstalk interference. CA. McLernon. 2667–2670. Blind capture and geolocation of stationary waveforms using multiplatform temporal and spectral selfcoherence restoral.46] [9. Exploitation of spectral correlation for signalselective direction ﬁnding. L. C. Honolulu. Castedo. Acoust. 1990.22] B. Speech. Tseng.V. Ding.V. An interferencetolerant algorithm for localization of cyclostationarysignal sources. pp. Part I: Theory and method IEEE Trans. M. W. Gardner. France.A. Spectralcorrelation based estimation of channel parameters by noncoherent processing. 38 (1990) 1367–1374. vol. Signal Process. Inform. IEEE Trans. 653–657. Method and apparatus for multiplexing communications signals through blind adaptive spatial ﬁltering. Xu.A.R.A. W.A. C. Rochester.A.25] V. Schell. Izzo. Speech.K. Department of Electrical Engineering and Computer Science.30] S. Gardner. 41 (1993) 1961–1964. pp. 1993. Gardner. Bose. IEEE Trans.43] [9. Paura. Spooner.260. pp. CA. CA.37] [9. Schell. Signal Process. in: Proceedings of China 1991 International Conference on Circuits and Systems. [9. Castedo.38] [9. Handbook of Statistics. IEEE Trans. 5. T. A cyclic SVDbased algorithm for multiple source localization.24] M. 1990. 1990. ´ S. Biedka. Systems and Computers. [9. Schell. L. 1990. pp.V. Gardner. Schell. Gardner et al. CA. in: Proceedings of International Conference on Acoustic. Signal Process. Paciﬁc Grove.A.40] [9.32] T. Biedka.G. W. 1993. 267–270. US Patent No. Gelli. D. High resolution direction ﬁnding. 354–357. Gardner.M. . in: Proceedings of International Symposium on Information Theory and its Applications (ISITA’90).D. A.36] [9. Napolitano. 1991. CA.33] G. W. W. L. Gardner.47] [9. Falconer.41] [9. 1993. 1991.V. Signalselective highresolution direction ﬁnding in multipath.48] ` Treizieme Colloque sur le Traitement du Signal et des Images (GRETSI’91). [9. 1990.23] H. The Cramer–Rao lower bound for parameters of Gaussian cyclostationary signals. Systems and Computers. NY.K. IEEE Trans. Electron. Rao (Eds. 1993. Davis. and Signal Processing (ICASSP’90). Gardner. 3. US Patent No.V. 1993. G. S. Comparison of autocorrelation and crosscorrelation methods for signalselective TDOA estimation. S. NM. pp. Kailath. 284–287.45] [9. C. Inverse of linear periodically timevarying ﬁltering. A new cost function for adaptive beamforming using cyclostationary signal properties. L. [9. pp. Behavior of adaptive beamformers based on cyclostationary signal properties in multipath environments. HI. Elsevier.LesPins. L. 40 (1992) 1682–1686. W. Amsterdam. Speech Signal Process. W. Young. in: Proceedings of 25th Asilomar Conference on Signals. Fahmy.39] [9. Gardner. October 19. Paciﬁc Grove. T. Subspaceconstrained SCORE algorithms.V.42] [9. University of California. W. Agee. A simple and effective algorithm for estimating time delay of communication signals.28] S. 1217–1220.26] S.K. in: Proceedings of IEEE International Conference on Acoustics. [9. Detection of the number of cyclostationary signals in unknown interference and noise. Albuquerque. Progress on signalselective direction ﬁnding.A.A. Signal Process.A.A. Theory 38 (4) (1992) 1418–1422. Poggi. D. W. November 9. Directionofarrival estimation via exploitation of cyclostationarity—a combination of temporal and spatial ﬁltering.G. G. Lett. Selfcoherence restoring signal extraction apparatus and method. and Signal Processing (ICASSP’93). Dissertation. Izzo. Paciﬁc Grove. Ph. Poggi. W.968. Paciﬁc Grove. [9. and Computers. Agee.). Systems and Computers. Gardner. Gardner. Speech. W. CA. Signalselective timedifferenceofarrival estimation for passive location of manmade signal sources in highly corruptive environments. Systems. [9. IEEE Trans. Hajivandi. pp. B. pp. 255–258. China.C. 1990. 1990. 40 (1992) 1168–1184. Systems and Computers. Paura. in: Proceedings of the 24th Annual Asilomar Conference on Signals.G. Gelli. Analysis of LMS algorithm with inputs from cyclostationary random processes. S. NL. Sheuzhen. IEEE Trans. Signal Process. Subinterval cyclic MUSIC— robust DF with inaccurate knowledge of cycle frequency.ARTICLE IN PRESS W.
Paura. [9. Allen. CA. Nandi. Selfcoherence restoring signal extraction and estimation of signal direction of arrival. IEEE Trans. Cyclostationarity in Communications and Signal Processing. Castedo. F. Yountville. Methods for constraining a CMA beamformer to extract a cyclostationary signal.54] N. US Patent No. Wu.D.K. 42 (11) (1994) 3043–3050.75] [9.B. Y. F.73] [9. 141 (6) (1994) 312–318. Signal Process. Signal selective localization of nonGaussian cyclostationary sources. Napolitano. Modiﬁed cyclic methods for signal selective TDOA estimation. Parameter estimation of cyclostationary AM time series with application to missing observations.B. in: C. C. L. Schell. 42 (11) (1994) 3294–3298. Gardner. Minimumredundancy linear arrays for cyclostationaritybased source location.71] [9. Signal Processing 79 (2) (1999) 43–60. Performance analysis of the cyclic MUSIC method of direction estimation for cyclostationary signals. Signal Process. S.E. 44 (11) (1996) 2757–2767.H. February 1994 (in Italian). Lee. A recursive programmable canonical correlation analyzer. [9. T. March 29. 1995.70] [9.62] Q.B. Fast algorithm for adaptive beamforming of cyclic signals. G. F. M. Gelli. 43 (9) (1995) 2217–2221.G. IEEE Trans. IEEE Trans. [9. Gini.74] [9. Signal Process.77] [9. 5. Dissertation. Performance analysis of cyclic statistics for the estimation of harmonics in multiplicative and additive noise. Wong.T. in: Proceedings of IEEE International Conference on Acoustics.A. M. [9. Shamsunder. IEEE Trans. Friedlander. Cyclostationarity in Communications and Signal Processing. Nonlinear leastsquares estimation for harmonics in multiplicative and additive noise. 1994.63] T. Lee.B.66] [9. Universita di Napoli Federico II. and Signal Processing (ICASSP’94). IEEE Press.A. 34 (25) (1998) 2378–2380.52] W. Zhang. Mazzenga.V. 1828–1831. [9. Gelli. Giannakis. [9. IEEE Trans.B. M. [9. G. M. Giannakis. Kahn. IEEE Trans. Signal Process. [9. vol. 42 (10) (1994) 2860–2864. Fully blind estimation of time delays and spatial signatures for cyclostationary signals. Hongyi. Gerr. Giannakis. Lett. Signal Process.72] [9. Signal Process. Gardner.60] S. Ph.). Linearly constrained adaptive beamforming using cyclostationary signal properties. A method for reducing computations in cyclostationarityexploiting beamforming. University of Virginia. 43 (6) (1995) 1445–1460.D. G. Biedka. in: W. Signal Process. An overview of sensor array processing for cyclostationary signals.65] [9. Signal Process.V. G. Speech. 249–252. Gardner et al. vol. IEEE Trans. IEEE Trans. FueirasVidal. A. 47 (1999) 3235–3249. Sensor arraybased parameter estimation of cyclostationary signals. L. IEEE Trans. 43 (7) (1995) 1637–1650. Robust adaptive array beamforming for cyclostationary signals under cycle frequency [9. J. W.T. pp.55] G. G. 1994. L. Schell. 240–266 (Chapter 4). G. Wu. Control and Dynamic Systems. 43 (2) (1995) 492–505. [9. Y.V. A. [9. Harmonics in multiplicative and additive noise: parameter estimation using cyclic statistics. Zhou. Ghogo. Harmonics in multiplicative and additive noise: performance analysis of cyclic estimators.79] . 43 (12) (1995) 2898–2908. Programmable canonical correlation analysis: a ﬂexible framework for blind adaptive spatial ﬁltering. M. Zhou. Parameter estimation in CDMA multiuser detection using cyclostationary statistics.50] L. B. 1994. Signal Process. Signal Process. Blind adaptive beamforming for cyclostationary signals. Ghogo.57] M. Kahn.A.L. Italy. Retrieval of selfcoupled harmonics. pp. Franks.A.). Leondes (Ed. January 1995. in: Proceedings of IEEE International Conference on Acoustics. A. Biedka.51] L. Agee. [9. Estimating random amplitude polynomial phase signals: a cyclostationary approach. IEEE Press. Random amplitude and polynomial phase modeling of nonstationary processes using higherorder and cyclic statistics. Zhou. 1994 (Chapter 1). IEEE Trans.67] [9. G. Lett. 4. in: Proceedings of Second Workshop on Cyclostationary Signals. 47 (2) (1999) 363–377. L. Hybrid FMpolynomial phase ´ signal modeling: parameter estimation and Cramer–Rao bounds.49] T.B.56] L. [9. K. Wong. G. IEE Proc. G. [9.). Schell. B. Polyperiodic linear ﬁltering. A. Signal Process. Napoli. pp. / Signal Processing 86 (2006) 639–697 Speech. Spatiotemporal ﬁltering and equalization for cyclostationary signals. C. Giannakis. W. 43 (5) (1995) 1173–1185. Gardner (Ed.R.V. Shamsunder. Garel. 1994.68] [9.ARTICLE IN PRESS 686 W. Digital Signal Process. B. in: W. Tseng. Gardner. Academic Press. B. [9. G. S. Electron. VA.59] S. Signal Process. 168–239 (Chapter 3). Swami. An adaptive beamforming technique based on cyclostationary signal properties. 1994.V.53] G. New York. Swami. A. S. Zhou. 64.148. Giannakis. Q. IEEE Trans. Signal Process. G. Signal Processing 42 (1) (1995) 97–102.69] [9. 1994.Radar Sonar Navig. G. Biedka. Charlottesville. A. Schell.64] [9. Electron. Gardner (Ed. Diparti` mento di Ingegneria Elettronica. R. Castedo. Timedelay estimation for harmonizable signals. pp.78] [9. Ph.58] S. Giannakis. Gelli. [9. Schell. 42 (9) (1994) 2408–2419. Ho. New York. IEEE Trans. Gardner. W. 4 (1994) 49–62. IEEE Trans. FigueirasVidal. Izzo. New York. Giannakis. Yountville.M.A. Vatalaro. CA. in: Proceedings of Second Workshop on Cyclostationary Signals. 32 (3) (1996). Power and timing parameter estimation of multiple cyclostationary signals from sensor array data.A.B. Izzo. G. Zhou.61] S. IEEE Trans. Dissertation.76] [9. Signal Process. Grifﬁths. Mow. and Signal Processing (ICASSP’95).A.299. 45 (10) (1997) 2605–2608.
