Noise Cancellation in OFDM Systems Using
Adaptive Complex Narrowband IIR Filtering
Georgi Iliev, Zlatka Nikolova, Vladimir Poulkov, Georgi Stoyanov
Dept. of Telecommunications, Technical University of Sofia, Bulgaria
Abstract – In this paper a very low sensitivity fourth-order
complex band-pass filter section with independent tuning of the
central frequency and the bandwidth is developed. A
narrowband adaptive filter structure is formed around this
section, using LMS algorithm to adapt the central frequency.
The developed filter structure is providing a low computational
complexity and a very fast convergence. We use this adaptive
complex filter for noise cancellation in an OFDM transmission
scheme and show that under certain conditions SNR gain and
better bit error rate (BER) performance can be achieved.
I. INTRODUCTION
Adaptive notch filters are indispensable for many digital
signal processing (DSP) applications to separate, enhance or
track narrowband and harmonic signals mixed with
broadband noise. If these signals consist of real and imaginary
parts, a complex coefficients adaptive filter could be
implemented.
The complex adaptive systems are subject of study in a
number of publications. In [1] a new complex adaptive notch
filter for parameter estimation of complex sinusoids and chirp
signals mixed with wideband noise is proposed. A complex
multirate filter-bank-based adaptive notch filter is developed
to solve the co-existence problem of CDMA and TDMA
systems [2]. Reference [3] investigates digital complex
adaptive filter algorithms to cancel the radio frequency
interference in VDSL system. In [4] cascaded realization of
complex adaptive infinite impulse response (IIR) notch filters
are discussed.
In all these papers, like in many others, the attention is
focused mainly on the adaptive algorithms, their efficiency
and methods of improvement. But the adaptive complex
circuits used and their properties shouldn’t be neglected as
they influence significantly the quality of the adaptive
process. These circuits are, however, merely mentioned in
[1]–[4] and they tend, generally to be quite complicated.
In this paper we develop a fourth-order complex band-pass
(BP) filter section with independent tuning of the central
frequency and bandwidth (BW). The new complex variable
filter conducts very well in finite word-length environment
and demonstrates very low coefficient sensitivity. We use this
BP filter section in an orthogonal frequency division
multiplexing (OFDM) transmission scheme for noise
cancellation and performance improvement. The concept is
based on the realization of a low intermediate frequency (IF)
quadrature receiver similar to the one suggested in [5] with
application to wireless communication systems [6].
The paper is organized as follows. The next section
describes the developed OFDM scheme employing fourth-
order complex adaptive bandpass digital filtering for noise
2859
1-4244-0355-3/06/$20.00 (c) 2006 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2006 proceedings.
cancellation. In section III the fourth-order variable complex
digital filter is derived and investigated in different word-
length environments. Section IV presents adaptive complex
narrowband filtering based on the developed fourth-order
filter using LMS algorithm. In order to verify all theoretical
results in section V intensive simulations are conducted
exploring the main dependencies between filter parameters
and system characteristics leading ultimately to better
performance in severe noise conditions. Finally, section VI
concludes the paper summarizing the main results of this
publication.
II. OFDM TRANSMISSION WITH NOISE CANCELLATION
Orthogonal frequency division multiplexing is a
promising technique for high speed digital transmission in
different applications such as digital audio and video
broadcasting, wireless local area networks and broadband
mobile communications. The concept of a OFDM system
with noise cancellation using an adaptive complex
narrowband filter for such wireless applications is the
following. At the transmitter the complex OFDM signal,
generated via an inverse Fourier transform (IFFT) using
digital signal processing (DSP) techniques, is used to
modulate a low IF carrier signal. Then the two routes of the
complex signal are translated to RF employing both in-phase
and quadrature (I/Q) modulation.
DSP module
I
A/D LPF
Complex D
Adaptive E
IIR FFT M
RF
Q Filter A
P
A/D LPF
IF
Generator
IF control
Figure 1. Principle of an OFDM receiver with noise cancellation.
At the receiver (Fig.1) the operations are essentially
reverse. The signal from the antenna, after filtering and
amplification by a low noise amplifier in the RF module, is
demodulated in two channels - inphase and quadrature. The
signals from these channels, corresponding to the real and
complex parts of the signal from the IFFT at the transmitter,
after A/D conversion are fed to the inputs of the adaptive
complex narrowband IIR filter for noise cancellation.
Conversion to baseband is performed by IF demodulation and
low-pass filtering (LPF). The complex multicarrier baseband
signal is fed to the fast fourier transform (FFT) circuit and
after demapping is demodulated to n-QAM and binary form.
An IF control signal from the filter is used to avoid any
frequency mismatch of the IF frequency in the demodulation
process. This control signal is a function of the coefficient θ
responsible for the central frequency of the filter. As seen
from Fig.1, all signal operations after the RF module are
performed digitally and could be implemented in one digital
signal processing (DSP) circuit.
coefficients into Taylor series with respect to β and truncate
after the linear term) to eliminate these loops.
As a result the multiplier coefficients a and b in the section
(Fig. 2) are replaced by the composite multipliers ( â and b̂ )
containing also β , after applying the LP to LP spectral
LP −V
transformation (1). Thereby, the transfer function H LS − II (z )
of the BW-variable LP filter is obtained. Thus, we have two
variable by β multipliers aˆ =a+β. f and bˆ=b+β.e , where f
and e are coefficients calculated by means of a and b as
follows: f =a(2a+b−4) and e=b(2a +b−2) . The structure of a
composite multiplier is shown in Fig 3.

