Advanced Communications & Signal Processing Laboratory AM1, October 1994

and three general problems of great interest will be

investigated:
EXPERIMENT: AM1 ì the detection problem;
ì the directions of arrival (DOA) estimation
Experimental Handout:
problem.
Array Communications ì the signal parameter estimation problem.

& Processing With respect to detection problems the array

estimates the number of emitting sources (co-
channel signals) present in the array environment.
Supervisor:
On the other hand, Spatial Spectral Estimation
Dr. A. Manikas (Room:811A, Ext.: 46266)
techniques are concerned with the DOA problem.
http://skynet.ee.ic.ac.uk/manikas.html
Classical spatial estimation techniques are based
on the Fourier Transform (Conventional
Equipment:
Beamformer). The main drawback of the Fourier-
Any Computer with Audio Card; MATLAB;
type methods is that they offer limited resolving
In-house Array Processing Toolbox for MATLAB.
capabilities. Thus, in the last decade the so called
Superresolution methods have been introduced,
Aims & Outline: their main objective being to improve the resolving
By distributing a number of receiving elements
capabilities by using a model for the signals better
(from now on called sensors) in a 3-dimensional
than that used by Fourier methods. These methods
Cartesian space, an array is formed; the region
have given fresh impetus to the array processing
over which the sensors are distributed is called the
problem by dealing with the question of resolution
aperture of the array. The general array processing
of the arrays in such a way that there is elimination
problem is the obtaining of information about a
of the effects of the Signal-to-Noise Ratio (SNR)
signal environment from the waveforms received
on resolution, in contrast to the conventional
at the array elements, where the signal
methods where the resolution is limited by noise.
environment consists of a number of emitting
sources plus noise. These emitting sources, in the
Thus, this experiment is concerned with a compact
case of radar-based systems, are often targets
examination of the above concepts and provides a
which either reflect transmitted signals (as in
tutorial mechanism regarding array processing
active radars) or emit their own signals, (as in
techniques with main objectives to:
passive radars). Other typical examples of
ì estimate the number of the emitting sources;
applications of arrays of sensors are:
ì provide complete information about the
ì radio telescopy where the sensors are
directions of the emitting sources;
antennas and the emitting sources are radio
ì investigate the use of an array of sensors for
sources;
detecting and rejecting in-band interfe-
ì sonar where the sensors are hydrophones
rences;
and the emitting sources are ships (ship
ì estimate relative powers, cross-correlations,
noise);
etc.
ì geophysics where the sensors are seismo-
meters and the sources are earthquakes etc.;
However, a new 'hot' application of arrays is in the Tasks:
area of digital communications where space-time Consider a linear array of 5 sensors uniformly
and not just time information associated with the distributed along the x-axis in the 3 dimensional
communication signals can be utilized to provide real space.
more sophisticated and powerful communication
systems. Thus, by integrating array theory with type:
digital communication theory, new array=[-2 0 0; -1 0 0; 0 0 0; 1 0 0; 2 0 0];
communication system architectures are being
proposed which have a considerable impact, for The array receives noisy data from three sources
instance, on the capacity and performance of a which transmit at the same frequency band. The
mobile communication network. These directions of the sources are (30°,0°), (35°,0°), and
improvements can be achieved in a number of (90°,0°), where the first number represents the
different ways, for example, by suppressing co- azimuth and the second the elevation angles of
channel interferences, combating fading effects, arrival.
locating/tracing the mobile users, estimating other
signal environment parameters, etc. type: directions œ [30, 0; 35, 0 ; 90, 0]
This experiment aims to provide a theoretical
framework for handling various problems in a Using a multi-cross-correlator the second order
number of different applications. Emphasis will be statistics of the received array-signal, that is the
given to the so called "superresolution" approaches covariance matrix ‘BB , can estimated.

