2008 IEEE INTERNATIONAL RF AND MICROWAVE CONFERENCE PROCEEDINGS 2008 IEEE INTERNATIONAL RF AND MICROWAVE CONFERENCE PROCEEDINGS 2008

2008 IEEE INTERNATIONAL RF AND MICROWAVE CONFERENCE PROCEEDINGS 2008 IEEE INTERNATIONAL RF AND MICROWAVE CONFERENCE PROCEEDINGS December 2 December 2 December 2 December 2- -- -4, 2008, Kuala Lumpur, MALAYSIA 4, 2008, Kuala Lumpur, MALAYSIA 4, 2008, Kuala Lumpur, MALAYSIA 4, 2008, Kuala Lumpur, MALAYSIA

978-1-4244-2867-0/08/$25.00 2008 IEEE
R F
M 08
Analysis of Performance for Multiple Signal Classification (MUSIC) in
Estimating Direction of Arrival

1
Z. I. Khan, M. MD. Kamal, N. Hamzah, K. Othman and
2
N. I. Khan
1
Faculty of Electrical Engineering, University Technology Mara Malaysia
2
Faculty of Accountancy, University Technology Mara Malaysia
Shah Alam 40450 Selangor


Abstract- Array processing involves manipulation of
signals induced on various antenna elements. Its
capabilities of steering nulls to reduce co channel
interferences and pointing independent beams toward
various mobiles, as well as its ability to provide estimates
of directions of radiating sources, make it attractive to a
mobile communication system designer. In this paper
MUSIC algorithm was implemented in Matlab to
optimize the number of samples and number of elements
with the presence of noise. The optimization was
conducted in different configurations to improve the
performance of the algorithm in estimating the direction
of arrival (DOA) of the incident signals. The data
captured was analysed in order to obtain the real
performance of the algorithm by using statistical analysis.

I. Introduction

There is a one-one relationship between the direction of
a signal and the associated receiving steering vector. It
should therefore be possible to invert the relationship
and estimate the direction of a signal from the received
signals. An antenna array therefore should be able to
provide for direction of arrival (DOA) estimation.
There exist a Fourier relationship between the beam
pattern and excitation at the array which allows the
DOA estimation problem to be treated as equivalent to
spectral estimation [1].

In practice, the estimation is made difficult by the fact
that is usually an unknown numbers of signals
impinging on the array simultaneously, each from
unknown directions and with unknown amplitudes.
Also, the received signals are always corrupted by
noise. Nevertheless, there are several methods to
estimates the number of signals and their directions.

The Multiple Signal Classification (MUSIC) method
has received the most attention and has been widely
studied [2]. MUSIC algorithm is applicable to arrays
with arbitrary geometry and the price paid for this
generality is that the array





response must be measured and stored for all possible
combinations of source parameters. The advantage of
this algorithm is that it exhibits high resolution.
However, it can be computationally intensive, since it
requires a search through the entire array manifold for
the steering vectors that are orthogonal to the noise
subspace [3]. The MUSIC has been implemented in
Matlab using 15 elements linear array. This paper
presents the DOA estimation in noisy environment by
varying number of samples.

II. Mathematical Model for Music Algorithm

Let an array of M sensors received d narrowband plane
waves from far-field emitters with the same known
center frequency
0
f . With the narrow band assumption
the ith signal complex envelop representation [4] can
be shown as:

d i e t u t s
t t f j
i i
i
,...., 2 , 1 ) ( ) (
)) ( 2 (
0
= =
+
(1)
Where ) (t u
i
and ) (t
i
are slowly varying functions
of time that define the amplitude and phase of ith
signal, respectively. Slowly varying means
) ( ) ( t u t u
i i
and ) ( ) ( t t
i i
for all
possible propagation delays between array sensors,
and as result of this the effect of a time delay on
received waveforms is simply a phase shift, i.e,

0
2
) ( ) (
f j
i i
e t s t s

(2)
Then ) (t x
k
, the complex signal output of the kth
sensor at time t, can be written as:
M k where
t n e t s a
t n t s a t x
d
i
k
f j
i i k
d
i
k i k i i k k
i k
,....., 2 , 1
) ( ) ( ) (
) ( )) ( ( ) ( ) (
1
( 2
1
) 0
=
+ =
+ =

(3)
where ) (
i k
is the propagation delay between a
reference point and the kth sensor for the ith wavefront
impinging on the array from direction ,
i
) (
i k
a is
524


the corresponding sensor element complex response
(gain and phase) at frequency
0
f , d is the number of
sources present, and ) (t n
k
is additive noise at kth
sensor element.