III. COMPLEX DIGITAL FILTER CIRCUIT DERIVATION β f

The narrowband complex digital filter in the OFDM â
receiver should satisfy the following requirements. The filter a +
must be of IIR type in order to ensure higher selectivity with
Figure 3. The composite multiplier obtained after Constantinides LP to LP
lower order transfer function. It should contain canonical spectral transformation and followed Taylor series truncation.
number of multipliers and delays and it must be free of
parasitic effects like delay-free loops. The last problem can be The rotation transformation:
avoided after applying a proper design procedure, like the z = ze− j θ or z −1 = z −1(cosθ+ j sin θ) , (3)
proposed in [7] new method of design of complex variable
BP/BS filters with independently tuneable central frequency LP −V
applied on H LS − II ( z ) generates the BP complex coefficient
and BW. That method provides wider range of tuning of the BP −V
BW and higher freedom of the central frequency tuning. The transfer function H LS − II ( z ) , which is variable with respect
new filter should have low coefficient sensitivity in order to to both, the BW and the central frequency. As complex
decrease the degradation of its magnitude characteristic after BP −V
function H LS − II ( z ) can be presented by two real coefficients
the coefficients are quantized. And we try to develop narrow-
fourth-order functions:
band variable BP filters because such filters are needed for
most of the practical applications. LP −V z = ze − jθ
We start with a real coefficients lowpass (LP) filter and H LS → H BP −V ( z ) = H R ( z ) + jH I ( z ) ,
− II ( z )  (4)
LS − II
make its BW variable by applying the popular Constantinides
and realized by the complex structure in Fig.4.
LP to LP spectral transformation:
0,5
z −1 −β In
z −1 → . (1) Re
1−βz −1 + +
Out Re
One of the best possible second-order sections to realize a + b̂ â + +
narrowband LP filter (poles near z=1) is the one shown in cos θ
Fig. 2 [7] [8]. Its transfer function is + z-1 + + z-1 +

LP
H LS − II ( z )=0,5a
(1+ z −1 )2 . (2)
sin θ sin θ sin θ sin θ

1+(−2+2a+b)z −1 +(1−b )z −2
+ z-1 + + z-1 +
0,5 cos θ cos θ
In + b̂ â + + Out Im
+ + + +
In 0,5
+ b a + Im

z-1 + + z-1 Figure 4. Complex-coefficient fourth-order LS-II-based BP/BS filter section.

+ Having two inputs and two outputs (both couples real and
Out
Figure 2. Real-coefficients second-order low-sensitivity LP filter section.
imaginary), the structure is described by four real-coefficient
BP transfer functions two by two equal with ±sign and related
And as the realization of (1) is producing delay-free-loops, BP −V
to H LS − II (z ) as follows:
we apply the procedure proposed in [9] (expand all

2860

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2006 proceedings.
 (z )= H IIBP (z )= H R (z )= 0,5aˆ1+ [(B2 + 2 )A−]1z +
BP
−1 magnitude and phase responses of H R (z ) (5) (the bold lines)
H RR
1+ 2 B2 Az + and H I (z ) (6) is tuned by trimming of θ , while β=0.13 is
(5)
[ 2
2
1
2 −2
]
+ 2 B + (1+ B )(A − C ) z + 2aˆ − bˆ A z −3 + B z − 4 [( )] 1
fixed. The tuning of the BW by changing β and fixed θ=π 3
[ ]
+ B22 + 2 B1 (A 2 − C 2 ) z −2 + 2 B1 B2 Az −3 + B12 z −4 is without problems for rather wide range as it is shown in
Fig.5b.
and Due to the very low coefficient sensitivity of the initial LP
− (B 2 − 2 )Cz −1 + section, the variable complex LS-II-based filter section behaves
H RIBP (z )= − H IRBP (z )= H I (z )= 0,5aˆ
1+ 2 B 2 Az −1 + very well even when the coefficients are severely quantized.
(6)
(
+ 2bˆACz − 2 + 2 aˆ + 3bˆ − 4 Cz − 3
,
) Obviously the coefficients could be quantized to very short
word-length without affecting considerably the filter
[ ]
+ B 2 + 2 B1 (A 2 − C 2 ) z − 2 + 2 B1 B 2 Az − 3 + B12 z − 4
2
characteristics (Fig.6). The magnitude response isn’t
where destroyed even for very short word-length independent of the
values of the tuning parameters θ and β .
B1 =1−bˆ; B 2 = 2aˆ +bˆ −2; A=cos θ ; C =sin θ . (7)