J. [9. Sonar Navig. W. G. Eisenstein. Signal Processing 83 (5) (2003) 1035–1046. Eur. Q.A. J. Signal Process. pp.8] W. [10. US Patent No. Weinstein. IEEE Trans. Lee. J.D. 5. 2001. Acoust. Xin.A.6] R. Compensating for distortion in a communication channel. Areas Commun. 48 (7) (2000) 1893–1901. 5.91] [9. 148 (4) (2001) 193–199. P. A. W. Wang.M. [10. Loubaton. M.D. Zhou. L. Tung. Signal Process. IEEE Trans. IEEE Trans.9] A.D. Gelli. December 7. Gardner. Y. Archer. Identiﬁcation of systems with cyclostationary input and correlated input/output measurement noise. 687 [9. Z. E 83A (8) (2000) 1624–1633. Lee. ´ P. J. Y. A new method of channel identiﬁcation.87] [9. US Patent No. IEEE Trans. May 1.82] [9. 4. IEEE Trans. Ph. 1992. An extended cyclic MUSIC algorithm. 11 (5) (2000) 495–500.H. Trans.Radar. Wouda.870. in: Proceedings of the 21st Century Military Communications Conference (MILCOM 2000).H. Signal Process. L. Dissertation.J.H. Signal Processing 85 (2005) 643–649. D. IEE Proc. Gardner.Z. 1991. [9. 1989.4] T. Gardner. Lee.A. J. Signal Process. 646. in: Proceedings of the 24th Annual Asilomar Conference on Signals. Paciﬁc Grove. New methods for identifying the Volterra kernels of a nonlinear system.R. Efﬁcient robust adaptive beamforming for cyclostationary signals. Blind wideband spatiotemporal ﬁltering based on higherorder cyclostationarity.89] [9. [10. Antennas Propagation 47 (2) (1999) 233–241. F. K. 52 (10) (2004) 2847–2861. 10 (3) (1992) 630–639. [10. IEEE Trans. Cyclostationaritybased coherent methods for widebandsignal source location. November 11.A.W. DuelHallen. Signal Process. 50 (1) (2002) 110–118. Sano.A. Fundam. Fractionallyspaced equalization: an improved digital transversal equalizer. US Patent No. Y. 999. Lu. 50 (3) (2002) 590–605.V. Control 35 (4) (1990) 449–452. Simpliﬁed methods for identifying the Volterra kernels of nonlinear systems. / Signal Processing 86 (2006) 639–697 error.1] R. Ronald. Relat. Syst.93] [9. Brown. Archer. Higherorder cyclostationarity based direction estimation of coherent narrowband signals.B. D. September 26. Lee.94] [9. Technique for robust characterization of weak RF emitters and accurate time difference of arrival estimation for passive ranging of RF emitters. S. Martone. Shearer III.3] J. Speech Signal Process. [9.A. Gelli.85] [9.T.657. Signal Process. CA. Gitlin. Shih. Bradley Department of Electrical Engineering. H. [10.H. 1990.T.922. 2003. Sel. [10. Joint estimation of Doppler and timedifferenceofarrival exploiting cyclostationarity property. M.506.96] K.L. Gardner. CA. 602.T.A.M. Virginia Polytechnic Institute and State University. Lee.J. P. Zhou.C. W. B. 47 (10) (1999) 2867–2873. Rosen. Robust technique for estimating the bearings of cyclostationary signals.. McCallister. Sano. T. [9.99] G. Signal Process. Y. Non dataaided estimation of the modulation index of continuous phase modulations. 4.T. Huang. Sirven. Bergmans. C. T. Lee.88] [9. Maximum likelihood estimates of the time and frequency differences of arrival of weak cyclostationary digital communication signals. Krikorian.90] [9. US Patent No. [10.L. Equalizers for multiple input/multiple output channels and PAM systems with cyclostationary input sequence. Analysis and development of blind adaptive beamforming algorithms. IEE Proc. Tech.2] R. van Gerwen. Automat. Tsuji. Linear prediction approach to direction estimation of cyclostationary signals in multipath environments.A. Technol. Mattera. Cerrato.ARTICLE IN PRESS W. Gardner et al. and Computers.095.H. J. D. IEEE Trans. IEEE Trans. Z. Sinica 30 (3) (2002) 372–375 (in Chinese). Virginia. Lee. Paura. 2000. Data signal transmission system using decision feedback equalization. Lee. R.D. M. Commun. J.L. D. 6. Izzo. Y. Zoubir. Deconvolution of cyclostationary signals. Streight. Sullivan.84] [9. . A novel direction ﬁnding method for cyclostationary signals. Bell.5] W.98] P.A. W. 957–961. Bianchi. [10.Y.T. ASSP25 (1977) 466–476. H. Blind adaptive parameter estimation for CDMA systems using cyclostationary statistics. Morelande. Charge. Huang. IEEE Trans. [9.K.95] G. Signal Process. W. IEICE Trans. 53 (4) (2005) 1282–1290. 39 (1991) 813–817.80] [9. 60 (1981) 275–296. 51 (7) (2003) 1695–1701. Biedka.H.10] G. IEEE Trans. IEEE Trans. Acta Electron. US Patent No. Directionofarrival of signal sources exploiting cyclostationarity property. Jiang. [10.H. Golden. A. Wideband cyclic MUSIC algorithm.D. in: Proceedings of the 34th Midwest Symposium on Circuits and Systems. 592–597. 51 (10) (2003) 2471–2482.86] [9. Signal Process. 1990. Systems. Jiang. Lott. On the performance of cyclic momentbased parameter estimators of amplitude modulated polynomial phase signals. Fan. Sillard. Yan.7] W. 49 (4) (2001) 710–720. [10. Wireless geolocation system.97] J. Y. Y.92] [9.81] [9. Estimating the bearings of nearﬁeld cyclostationary signals. Adaptive multistage beamforming using cyclic higher order statistics.495. IEEE J. March 10. Lee.100] H. IEEE Trans. Use of cyclostationary signal to constrain the frequency response of a fractionally spaced equalizer.83] [9. Signal Processing 81 (6) (2001) 1317–1323. A. Robust adaptive array beamforming with random error in cycle frequency. Telecommun. 131. Mazzenga. Monterey. Xin. (4) (2002) 161–165. J. F.H. 1999. IEEE Trans.
