See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/283516900

Adaptive Multifunction Filter for Radar Signal Processing

Conference Paper · August 2013

DOI: 10.1109/ICCEEE.2013.6633904

CITATIONS READS

4 579

1 author:

Moutaman Mirghani
Institute of Space Research and Aerospace (ISRA), Sudan
30 PUBLICATIONS   22 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Evaluation of Quality of Encoded Quran Digital Audio Recording View project

ISRASAT1 Cube Satellite View project

All content following this page was uploaded by Moutaman Mirghani on 29 May 2017.

The user has requested enhancement of the downloaded file.

 Adaptive Multifunction Filter for Radar
Signal Processing
Moutaman Mirghani Daffalla
Department of Electrical and Computer Engineering
Engineering College – Karary University
Khartoum, Sudan
mtnmir@gmail.com

Abstract—Filtering of radar signals frequently takes place to II. MATCHED FILTERS

realize a certain task, such as interference reduction or Doppler
processing to remove clutter. However, in this paper a digital
Matched Filter (MF) is a filter that is designed according to
filter is proposed to be designed as a multifunction processor, the waveform of the transmitted radar signal. In presence of
achieving several sorts of signal processing within the same white noise only, the MF achieves in its output the maximum
module. Processing is easily done in the baseband rather than in SNR, compared to all other linear filters [1]. It is used in pulse
passband, for a moderately slow DSP processor. Hence, a lowpass compression to attain very narrow pulse out of long pulse of
filter will perform the task of rejection of the out-of-band low peak power [2]. The transfer function of a filter matched to
interference. Higher SNR can be obtained to enhance the a signal s(t) is given as the complex conjugate of the spectrum,
detection of targets inside noise if that filter is matched to the or Fourier transform, of that signal S(f) [3], i.e.
expected radar signal, even if it caused signal distortion. Doppler
processing makes use of digital filters to cancel signals from fixed H ( f )  kS * ( f )e  j 2fto (1)
and slow targets. The proposed filter would be designed as an where k is an arbitrary scalar that depends on the gain of the
adaptive filter to follow interference and clutter changes, and filter, and to is a delay required for best sampling instant.
eliminate blindness to some moving radar targets that arise in
digital Doppler processing due to aliasing caused by targets of
Equivalently, the impulse response of the MF can be found in
higher speeds. Matlab simulations are made to validate the idea terms of the signal as follows
and observe features and limitations of the proposed adaptive h(t )  ks* (t  t o )  ks(t o  t ) (2)
filter. Adaptive filter that is based on the least-mean-square
algorithm is designed to achieve the best outcome. Since not all waveforms have realizable matched filters,
new waveform designs are sought after. The waveform design
Keywords—Signal, Filter, Matched, Noise, Doppler, Adaptive, should satisfy a number of requirements that depend on the
Processing, Blindness. application and the nature of noise or clutter exists. However,
correlation between the transmitted and received signals can be
I. INTRODUCTION used instead of matched filtering.
When a correlator is used, the delay to is essential to be
The purpose of surveillance radar is target detection, in
synchronized so as to obtain the maximum SNR, while it is not
addition to other tasks, such as Tracking-While Scanning
critical in a MF. Matched filter is active in the presence of
(TWS). In target detection, the requirement is to detect the
white noise rather than colored-noise. Therefore in presence of
presence of a moving target like aircraft in the presence of
colored-noise, a whitening filter has to be applied prior to the
unwanted signals in a reliable manner.
matched filter or correlator, in order to alter the spectrum of
Those unwanted signals consist of interference caused by
noise and avoid mismatch with the signal.
