SPACES-2015, Dept of ECE, K L UNIVERSITY

DOA Estimation of a System Using MUSIC Method

M. Devendra Dr. K. Manjunathachari
Associate professor in ECE dept Professor & Head of ECE Dept
G. Pulla Reddy Engineering College GITAM Institute of Technology, GITAM University
Kurnool, AP, INDIA. Hyderabad, INDIA

Abstract—Very accurate estimation of the signal direction of estimates directly. One of the subspace method used for DOA
arrival (DOA) has received a tremendous interest in estimation is ESPRIT (Estimation of Signal Parameters via
communication and radar systems for commercial and as well as Rotational Invariance Techniques) [5]: The ESPRIT algorithm
for military applications. This paper proposes the requires “translational invariant” arrays, i.e., an array with
implementation of finding direction of arrival of the signal to an identical copy displaced in the space. The geometry and
array system using MUSIC (MUltiple SIgnal Classification) response of these arrays need not be known; only the
algorithm. The solution includes finding general Eigen values and measurements from these arrays and the displacement between
used Jacobi algorithm for the calculation of Eigen values and the identical arrays are required. The computational
Eigen vectors, which uses rotation mode to realize Eigen value
complexity of ESPRIT is less compared to search-based
decomposition in order to reduce computations and finally
methods. In Root-MUSIC, the array is required to be uniform
achieves real time array direction finding. Performance
estimation of music algorithm by varying number of antenna and linear with search procedure in music is replaced by root
elements in an array is performed and Simulation results are finding approach.
obtained using MATLAB MUSIC (Multiple Signal Classification) method has high
resolution but it is a non - linear spectral method. This method
Keywords—Direction finding;Music; Jacobi; Eigen values; is basically implemented in spatial domain as a method for
I. INTRODUCTION Direction of Arrival (DOA) estimation and parameter
estimation of superposed radio signals on the antenna array.
DOA estimation is one of the most important research
problem in various applications like radar, sonar, MUSIC method is of specific theoretic and practical
communications etc., [1]. Estimation algorithms of DOA interest to the researchers, reason being its high resolution
depend upon maximum likelihood and subspace decomposition properties and accurate performance. MUSIC method is a
approaches [2,3]. Subspace-based (or super-resolution) subspace method. The high resolution properties of the
approaches have attracted much attention, after the work in [4] MUSIC method are based on its special properties and ability
is computationally simple as compared to the ML approach. to decompose into two orthogonal subspaces (signal subspace
There are number of algorithms for subspace DOA estimations and noise subspace).
which are shown in the below figure. MUSIC method provides asymptotic unbiased estimation
of parameters in the spatial model of multiple signal
superpositions. It means that, in case of direction of arrival
estimation problem, when the number of spatial-time samples
of the signals in the asymptotic case if tends to be infinity,
implies the Standard deviation of estimation error of the DOA
estimation tends to lower Cramer-Rao's bound and mean value
of the estimation error tends to zero. MUSIC method can be
applied to the DOA estimation when the antenna array is of
non-uniform geometry, so that the spatial sampling of the
wavefront is non-uniform.
A good advantage of this method compared to the DFT
method is that the spatial (time) sampling of wavefront
Fig.1. Different methods of DOA estimation (signal) can be non-uniform. By using this method to the
systems for radiofrequency spectrum monitoring, a problem of
In signal subspace methods, only signal subspace signal detection and parameter estimation in multiple incident
information is retained while in noise subspace based signal scenario can be solved in a new qualitative approach.
methods, only noise subspace information is retained. These These problems are difficult to solve using classic spectrum
subspace methods are divided into search based and noise analysis, especially in the case when radio signals are fully or
based methods. In search-based methods, it is assumed that the partially overlapping both in time and frequency domains.
response of the array to a single source, the array manifold
a(θ), is either known analytically as a function of arrival angle, By using the MUSIC method in the analysis of spatial
or is obtained through the calibration of the array. Algebraic samples of wavefront, the following unknown parameters of
methods don’t require a search procedure and yield DOA superposed radio signals can be estimated: the number of