IEEE Trans. (Almost) periodic moving average system identiﬁcation using higher order cylicstatistics. C.). Giannakis. Tanda. Part II: A timedomain approach.33] [10.ARTICLE IN PRESS 688 W. Blind channel identiﬁcation based on second order statistics: a [10. 44 (3) (1996) 324–336. L. Izzo. / Signal Processing 86 (2006) 639–697 frequency domain approach. 1994.11] G. J.K.36] [10. in: W. Cyclostationarity in Communications and Signal Processing. Paris. Signal Process. Napolitano. Tong. Liang.16] Z. Department of Signal Processing. Signal Processing 50 (1996) 83–99.18] L. Tong. Dipartimento di Ingegneria Elettronica. Blind identiﬁcation of linear subsystems of LTIZMNLLTI models with cyclostationary inputs.044. Chevreuil. Wong. Li. in: Proceedings of VIII European Signal Processing Conference (EUSIPCO’96). Theory 41 (1) (1995) 329–334.A. Luschi. France.L. T. Nikias. Hatzinakos.19] Y. Lett. Gardner (Ed. G. Chevreuil. New York. Xu. IEEE Trans. 417–436. Giannakis. Use of a fractionally spaced equalizer to perform echo cancellation in a fullduplex modem. Kailath. Signal Process. A. L. T. IEEE Trans. Zeng.23] W. Xu. 32 (19) (1996) 1763–1764. IEEE Press. J. ` Universita di Napoli Federico II.B. P. ARMA system identiﬁcation based on secondorder cyclostationarity. H. M. Paura. 45 (7) (1997) 1785–1794. F.D.28] [10. Andrieu.25] L. Eur. Lett. [10. Gardner (Ed. Duvat. Cyclostationarity in Communications and Signal Processing.A. T. Paura. Chambers. Signal Process. Kailath. Leyman.R. IEEE Trans. Z. D.35] [10. Theory 40 (2) (1994) 340–349. 1994. Chen. 46 (11) (1998) 3099–3104. [10.E. IEEE Trans. Giannakis. Napolitano.A. Gelli. Inform.38] [10. Identiﬁcation of polyperiodic nonlinear systems. Spalvieri. Liu. Signal Process. Trieste. 45 (8) (1997) 1919–1930. [10.41] .40] [10. Spectralcorrelation based methods for multipath channel identiﬁcation.12] L. Nonminimum phase channel deconvolution using the complex cepstrum of the cyclic autocorrelation. Y. Kailath. Cyclostationarityexploiting methods for multipathchannel identiﬁcation. Blind equalization in a cyclostationary context. L. 45 (8) (1997) 2023–2036. Blind channel identiﬁcation and equalization using spectral correlation measurements. Tugnait.K. 5. Tong. Exploitation of cyclostationarity for identifying the Volterra kernels of nonlinear systems. May 12.A. Twostage interference immune blind equaliser which exploits cyclostationary statistics.A.163. L. 5. Blind equalization and estimation of FIR communication channels using fractional sampling. Polyspectral and cyclostationary approaches for identiﬁcation of closedloop systems.27] [10. 4 (1997) 204–206. Fractionality spaced equalizing circuits and method. H. 1998. Dissertation. D. Izzo. Izzo. IEEE Trans. Paura. IEEE Trans. Automat. T. Guglielmi. Spectralcorrelation based methods for multipath channel identiﬁcation.26] [10. B. G.C. Hatzinakos. Doucet. IEEE Trans. S.13] W. 42 (11) (1994) 3026–3042. Prakriya. G. pp. L. Kailath. Signal Process. Tong. Electron.A. Ph. IEEE Press. [10. Identiﬁability of a band limited system from its cyclostationary output autocorrelation.29] [10. Ding.C. P. Soong.20] A.D. [10. Blind identiﬁcation and equalization based on second order statistics: a time domain approach. Blind channel identiﬁcation and equalization using spectral correlation measurements. Inform. Electron. 33 (5) (1997) 356–357. G.15] Y. L. in: W. 42 (2) (1994) 483–485. Recent developments in blind channel equalization: from cyclostationarity to subspaces.751. [10. Technol.21] L. Gardner (Ed. 1994. Giannakis. A. Blind secondorder identiﬁcation of FIR channels: forced cyclostationarity and structured subspace methods. 45 (7) (1997) 1815–1830.). A. in: W. New York. 91–416. Higherorder cyclostationaritybased methods for identifying Volterra systems by [10.14] G. pp. Trans. Mattera. Napoli. 42 (12) (1994) 3483–3494.B. Commun. 3 (1) (1992) 341–348. IEEE Signal Process.D. Xu. [10.D. Hassibi. [10. IEEE Trans. Tong. New York. pp. C. A. US Patent No. 1992. IEEE Trans. A.).L. Tsatsanis. IEEE Trans. Archer. A. Part I: Frequencydomain analysis. Signal Process. Gardner. G. J. Signal Processing 31 (1) (1993) 17–29. Theory 41 (1) (1995) 308–311. February 1994 (in Italian). IEEE Trans. Gardner. A. B.H. 1996. Theory 39 (3) (1993) 535–542.17] D. Italy. A. A. Napolitano. Bayesian deconvolution of cyclostationary processes based on point processes.A. Loubaton.34] [10. M. Serpedin. IEEE Trans. Ph. Liang.32] [10. Leyman.R.31] [10. Ding.39] [10. Inform. Transmitter induced cyclostationarity for blind channel equalization. IEEE Press. E. [10.B. 1997 (in French). 46 (3) (1998) 779–783. Napolitano. Multipath channel identiﬁcation by an improved Prony algorithm based on spectral correlation measurements.B. Paura. Asymptotically optimal blind fractionally spaced channel estimation and performance analysis. S. Gardner et al. Telecommun. Relat.24] G.768. November 10. Signal Processing 46 (1995) 75–83 [10. Cyclostationarity in Communications and Signal Processing. Signal Process. Blind identiﬁcation of ARMA channels with periodically modulated inputs. L. H. Lett. 437–454. [10. On blind identiﬁability of multipath channels using fractional sampling and secondorder cyclostationary statistics. Signal Process. C. Golden. Halford. Blind channel estimation using the secondorder statistics: algorithms. Dissertation. IEEE Trans. [10. US Patent No. [10.30] [10. T. (Almost) periodic FIR system identiﬁcation using thirdorder cyclicstatistics.22] L. Tugnait. Y. G.37] [10. Control 40 (5) (1995) 882–885 [10. Xu. Signal Process. IEEE Trans. G. Inform. ENSTTelecom Paris.H. Italy.
7] W. Loubaton. G. Otbor Peredacha Inform. Chevreuil. E. 69–72. 419–424.S. [10. IEEE Trans. Chevreuil. 25.5] W. Davis. European Trans. Bradaric. [11. Yavorskii. Blind channel identiﬁcation and equalization using nonredundant periodic modulation precoders: performance analysis. A. 72 (1982) 1827–1830.O. Mazzenga.J.47] [10. Henrio. Giannakis. Mezentsev. Signal Process. [11. Petropulu.46] [10. Radio Eng. 47 (2) (1999) 572–578. Giridhar. Inform. Signal Process. Linear prediction error method for blind identiﬁcation of periodically timevarying channels. Bolcskei. Nonparametric signal detection and identiﬁcation of modulation type for signals hidden in noise: an application of the new theory of cyclic spectral analysis. Blind channel identiﬁcation and equalization with modulation induced cyclostationarity.56] S. Coherent and noncoherent detection of signals in cyclostationary noise and of cyclostationary signals in stationary noise. I. J. Twostage interferenceresistant adaptive periodically timevarying CMA blind equalization. 47 (3) (1999) 848–856.C.A. F. [11. IEEE Trans.L. Gardner et al. 50 (1) (2002) 96–109.43] [10. Mattera.A. [11.M. Theory IT28 (1982) 924–932. H. Signal interception: a unifying theoretical framework for feature detection. IEEE Trans.782.P. Exploiting input cyclostationarity for blind channel identiﬁcation in OFDM systems. Electron. Technol. Hurd. Blind equalizationcase of unknown symbol period. IEEE Trans.50] [10.45] [10.S. 48 (6) (2000) 1570–1586.B. Celebi. 50 (3) (2002) 662–672. Houcke. A. 1987. Commun. pp. 53 (2) (2004) 457–465.54] [10.N. Gardner.55] [10. Signal interception: a unifying theoretical framework for feature detection. 1979. 6. Graphical methods for determining the presence of periodic correlation. Verde.K. Ph. 119–124. IEEE Trans. Otbor Peredacha Inform. L. 9 (1) (1998) 27–31.2] D. US Patent No. 11 (3) (2000) 293–296. Luo. 46 (7) (1998) 1930–1944. [11. Soc. pp. Hleiss. J. awarded S. Eur. COM36 (1988) 897–906. Anal.48] [10. in: Proceedings of the 11th GRETSI Symposium on Signal and Image Processing. Spooner. Signal Process. Tanda. P. The problem of veriﬁcation of a stochastic rhythmicity model. IEEE Trans.6] H. 2004. 1983. 1988. Ocean. Rice Prize Paper from the IEEE Communications Society. Methods and devices for simplifying blind channel estimation of the channel impulse response for a DMT signal. P.A. Eng. Wagstaff.A. San Diego.L. 1198–1202).G. F. Rybakov. Dragan. W. Tugnait. Periodic nonstationary processes. R.A. Telecommun.8] D. Eur. G. Optimum detection of a class of random Gaussian pulsed signals in the presence of noise with a priori parametric uncertainty. Signal Processing 85 (7) (2005) 1275–1286.B. [11. Izzo. E. IEEE Trans. A. Paura. Cyclic spectral analysis for signal detection and modulation recognition.10] W.11] W. Time Ser. Nice. in: Proceedings of MILCOM 88. Optimum and suboptimum detector performance for signals in cyclostationary noise. A. Signal Processing 75 (1) (1999) 41–50. G. Loubaton. Gelli. Department of Electrical and Computer Engineering. MIMO blind secondorder equalization method and conjugate cyclostationarity.13] N. Signal Process. D.P. R. IEEE Trans. Signal Process. Phys. IEEE Trans. August 24. Giannakis. Duhamel.D. Heat. 51 (3) (2003) 781–793. J. Bukofzer. Chevreuil.52] [10. H.W. Meas. Bolcskei. Signal Process. 50 (1977) 28–34 (in Russian). R. A subspacebased ¨ approach to blind channel identiﬁcation in pulse shaping OFDM/OQAM systems. [11. CA.W.51] [10. Health. Signal Process. Structural characterization of locally optimum detectors in terms of locally optimum estimators and correlators. IEEE J. I.57] S.3] Y. P. Narasimhan. 49 (7) (2001) 1594–1598. [11. Trans. IEEE Trans. Antoni. [11. Signal Processing 67 (1) (1998) 77–98.I. Relat. Gardner. Technol. IEEE Trans. Blind channel ¨ identiﬁcation and equalization in OFDMbased multiantenna systems. K. 689 [10. Hurd.4] B.V.042. G. H. 25 (1980) 62–66 (English translation of Radiotekhnika Electron. Serpedin.44] [10. [11. University of California. Gardner. in: Proceedings of Ninth Symposium on Signal Processing and its Applications. Bilyik. 59 (1980) 3–12 (in Russian).49] [10. Gardner.1] M. A. C. P. Diamantaras. Gerr. Gardner. Loubaton. V. M. Amer.L. Blind MIMO FIR channel identiﬁcation based on secondorder spectra correlations. IEEE Trans. 1 (3) (1990) 233–238. [11.9] W. Instrumentat.A. Signal Process. Channel blind identiﬁcation based on cyclostationarity and group delay. 51 (6) (2003) 1668–1674. 12 (1991) 337–350. . Mazzenga. Serpedin. [11. Signal Process. OE12 (1987) 97–115. Telecommun. Technol. Telecommun.12] L.53] [10. Trans. Optimum and suboptimum detection of weak signals in cyclostationary nonGaussian noise. pp. Giannakis.42] [10. Dissertation. Identiﬁcation of polyperiodic Volterra systems by means of input–output noisy measurements.P. 50 (12) (2002) 3070–3082. F. Hazarataiah. On the identiﬁability of a channel transfer function from cyclic and conjugate cyclic statistics. Bukofzer. Paulraj. S. Clipping degradation in a system used to detect timeperiodic variance ﬂuctuations. Relat. / Signal Processing 86 (2006) 639–697 input–output noisy measurements. Acoust. P.V. France. Signal Process. Blind multipath channel identiﬁcation for wideband communication systems based on cyclostationary statistics. [11.B. Relat. P. J. IEEE Trans.ARTICLE IN PRESS W. M. J. H a —A consistent estimator for frequency response function with input and output noise.