signals produced by other nearby transmitters that could be
Integration is a class of matched filtering techniques that is
operating in the same band as the radar transmitter itself, and
used in Radar Signal Processing (RSP) in order to enhance the
the ubiquitous noise produced by electronic devices at the front
SNR of the received echo signal. The integrator accumulates
end of the receiver. Other source of unwanted signals is clutter,
the energy of subsequent echo pulses that are reflected by the
which is radar backscatter from objects other than the target
same target. On the other hand, noise signals are not correlated
that lie in the path of the transmitted signal.
and thus their integrated energy tends to vanish, and hence the
Filtering of signals is the major tool used in the signal
SNR at the output of the integrator is going to increase. Pre-
processing of radar signal returns. Most of analog filters have
detection integration takes place within the IF stage of the
been replaced by digital filters, particularly in baseband. Filters
receiver, realizing better SNR compared to post-detection
are used to shape the spectrum and rejects interference, while
integration.
passing the wanted echo signal with minimum distortion and
realizing more desirable higher Signal-to-Noise Ratio (SNR). III. DOPPLER PROCESSING
However, the same filter performing the task of rejection of Doppler processing of radar signals is applied to remove
the out-of-band interference could be adaptive to manage other fixed targets and raise the Signal-to-Clutter Ratio (SCR).
tasks. Here, the intended adaptive multifunction filter serves as Currently, there are two major processing techniques to
a matched filter as well as a Doppler processor. enhance the SCR in pulse radar, which are the Moving Target
Indication (MTI) used in short pulse radar, and Moving Target cos ωot and sin ωot, where ωo is the IF frequency. Each channel
Detection (MTD) that is used in pulse Doppler radar [4]. If the has its own MTI filter, whose output is orthogonally summed
wavelength of radar signal is λ, the Doppler frequency shift in to that of the other channel filter [3].
the signal received from a target moving with a relative
velocity vr (to or from the radar) is given as IV. DIGITAL FILTERS
2v Generally speaking, digital filters are used in separation and
fd  r (3)
 restoration of signals. Signal separation is needed when the
An MTI filter is easier to be implemented and costs a little, signal has been contaminated with interference and noise,
compared to MTD processors. In baseband, an MTI filter is while restoration of signal is used when a signal has been
Highpass Filter (HPF) that filters out signals from fixed targets, distorted in some way.
which are of zero Doppler frequency, and attenuates signals All of those filters used in radar signal processing and
come from slow targets. Unfortunately, MTI filter is either a mentioned earlier, i.e. Lowpass Filter (LPF), MF and MTI
digital filter or a discrete analog filter, and both of them have filters will be more reliable if they are implemented as digital
periodic frequency response that repeats each multiple of the filters. Digital filters are replacing analogue filters because of
sampling frequency, i.e. the Pulse Repetition Frequency (PRF) their reliability and their compatibility to Digital Signal
of the pulse radar. Processing (DSP). They are available in two flavors, Finite
MTI filter acts a proper HPF only within the Doppler Impulse Response (FIR) and Infinite Impulse Response (IIR)
frequency range -PRF/2 ≤ fd ≤ PRF/2, as illustrated in Fig. 1 filters [6]. Although IIR filters are faster and have fewer
below, where fd is the Doppler frequency of the echo signal, as components compared to FIR filters, they should be tested for
defined in equation (3). That complies with the Nyquist rate stability per each design changes. On the other hand, FIR filter
that restricts fs ≥ 2 fd , where fs is the sampling frequency. For is stable by definition and has linear phase response.
targets of higher Doppler shifts, velocity ambiguity occurs.