309
SPACES-2015, Dept of ECE, K L UNIVERSITY
superposed (active) radio signals in the given frequency band noise or colored noise with distribution not known.
and parameters of each superposed signal such as: spectral Covariance matrix R is given by
bandwidth, direction of arrival (both azimuth and elevation)
and polarization. The process of finding estimate of spectral }= ARsAH + Rn (3)
bandwidths of multiple incident signals in open references is Where is p x p signal covariance
called ‘band or spectrum segmentation’. matrix. is m x m noise covariance
Section II gives the information on MUSIC algorithm. matrix. If the noise is white then Rn= σ2I. But, in reality the
Section III deals with various implementations of music correlation matrix is approximated by uniform averaging by
algorithm. Section IV highlights advantages and disadvantages some number of snapshots as
of music algorithm. Section V gives simulation results related ∑ (4)
to performance estimation of music algorithms.
II. MUSIC ALGORITHM Music in [6] is a search base method of noise subspace in
which noise information is retained based on property that the
The subspace properties of spatial covariance matrix are steering vectors are orthogonal to any number of linear
used in estimation of DOA. The auto correlation or auto combination of noise subspace eigenvectors.
covariance of a signal plus noise model is estimated and this
model is used to form a matrix whose Eigen structure gives the Principle of search based method is with an assumption
signal to noise subspace. This technique is used in antenna that the array manifold is known, and the arrival angles are
array for DOA estimation which will operate on spatial estimated by locating the peaks of the function S(θ)=1/a(θ)H N
covariance matrix. a(θ) with N being a matrix formed using the noise space
eigenvectors.
(1)
The steering vectors corresponding to incoming signal
a H Steering vector corresponding to array Qn noise lies in the signal subspace and are orthogonal to noise
subspace vector matrix. Let consider an array of M antenna subspace. In this method the estimates of DOA’s is to search
elements receiving a set of plane waves emitted by P (P < M) through a set of all possible steering vectors and find the one
sources in the far field of the array. We assume a narrow-band that are orthogonal to noise subspace. If a(θ) is steering vector
propagation model, i.e., the signal envelopes do not change corresponding to one of the incoming signals then it has to
while the time their wave fronts start to travel from one sensor satisfy a(θ)H * N=0. Where N is the matrix formed using noise
to another. Suppose that these signals have a common subspace Eigenvector. In practice a(θ) will not be orthogonal
frequency of fo; then, the wavelength is given by λ= c/fo where to noise subspace due to errors in estimation of N. However,
c is the speed of propagation. The received M-vector r(t) at the power spectrum is given as
time t is
(5)
r(t)= A*s(t) + n(t) (2)
s(t)=[s1(t) s2(t). . . . . . sP(t)]T is a P vector source A=[a(θ1) N=∑ (6)
a(θ2) . . . . a(θP)] is a steering matrix of size M X P. In which ei = Eigenvector corresponding minimum value of R.
a(θi) is the ith steering vector response of the array to the ith
source arriving from θi direction. n(t)=[n1(t) n2(t) . . . . . III. IMPLEMENTATION OF MUSIC ALGORITHM
nm(t)]T is the additive noise process. Figure 2 shows the
problem DOA. A. Eigen vector decomposition using Jacobi
Usually the correlation matrix is complex valued.
Computation of EVD for a complex matrix is atough task.
Reducing the computational complexity via unitary transform
allows real-valued Eigen value decomposition of the
transformed real number correlation matrix. Since the EVD
method has a large portion of the whole computational load of
MUSIC based algorithms, the real-valued Eigen value
decomposition will provide the fast and compact computation.
If the steering vectors are arranged in conjugate centro-
symmetric as
π π
a(θ)=[ ,……, ] (7)
Fig 2: DOA problem of uniform linear array

the correlation matrix R becomes centro-hermitian. The real-

The source signals are statistically independent & partially valued correlation matrix is obtained via any unitary
correlated or completely correlated. The array may have the transform Q as
arbitrary shape and response. The noise process is independent
of the source, having zero mean and it may be either white (8)

310
SPACES-2015, Dept of ECE, K L UNIVERSITY
The unitary Q can be chosen as, eigenvectors returned to complex values by the unitary inverse
transform as
(9)
√ , ∑ | . | (14)
0 Where Ei is the ith Eigen vector related to the noise subspace,
0 √2 0 (10) K and L are the number of antenna elements and the number
√
0 of waves.
according to the even and odd number of arrays respectively, C. Music spectrum via min norm
where the vector 0=[0,0,….,0]T, and I and Q are the identity
matrix and column flipped identity matrix in the left right . According to Eigen values and Eigen vectors of EVD, we
direction respectively. The selection of the real part only have constructed the pseudo spectrum equation as follows
provides FB(Forward–Backward) averaging. The unitary (15)
MUSIC computational flow is involved in 4 steps largely like; H
Estimation of correlation matrix including unitary transform & In the above equation En is the noise subspace and a(θ) is the
spatial smoothing if required, EVD (Eigen Value direction vector of the array. A sorting algorithm is used to
Decomposition) of the correlation matrix, Computation of find the minimum norm values and the corresponding values
MUSIC spectrum and 1-dimensional peak search (local of θ in the array signal DOA.
maximum detection). In this section, the DSP concepts for the
dominant procedures in the unitary MUSIC algorithm are IV. ADVANTAGES AND DISADVATGES
explained. In the computation for music DOA estimator, the Music algorithm has a very good resolution and can
circuit design of Eigen vector decomposition is studied [7]. It be applied for a variety of array geometries. Music will not be
has used CORDIC based Jacobi method, and is suitable for able to identify DOA of correlated signal. It is very
hardware implementation with fast parallel processing. Cyclic computationally complex and expensive because of its
Jacobi processor computes real symmetric Eigen value involvement in searching peaks in music spectrum. But,
problems by applying a sequence of orthonormal rotations to however, the above problem can be made overcome using
the left and right sides of the correlation matrix as, ET*R*E=D, spatial smoothing by applying spatial smoothing to the music
Where R is a correlation matrix and D is diagonal matrix with algorithm. It will also work for multipath propagation
non diagonal element very much equal to zero. environment.
1 0 0 0 V. SIMULATION RESULTS
0 0
(11) The below figure shows the simulation result of DOA
0 0
estimation using Music search based algorithm for a uniform
0 0 0 1
linear array. Number of antennas in an array are 8 which are
Eij is orthogonal matrix and θ is angle of rotation. Here we separated by a distance d=2.15 with a narrow band signal
have chosen the Jacobi algorithm for orthogonal impinging array with an angles 200 and 500. Received signal
transformation, which has a large degree of parallelism, and is has 20000 bits.
very easy to implement in FPGA. As high dimension array
system can be achieved by a combination of 2-dimension sub-
matrix, we have focussed on the implementation of 2-
dimension sub-matrix; equation (12) shows the singular value
decomposition.
1 0
(12)
0 2
In equation (12) 1 and 2 are the corresponding Eigen
values of the decomposition. Whereas is the rotation angles
and then calculations are shown as follows
2 tan (13)
jacobi algorithm is iterative process which is performed until
correlation matrix reduces to diagonal matrix
B. MUSIC spectrum via DFT
The first step is to transform input matrix to unitary
matrix by unitary transform Q in equations (9) and (10). Fig.3. simulation result of music spectrum
Transformed matrix is divided by M for spatial smoothing.
The next step after spatial smoothing is EVD using Jacobi The figure below shows the simulation result of DOA
algorithm. After the EVD step, the reciprocal music spectrum estimation using Music via min norm with complex Eigen
in equation (5) is found via spatial DFT of the noise values.