Rostaing. Tanda. Y. Yeung. Izzo. [11. Gardner. Ph. Dissertation. Marchand. Dipartimento di Ingegneria Elettronica. [11.W.and secondorder cyclostationary sources—application to AM. Telecommun. 41 (6) (1993) 905–916. 5.36] [11. 1995. Subotic. F.24] L. P. Spooner. M. 47 (1) (1999) 10–13. Napoli. P. 1980. Vecchia. B. Ferreol.21] C. 1995 (in French). and deterministic sources. in: Proceedings of the Third International Symposium on Stochastic Hydraulics. [11. [11.J. W. Antoni. US Patent No. Signal Process. Signal Process. [11. IEEE Trans.25] G. Eur.20] A. A.239. Sills. 1998. Simple test for distinguishing constant from time varying amplitude in harmonic retrieval problem. Lacoume. US Patent No.15] W.A. IEEE Trans. Dissertation. B. in: Proceedings of International Conference on Acoustics. Ferreol. Ph. L. Gardner.S. Ann. L. Carrara.19] W. Asymptotic performance analysis of cyclic detectors. 3 (1) (1992) 39–43.6] . Biometrika 67 (1980) 365–373. Tokyo.A. D. Paciﬁc Grove. Spooner. F.M. bilinear and nonstationary. Paura. MD. [11. Cyclic spectral analysis and timefrequency analysis for automatic transmission classiﬁcation. Paura. 4. Giannakis. Thelen.455. [11. Robust feature detection for cyclostationary signals. IEEE Trans. Commun. IEEE Trans. Commun.3] [12. Spooner. J. 1967. Baltimore. [11. Izzo.K. ´ A. Detection of a digitally modulated signal using binary line enhancement. 47 (4) (1999) 1137–1141. Blind source separation using secondorder cyclostationary statistics. Grenoble. Brelsford. Gournay. Speech. C. P. Napolitano. France.J. Signal Process. August 6. Signal Processing 85 (2005) 51–66. 4. 42 (9) (1994) 2355–2369. 6.D. Bonnardot.1] [12. Cyclostationary signal models for the detection and characterization of vibrating objects in SAR data. Process of cyclic detection in diversity of polarization of digital cyclostationary radioelectric signals. vol. Troutman.22] R. 1998. C. C. CA. [11.C.32] [11. 2004. Delayandmultiply detectors for signal interception in nonGaussian noise. France. IEEE Trans. [11. Biometrika 66 (1979) 219–228. Signal recognizer for communications signals.16] L. Black. Hidden periodic autoregressivemoving average models in time series data. IEEE Trans.38] [11. in: Proceedings of International Conference on Acoustics. pp.14] A. October 3.B. and Signal Processing (ICASSP’98). J. IEEE Trans. C. in: Proceedings of Quinzieme Colloque GRETSI. University of NiceSophia Antipolis. / Signal Processing 86 (2006) 639–697 Dissertation.27] G. Biometrika 78 (1) (1991) 53–63.A. US Patent No. 44 (3) (1996) 368–376. ´ [11. Y. R. Commun. 1998 (in French). Ph. Detection of modulated signals by exploiting their cyclostationarity properties: application to sonar signals.R.746. 2002. J. Statistician 27 (3–4) (1978) 237–254.34] [11. 2165–2168. Xiang. JuanlesPins. Multiple hypothesis modulation classiﬁcation based on cyclic cumulants of different orders. Tanda.39] [12.D. Pitarque. R. February 1992 (in Italian). P. Relat. 1304–1308. Tanda. Inst. E. M. T.V. and Signal Processing (ICASSP’98).107–115 G. Besson. Systems. Manton. F. Smith. Grenoble. Signal interception in nonGaussian noise. vol. 6 (1978) 1310–1317. Boiteau. 48 (6) (2000) 1712–1725. Paura. Izzo.35] [11. Signal Process. 1996 (in French). Gardner. J. Dissertation.33] [11.H. On periodic and multiple autoregressions. IEEE Trans. M.M. Grupe. [11.846. L. ´ A. Hua. Johns Hopkins University.A. Dandawate. 1998. L. Uni` versita di Napoli Federico II. Granger. and Computers. pp. Statistical test for presence of cyclostationarity. Detection and classiﬁcation of digital modulations by higher order cyclic statistics. Signal Process. C. 44 (5) (1996) 1214–1223.23] P. IEEE Trans. Gelli. February 10. El Badaoui. Rostaing.A.37] [11. Trans. Commun. Q.M. Signal detection in cyclostationary generalized Gaussian noise with unknown parameters.4] [12. Gardner et al. M.A.M. G.D. Searchefﬁcient methods of detection of cyclostationary signals. Guillet.D. Albera. Some new time series models: nonlinear. Statist.690. Ph. 2157–2160. Commun.ARTICLE IN PRESS 690 W. Tiao. 40 (1992) 1030–1037.A. W. J. Some results in periodic autoregression.W.30] [11. Ballerini. Chevalier. Secondorder blind separation of ﬁrst. [11. JuanLesPins.L.430. K. France. L. Testing for periodic autocorrelations in seasonal time series data.18] M. O.2] [12. [11. P. Thierry. M. Gardner. Speech. N. Marchand.29] P. 52 (4) (2004) 845–861. Signal Process. A general maximum likelihood framework for modulation classiﬁcation. P. Cyclostationaritybased algorithms for signal interception in nonGaussian noise. Le Martret. Stoica. [11. IEEE Trans. pp. Nicolas. 1995. Probability predictions and time series with periodic structure. [11. 49 (4) (2001) 694–701. Ferreol.26] P. [11.31] [11.28] D. Pagano. Italy. On the behavior of current second and higher order blind source separation methods for cyclostationary sources. Tanda. in: Proceedings of 15th GRTESI Symposium on Signal and Image Processing.D. Nice. Blind separation of convolved cyclostationary processes. W. AbedMeraim. in: Proceedings of the 32nd Asilomar Conference on Signals. Poly. IEEE Trans. Gardner. M.5] [12. Technol. Chevalier. 40 (1) (1992) 149–159. 42 (5) (1994) 2165–2173. Salas. Detection and source location of weak cyclostationary signals: simpliﬁcations of the maximumlikelihood receiver.M.A. Cyclostationaritybased signal detection and source location in nonGaussian noise. Le Martret. 6. Signal interception: performance advantages of cyclic feature detectors. Correlation properties of periodic ARðpÞ models. Commun.V.R. INPGNat. IEEE Trans.17] A.
24 (1993) 703–711. Boes. Periodic autoregression with exogenous variables and equal variances. Ph. Gardner et al. P. [12. 26 (1990) 855–861.27] H. [12.A. Sakai. Cipra.1) processes. Dordrecht.K. model adequacy and periodic correlation in time series forecasting. Andel. Bolzern. 8 (1987) 293–300. pp. De Nicolao.30] H. On interpolation on periodic autoregressive processes. [12. [12. Hui. IEEE Press. Acoust.25] W. Estimation in multiple autoregressivemoving average models using periodicity. [12.). Math.34] A. G. Parameter estimation for periodic ARMA models.D. Hallin. IEEE Press. Rubio. Anderson. Math. M. France. [12.12] J. Salas. 1994. Boston. Commun. Maximum likelihood estimation for periodic autoregressive moving average models. Manage. Salas. Int. Anal. On periodic autoregression with unknown mean. Statist. McLeod. Vecchia. 237–245.D.T. D. [12. J. 22 (1986) 1385–1396. J. [12. in: Proceedings of the NATO Advanced Research Workshop. prediction. J. Robinson (Eds. Forecasting of multivariate periodic autoregressive movingaverage processes.9] J.K.A. A minimum prediction error algorithm for estimation of periodic ARMA models. Water Resour.V. J. J. Bittanti. et al. MA. Aplikace Motemotiky 34 (1989) 387–395.23] A. 1986. Ved.8] H. Thompstone. Time Series Analysis. Sakai. Periodic moving average process. Periodic autoregressivemoving average (PARMA) modeling with applications to water resources. J. Tusty.38] G. Boshnakov. 14 (6) (1993). Appl.I. M. Time Ser. Grossman. Thompstone.A. 35–49. 61 (1993) 387–393. IEEE Trans. J. Salas. D. Multivariate periodic ARMA(1.11] A. Recursive computation of the parameters of periodic autoregressive moving average processes. 30 (1985) 218–229.M. Appl.I. Anderson. 14 (1993) 1–18.J. Anal. Sihka. Time Ser. [12. Piroddi.V. [12. Colorado State University. Water Resour. [12. On the invertibility of periodic movingaverage models. [12. Anal. in: Proceedings of the European Control Conference. Vecchia. Modeling 5 (1981) 241–245. J. Gardner (Ed.32] M. Math. J. [12.37] A. 12 (1991) 73–82. 17 (1988) 1483–1494. Anal. (Eds.K.J. ´ [12. J.22] T. D. Modeling of aggregated hydrologic time series. Cyclostationarity in Communications and Signal Processing.C. Asymptotic results for periodic autoregressive movingaverage processes. Statist.7] S. De Nicolao. E. 1985. 1985.N. 267–294 (Chapter 5). K. Time Ser. Circular lattice ﬁltering using Pagano’s method.16] A.D. Combining hydrologic forecasts. R.13] J. Obeysekera. [12. Dandawate. pp. pp. 44 (3) (1996) 673–684.G. Ann. 86 (1986) 197–219. in: O.28] T. Modeling periodic moving average processes using cyclic statistics. 17 (1995) 333–349. [12. Hallin. / Signal Processing 86 (2006) 639–697 [12. G. Ord. Water Resour.L. Anal. 239–249. Smith. Locally optimal tests against periodical autocorrelation: parametric and nonparametric approaches. Speech Signal Process. Periodic gamma autoregressive processes for operational hydrology.B. [12. [12. [12. Signal Process. Math. J. Comput. Res. [12. [12. CO. Periodic autoregression with exogenous variables and periodic variances.W. Hipel. 18 (1982) 1006–1010. McLeod. Grenoble.K. Rev. Anal. A. Vecchia. ASSP (1982) 279–287. Andel. 480–493.C.D. Akod.35] T. Reidel. Technometrics 27 (1985) 375–384. Simultaneous conﬁdence bands for the spectral estimate of twochannel autoregressive processes. [12.10] H.31] P. 15 (1993) 263–268. Shell. Appl. On recent developments in prediction theory for cyclostationary processes. Piroddi. Plan. L. 30 (1985) 126–139. Statist. in: W. in: W. Purassanta. Time Ser.M. Res. Math. [12. Bentarzi.A.14] T.19] S. Andel. J.). Aggregation and estimation for periodic autoregressive moving average models. Cesk. Sakaguchi. Sakai. [12. Groningen. [12. Li. Time Ser. A. Modelling and predictions of stochastic processes involving periodicities. Res.V. Miamee. Anal.39] G. 16 (1995) 127–146. Miamee. A.D. New York. Noakes. Time Ser. Probability Theory and Mathematical Statistics with Applications.18] J.36] S. 691 [12. Biowas.40] M. 31 (1986) 480–485. An algorithm for the exact likelihood of periodic autoregressive moving average models.). [12.W. Holland. Periodic covariance stationarity of multivariate periodic autoregressive moving average processes. 1983. Bolzern.21] J.20] B.ARTICLE IN PRESS W. Estimation of ARMA models with seasonal parameters. pp. Explicit formulas for the best linear predictor and predictor error matrix of a periodically correlated process. Bentarzi. Representation. Anal. Amsterdam.T. Obeysekera. Goodwin. [12. [12. On the spectral density matrix of a periodic ARMA process. Y. Anal. Cipra. Fort Collins. Adams. 21 (1985) 721–730. Bartolini.A.26] J.B. vol. Dissertation. 11 (1990) 49–56. Covariance matrices characterization by a set of scalar partial autocorrelation coefﬁcients. A.29] S. Res. Econometric Theory 12 (1996) 88–112. 6. Appl. Grouping of periodic autoregressive models. Hipel.17] A. 113 (1987) 29–41.41] A. P.B. J. Giannakis. G. Time Ser. The periodic prediction problem for cyclostationary processes—an introduction. [12. Elsevier. L. New York. Bittanti. Gardner (Ed. Fernandez. McLeod. . Time Ser. IEEE Trans. Parsimony. Ula. [12.V.).A. [12. SIAM J. 1994. Sakai. Andel. Vecchia.G.D. P. Ula. F. Water Resour. K. J.24] P. Water Resources Bull. [12. Water Resour. G. in: W. R. 24 (1988) 1237–1246. K. NL.V. Salas.I. J.33] A.A. Hydrol. 1991 [12. 11 (1983) 337–340. Bittanti. Simulat.15] R.