Fig. 1. Frequency Response of MTI Filter. Fig. 2. FIR Filter.

Consequently, MTI filter rejects some moving targets of From Fig 2 above, we can notice that the current sample of
nonzero Doppler frequencies. Those targets have frequencies the output signal of the FIR filter y(n) is the sum of the current
multiples to the PRF and they are known to move with blind sample of the input signal x(n) and previous samples multiplied
speeds, as the radar will be blind and wont detect them. The by the filter coefficients bk as follows
first blind speed of MTI radar is related to its PRF and N 1

wavelength as follows y (n)   bk x(n  k ) (5)

k 0
PRF 
vB  (4) where N is the number of taps or filter coefficients. N−1 is the
−1
2 filter order, which equals the number of delays marked Z .
Generally, MTI filters avoid blind speeds difficulty by
using a staggered PRF, rather than using fixed PRF. Hence, the V. ADAPTIVE FILTERS
probability that a target moves with one of the blind speeds will Due to its stability for all designs, FIR filter shown in Fig. 2
be considerably reduced [5]. above could be adaptive and its impulse response can be
In pulse Doppler radar, a bank of filters is used to achieve modified to adapt to the required signal or the unwanted noise
larger dynamic range in the Doppler frequency domain. This and clutter. For any FIR filter, the filter coefficients bk are the
procedure avoids filtering out moving targets due to blind samples of its impulse response h(n).
speeds and minimizes ambiguity in the velocity measurement. For instance, a LPF may be controlled so as to reshape its
Commonly, both MTI and MTD take place in the video frequency response with variation in the received signal. A MF
stages of the receiver (baseband), after the signal demodulation can be adapted to different waveforms transmitted by radar in
by means of phase detection, rather than using envelope order to enhance diversity of detection or to avoid jamming.
detection. Coherent demodulation is essential so as to preserve An MTI filter might change its notch to compensate for a
the phase information of the received signal. However, the IF moving radar platform, like in Airborne MTI (AMTI) radar [4].
signal should be demodulated with the aid of quadrature basis The basic concept of an adaptive filter is shown in Fig. 3
in order to avoid phase angle ambiguity. It is resolved into below. The objective is to filter the input signal x(n) with an
inphase and quadrature channels using two coherent signals, adaptive filter, in such a manner that it matches the desired
signal d(n). The desired signal is subtracted from the filtered The design of highpass, bandpass or bandstop filter typically
signal y(n) to generate an error signal. The error signal e(n) starts with a prototype lowpass filter, which is then converted
drives an adaptive algorithm, which generates the filter to the required filter [10].
coefficients bk or tap weights w(k) in a manner that minimizes For better filtering, signals within the passband are required
the error signal according to the adaptive algorithm. Mean to propagate through the filter, while those signals within the
Square Error (MSE) algorithms, which includes the Least- stopband should be adequately blocked or at least face high
Mean-Square (LMS) and the Recursive-Least-Squares (RLS) attenuation. Moreover, minimum level of ripples should exist
algorithms is one of the most popular known algorithms [7]. in both passband and stopband, and the transition band between
them is to be as fast as possible.
The impulse response of a HPF (such as MTI filter) is
formed by changing the sign of each sample of the impulse
response of the prototype LPF, and adding one to the sample at
the center of symmetry. This action in the time domain inverts
the frequency spectrum as stated in the conversion equations
below
H HPF (e j )  1  H LPF (e j ) (8)
hHPF (n)   (n)  hLPF (n) (9)
jω
where H(e ) and h(n) represents the transfer function and the
impulse response, respectively. The combination of cascading
Fig. 3. Adaptive FIR Filter. the noise reduction LPF with the MTI HPF can be replaced by
one BPF, which has an impulse response that is given through
In Fig. 3 above, e(n) is the estimation error at time n, which
the convolution of the two impulse responses as follows
is fed back into the adaptive algorithm to minimize a function
of this difference, known as the cost function [8]. In the case of hBPF (n)  hLPF (n) * hHPF (n) (10)
radar, the optimal output of the adaptive filter is equal in value The design of an FIR filter involves finding the coefficients
to the wanted echo signal. When the adaptive filter output is h(n) that result in a frequency response that satisfies given
equal to the desired signal, the error signal goes to zero. filter specifications. For an ideal LPF, its impulse response is
The MSE algorithms aim to minimize a cost function equal the sinc function, which extends to infinity. Convolving an
to the expectation of the square of the difference between the input signal with this response provides perfect low-pass
desired signal and the actual output of the adaptive filter y(n) filtering. However, the sinc function should be truncated to
 (n)  E{e 2 (n)}  E{[d (n)  y(n)]2 } (6) have h(n) of a finite number of samples. Due to truncation, the
frequency response of the filter involves ripples and slower
The LMS algorithm is a widely used algorithm in adaptive transition band. This results from the discontinuity at the ends