311
SPACES-2015, Dept of ECE, K L UNIVERSITY

Fig.4. simulation result of reciprocal music spectrum

Fig.6. simulation result of reciprocal music spectrum with real Eigen values
The figure below shows the MATLAB simulation result of
For different number of antennas, the MATLAB simulation
DOA estimation with hardware compatible Jacobi algorithm
results are shown below
in which complex Eigen values are replaced with real Eigen
values

Fig.5. simulation result of music spectrum with real Eigen values

The figure below shows the MATLAB simulation result of Fig.7. simulation result by varying no of antennas N= [8 10 12 14]
DOA estimation using music via min norm with hardware
compatible Jacobi algorithm in which complex Eigen values As the number of antenna elements in an array increases the
are replaced with real Eigen values. accuracy of DOA estimation also increases. The figure below
shows the performance of music algorithm for different
number of antennas

312
SPACES-2015, Dept of ECE, K L UNIVERSITY

Fig.8. performance of music for different no of antennas

VI. CONCLUSION
According to the basic principles of direction finding,
this paper discussed the MUSIC algorithm, and introduced a
direction finding system, which has good resolution and can
be used to any array geometries. In the references related to
high-resolution spectral methods it is usually stated that
MUSIC method does not have wide practical application in
systems for direction of arrival estimation, because of its
numerical complexity which has been overcome in this paper.
This can overcome the problem of correlated signal as we
have used spatial smoothing. Performance estimation of music
algorithm is also carried in this paper i.e., as number of
antenna elements in an array increases, the performance of the
system also increases.
REFERENCES
[1] P. Chevalier, A. Ferreol, and L. Albera,“High-Resolution Direction
Finding From Higher Order Statistics: The 2q-MUSIC Algorithm,”
IEEE Trans on Signal Processing, vol. 54, pp. 2 986-2 997, August
2006.
[2] Liu Xuebin, Wei Gang, and Ji Fei, “A Design of Dual Uniform Linear
Array Based on the Method of DOA Estimation with FFT,” Journal of
South China University of Technology: Natural Science Edition, vol. 32,
pp. 5-10, April 2004.
[3] Yang Xueya, Chen Baixiao, and Zhu Gen-sheng,“Novel method for
DOA estimation of coherent signals by eigen-subspace iteration,”
Journal of Xidian University: Natural Science Edition, vol. 37, pp. 242-
247, April 2010.
[4] Schmidt, R.O., Asignal subspace approach tomultiple emitter location
and spectral estimation,Ph.D. dissertation, Stanford University, Stanford,
CA, Nov. 1981.
[5] Paulraj, A., Roy, R. and Kailath, T., Estimation of signal parameters via
rotational invariance techniques-ESPRIT, Proc. 19th Asilomar Conf. on
Signals, Systems, and Computers, Asilomar, CA, Nov. 1985
[6] Kaveh, M. and Barabell, A.J., The statistical performance of the MUSIC
and the Minimum-Norm algorithms in resolving plane waves in noise,
IEEE Trans. on Acoustics, Speech and Signal Processing, 34, 331–341,
Apr. 1986.
[7] M. Kim, K. Ichige, H. Arai, ”Design of JACOBI EVD Processor Based
on CORDIC for DOA Estimation with MUSIC Algorithm”, The 13th
IEEE Symposium on Personal, Indoor and Mobile Radio
Communications (PIMRC2002), Lisbon, Portugal, Sept. 2002

313