Anderson.V. The cumulant theory of cyclostationary timeseries.A.D.10] [13. M. [12. [12. New York. Ankara. LambertLacroix. Wickert. Leondes (Ed. in: Proceedings of VII European Signal Processing Conference (EUSIPCO’94). 1994 C. [12. Gardner. Shao. New York.M. Turkey. W. 19 (2) (1999) 287–322. Statist. Anal. Stochastic Process.). Appl. Meerschaert.42] P. Albuquerque. 42 (12) (1994) 3409–3429. Recursive prediction and likelihood evaluation for periodic ARMA models.T. IEEE Trans. 1990. On AR(1) models with periodic and almost periodic coefﬁcients. UK. Giannakis. G. R.D. in: W.M.G. 2. Asymptotic theory of mixed time averages and Kthorder cyclic moment and cumulant statistics. Theory 39 (6) (1993) 1864–1876.13] [13.1] V. Anderson.A.5] W. Symmetry properties of highorder spectral densities of stationary and periodicnonstationary stochastic processes.M.15] [13. cyclic cumulants. Anal.D.3] A. W. in: A.M. Statist.A. Dandawate. Basawa. [12. 25 (3) (2004) 359–372. pp. vol. Parameter estimation for periodically stationary time series. Honolulu. Gosh (Ed.B.48] R.). Gardner. 35 (1) (1998) 46–54.M. 26 (4) (2005) 489–518. Proceeding of the Workshop on Nonstationary Stochastic Processes and their Applications. C.49] I.V. Signal Process. Paura.52] Q. Ann. 83 (1999) 149–169. J. Systems. A.O. Rochester. Computation and characterization of autocorrelations and partial autocorrelations in periodic ARMA models. IEEE Trans. Nonparametric polyspectral estimators for kthorder (almost) cyclostationary processes. Lund. Asymptotics. Part II: Development and applications. A. [12. R. Large sample properties of parameter estimates for periodic ARMA models. Vecchia. Alekseev. Appl.9] [13.L. Theory 40 (1) (1994) 67–84. Math. Reed. Spooner.17] [13. 48 (6) (2000) 1800–1803. Gardner. / Signal Processing 86 (2006) 639–697 Information Theory and its Applications (ISITA’90). Estimation of cyclic polyspectra. HI. J. and cyclic polyspectra. Paciﬁc Grove. Inform.18] [13. Time Ser. Appl. Spooner. Ph. Control and Signal Processing.14] [13. A. A. A.D. Charlottesville. Spooner.7] [13. [12. M. Hurd. Gardner.M. pp. Inform. Modeling and inference for periodically correlated time series. Zivanovic. 55 (1) (2003) 41–68.). C.ARTICLE IN PRESS 692 W. Theory 41 (1) (1995) 216–232. Cartaxo. Time Ser. M. Department of Electrical and Computer Engineering.L.M. Gardner (Ed.53] P. Dandawate. 1991.2] D. G. World Scientiﬁc. Problems Inform. Makagon. C. 1548–1552 [13. 1994. in: Proceedings of 25th Annual Asilomar Conference on Signals.A. IEEE Trans. Probab. University of Virginia.). NY. Hallin. Stochastic Process. 65. On periodic autoregressive processes estimation. L. J. Spooner. Amblard. Amsterdam. Dandawate. Gardner. and Computers.V. in: Proceedings of the Fifth IEEE/ ASSP Workshop on Spectrum Estimation and Modeling. Spectral characterization of nth order cyclostationarity. Control and Dynamic Systems. I. Inst. C. January 1993 ´ A. Giannakis. Spooner. 1990. P. Part I: Foundation. Marchand. Academic Press.45] P. Franq. IEEE Press. Elsevier Science. L. IEEE Trans. Lund. Gardner et al. Signal Process.A. 251–255 [13. Anal. [12. Meerschaert. M. in: S. Transmission 23 (1987) 210–215.16] [13. 355–358 C. Higherorder statistics for nonlinear processing of cyclostationary signals. Time Ser. Virginia. [12. and Signal Processing (ICASSP’91). Theoretical prediction of periodically correlated sequences. J.M. Dissertation. Theory 35 (1989) 700–703. Bentarzi.11] [13. Napolitano. MIMO Volterra system input/output relations for cyclic higherorder statistics. 1994 (Chapter 1). Proceedings of the 1990 Bilkent International Conference on New Trends in Communication. Periodic moving averages of random variables with regularly varying tails.A.51] A.6] [13. Makagon. vol. Cyclostationary signal processing. Theory and application of higherorder cyclostationarity.M. P.19] . University of California.A. IEEE Trans. 100 (2002) 167–185. Dissertation. [13. Nonparametrics and Time Series. IEEE Trans. IEEE Trans. J. 25 (1997) 771–785. Anal. VA.4] W. [12. Miamee (Ed. 1990. Dandawate. I. Gardner. 1997 [12.A. Arikan (Ed. Inform. Spooner. Basawa. W. Statist. in: C. Netherlands. A general property of the norder moment generating function of strictsense cyclostationary processes. Spooner. W. Consistent and asymptotically Normal estimators for cyclically timedependent linear models. Signal Process.46] A. ´ A.A. ´ A. Speech. Higher than secondorder statistics for complexvalued [12. [13.A. 1992.44] M. Lund.V. Inform. 110–125. in: Proceedings of the International Conference on Acoustics.43] R.B.T. Izzo. Gardner. pp.8] [13. 42 (12) (1994) 3387–3408. Spooner. The cumulant theory of cyclostationary timeseries.B. Nonparametric cyclicpolyspectral analysis of AM signals and processes with missing observations. J. C. Nonstationary moments of a random binary pulse train.). Davis. Meerschaert.E. Probab. VA. W.M. Giannakis. 22 (6) (2001) 651–664.47] S. 91–167 (Chapter 2). 21 (1) (2000) 75–93. CA. M.V. Some aspects of the higherorder cyclostationary theory. Innovations algorithm for periodically stationary time series.12] [13.G.A.V. pp.L. pp. 1991 ˘ ´ G. Cyclostationarity in Communications and Signal Processing. ´ A. Edinburgh. Basawa.O. CA. Anderson. Lacoume.L.50] H. G. Time Ser. Bibi. An overview of higherorder cyclostationarity. Ann.M. [13. Math. in: E.L. Miamee. in: Proceedings of the 1990 International Symposium on [13. Lund. C. Ph. [12. Higher order cyclostationarity. 1991 C. Exploiting cyclostationarity and higherorder statistics in signal processing.