filtering due to its simplicity of computations. It is known as of the truncated sinc function. Increasing samples of h(n) does
stochastic gradient-based algorithm, which utilizes the gradient not reduce problems since discontinuity is of a major effect.
vector of the filter tap weights to converge on the optimal
Several windows are designed for FIR filters, so as to
Wiener solution [9]. Its relative simplicity has made it the
reduce side-lobes that occur due to rectangular windowing [6].
benchmark against which all other adaptive filtering algorithms
Hamming window is given for an FIR filter of length N as
are judged. For each iteration of the LMS algorithm, the filter
follows
tap weights of the adaptive filter are updated according to
2n
w(n  1)  w(n)  2e(n) x(n) (7) win (n)  0.54  0.46 cos( ) 0  n  N 1 (11)
N
The parameter μ is known as the step size, which is a small To design a windowed FIR filter, two parameters must be
positive constant. This step size controls the influence of the selected: the cutoff frequency fc and the length of h(n), N. The
updating factor. Selection of a suitable value for μ is essential frequency is expressed as a fraction of the sampling rate fs and
to the performance of the LMS algorithm. If the value is too lies between 0 and 0.5 (the folding frequency). Value of N sets
small, the time the adaptive filter takes to converge on the the roll-off according to an approximation of N = 4/BW, where
optimal solution will be too long. On the other hand, if μ is too BW is the transition bandwidth that is measured from where the
large, the adaptive filter becomes unstable and its output curve just leaves 1 to where it almost reaches 0. The transition
diverges from the desired signal [7]. bandwidth as well is expressed as a fraction of the sampling
frequency, between 0 and 0.5. Matlab is frequently used to
VI. DESIGN OF DIGITAL FILTER
simplify filter design. The FIR1 built-in function offers FIR
Now, we are going to design an FIR digital filter as that filter design using the window method to design an N th order
performs as a multifunction adaptive filter for radar signals. FIR digital filter of any of the four types, and returns the filter
Firstly, we are going to focus on the design of FIR filters in the coefficients in vector b of length N+1 as follows
frequency domain. An FIR filter is totally described by its
b  fir1( N , c , ' ftype ' , window) (12)
impulse response h(n) that is formed by the filter coefficients.
 The text ftype indicates the type of the filter, where window contrary, it distorts the received echo signal waveform to
indicates the window used in the design of the FIR filter. The realize better SNR that assists the decision of detection. Hence,
cut-off frequency ɷc must be between 0 and 1.0, where 1.0 is if we are going to integrate that filter in the proposed multi-
the folding frequency. For BPF, ɷc is a vector of two samples, function filter, we should accept compromises.
which are the lower and the upper cutoff frequencies. Therefore, we have selected that proposed filter to be an
adaptive filter. Rather than using three FIR filters each of self-
VII. DESIGN OF MATCHED FILTER determined coefficients, only one adaptive filter is expected to
Making use of Equation (2) above, the impulse response of perform the whole task. Those coefficients will be adjusted
an FIR filter acting as a MF to some required signal s(n) of according to the deviation in the received signal from the
length N, can be given in terms of the signal as follows wanted echo signal that is supposed similar to the transmitted
h(n)  s( N  n) (13) signal. This proposal provides a solution that applies linear
system, rather than searching for a nonlinear processor.
This filter will produce in its output a signal y(t) given by Matlab with Simulink provides very constructive signal
the convolution of its impulse response h(n) with the input processing toolbox. In order to design and simulate the
signal x(n). If the input signal is x(n) = s(n) , the output signal intended adaptive filter, we can make use of the LMS Filter
would be its Autocorrelation Function (ACF), which realizes block that comes within the signal processing toolbox of
the crest SNR that indicates target detection. Simulink. Later, a Matlab code can be organized to carry the
Hence, to obtain a higher SNR in the output signal, the whole computations explicitly. Such code is easily converted to
transmitted radar signal s(n) should have an ACF of higher C code, which is more appropriate for digital signal processors.
peak-to-sidelobe ratio. That requires designing special signal Fig. 5 at the end shows the required adaptive filter that is
waveform that extends in the frequency domain and has based on the LMS algorithm, designed as part of a Simulink
considerable randomness. Pseudorandom sequences are used to model. The LMS Filter parameters that are set include filter
modulate the radar carrier instead of conventional rectangular length N=16, step size μ= 0.2 and initial value of filter weight
signal. Examples of famous codes used in radar waveform w(0)=0. The required echo signal is pulse-like, while the input
design are Barker codes and Pseudo Noise (PN) codes, which signal is modeled as the sum of that signal in addition to white
are known also as maximal length sequences [2]. noise and sinusoidal varying clutter. The amplitudes used for
The FIR filter of the impulse response stated in equation the signal, noise and clutter is chosen in a way that realizes
(13) can be implemented as a transversal filter, such as that reasonable SNR, so as to make it easy for us to distinguish