Using vector notation for the outputs of M sensors, the
data model becomes;

=
+ =
d
i
i i
t n t s a t x
1
) ( ) ( ) ( ) ( (4)

where
[ ]
T
M
T
M
T
f j
i M
f j
i i
t n t n t n
t x t x t x
e a e a a
i M i
)] ( ),....., ( [ ) (
) 5 ( )] ( ),...., ( [ ) (
) ( ,...., ) ( ) (
1
1
) ( 2 ) ( 2
1
0 1 0
=
=
=



The M x 1 vector ) (
i
a is known as array response or
array steering vector for direction
i
. With defining
matrix )] ( ),...., ( [
1 d
a a A = and d x 1 vector
T
d
t s t s t s )] ( ),......, ( [ ) (
1
= the relation (4) can be
written as

) ( ) ( ) ( t n t s A t x + = (6)

The DOA estimation problem is finding
d i
i
,..., 1 , = from a finite number (N) of data
samples or snapshots of ) (
j
t x taken at times
. ,..., 1 , N j t =
For solving the DOA estimation the following
assumptions are made:
1. The number of sources is known and is less
than the number of sensors, i.e, d<M.
2. the sources are uncorrelated zero mean
stationary processes with the d x d diagonal
covariance matrix;
} ,...., , { } {
2 2
2
2
1 d
H
s
diag s s E R = = (7)
Where
2
1
denotes the power (variance) of
ith source, E is mathematical operator and H
denotes conjugate transpose.

3. The additive white noise at each sensor is a
stationary zero mean complex white Gaussian
noise process. The noise processes of different
sensors are uncorrelated and
MxM n
H
n
I n n E R
2
} { = = (8)
where
2
n
is the noise power at each sensor
and
MxM
I is an identity matrix of order M.

4. The noise and signal waveforms are
uncorrelated.
5. The array response vector ) ( a is known for
all and the array is configured in such a
way that the matrix A in relation (6) has full
column rank, i.e, rank(A)=d.

From the above assumptions and (6) the M x M
covariance matrix of received data can be written as
I A AR x x E R
n
H
s
H
x
2
} { + = = (9)
In practice, only a sample covariance matrix is
available, i.e. an estimation of
x
R based on a finite
number (N) of data samples or snapshots

=
=
N
j
H
j j x
t x t x
N
R
1
) ( ) (
1
(10)
This mathematical model will be used in implementing
the MUSIC algorithm in the Matlab and hence is used
in optimizing the number of samples in the algorithm.

III. Implementation of Music Algorithm

The signal covariance matrix, R
s
, is clearly a N x N
matrix with rank M. It therefore has N-M eigenvectors
corresponding to the zero eigenvalues [5]. Let q
m
be
such an eigenvector. Therefore,

0 = =
m
H
m s
q SAS q R (11)
0 =
m
H H
m
q SAS q (12)
0 =
m
H
q S (13)
where this final equation is valid since the matrix A is
clearly positive definite. Equation (14) implies that all
N M eigenvectors (q
m
) of R
s
corresponding to the
zero eigenvalues are orthogonal to all M signal steering
vectors [5].
This is the basic for MUSIC, Call Q
n
the N x (N-M)
matrix of these eigenvectors. MUSIC plots the pseudo-
spectrum as shown in equation below
525


