MODELLING AND PERFORMANCE ANALYSIS OF DOA

ESTIMATION IN ADAPTIVE SIGNAL PROCESSING ARRA'YS

Fahad A. Belhoul, Raed M Shubair and Mohammed E. Al-Mzialla

Etisalat College of Engineering,

Emirates Telecommunications Corporation (ETISALAT)
P.O.Box: 980, Sharjah, United Arab Emirates
Tel: +971-6-5611333, Fax: +971-6-5611789
e-mail: {rshubair, almualla}@ece.ac.ae

ABSTRACT In addition to its usefulness in the design and analysis of

adaptive signal processing arrays, the developed tool can also
This paper presents a simulation tool for Direction-of-Arrival be used for computer-aided learning of modem communication
(DOA) estimation in adaptive signal processing arrays. The tool systems utilizing smart antennas.
implements the Multiple Signal Classification (MUSIC) DOA
algorithm. A user-friendly Graphical User Interface (CUI) is 2. DOA ESTIMATIOlN USING MUSIC
also developed. In addition to its usefulness in the design and
analysis of adaptive signal processing arrays, the tool can also One of the most popular and simplest algorithms used for DOA
be used for computer-aided learning of modern communication estimation is the MUSIC algorithm. The standard MUSIC
systems utilizing smart antennas. To demonstrate the versatility algorithm, also known as spectral MUSIC, was introduced by
and accuracy of the developed tool, it is used to carry out a Schmidt [l]. In its simplest form, MUSIC involves estimating
detailed performance analysis of DOA estimation using the noise subspace from the correlation matrix on which the M
MUSIC. This includes investigating the effect of the number of array steering vectors are projected. These steering vectors are
mobile users and their spatial distribution, the number of array also known as direction vectors and they represent the response
elements and their spacing, and the number of signal snapshots. of an ideal array to the signal sources. The signal sources can be
derived from the direction vectors which are as orthogonal to
1. INTRODUCTION the noise subspace.
The algorithm starts by constructing <I real-life signal
Adaptive signal processing arrays, known also as smart model. Consider a number of plane waves froim Mnarrow-band
antennas, have been widely adopted in third-generation (3G) sources impinging from different singles 4,i == 1,2, ..., M, into
mobile systems because of their ability to locate mobile users a uniform linear array (ULA) which consists of N equi-spaced
with the use of Direction-of-Arrival (DOA) estimation elements or sensors. At a particular instant of time t, t=1,2, .. .,
techniques. Adaptive antenna arrays also improve the K, where K is the total number of snapshots taken, the array
performance of cellular systems by providing robustness against output will consist of the signal plus noise components. The
fading channels and reduced collateral interference. signal vector x(t) can be defined as:
The goal of DOA estimation is to use the data received at M
the array to estimate the direction of arrival of the signal. The x(t)= Ca(e,).!r,(t), (1)
results of DOA estimation are then used by the array to design m=l
the adaptive beamformer which is used to maximize the power where s(t) is an Mxl vector of s,ource waveforms, and for a
radiated towards users, and to introduce nulls to combat particular source at direction B from the array boresight; a( B) is
interference. As a result, a successful design of an adaptive an Nx 1 vector referred to as the array responsi: to that source or
array depends highly on the performance of the DOA array steering vector for that direction. It is given by:
algorithm.
&)=[I e-14 ...... e
-,(N-lk 11', (2)
DOA algorithms are usually complex and their
performance depends on many parameters. Thus, there is a need where T is the transposition operator, and 4 represents the
for tools that simulate DOA algorithms and ease the task of electrical phase shift from element to element along the array.
analyzing their performance. This can be defined by:
In this paper, we develop a tool that simulates the Multiple
SIgnal Classification (MUSIC) DOA algorithm. A graphical 4 = (2n/il)dccbse, (3)
user interface (CUI) is also presented to ease the use of the tool. where d is the element spacing arid 1 is the wavelength of the
To demonstrate the versatility and accuracy of the received signal.
developed tool, it is used to carry out a detailed performance The signal vector x(t) of size Nxl can be written as:
analysis of DOA estimation. This includes investigating the
effect of various parameters such as the number of mobile users
~ ( t=) A .~ ( t, ) (4)
and their spatial distribution, the number of array elements and where A=[a(O,) ... a(&)] is an NxMmatrix oFsteering vectors.
their spacing, and the number of signal snapshots. The array output consists of the signal plus noise
components, and it can be defined as:

0-7803-8163-7/03/$17.00 0 2003 IEEE ICECS-2003

340
 u(t) = x ( t ) + w(t) , (5)
4.2. Spatial Distribution of Mobile Users
where x ( f ) and w(f)are assumed to be uncorrelated and W ( Z is
)
modelled as temporally white and zero-mean complex Gaussian When the mobile users are close to each other, there will be a
process. Equation 5 can be written in matrix form of size NxK strong correlation between the signals. In this case the MUSIC
as: algorithm will find it difficult to resolve the users. This is
illustrated in Figure 3(a) for 3 adjacent users. However, the
U = A.S+W, (6)
DOA resolution improves significantly as the users move away
where S=[s(l) ... s(K)] is an MxK matrix of source waveforms from each other, as shown in Figure 3@).
and W = [ w ( l ) ... w(K)] is an NxK matrix of sensor noise.
The spatial correlation matrix R of the observed signal 4.3. Number of Array Elements
vector u(t) can be defined as:
Figure 4(a) shows the MUSIC spectrum generated with 4 array
R = E[u(t).~ ( t ), ~ ] (7) elements, whereas Figure 4@) is the spectrum generated with
where E[ ] and H are the expectation and conjugate transpose 16 array elements. It is evident that using more elements
operators, respectively. Substituting (5) into (7), the spatial improves the resolution of the MUSIC algorithm in detecting
correlation matrix R can now be expressed as: the incoming signals. This is achieved, however, at the expense
of computational efficiency and hardware complexity of the
) ~+ E[w(t).~ ( t ) ~ ] .(8)
. ( t.AH]
R = P[A * s ( t ) ~
sensor array.
For this signal model, the correlation matrix R will have M
signal eigenvalues, and N-M noise eigenvalues. Let E, be the 4.4. Element Spacing
matrix constructed of the corresponding M signal eigenvectors
E,=[el e2 ... e ~ ]and
, E,, be the matrix containing the remaining Figure 5(a) shows the MUSIC spectrum for an element spacing
N-M noise eigenvectors En=[eml ew2 ... eN].The peaks in the of +0.125h, whereas Figure 5(b) shows the spectrum obtained
MUSIC angular spectrum occur whenever the steering vector when the elements are farther away from each other with a
E(+,) is orthogonal to the noise subspace. The angular spectrum spacing of +OSh between adjacent elements. When the
is, thus, given by: sensors are placed too close to each other, mutual coupling
occurs and this may lead to inaccuracies in the estimated DOAs.
Mutual coupling effects for closely spaced elements must,
therefore, be taken into account when designing the sensor
The angles of arrival are estimated by detecting the peaks array. Increasing the spacing between the elements of the sensor
in this angular spectrum. array will result in better resolution of the estimated peaks. This
is achieved, however, at the expense of a larger size array.
3. GRAPHICAL USER INTERFACE
4.5. Number of Signal Snapshots
The MUSIC algorithm has been implemented using MATLAB
version 6.5. A GUI has also been built to ease the simulation. A The number of signal snapshots (samples) used to generate a
layout of the GUI is depicted in Figure 1 . The user can input the realistic signal model is a key factor in the realization of
signal parameters including the number of snapshots K , the practical antennas. Figure 6(a) shows the MUSIC spectrum
number of mobile users M, and their angle(s) of arrival 4. This obtained when only 10 snapshots were taken, whereas Figure
information is used to generate a realistic signal model. As for 6(b) is obtained when 1000 snapshots were used. It is evident
the sensor array, the user may input the number of array that increasing the number of snapshots leads to sharper
elements N and their spacing d. Default values for these spectrum peaks, indicating better resolution and more accurate
parameters can be retrieved at any time by pressing the detection. This is achieved, however, at the expense of a greater
“Default” button. The simulation can be started by clicking on processing time.
the “Run” button. This will produce a plot .of the angular
spectrum. This plot can be saved to a file by pressing the 5. CONCLUSIONS
“Save” button. Alternatively, the plot may be sent to a printer
by pressing the “Print” button. In this paper, a versatile simulation tool that implements the
MUSIC algorithm was developed together with a user-friendly
4. PERFORMANCE EVALUATION GUI. A number of numerical experiments were conducted to
investigate the effect of various parameters on the performance
To demonstrate the versatility and accuracy of the developed of the MUSIC algorithm, and its ability to resolve incoming
tool, it is used in this section to study the effect of various signals accurately and efficiently. This analysis tool can be
parameters on DOA estimation using MUSIC. used, for example, to improve and accelerate the design of
wireless networks. The developed tool can also be used for
4.1. Number of Mobile Users computer-aided learning of modern communication systems
utilizing smart antennas.
As can be seen from Figure 2(a), the MUSIC algorithm gives a
very sharp resolved peak when only one mobile user is present 6. EFERENCES
in the vicinity of the base station. However, the performance of
the algorithm degrades when there are many mobile users and [l] R. Schmidt, “Multiple emitter location and signal
correlation between them exists, as can be seen from Figure parameter estimation”, Proceedings of RADC Spectrum
2(b) for 7 users. Estimation Workshop, pp. 243-258, 1979.

341
[2] J. Liberti and T. Rappaport, Smart Antennas for Wireless [4] K. AI-Midfa, Investigation of Direction-of-Arrival
Communications. Prentice Hall, New Jersey, USA, 1999. Algorithms. Draft Ph.D. Thesis, to be presented to the
[3] S.C. Swales, Spectrum EfJicient Cellular Base-Station Of 2003.
Antenna Architectures. Ph.D. Thesis, University
of Bristol, November 1990.

Figure 1. Layout of graphical user interface.

(4 (b)
Figure 2. Effect of number of mobile users Mon MUSIC spectrum: (a) &I user, and (b) &7 users.

A,,?" "% ,,*I ,deld'cdr. r ,s 0 &,ti+ 9 dil'E13

(4 @)
Figure 3. Effect of spatial distribution of mobile users on MUSIC spectrum: (a) adjacent users, and (b) separated users.

342
 --
....................

(a) @)
Figure 4. Effect of number of array elements Non MUSIC spectrum: (a) N=4 elements, and (b) N=16 elements.

@)
Figure 5. Effect of element spacing don MUSIC spectrum: (a) C+ 0.125h, and (b) hO.5h.

. . 1 *tn.rr.

(a) @)
Figure 6. Effect of number of snapshots K on MUSIC spectrum: (a) K=10, and (b) K=IOOO.

343