shown in Fig. 2 above. The filter coefficients are arranged in target echo signal inside noise and clutter on the scope.
the reverse order of the samples of the transmitted signal s(n). Fig. 4 presents the signals observed on the multichannel
However, these coefficients can be changed each time the radar scope used in the Simulink model. The most upper trace shows
operator decides to change the waveform of the transmitted the noisy cluttered signal, the second shows the final output
signal. signal of the adaptive filter, the third presents the desired echo
There are several reasons for that operational procedure, signal, while the lower trace shows output signals for different
such as adding waveform diversity to enhance the probability changing coefficients. Similarly, a numerical display shows the
of detection of targets in noisy and cluttered environment. final coefficients computed by the LMS algorithm. A code in
Another reason is of military nature for security purposes so as Matlab that include the LMS adaptive algorithm is designed to
to avoid being jammed by adversary electronic attacks, such as perform the task of the adaptive FIR filter and support the
in frequency hopping radar that changes its frequency results obtained by Simulink.
continuously. Instead of doing that in RF band, it is easily done
in baseband prior to Digital Up Conversion (DUC) with the aid
of DSP. Frequently, code diversity is applied without change in
the RF signal frequency, which leads to easy RF design.
VIII. DESIGN OF THE MULTIFUNCTION FILTER
Regarding the noise reduction LPF and the MTI filter,
which is a HPF, they are designed in a similar manner for
design of windowed FIR digital filter described earlier. The
resultant BPF is not essential to be of symmetry in the lowpass
and highpass parts of its response, since each part may have a
different prototype LPF. The foundation for that; interference
signals are characterized by different attributes from signals
vary due to movement of targets relative to radar.
Concerning the MF, its coefficients can be assumed to be
constant for regular radar. Otherwise, it need be updated by the
transmitting section of the radar each time the code is changed.
It worth saying that while the two filters above act to preserve
the signal waveform, the MF doesn’t operate on that. On the Fig. 4. Results Observed for the Simulink Filter Model.
 IX. CONCLUSION ACKNOWLEDGMENT
In this paper, an adaptive multifunction FIR filter has been I would like to thank the staff of the electrical engineering
proposed to perform different tasks, such as filtering of the department of Karary University, and colleagues in the Nile
received signal to reject out-of-band interference, moving Centre of Technology Research (NCTR) for their continuous
target indication and matched filtering. support and everlasting help. Thanks also are to postgraduate
Matlab program is used in the design of each of the students of the telecommunications MSc program in Karary
individual filters and the resultant multifunction filter, making University and researchers in the DSP group in NCTR to their
use of the built-in functions and the filter design and analysis direct and indirect contribution in the knowledge base used in
tools. Simulink was used to test an adaptive LMS filter that this paper.
performs as the intended multifunction filter. Simulation of the
designed filter is done for input signals of diverse waveforms REFERENCES
and different SNRs and SCRs. [1] M. M. Daffalla, Signal Analysis and Processing, Communication and
A code written in Matlab has been designed and simulated Navigation Research Center, Karary University, 2012.
for the same adaptive FIR filter, which followed the LMS [2] Fred E. Nathanson, Radar Design Principles, Signal Processing and the
adaptive algorithm step by step. Loops are used within the Environment, 2nd ed., McGraw-Hill, 1991.
program to generate the required filter coefficients. When it [3] Merrill I. Skolnik, Introduction to Radar Systems, McGraw-Hill, 3rd ed.,
was subjected to the same conditions and same input signals, 2001.
that Matlab code performed identical to the previous adaptive [4] Merrill I. Skolnik, Radar Handbook, McGraw-Hill, 2nd edition, 1991.
FIR filter and provided same observations that had been [5] M. M. Daffalla, Radar Principles, Karary Academy of Technology,
2006.
obtained by Simulink.
[6] V. K. Ingle, J. G. Proakis, Digital Signal Processing Using Matlab,
I recommend that this simulated processing scheme to be Brooks-Cole Publishing, 2000.
realized and applied in a real-time digital signal processor. [7] Simon Haykin, Adaptive Filter Theory, Prentice Hall, 4th edition, 2002.
Other adaptive algorithms are to be investigated to come up [8] S. Haykin, B. Widrow, Least-Mean-Square Adaptive Filters, John
with a better design of filter. The author of this paper has tried Wiley & Sons, 2003.
to establish a solid ground for other researchers, in order to add [9] S. Haykin, Adaptive Radar Signal Processing, Wiley & Sons, 2007.
on extra improvements for this significant technique on the [10] Vijay K.Madisetti, Douglas B.Williams, Digital Signal Processing
topic of radar signal processing in future. Handbook, CRC netBase, 1999.

Band-Limited
White Noise Output
Input

LMS Error

Sine Wave
Desired
Wts

Scope1
LMS Filter

0.7456
Desired Target
CONV 0.1521
Echo Signal
0.03149
Convolution
-0.02055
-0.0463
-0.01985
Display

Fig. 5. Simulink Model for Adaptive Filter.

View publication stats