θ=π /10 θ=3π /10 θ=π /2 θ=7π /10 θ=9π /10 β=0.05
θ=π/4

β=0.2
θ=4π/5

θ=π /10 θ=3π /10 θ=π /2 θ=7π /10 θ=9π /10

Figure 6. Magnitude response of the variable complex BP fourth-order filter

section for different word-length and θ and β .
(a)
In Fig. 7 the worst-case sensitivity of the complex LS-II-
based filter is compared to two other well-known second-
β= 0
β= 0.25
order complex structures – the direct-form and the minimum-
β= 0.2 norm-based for the same pole position p1,2= 0,939±j0,1997
β=−0.25 near the unit cycle. The LS-II-based filter demonstrates about
10 times lower worst-case sensitivity in comparison to the
minimum-norm-based and about 50 times lower than the
direct-form-based.

β= 0
β= 0.25
β= 0.2
β=−0.25

(b)
Figure 5. Magnitude responses of the variable complex BP fourth-order filter
for (a) different values of θ ; (b) different values β .

The BW of the obtained variable complex filter can be

tuned by changing of β , while the central frequency can be
varied by changing of θ .
In Fig.5a it is shown how the central frequency of the Figure 7. Worst-case sensitivity comparison of narrow-band complex BP
direct form, minimum norm and LS-II-based filter sections.

2861

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2006 proceedings.
 IV. ADAPTIVE COMPLEX NARROWBAND FILTERING situations K is approximately equal to 0.1.
In Fig. 8 the block-diagram of the new adaptive filter is
V. EXPERIMENTS
shown. In the following we consider the input/output relations
for corresponding BP/BS filters - equations (8)-(11). First we test our filter for detection of narrowband complex
ADAPTIVE signals. Input signal is a mixture of additive white Gaussian
ALGORITHM noise (AWGN) and OFDM modulated signal simulated
+ eR(n) according to the scheme presented in section II.
xR(n)
FOURTH-ORDER yR(n)
COMPLEX
FILTER yI(n)
xI(n)
+ eI(n)
Figure 8. Block-diagram of a versatile adaptive complex narrowband filter.

For the BP filter we have the following real output:

y R (n)= y R1 (n)+ y R 2 (n) , (8)
where y R1 (n) is the part of the real output signal when the
input signal is x R (n) , while y R 2 (n) is obtained when x I (n)
is the input signal. Similarly, for the imaginary output we get:
yI (n)= yI 1(n)+ yI 2 (n) . (9)
For the band-stop filter we have - real output: Figure 9. Trajectories of filter coefficient θ for different step size µ, β = 0.1.
eR (n)= xR (n)− yR (n) , (10)
In Fig. 9 the learning curves for different values of the step
and imaginary output size µ are shown. It is seen that for µ=0.005 the convergence
eI (n)= xI (n)− yI (n) . (11) is so fast that it takes less than 100 iterations to adapt.
In Fig. 10 the results for different filter bandwidths are
The cost-function is the power of band-stop filter output presented. It is obvious that the filter is behaving quite
signal: conventionally. It should be mentioned that if some other
[e(n)e ∗ (n)] , (12) (non low-sensitivity) LP-prototype were used, the coefficient
where β could not take values smaller than - 0.1 without destroying
the magnitude shape. Thus we can always obtain a faster
e(n)=e R (n)+ je I (n) . (13)
convergence because of the wider BW.
We apply a Least Mean Squares (LMS) algorithm to up- Next, in Fig. 11 we show the behaviour of our filter for
date the filter coefficient responsible for the central frequency quite a wide range of sinusoidal frequencies. It is seen that in
as follows: all the cases our filter converges to the proper frequency
∗ value.
θ(n+1)=θ(n)+µ Re[e(n) y ' (n)] , (14)
µ is the step size controlling the speed of convergence, (*)
denotes complex-conjugate, y′(n) is the derivative of
y(n)= y R (n)+ jy I (n) with respect to the coefficient subject of
adaptation.
In order to ensure the stability of the adaptive algorithm we
should set the range of the step size µ. We use the results
reported in [10]:
K
0<µ< , (15)
Trace( R)
or in a more convenient form:
K
0<µ< 2 . (16)
Nσ
In our case σ2 is the power of the signal y′(n), N is the filter
order and K is a constant depending on the statistical
characteristics of the input signal. In most of the practical Figure 10. Trajectories of filter coefficient θ for different BW, µ = 0.001.