Dr A. Manikas  Array Communications & Processing 1

Advanced Communications & Signal Processing Laboratory AM1, October 1994

By using the matrix ‘BB solutions to the following ì Comment on the form of the matrix ‘BB .
three basic problems will be addressed: Check if your comments are valid for any
array geometry.
ì detection problem
ì estimation problem 3. Practical Covariance Matrix:
ì interference suppression problem.
ì In this case the three signal are initially
The experiment will involve three covariance three audio signals and then three images
matrices. The first will be the theoretical which are transmitted at the same frequency
covariance matrix for three equi-power sources, band in the presence of noise. The received
the second will involve three audio signals while array-signal is the matrix —_au for the
the third three colour images. audio signals and the matrix —_im for the
images
1. Form the pattern of the above array.
type: load —audio;
type: load —image;
Z=pattern(array);
plot2d3d(Z,[0:180],0,'gain in dB','initial pattern'); Listen to the audio signal associated with
the 2nd antenna by using your audio
Check the gain provided by the array for the three facilities.
sources.
type: sound(abs(X(2,:)), 11025);
2. Theoretical Covariance Matrix Formation:
Then observe the image at the output of the
ì For each source there is a vector, called the second antenna by typing:
Source Position Vector (or manifold displayimage(X_im(2,:),image_size, 201,
vector), which is a function of the position 'The received signal at the 2nd antenna');
of the source and the geometry of the array.
This means if you know the location of the ì Form the covariance matrix of the received
sources and the array you may form the signal.
SPVs. Consider the matrix ’ with columns
the Source Position Vectors (manifold i.e. type
vectors) W3 3=1,2,3. This matrix can be Rxx_au=—_au*—_au'/length(—_au(1,:));
formed by typing: Rxx_im=—_im*—_im'/length(—_im(1,:));

’=spv(array, directions) ì N.B.: Remember that ‘xx_theoretical is

the theoretical covariance matrix while
Comment on the form of the vectors W3 ‘xx_au and ‘xx_im are two practical
covariance matrices for the situations
ì Consider that you know that the above described above.
mentioned three sources are uncorrelated
and of equal powers (say 1, normalised). 4. Forget that you know that there are 3 sources
Form the covariance matrix of the 3 sources present and in addition forget their directions and
‘77 Þ their powers
i.e. type: directions=[];
ì Assume that the noise present in the array ‘mm=[];
is: additive isotropic noise, uncorrelated ’=[];
with the transmitting sources and of power sigma2=[];
5 # =40dB below the power of the signals
5. Detection Problem:
i.e. type: sigma2=0.0001
5.1. Consider that the only information provided to
ì Form the covariance matrix of the signals at you is the matrix ‘BB,theoretical at the input of the
the input of the array ‘BB,theoretical . This is above array. Form the eigendecomposition of the
known as the theoretical covariance matrix matrix ‘BB,theoretical
and is given by i.e. type: eig(‘xx_theoretical)

‘BB,theoretical œ ’Þ ‘77 Þ’H  5 # ˆ& Is there any way to estimate

ì the power of noise, and
i.e. type ì the number of sources which are present
Rxx_theoretical=S*Rmm*S'+sigma2*eye(5,5)

Dr A. Manikas  Array Communications & Processing 2

Advanced Communications & Signal Processing Laboratory AM1, October 1994

simply by observing the eigenvalues of 11. Multipaths - Coherent Sources:

‘BB,theoretical ? ì Consider the cases where two sources i.e.
30°, 35° are fully correlated (or coherent).
5.2.Write your conclusions from 5.1 as a formal Repeat instructions 5.1, 5.2 and 9.
STATEMENT (or THEOREM). ì Compare your results from the uncorrelated
and correlated cases.
5.3. Repeat instructions 5.1 but this time use the ì Is it possible to overcome the problem
practical covariance matrices ‘xx_au and ‘xx_im resulting when the sources are coherent?
rather than the theoretical covariance. Comment The answer is YES by applying the so
about the validity of the conclusions of 5.2. called spatial Smoothing technique [2,5]
before using the MuSIC algorithm.
6. Estimation Problem - based on the 'array pattern': Can you apply this technique? Write a small
program in matlab.
ì Let us consider that the source located at
90° is the 'desired' source and this direction 12. Reception Problem
is known to you while the remaining two
sources (i.e. 30°, 35°), known as interfe- 12.1. a) Consider the signal —_au at the input of
rences, are completely unknown. the array, having the covariance matrix ‘xx_au
(see instruction 3). Design an optimum Wiener-
ì Let us estimate firstly the SPV W. which Hopf beamformer (i.e. a linear receiver) to
corresponds to the desired source isolate/receive the 90° signal by typing

i.e. type: Sd=spv(array,[90,0]) Sd=spv(array,[90,0])

wopt=inv(‘xx_au) *Sd
ì Next let us weight the array elements by the yt=wopt'*—_au;
following vector A9:> œ a ‘" BB,theoretical W.
b) Listen to the 90° signal at the output of this
i.e. type wopt=a inv(‘xx_theoretical) *Sd optimum linear receiver (beamformer) by typing

where a is a constant (gain factor). sound(yt, 11025);