2
1
2
) (
1
) (
) ( ) (
1
) (
) (
1
) (

s Q
P
s Q Q s
P
q s
P
H
n
MUSIC
H
n n
H
MUSIC
M N
m
m
H
MUSIC
=
=
=

=
(14)
Since the eigenvectors making up Q
n
are orthogonal to
the signal steering vectors, the denominator becomes
zero when is a signal direction. Therefore, the
estimated signal directions are the M largest peaks in
the pseudo-spectrum. However, in any practical
situation, the signal covariance matrix R
s
would not be
available. The most we ca expect is to be able to
estmate R the signal covariance matrix. The key is that
the eigenvectors in Q
n
can be estimated from the
eigenvectors of R.
For any eigenvector q
m
Q,
m m s
q q R =
m m s m
Iq q R Rq
2
+ =
m m
q ) (
2
+ = (15)
Any eigenvector of R
s
is also an eigenvector of R with
corresponding eigenvalues +
2
. Let R
s
=QQ
H
Therefore,
H
Q I Q R ] [
2
+ =

+
+
+
=
2
2
2 2
2
2
2
1
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0

L L
M O M M M M M
L L
L L
M M M M O M M
L L
L L
M
Q
(16)

Based on this eigendecomposition, we can partition the
eigenvector matrix Q into a signal matrix Q
s
with M
columns, corresponding to the M signal eigenvalues,
and a matrix Q
n
, with (N M) columns, corresponding
the noise eigenvalues (
2
). Note that Q
n
, the N x (N
M) matrix of eigenvectors corresponding to the noise
eigenvalues (2), is exactly the same as the matrix of
eigenvectors of R
s
corresponding to the zero-
eigenvalue. This is the matrix used in MUSIC
spectrum. Q
s
defines the signal subspace, while Q
n

defines the noise subspace.

Using the final two observations, we see that all noise
eigenvectors are orthogonal to the signal steering
vectors. This is the basic for MUSIC. Consider the
following function of :

) 17 (
) ( ) (
1
) (
1
) (
1
2

s Q Q s
s q
P
H
n n
H N
M m
H
m
MUSIC
= =

+ =

where q
m
is one of the (N-M) noise eigenvectors. If is
equal to DOA one of the signals, s() q
m
and the
denominator is identically zero. MUSIC, therefore,
identifies as the directions of arrival, the peaks of the
function P
MUSIC
().

IV. Data Analysis

In practice, the number of sources an array may be able
to resolve depends not only on the number of elements
in the array but also on the array geometry, available
number of snapshots (samples) and spatial distribution
of sources [6]. The optimization for the number of
elements was done earlier in [7] with constant
parameters. The number of sample used was 100 with
displacement of /2 and two incident signals at 20 and
70 as the reference angles. The percentage of error for
MUSIC algorithm was estimated in there different
cases.

A. Performance of MUSIC using different number of
incident signals.

The simulation was to test the program whether the
direction of arrival (DOA) of the incident signals was
obtainable. The program was tested with different
numbers of incident signals as the angles of the sources
were varied. This study is using 100 as samples
because MUSIC is capable of estimating the DOA with
a minimum number of samples which is also stated in a
study by Bellofiore, et. al [3], However, as for the
elements array the study assumes 15 numbers of
elements to be used that acts as the antenna array with
displacement of 0.5 and assuming the signal is
Gaussian. The steering vector is used to define the
eigenvalues of the signals and hence tabulate the matrix
shown in equation 16.

526


Number of Incident
Signals
DOA MUSIC % of Error
1 20 20.00 0%
-15 -14.00 7%
2
20 20.00 0%
-15 -14.00 7%
10 10.00 0% 3
30 30.00 0%
-25 -26.00 4%
-15 -16.00 7%
20 20.00 0%
4
70 70.00 0%
-25 -24.00 4%
-15 -14.00 7%
20 20.00 0%
50 50.00 0%
5
70 70.00 0%


Table 1: Percentage of error for MUSIC using different number of
incident signals

The performance of the algorithm was then being
analysed using SPSS software. This is to test the
acceptable range and the normality of error in the
estimated DOA. By referring to Figure 1 below, the
plot shows the percentage of error during the
estimation is within the reasonable normal distribution
straight line. This indicates that MUSIC has a high
accuracy in performing the estimation of DOA.

