‘Tivee properties shared by many manmade signals are described: (1) the property that enables generation. Of spectral lines by quadratically transforming the signal: (2) the statis- ‘teal property called second-order ‘eyelostationarity, whichmeans that the autocorrelation function fluctuates periodically with time: and (@) the ‘correlation property for signal compon- fenis (n distinct spectral bands. It is ‘shown that these three properties are ‘Afferent manifestations ofa single at tribute called spectral redundancy. This intriguing attribute is studied. ‘and a vartety of ways for exploiting it to perform signal processing tasks in- woluing detection and estimation of highly comupted marunade signals are explained. as sano 28 © WH Exploitation of Spectral Redundancy in Cyclostationary Signals any conventional statistical signal processing ‘methods treat random signals as if they were sta istically stationary, in which case the parameters of the underlying physleal mechanism that generates the Signal would not vary with time. But for most manmade Signals encountered in communication, telemetry. radar land sonarsystems, some parameters do vary periodically ‘with time, In some eases even multiple incommensurate {not harmonically related) periodicities are involved. Ex amples include sinusoidal carriers in amplitude. phase fand frequency modulation systems. periodic keying ofthe amplitude, phase, or irequency in digital modulation sys. tems, and periodic scanning in television, facsimile, and some radar systems. Although in some cases these per fodiities ean be ignored by signal processors, such as re teivers which must detcet the presence of signals of in terest, estimate their parameters, and/or extract their messages. in many cases there can be much to gain in| terms of improvements in performance of these signal processors by recognizing and exploiting underlying pe- Fodicity. This typically requires that the random signal bbe modeled as cyclostationary. im which case the statis tieal parameters vary in time with single or multiple pe Fodieities, “This article explains that the cyclostationarity attr: bute, as it is reflected in the periodicities of (second-order) moments of the signal, ean be Interpreted in terms of the property that enables generation of spectral lines from the vest ensignal by putting it through a (quadratic) nonlinear trans. formation. It also explains the fundamental link between the spectral-line generation property and the statistical property called spectral correlation, which corresponds. to the correlation that exists between the random Nuc tuations of components of the signal residing in distinct spectral bands. The article goes on to explain the effects fn the spectral-correlation characteristics of some baste signal processing operations, such as filtering, product ‘modulation, and time sampling. It is shown how to use these results to derive the spectral-correlation character {sties for various types of manmade signals. ‘Some examples of signals that can be appropriately modeled as eyelostationary can be interpreted as the re sponse ofa linear or nonlinear system with some period: ically varying parameters to stationary random excita: tion, Specific examples include stationary random ‘modulation of the amplitude. phase, o frequency of & sinewave; stationary random modulation of the ampll: tudes, widths, or positions of pulses in an otherwise pe Flodie pulse train: periodically varying Doppler effect on aa stationary random wave: and periodic sampling. mul: Uiplexing, oF coding of stationary random data. In addi tion to these examples of manmade signals, some natal signals also exhibit eyclostationarity due, for example, (o Seasonal effects in time-series data sets obtained in me: teorology, climatology. atmospheric science, oceanogra phy. and hydrology, as well as astronomy. Numerous ex: amples and references are given in (1, Chapters 12 and 14], [2, Chapter 12) Finally, and most importantly, this article describes some ways of exploiting the inherent spectral redun- ancy associated with spectral correlation to perform various signal processing tasks. These include detecting the presence of signals buried in noise andlor severely masked by interference: recognizing such corrupted sig nals according to modulation type: estimating parame ters such as timediflerence-of-arrival at two reception platforms and direction of arrival at a reception array on a single platform: blind-adaptive spatial Gitering of sig nals impinging on a reception array; reduction of signal corruption due to cochannel interference andlor channe! fading for single-receiver systems: linear periodically time-variant prediction: and identification of linear and nonlinear systems from input and output measurements ‘The deseriplions include brief explanations of how and why the signal processors that exploit spectral redun- dancy can outperform their more conventional counter: parts that ignore spectral redundaney or, equivalently. Ignore cyclostationarity In the next section, the