This is the optimum Wiener-Hopf solution.
and describe it in your log-book.
ì Estimate once again the array pattern of the
array when it is weighted by A9:> 12.2. Repeat instruction 12.1.a but this time use
—_im and ‘xx_im.
i.e. type:
Z=pattern(array, wopt); Observe the 90° signal at the output of the array
plot2d3d(Z, [0:180], 0, 'gain in dB', (i.e signal y(t)) by typing
'W-H array pattern');
displayimage(yt, image_size, 202,
ì Can you distinguish the directions of the 'The received signal at o/p of W-H
two interferences from that array pattern? beamformer');

7. Repeat instructions 2, 4, 5.1 and 6 but with and describe it in your log-book.
noise level 10dB below the level of the sources.
12.3. Consider the signal —_im at the input of the
8. What conclusions can be drawn from 6 and 7 ? array. Using the directions estimated by MUSIC
algorithm (see instruction 10) design a
9. Estimation Problem - Superresolution Approach: superresolution beamformer. This beamformer
Study and implement the MuSIC algorithm [1] in should isolate/receive one of the signals and, at
order to estimate the directions of the three the same time, suppress completely
sources. (asymptitically) the effects of the other two co-
channel signals (i.e. to provide complete
N.B. please use the following format interference cancellation).
Z = music(array, Rxx_theoretical) Generilize this to a multi-beam beamformer for
plot2d3d(Z,[0:180],0,'dB', 'MuSIC spectrum'); receiving the three images/signals. Observe the
three images at the output of the beamformer using
10. Repeat instruction 9 by using ‘xx_au and the displayimage command.
‘xx_im rather than ‘xx_theoretical. Plot the array patterns.

Dr A. Manikas  Array Communications & Processing 3

Advanced Communications & Signal Processing Laboratory AM1, October 1994

N.B.: Study the material provided in Appendix-1 Appendices:

in order to give a reasonable answer to this
question. Appendix-1: "Support Notes - Experiment AM1",
Department of Electrical & Electronic
12.4. Comment on the results of 12.1, 12.2 and Engineering, Imperial College, 1994.
12.3.
Appendix-2: "Some Introductory Notes on
13. Practical Detection Criteria: Diversity Techniques", Department of Electrical &
Electronic Engineering, Imperial College, 1989.
Simulate 250 snapshots for the environment
decsribed in instructions 1 and 2. Then form the References:
practical covariance matrix ‘BB . 1. Schmidt R.O., "Multiple Emitter Location and
Signal Parameter Estimation", IEEE
Apply the detection criterion (Question 5) in order Transactions on Antennas and Propagation,
to detect the three sources. Identify any problem Vol. AP-34, No.3, pp 276-280, March 1986.
and write it down in your Log-book together with
your conclusions. Then search the EEED Library 2. Shan T., Wax M., Kailath T., "On Spatial
on Level 6 to find papers associated with the Smoothing for Direction-of-Arrival Estimation
following criteria: of Coherent Signals", IEEE Transactions on
Acoustics, Speech and Signal Processing, Vol.
ì AIC (Akaike Information Criterion) ASSP-33, No.4, pp 806-811, August 1985.
ì MDL (Minimum Description Length)
3. Monzinco A.A., Miller T.W., Introduction to
Apply these criteria to detect the three sources. Adaptive Arrays, John-Wiley, 1980.

What conclusions can be drawn from these 4. Hudson J.E, Adaptive Array Principles, IEE-
criteria? Peter Peregrinus, 1981.

14. Design an adaptive beamformer using the 5. Pillai S.U., Array Signal Processing, Springer-
following two adaptive algorithms: LMS and RLS Verlag, 1989.

Figure-1: Baseband signal received at the 2nd antenna of a linear array of 5 antennas operating in
the presence of 3 co-channels signals (images) of bearings 30°ß 35° and 90°.

N.B.: This handout, the two associated appendices and the MATLAB library of the experiment can
be found in Dr Manikas' web site (click software).

Dr A. Manikas  Array Communications & Processing 4