2862

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2006 proceedings.

Figure 11. Trajectories of filter coefficient θ for different carrier frequency,
µ = 0.001, β = 0.1.
Finally, the performance of the OFDM scheme is tested. A
simulation model of the OFDM system depicted in section II
was developed to evaluate the bit error rate as a function of
the signal to noise ratio (SNR). For the simulation model a 16
QAM OFDM scheme is employed, with coding rate ½ and
number of 52 subcarriers and 4 pilots. For the case of
comparison the estimations of the BER/SNR curves were
conducted relatively to baseband in the cases with and
without adaptive filtering. The results in Fig. 12 show that an
average SNR gain of 10 dB is obtained for the proposed
OFDM scheme employing the developed adaptive complex
narrowband filter.
Figure 12. BER versus SNR.
VI. CONCLUSIONS
In this work an OFDM transmission scheme with noise
cancellation based on the realization of an IF quadrature receiver
and employing adaptive complex narrowband IIR filtering was
developed. The simulation experiments show that the scheme
manifests a better BER performance due to the improved method
2863
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2006 proceedings.
of design of the filter. The very low-sensitivity variable complex
fourth-order BP filter section has a very precise and completely
independent tuning of the BW and central frequency. The BW
can be tuned in much wider frequency range compared to other
well known methods and without any requirements to fix some
of the cut-off frequencies. By changing the transformation factor
θ , the central frequency can be tuned over the entire
frequency band also adaptively by applying an LMS
algorithm. The accuracy of tuning is very high and it is
possible to use coefficients with shorter word-length
decreasing thus the power consumptions for both, adaptive
filtering and updating the coefficients.
The convergence of the adaptive algorithm for the deve-
loped variable complex filter is investigated experimentally
and the efficiency of the adaptation is indisputable proved.
The main advantages of the proposed adaptive complex
structure and its application for noise cancellation in an
OFDM scheme could be summarized as follows: very low
coefficient sensitivity providing resistance against
quantization effects, very high accuracy of tuning and
adaptation, very fast convergence and low computational
complexity.
REFERENCES
[1] Soo-Chang Pei and Chien-Cheng Tseng, “Complex adaptive IIR notch
filter algorithm and its applications”. IEEE Transactions on Circuits
and Systems II: Analog and Digital Signal Processing, vol. 41, No. 2,
pp. 158-163, Feb. 1994.
[2] S. Hara, T. Matsuda, K. Ishikura and N. Morinaga, “Coexistence
problem of TDMA and DS-CDMA systems-application of complex
multirate filter bank”, Global Telecommunications Conference, 1996.
GLOBECOM '96. 'Communications: The Key to Global Prosperity, vol.
2, pp. 1281–1285, 1996.
[3] Yaohui Liu, Timo I. Laakso, Paolo S.R. Diniz. ”Adaptive RFI
Cancelation in VDSL Systems”, ECCTD’01- European Conference on
Circuit Theory and Design, August 28-31, 2001, Espoo, Finland
[4] S. Nishimura and Hai-Yun Jiang, “Cascaded realization of complex
adaptive IIR notch filters”, Proc. of the 40th Midwest Symposium on
Circuits and Systems, vol.2, pp. 691–693, 1997.
[5] V. Dias, “Complex signal sigma-delta modulators for quadrature
bandpass A/D conversion”, Microelectronics Journal, p. 505-524,
1996.
[6] A. Bahai, B. Saltzberg, M. Ergen, Multi-carrier Digital
Communications. Theory and Applications of OFDM, NY: Springer
Science, 2004.
[7] G. Stoyanov and Zl. Nikolova, “Improved Method of Design of
Complex Coefficients Variable IIR Digital Filters”, TELECOM’99,
Varna, Bulgaria, vol. 2, pp.40-46, Oct. 1999.
[8] G. Stoyanov, M. Kawamata, Zl. Valkova. “New first and second-order
very low-sensitivity bandpass/ bandstop complex digital filter sections”.
Proc. IEEE 1997 Region 10th Annual Conf. "TENCON’97", Brisbane,
Australia, vol. 1, pp. 61-64, Dec. 2-4, 1997
[9] S. Mitra, Y. Neuvo, H. Roivainen. ”Design of recursive digital filters
with variable characteristics”. Int. J. Circuits Theory Appl., vol.18,
pp.107-119. 1990.
[10] S. Douglas, “Adaptive filtering,” in Digital Signal Processing
Handbook. D. Williams and V. Madisetti, Eds. Boca Raton: CRC Press
LLC, pp. 451–619, 1999.