Lindquist (Eds. Kadiman. Napolitano. pp. Paciﬁc Grove. received EURASIP’s Best Paper of the Year award S. CA.M.N. Izzo. K.ARTICLE IN PRESS W. Brown. Izzo. A.30] [14. B 32 (1982) 1384–1398. Systems 28 (1985) 63–65 (English translation of Izv. Nauka.16] [13. Frequency ﬂuctuations in isochronous selfoscillators at periodically nonstationary random actions.7] [14. Strom. in: Proceedings of 15th GRETSI Symposium. Cyclic higherorder statistics: input/ output relations for discrete. Acoust. Appl. Khotuntsev.22] [14. Eng. Izzo. A. A. Optimal control of servicing in a multichannel system. Flagiello. A.B. Silaiev. V. Williamson. Commun. IEEE Trans. IEEE Trans. K. E.P.V. Byrnes. A. Spooner.V. Commun. Isokawa. Periodic nonstationarity of noise in harmonic systems. Paciﬁc Grove. Signal Processing 56 (1997) 279–292. K. Radio Eng. IEEE Trans. Systems Sci.N.L. Pechinkin. pp. Systems 29 (1986) 17–21. France. Y. A.I. Radiophys. P. Circuits Systems CAS (1977) 531–541.V. Dragan. 297–309.14] [14. Amblard. Sov. Izdat.8] [13. C.28] [14. .15] [13. Automatica 23 (1987) 741–747. Semenov. 1995 L. Correlation relations between the complex amplitudes of the noise components in bipolar and ﬁeldeffect transistors in the presence of strong signals. Solids 18 (1983) 14–20. Speech Signal Process. G. pp. CA. 2001 A. On the stability of linear stochastic systems with periodically nonstationary parametric excitation. S. Frequency Domain and State Space Methods for Linear Systems.29] [13. Fluctuations in autooscillators for periodically nonstationary shot noise.A.R. Moskow.3] [14. Obrezkov. Yakimov. Yavorskii.M.31] [14. Systems. Higherorder cyclostationarity properties of sampled timeseries.I.I. VUZ Radioelectron. 48 (4) (2000) 1040–1052.A. Ghozati. Akashi (Ed.23] [14.P.V. 48 (2000) 1462–1466. 2. D. Rozenvasser. An identiﬁcation problem in almost and asymptotically almost periodically correlated processes. 39 (1974) 23–31 (in Russian). Radioelectron. vol. J. Masson. V.L. Spooner.9] [13. V.26] [14. Ferrara Jr. B. Signal Process.Y.M.). Polyspectral analysis of mixed processes and coupled harmonics. ASSP (1981) 413–422. Y. 46 (2) (1998) 429–439.). Hydroacoustic communication channels with surface scattering represented in 693 [13. Lacoume.I. Paciﬁc Grove.24] [13. Kyoto. Mech.M. Japan. Otbor Peredacha Inform. Napolitano. C. Bruyevich. Spooner. Cybernet. / Signal Processing 86 (2006) 639–697 cyclostationary signals. S. F. 32 (1987) 126–130. Quantum Electron. 1985. J. Zhou. A. Multirate processing of timeseries exhibiting higherorder cyclostationarity. Signal Processing 54 (1996) 303–307. Giannakis. in: Proceedings of the 35th Annual Asilomar Conference on Signals. 2000 C. 69–72 A. Napolitano. P.18] [14. Stability properties of periodic ﬁlters. A. E. Common. 23 (1968) 91–96. 1973 (in Russian). Amsterdam.2] [14. Approximate analysis of pulsed tracking systems when there is ﬂuctuating interference. in: Proceedings of the 34th Annual Asilomar Conference on Signals. NorthHolland. Napolitano. Cyclostationarity in the digital regulation of continuous time systems.4] terms of linear timevarying ﬁlters. Commun.27] [14.. Signal Processing 42 (2) (1995) 147–166. and Computers. Acoust. G. Optimal cyclostationary control: a parameteroptimization frequencydomain approach. 857–862 Y. T. and Computers. Technol. L. Int. in: Proceedings of the 29th Annual Asilomar Conference on Signals.K.1] [14. Pelkowitz. in: H. G. CA. Frequencydomain implementations of periodically timevarying ﬁlters.11] [13. Frequency domain analysis of wraparound error in fast convolution algorithms. J.A. Phys. Izzo.13] [13. Rootenberg. IEEE Trans. Napolitano. J. Guardabassi. Systems. IEEE Trans. and Computers. G. S.6] [13.G. A model of the test signal noise in channels with digital signal processing. Yeung. ASSP (1985) 883–892. Leonov. Y. L. Signell. Williamson. Signal Process. in: C.20] [14. IEEE Trans. On the utility of sixthorder cyclic cumulants for RF signal classiﬁcation. Periodically Nonstationary Control Systems. I.V. Signal Process. Electron. Kadiman.25] [14. Dragan. Inform. 1982.10] [13.12] [13. Speech Signal Process. Radioelectron. L. 24 (1979) 1806–1811 (in Russian). Zaytsev. Napolitano.A. L. 1997 (in French).21] [14. Spooner. D. Mogilevskaya. Gardner et al. Proceedings of the Eighth Triennial World Congress of the International Federation of Automatic Control. JuanLesPins. Systems. J. Analysis of periodically switched linear circuits. Smirnov. Y.and continuoustime MIMO linear almostperiodically timevariant systems. W. 1995 C. 17 (1979) 36–42 L.19] [14. A.).N. Frolov. A. A computationally efﬁcient and interference tolerant nonparametric algorithm for LTI system identiﬁcation based on higherorder cyclic statistics. Higherorder statistics for Rice’s representation of cyclostationary signals. Classiﬁcation of cochannel communication signals using cyclic cumulants. Schell. G.M. Higher Order Statistics for Signal Processing. 19 (1982) 456–462. Voloshin. Sizyakov. in: Proceedings of the 29th Annual Asilomar Conference on Signals. Higherorder cyclostationarity properties of coded communication signals.17] [14. Paris. Czech. J. Discrete minimax linear quadratic regulation of continuoustime systems.O. Radiotekhnika Elektron.A. A. Yavorskii. Application of the group pulse processes in the theory of Barkhausen noise. Theory 42 (3) (1996) 943–958.A. Vokurka. Systems. Paciﬁc Grove. CA. Medianbased cyclic polyspectrum estimation. and Computers.5] [14. 8 (1977) 953–959. Automatic radiofrequency environment analysis. Probab. 1995. Bittanti. V. IEEE Trans.
Acoust. [15. Tolouse.28] F. J. Randall. Cyclostationary processes: application in gear faults early diagnosis. 14 (3) (2000) 371–385. [15. S. Antoni. Automatica (1992). Kato.B. Long.20] J. pp. NJ. Systems Signal Process. Capdessus. G. PrenticeHall.19] J. S. applications to early diagnosis for helicopter gearbox and classiﬁcation method based on HOCS. Systems Signal Process. Kato. 149–152.026.7] T.30] P.R. G. 14 (3) (2000) 387–412.F. in: Proceedings of IEEE SouthEast Conference ’00. D. Acoust. Electron. 6.15] J. McCormick. A. [15. US Patent No. US Patent No. Goette. Bouillaut. 131–138. Gonzales. Ghione. Daniere. [15.S. Hurd. DusserreTelmon.P. [14. Systems Signal Process. US Patent No. 257 (5) (2002) 839–856. Donati. 1994. Mech. [15. Zerbi.12] G.P.B. [14. Boehme. Soc. in: Proceedings of XI European Signal Processing Conference (EUSIPCO’02). De Nicolao. Thirdorder cyclic characterization of vibration signals in rotating machinery. Sound Vib. May 21. Tanimoto. Circuits Systems (1993) 581–590. Nashville.A. 104–109. 2000. Devices 49 (9) (2002) 1640–1647. P.C. J. B. 15 (5) (2001) 923–943. 1991. 2004. R.687. Mech. Antoni. [14. European Patent No. Hernandez. [15.25] H. IEEE Trans. Sidahmed. 248 (5) (2001) 829–845. 2003. Braun. in: Proceedings of IEEE SouthEast Conference ’00. Exact noise analysis of SC circuits and an approximate computer implementation.K.11] L. 2 (2003) 455–458. [15. M. F.6] T. Digital Control and Implementation. March 18. Effective vibration analysis of IC engines using cyclostationarity. 072. Pitch determination of music signals using the generalized spectrum. Starikov. Inc. E. [14. M. Bonani. [15. Vib. R.. Spectral factorization of linear periodic systems with application to the optimal prediction of periodic ARMA models.1] S. A prediction method for probability distribution of road trafﬁc noise at an intersection. Amer.23] S. ASME J. [15. F. G. [15. Roychowdhury. K. Glossiotis. Cyclostationarity in rotating machinery vibrations. June 27. [15. Comput. Raad. [14. New York. 2000. Antoniadis.17] G.536. Long. Adelaide.D. US Patent No. Sugawara. Randall. Prieux. C.29] K. Application of cyclostationarity and timefrequency analysis to engine car diagnostics. J. Electron. Differential diagnosis of gear and bearing faults. 2002. Donohue. The relationship between spectral correlation and envelope analysis in the diagnostics of bearing faults and other cyclostationary machine signals.8] C. T. T.F. Mech. pp. Gobet. Black. J. Katz (Ed. Mech. ˘ ` ´ L.13] R.26] J. K. Nashville.D. Feldmann.K. Australia. / Signal Processing 86 (2006) 639–697 [15. 6. 257 (5) (2002) 815–837. A stochastic model for simulation and diagnostics of rolling element bearings [14.L. D. Antoni. J. 70 (1980) 513–526. H. Noise source modeling for cyclostationary noise analysis in largesignal device operation. 1111364. J. [15. AIP Press. 6.B. Williamson. [14.2] S. 2001. L. Effectiveness and sensitivity of vibration processing techniques for local fault detection in gears. Roychowdhury. A.C. J. Black. Randall. 12 (2) (1998) 225–242. J.L. Bittanti. Process for the detection of damage to components of an engine. Yamaguchi. J. D. Cyclostationary analysis of rollingelement bearing vibration signals. Systems Signal Process. R. February 3.B. Effective vibration analysis of IC engines using cyclostationarity. Sidahmed. France. [15. Dalpiaz. M. Australia 18 (1990) 46–50.18] A. Dynamical systems with cyclostationary orbits. Sidahmed. [14.). Rivola. 947.16] J. [14. Mech.24] M. Frequency correlation analysis for periodic echoes. Vaissiere.4] D. IEEE Trans. June 6. Lacoume. A practical method of predicting noise produced by road trafﬁc controlled by trafﬁc signals. 887. Antoni. J. 328 (1991) 299–315. [15. Y. Seth.9] G. Okumura.R. Guillet. J. Apparatus and method for reducedorder modeling of timevarying systems and computer storage medium containing the same. Systems Signal Process. Guillet. Jones. Solid (3) (1998) 324–336. J. IEEE J. Sound Vib. Method of making an integrated circuit including noise modeling and prediction. in: R. .P. 389.H. Antoni. Numerical noise analysis for nonlinear circuits with a periodic large signal excitation including cyclostationary noise sources.B. TN. [15. 2000. 15 (5) (2001) 945–962. 1994. DusserreTelmon. Gullapalli.T.5] A. Cyclostationary noise analysis of large RF circuits with multitone excitations. Speech. Englewood Cliffs. Part I—A methodology for condition monitoring. in: Proceedings of International Conference on Acoustics. Franklin Inst. 86 (1989) 2206–2214. Part II —New results on the reconstruction of the cylinder pressure. V. Monte Carlo simulation in electronic noise in semiconductor materials and devices operating under cyclostationary conditions. S. Donohue. Flores. Shiktorov. Daniere.14] J. Gardner et al. The Chaos Paradigm. J. Roychowdhury. Varani. Prieux. [14. Damage detection of motor pieces. TN. S. Itakura. 6. R. P. D.10] I. F. Rubini. Method and apparatus for analyzing small signal response and noise in nonlinear circuits.22] D. J.27] J.31] J. Yamaguchi. Circuits Systems 36 (1989) 508–521. Reggiani. D.658. C. [15. Flores. Analysis of repetitive mechanism signature. Randall. 124 (2) (2002) 165–171. Gruzinskis. IEEE Trans.3] Y. Feldmann. Nandi. Antoni. and Signal Processing (ICASSP). Chobsaard. Bittanti. G. [15.E. Acoust. F. Sound Vib. The simple pendulum and the periodic LQG control problem. [15. [14. Koenig. Sound Vib. Cyclostationary approach and bilinear approach: comparison. [14. 2002.ARTICLE IN PRESS 694 W.21] S.