Fig 1: Q-Q Plot for MUSIC using different number of incident
signals.

B. Performance of MUSIC using different number of
elements.

The second evaluation concentrates on the performance
of MUSIC using different number of elements. The
error percentage decreases as the number of elements
reaches 15 as illustrated in Table 3, which shows the
optimized number of elements used for MUSIC
algorithm.

MUSIC
Noise (dB)
20 %error 70 %error
1 20 0% 68 1%
3 20 0% 72 1%
5 20 0% 72 1%
8 20 0% 74 6%
12 20 0% 64 9%
16 22 10% 76 9%
20 12 40% 70 0%

Table 2: Percentage of error for MUSIC using 10 number of
elements.


Figure 2 and 3 shows the Q-Q plot for 10 and 15
numbers of elements. The error performance for 20 of
DOA estimation is within the linear distribution line.
However in comparison with the numbers of elements,
Fig. 3 shows a better linearity. This indicates the level
of error is within the acceptable range in evaluating the
performance of the algorithm.

Fig 2: Q-Q Plot for MUSIC for 20 with 10 elements with noise
variation


527




Fig 3: Q-Q Plot for MUSIC for 20 with 15 elements with noise
variation

Fig. 4 and 5 shows the Q-Q plot for the error estimation
for 70 of DOA. The distribution of errors is skewed to
the right which is far from the normal linear
distribution. It can be said that the level of error at 20
of DOA at 10 and 15 numbers of elements are more
acceptable rather than at 70 of DOA.


Fig 4: Q-Q Plot for MUSIC for 70 with 10 elements with noise
variation


MUSIC
Noise(dB)
20 %error 70 %error
1 20 0% 70 0%
3 20 0% 70 0%
5 20 0% 70 0%
8 20 0% 70 0%
12 20 0% 70 0%
16 20 0% 72 3%
20 20 0% 66 6%

Table 3: Percentage of error for MUSIC using 15 number of
elements.


Fig 5: Q-Q Plot for MUSIC for 70 with 15 elements with noise
variation


V. CONCLUSION

This paper shows the performance of MUSIC
algorithm based on the error evaluation using SPSS.
The result indicates, the algorithm has successfully
estimated the Direction of Arrival of the incidents
signals impinging on the antenna array. The evaluation
based on the optimum number of elements used shows
a significant percentage of error which is acceptable in
evaluating the algorithm performance and hence
indicates the accuracy of the algorithm


REFERENCES

[1] Raviraj Adve, Direction of Arrival Estimation, Master
Thesis, Toronto University, Canada, 2003.

[2] Ali Olfat, S. N. Esfahani, High resolution direction of arrival
estimation, Master Thesis, Department of Electrical
Engineering, Tehran University, Tehran, Iran 2002,

[3] S. Bellofiore, C. A. Balanis, J. Foutz and A. S. Spanias, Smart
Antenna System for Mobile Communication Networks Part 2:
Beamforming and Network Throughput, IEEE Antennas and
Propagation Magazine (Vol.44), No 4, pg. 106-114, August
2002.
528



[4] H. L. Trees, Detection, Estimation and Modulation Theory,
New York, Wiley, 1971.

[5] Ali S. Dakdouki, M. Tabulo, On the eigenvalues distribution of
smart antenna arrays in wireless communication systems,
IEEE Antennas and Propagation Magazine, Vol. 46, No. 4,
August 2004.

[6] L. C. Godara, Application of Antenna Arrays to Mobile
Communications, Part 2: Beam-Forming and Direction of
Arrival Considerations, Proceeding of the IEEE, Vol. 85, No.
8, pp. 1195-1234, August 1997.

[7] Z. I. Khan and Md. R. Islam, Direction of Arrival Estimation in
Multiple Classification Signal (MUSIC) in Matlab, accepted
for publication in International Conference on Computer and
Communications Engineering, IIU Malaysia, May 2006.


529