possibility of generating spectral lines by simply squaring the signal is illustrated for (wo types of signals: the random-amplitude modulated sine wave and the random-amplitude modulated periodic pulse train. Then itis explained that the property that sors espace‘enables spectralline generation with some type of ‘quadrae time-invariant transformation 1s called ‘yclostationarity and Is characterized by the cyelic ‘autocorrelation function, which Isa generalization of the conventional autocorreiation function. Following this, {is shown that a signal exhibits cyclostationarity and only if the signal is correlated with certain frequency- shifted versions of itself. In the third section, the correlation of frequeney shifted versions of signal is localized in the frequeney domain and this leads to the definition of a spectral correlation density function (SCD), It is then ex: plained that this finetion i the Fourer transform of the cyclic autocorrelation function. This Fourier. transform relation between these two functions in cludes as a special ease the well-known Wiener rela tion between the power spectral density funetion and the conventional autocorrelation function, A normal: ‘zation of the spectral-correlation density function that converts it intoa spectral correlation coefficient, whose ‘magnitude is between zero and unity. #8 then intro: duced as a convenient measure of the degree of spec: ‘yal reduadaney in a signal, Continuing in this SCD section, the effects on the spectral correlation density function of several signal processing operations are deseribed. These include fl tering and waveform multiplication, which in turnin clude the special eases of time delay and multipath propagation, bandlimiting, frequency conversion, and lume sampling. These results are used to derive the spectralorrelation density function for the random: amplitude modulated sine wave, the random-ampl: ‘ude madulated pulse train, and the binary phase-shift keyed sine wave, The spectral-correlation density functions for some other types of phase-shitt keyed signals are also described graphically. Finally in this, Section, the measurement of estimation of the ideal) speciral-correlation density function is brielly dis. cussed and illustrated with a simulation of a phase shift keyed signal ‘The fourth section of this article contains the payoff for working through the preceding two sections. I pro- vides explanations of how the spectral redundancy that is Inherent in signals that exhibit eyclostationar: ity can be exploited in a variety of statistical signal processing tasks. The spectral redundancy can gen trally be exploited to enhance the accuracy and tell ability of information gleaned from measurements of corrupted signals. Such information includes the fol lowing: 1) A decision as to the presence or absence of a ran dom signal with a particular modulation type in a background of noise and other modulated signals, A classification of multiple received signals in nois: according (o their modulation types, An estimate ofa signal parameter, such as carrier phase, pulse timing, or direction of arrival. or of the number of signals being received simulta neously. based on noise-and-interference-cor- rupted measurements, Anestimate ofa message being communicated by ‘signal over a channel corrupted by noise. inter 2 3 w vesewacat ference, and distortion, 5) A prediction of a future value of a random signal, and 6) Anestimate ofthe input-output relation ofa linear for nonlinear system based on measurements of the system’s response to random exeitation. ‘The article concludes with a brief section indicating how the theory of second-order cyclostationarity sur veyed in the preceding sections ean be generalized to higher-order eyclostationarity. which corresponds to the property that enables spectralline generation using higher order nonlinearities, such as cuble, quar tie, and so on. References to more in-depth treatments, of the theory and its applications are given throughout the article. CYCLOSTATIONARITY Spectral line genera A signal xt) contains a finite strength additive sine- wave component (an ae component) with frequeney a, say acosQrat +9) witha #0 o AF the Fourler coeiMicient ine) o {snot zero, in which ease (1) gives In (2), the operation (+) Is the time-averaging opera: Hon 1 > # sim 5 Im this case, the power spectral density (PSD) of x40 Includes a spectral line at frequency f=0 and ils image {f=-a. {The PSD is defined in the next section.) That is, ‘the PSD contains the adltive term! Inca ~ a) + 1 + a o where 4} is the Dirac delta, or impulse, function. For ‘convenience in the sequel, itis sald thal such a signal exhibits first-order periodicity, with frequency a. Let x{t) be decomposed into the sum of ils finite: strength sine-wave component, with frequency a, and lts residual. say n(0, M0) = ac0s rat + 0) + mi) a where nlt) 8 defined to be that which is left after sut traction of 1) from a(t) It fs assumed that nt is ran dom. Here, the term random is used to denote nothing saci