Assoc.23] [17.10] [16.R. Donohue. J.5] [17. 96 (6) (1994) 3504–3515. K. 2685–2686.M. On detecting and modeling periodic 695 [15. K. Varghese. 5.P. Forsberg. Smith. Les´ kow.9] [16. 725–728. / Signal Processing 86 (2006) 639–697 with localized faults. J. NV. V.H.D. Statist. T. H. F. P. Genis. UK. H. Cyclostationary modelling of rotating machine vibration signal. On periodic structures and testing for seasonal unit roots. Piccoli.D. Statist.11] [16. 267 (2003) 253–265. Bus.7] [17. Cyclic statistics in rolling bearing diagnosis. P.D.M.R. Donohue. D. Raad.D. Weron. D. R. A. E. Osborn. K. Econometrics 48 (1991) 373–384. Haung. F. Bus. Chatterjee. Tardu. S. H.D.H.3] [15. Systems Signal Process. Ghysels. Varghese.7] [16. Statist. Econom. Antoni. N. Ghysels. Franses.9] [16. Donohue. 13. L. in: Review of Progress in Quantitative NonDestructive Evaluation. T.4] [17. Statist. 1995. 1998. Periodic autoregressive conditional heteroscedasticity. Contributions to the higherorder cyclic statistics: applications to gear faults. Varghese. Varghese. T. Econom. Seattle. C. A cyclostationary least mean squares algorithm for discrimination of ventricular tachycardia from sinus rhythm. 48 (1995) 241–248. IEEE Trans. J. Donohue. 19 (1995) 683–704. K. Testing stationarity for stock market data. 19 (2005) 467–482.B.1] [16.R.J. pp.D. J. Ph. Testing for periodic integration. 1991. Spectral redundancy in tissue characterization. 56 (1994) 421–439. R. Ferroelectrics Frequency Control 40 (3) (1993) 330–337. Assoc.H. K. Seattle. Sound Vib. Dehay. Analysis and classiﬁcation of tissue with scatterer structure templates.D. in: Proceedings of the International Congress on Acoustics and Acoustical Society of America. Varghese.M. WA. pp. US Patent No. C. J. E. Comput. Econom. Kong. ´ Wy"omanska.12] [16. M. IEEE Trans.D. Universite de Technologie Compiegne. 1994. K.D. Econom. K. 336 (1–2) (2004) 196–205. Franses. Seattle WA. Osborne. Cheah.P. Parzen. Acoust. Goldberg. The effects of seasonally adjusting a periodic autoregressive process. J. Qu. . Makagon. Econom. Bilgutay. Brunswick. J. Lee. Cambridge. F.8] [17. pp.H. 7 (1989) 117–127 (see also J. pp. V. Imaging 22 (3) (2000) 137–152. F. P. F. Ultrason. 951–958. Amer.K.4] [17. A. Statist.6] [16. L. in: Proceedings of 1995 Ultrasonics Symposium. J. 50 (1996) 205–212.M. Duct detection and wall spacing estimation in breast tissue. A. The performance of periodic autoregressive models in forecasting seasonal UK consumption. Survey 12 (1998) 651–678. A. November 2003 (in French). B. 91 (1996) 1551–1559. Bonnardot.5] [16. 18 (6) (2004) 1285–1314. L.10] [17. E. Todd.D. Dyn. Bressler. Appl. Mech.24] [17. Z. 125 (3) (2003) 282–289.20] [15.22] [17. 1315–1318. Franses. France.H. N. Physica A Statist. Ultrason. F. Systems Signal Process. Z. Bilgutay.W. Donohue. Mech. Donohue.D. Donohue. Econometrics 3 (1989) 255266). Varghese. Model selection in periodic autoregressions. Characterization of unsteady time periodical turbulent ﬂows. pp. Ferroelectrics Frequency Control 42 (3) (1995) 451–463. Control 14 (1990) 763–796. Spectral correlation in ultrasonic pulseecho signal processing. Spectral correlation ﬁlters for ﬂaw detection. Cyclostationarity analysis for gearbox condition monitoring: approaches and effectiveness. Halpern. D. Mech.6] [17. Donohue. T. J.11] [17. Forsberg. A. McAleer. Varghese.14] [16. J.13] correlation in ﬁnancial data.D. Pagano.13] [16. 1999. 2001. Donohue. Donohue. BroszkiewiczSuwaj.ARTICLE IN PRESS W. 1995.1] [15.2] [15. Osborn. Data Anal. Donohue. Lett. Periodic linearquadratic models of seasonality. Reno.D. July 12. 1993. Ghysels. Disserta´ ` tion. Ultrasound Imaging 15 (1994) 238–254. T.A. K. Ferroelectrics Frequency Control 46 (2) (1999) 300–310. Specular echo imaging with spectral correlation. Genis. J. M.21] [17. D. Comptes Rendus Mecanique 331 (2003) 767–774 (in French). Maine. J. in: Twentieth International Symposium on Ultrasonic Imaging and Tissue Characterization. Com` piegne. 1995. The Econometric Analysis of Seasonal Time Series. J.893. Characterization of tissue microstructure with spectral crosscorrelation. Spectral redundancy in characterizing scatterer structures from ultrasonic echoes. Raad. T. Donohue. E. Econom. K. ASME J. Zhu. Themes in Modern Econometrics. Bollerslev.3] [16. Econometrics 9 (1979) 137–153. Paap. El Badaoui. T. T. Econom. Franses. Vib. Ultrason. Amer. Savic. 1353–1356. Varghese. Lett.P.327. Finelli.8] [16. M. E. An approach to modeling seasonally stationary timeseries. Appl. Jenkins. Amer. in: Proceedings of 1995 Ultrasonics Symposium. Duct size estimation in breast tissue. The analysis and classiﬁcation of smallscale tissue structures using the generalized spectrum. vol.S. Statist. 13. Oxford Bull. Estimating mean scatterer spacing with frequencysmoothed spectral autocorrelation function. Meanscatter spacing estimates with spectral correlation. M. The implications of periodically varying coefﬁcients for seasonal timeseries processes. R. J. K. Foresberg. 14 (2) (1996) 139–151.Y. vol. Soc. Detection of cholestrol deposits in arteries. P. E.J. Hall. T. Feng. Cointegration analysis of seasonal time series. K. K. Acoust. WA. L. Ultrason. Huang.R. Gardner et al. Li. Boswijk. Cambridge University Press. in: Proceedings of 1999 Ultrasonics Symposium.2] [17. in: Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Trans. Forsberg.12] [17. J.
Habibi. J. J. A. 39 (1974) 3–7 (in Russian). Math. F.3] Y. Virginia. Axis crossings and relative time of existence of a periodically nonstationary random process. [18.6] M. 1998. Donohue. Statist. Med. M. 1991. Hardin.A. Huang. Markelov.D.A. C.C. CA.J. Properties of counts of periodically correlated random processes. K. 11 (2) (2002) 135–151. World Scientiﬁc.15] K. G. Oper. Izv. Spectral estimation for strongly periodically correlated random ﬁelds deﬁned on R2. Burks. Rubin. [18.4] M. 39 (1991) 424–430. Paciﬁc Grove.ARTICLE IN PRESS 696 W. N.M. Piccoli. 58 (1979) 279–300. / Signal Processing 86 (2006) 639–697 IEEE INFOCOM’83. A. Petropulu. Stationary and cyclostationary ﬁnite buffer behaviour computation via Levinson’s method. in: Proceedings of Fourth International Conference on Signal Processing (ICSP’98).G. Dragan.H. L. AT (1984) 2159–2170.H. Models and level crossings of periodically nonstationary processes.G. C.G. 90 (2) (2004) 359–383. IEEE Trans. Huang. 36 (1973) 28–31 (in Russian). Fourier analysis for modelling some cyclic behaviour of networks. IEEE Trans. Tretiak. in: Proceedings of [19. Otbor Peredacha Inform. I. Clare.A.J. Detection of periodic autocorrelation in time series data via zero crossing. Periodic property of streetcar congestion at the ﬁrst station. Sov.G.C. Martin. Giannakis.2] [21. Anderson.4] [21. Y. Bell System Tech. F.P. Shankar. 19 (1996) 426–434. Hurd.G. Miamee. [18.4] [20. [17. Gefen. Dehay.6] [20. G. Gardner. Comput.L. Miamee. A singleserver queue with cyclostationary arrivals and arithmetic service. VUZ Radiotekhnika 9 (4) (1996) (in Russian). C.B. Otbor Peredacha Inform. T. High crossings of periodically nonstationary Gaussian process.A. H. Soc. Statistical properties of level crossings of a periodic nonstationary process and a periodic inhomogeneous ﬁeld.C. Image Process. 52 (1977) 3–8 (in Russian).G. I. Bilyik. 10 (1996) 1448–1461. Huang. Dumane. J. V. Li. 38 (1974) 15–21 (in Russian). 142 (1990) 191–202. Radiophys. Otbor Peredacha Inform.1] [20.G. 48 (1976) 27–33 (in Russian).16] K. Makagon. Kallianpur. in: Proceedings of the 29th Annual Asilomar Conference on Signals. UT. Oper. Otbor Peredacha Inform. Signal Process. D.G. B.S. Correlation estimation and timeseries modeling for nonstationary processes. Donohue. ROC analysis of classiﬁers based on ultrasonic tissue characterization features. Markelov.G.P. Kutay. Imaging 22 (2) (2003) 170–177.). San Diego. L.3] [19. Hurd. Bilyik. J.W. [19. Dragan. IEEE Trans. Yuan.10] D. [18. [18.D. [18. Probab. 165–175. W. J. Correlation autoregressive processes with application to helicopter noise. [18.5] M. Sound Vib. O. Speech and Signal Processing. Correlation autoregressive sequences: a summary. Strong laws of large numbers for two dimensional processes. [17. Hurd. W. [18. 1983. Spatiotemporal approach for timevarying image motion estimation. Georgiou.2] V.3] [21. Nakatsuka. 229–238.B. J.G.17] K. F. Q. Kaplan. Piccoli.18] S. VA. On a class of nonstationary stochastic processes. Otbor Peredacha Inform. [18. Correlation and spectral theory for periodically correlated random ﬁelds indexed on Z2.A. Chen. Donohue. Bilyik. Lewis. in: A. Ferroelectrics Frequency Control 50 (6) (2003) 724–729. Cohen. 9 (1966) 440–443 (in Russian). Ultrasound Med.1] V. J. G.T. L. On the spectral coherence of nonstationary processes. Miamee.5] [19. Huang. J. [17. W. F. T. 2001. Ultrason.6] [20. Goldberg. Farshidi. A queuing model for a hybrid data multiplexer. Ser. Forsberg.K.A. Gopinath. Ackroyd. D. 59 (1980) 24–31 (in Russian). Otbor Peredacha Inform. pp. 11 (2) (1991) 157–167. Signal Processing 15 (1988) 31–41.3] [20. Biol.P. Bilyik. V. IEEE Trans. P.7] M. H. and Computers.G. Ultrasonic scatterer structure classiﬁcation with the generalized spectrum. Foschini.8] M. 31 (1983) 184–205. Sankhya: Indian J. Nandhakumar. The periodic correlationrandom ﬁeld as a model for bidimensional ocean waves. Otbor Peredacha Inform. H. M. Forsberg. in: Proceedings of the IEEE International Conference on Acoustics. Some properties of periodically nonstationary processes and inhomogeneous ﬁelds. Forsberg.R. Gardner et al.4] [19. Yavorskii.9] M. Cheng.M. Systems. Statist. J.5] . Methods Statist. Hardin. CA. On extrusions and relative time of staying of the periodically nonstationary random process. B. Reid. Anal. 33 (1972) 9–12. Tissue classiﬁcation with generalized spectrum parameters. Gardner. Miamee (Ed. H. M.7] [21. L. Upcrossings of periodically nonstationary processes. Time Ser. 51 (1982) 15–21 (in Russian). Spectral correlation of randomly jittered periodic functions of two variables. 27 (11) (2001) 1505–1514. Res. Multivariate Anal. On the theory of upcrossings of periodically nonstationary processes.2] L. Bilyik. Genis. A (Part 2) 52 (1990) 145–156. Hardin.N. On spectral density estimates of Gaussian periodically correlated random ﬁelds.E. Piccoli.2] [20. B. A. A.A.1] R. Bilyik.1] [21. J. Proceeding of the Workshop on Nonstationary Stochastic Processes and their Applications. pp. Commun.W. 20 (4) (1999) 435–452. D. Res. Queueing analysis of TDMA with limited and unlimited buffer capacity. A. Salt Lake City. [18. Malignant and benign breast tissue classiﬁcation performance using a scatterer structure preclassiﬁer. Otbor Peredacha Inform.5] [20. Japan 29 (1986) 1–20. Donohue. 1995. V. Math. [17. G.D. [19.D. Alekseev.
Amerio. A. Chernenko.G. Napolitano.6] L. 1967 (Revised English edition translated by R. Gordon and Breach. Izzo.4] N. [22. in: Proceedings of XIII European Signal Processing Conference (EUSIPCO 2005). 1933 (also New York: Dover. Makagon. 50 (12) (2002) 2947–2961.11] L. 1974). 1933 (also Chelsea. in: Proceedings of the 26th Asilomar Conference on Systems and Computers. Inform. submitted. 1954). T. Cambridge University Press. Further references [23.5] B. [21. 1042–1047. Almost Periodic Functions. Leskow. Signal Process.A. Signal Process. et al. Generalized almostcyclostationary signals. C.9] L. Corduneanu. 43 (1988) 23–69. [21. 1932 (also New York: Dover. 1992.6] P. Linear periodically timevarying discretetime systems: aliasing and LTI approximations. 44 (5) (1957). New York. [21. IEEE Trans. T. Napolitano. New York. A dynamic multilocation supply model with redistribution between the stores. Chelsea. Foundations of the nonstochastic approach for time series analysis. Ann. Van Nostrand. Lii. [22. Timefrequency representations of generalized almostcyclostationary processes. Topics in the Theory of Random Noise.4] V. Hawkes (Ed. Cours de Mathematiques.7] G.S. in: Proceedings of XII European Signal Processing Conference (EUSIPCO 2004). [23. Bohr. 69 (1997) 179–193. [23.S. Model. vols. 1987.8] A. The higherorder theory of generalized almostcyclostationary timeseries. Stat. Wiener. Y. 2005. [22. Advances in Imaging and Electron Physics. vol. New York.11] C. S. Theory IT21 (1975) 605–611.12] K. Rosenblatt. pp.5] M. 697 [22. Prouse. Inform.16] L. Stratonovich. Paciﬁc Grove. . AlmostPeriodic Functions and Functional Equations. Signal Process. Springer. ´ [23.14] A. Almost Periodic Functions. IEEE Trans. A. A. A. The Fourier Integral and Certain of its Applications. IEEE Trans. Prague. V. Signal Process. Zemanian. vol. A.14] A. pp. Math. Napolitano. Sampling of generalized almostcyclostationary signals. 1972 (in Russian). NJ. Bass. Part II: Development and applications. Besicovitch. Res. Coll. Napolitano. Allen. Loeve. [21. 1989. Sci. IEEE Signal Process.18 (in Czech). Cambridge University Press. Linear timevariant transformations of generalized almostcyclostationary signals. [21. Princeton. Theory 49 (9) (2003) 2172–2191. [23. Austria. London. Napolitano.9] E.3] K. Uncertainty in measurements on spectrally correlated stochastic processes.8] R.S. [21. 50 (12) (2002) 2962–2975.E.S. 30 (1) (2002) 258–297. 50 (9) (2002) 2221–2229. 51 (6) (2003) 1546–1556.). Ann. [21.7] J. Napolitano.W. Elsevier. [22.10] L. Zalud. Circuits Systems CAS (1975) 162–168. 55 (1930) 117–258. A uniﬁed analysis of multirate and periodically timevarying digital ﬁlters. 2005. Section 1.G. Detecting target motion by frequencyplane smoothing. [23.10] A.13] L.15] J.3] H. Koronkevich. in: P. Gartshtein.2] R.H.13] A. Lett. Napolitano. III. Pfaffelhuber. / Signal Processing 86 (2006) 639–697 [21. Izzo. A. Hobbs. 1963. Assessing characteristics of periodic random signals. M. P.S. Almost Periodic Functions. Acta Math.A. A. Asymptotic Normality of statisticalfunction estimators for generalized almostcyclostationary processes. [22. IEEE Trans. [23... Kuznetsov. Masson & Cie. I and II.1] N. Miamee. Representation of linear periodically timevarying multirate systems.N. Stochastic Process. Statist.2] A. AN USSR. L. Amsterdam. LowNoise Semiconductor Circuits.7] L. 30 (1997) 225–235. [23. Izzo. Dover. New York. Ser. Napolitano. Algorithmization of Industrial Processes.ARTICLE IN PRESS W. IEEE Trans. [23. Napolitano. Napolitano. [23. 13 (1982) 267–286 (in German). 46 (11) (1998) 2975–2989. Chen. Generalized harmonic analysis.S. 103–223 (Chapter 3). Mathematical model and device for diagnostics of periodically nonstationary random processes. Meyer. Van Nostrand.17] A. 2004. 1997. Appl. 1971. [23. The Zak transform: a signal transform for sampled timecontinuous signals. New York. A.P. [21.J. Oper. ` [23. Distribution Theory and Transform Analysis. Berlin. Gardner et al. Chen. Signal Process. Burrus. CA. [22. On the spectrum of correlation autoregressive sequences.L. Antalya.M. Dragan. Janssen. Spectral analysis for harmonizable processes. Philips J. [21. Mehr. Smelyakov. Silverman). Cybernetics Institute. Chen. in: Proceedings of 16th GRETSI Symposium on Signal and Image Processing.A. Izzo. 135. [23. Turkey. 1958). Wiener.N. Qiu. SNTL. Mehr. [22. Kochel. Vienna. Voychishin.6] J. Generalized harmonic analysis for distributions. [23. IEEE Trans. Radiotekhnika 60 (1982) 116–123 (in Russian).8] A. 6 (1984) 45–48 (in Russian). V.I. Linear timevariant transformations of generalized almostcyclostationary signals. 1971. [21. Elektron. Paris. London.12] T. Systems Control Lett. On aliascomponent matrices of discretetime periodically timevarying systems. Signal Process. IEEE Trans.V. Izzo.15] A. L’vovskii Inst. Meansquare consistency of statisticalfunction estimators for generalized almostcyclostationary processes. Kulesov. Probability Theory. [23. [21.A. Part I: Theory and method.1] O. 1980. IEEE Trans. Optim. Izzo. 8 (2001) 114–116.
This action might not be possible to undo. Are you sure you want to continue?