Beamforming

Siham HADOUDOU Mustafa HIRI

School of Applied Sciences Tetouan School of Applied Sciences Tetouanh
siham.hadoudou@gmail.com mstcr766@gmail.com

Abstract

Blind sources separation is made for the estimation of mixed sources without knowing any information about
the original sources separate. However, beamforming has a similar role as the BSS but this time beamforming
serves to estimate the arrival angle DOA of the audio signals. The main objective of this paper is to provide a
comparative study of various classical multi-channel DOA algorithms and high-resolution, subspace-based,
multi-channel DOA algorithms for multiple signals arriving at different DOA.

I. Introduction

lind sources separation is a technique

B among the known techniques in the

field of telecommunication, this tech-
nique is used to make an approximation for
Figure 1: Smart antenna transceiver architecture

in this papier we present in section2

estimating the original source signals using the beamforming principale and the
only the information of the mixed signals Doa estimation than the ESPRIT and MUSIC
observed in each input channel. The BSS algorithms , in section 3 âĂe˛ and we summrizs
applies especially to the realization of high with a conclusion.
quality hands-free telecommunication systems
using speech recognition and noise reduction.
The BSS also uses an analysis called ICA
II. Beamforming
Principal Component Analysis [1], which
i. Definition and principle
relies mainly on higher order statistics and
information theory [2]. spatial position of the Beam formation is a term that comes from the
sources and applies a suitable spatial filter fact that the first spatial filters designed to
to the mixing channels so as to improve our form pencil beams aim at receiving a signal
target source and cancel the other sources. radiating from a specific location and attenu-
ating signals from other locations. "forming
Smart antennas have great potential to beams" to indicate energy radiation. However,
overcome the degradation of these systems beam formation is applicable to radiation or
by exploiting the space domain to reduce the energy reception.[4] But there are several types
effects of interference, extend the range and of beamforming such as scannig, convolutional
coverage of wireless networks, increase system beamforming and adaptive beamforming, in
capacity, and obtain achievable data. It is an our study we are interested in this last type.
interdisciplinary subject that requires knowl- Adaptive beamforming is a commonly used
edge and skills in areas such as antenna arrays, technique that allows the system to operate
signal processing, digital communications, RF in a disturbed environment by adaptively
engineering and wave propagation.[3] modifying the array diagram of the system, so

1
that null values are generated in the angular lo-
cations of the interference sources Figure 1 [5].

Adaptative beamforming have two ref-

erences : temperal reference and spatial
references,for the first reference we have
different adaptative algorithms, but in the
second reference we have DOA algorithms,that
we are interessting. In the next part we have
more details about DOA estimation. Figure 3: The system structure of DOA estimation

• The target step is a step consisting of sig-

nal source parameters and a complex envi-
ronment. To estimate them, we use certain
specific methods
• The observation stage is a stage that re-
ceives the radiation signals from the target
stage.
• The estimation step is a step that uses spa-
tial spectral estimation techniques to ex-
tract the character parameters of the spec-
trum. signal of the complex environment.
This step is equivalent to the reconstruc-
Figure 2: Adaptive Beamformer tion of the target stage.

- Basic Principle of DOA Estimation for :

ii. Direction of arrival (DOA) Estima- The angle between the normal of the
tion matrix and the direction vector of the plane
wave is the Direction of arrival (DOA). The es-
timated DOA target provides N instantaneous
In the theory of network signal processing, data: X (1) ... X (N), using an algorithm to
there is an important concept called the spatial estimate the value of multiple DOA (θ) signals.
spectrum, which presents the distribution of For signals generally distant from each other,
signals in all directions of space. If the spatial there is a difference in wave direction when
spectrum of the signal can be obtained, the di- the same signal reaches different elements of
rection of arrival (DOA) can also be obtained. the matrix. This channel difference leads to a
Therefore, the estimation of the spatial spec- phase difference between the elements of the
trum can also be called estimation of the DOA. arrival matrix. Using the phase difference be-
-Structure of spatial spectrum estimation tween the elements of the signal network, the
system: basic principle of the estimation of the DO.[6]
The space can be divided into three corre-
sponding spaces, namely target stage, observa- We choose in our study to working with
tion stage, and estimation stage [6]. michrophones array.

2
Figure 4: Uniform Linear Array with Far Field Source Figure 5: M element antenna array with D arriving
signals

The first microphone M1 is chosen as the • Calculation of eigenvectors a eigenvalues

reference microphone. The distance between of the autocorrelation matrix[7]:
any microphone pair is constant, say d. The R xx = A ∗ Rss ∗ A H + σ2 ∗ I
distance to be travelled by sound waves from were: A = [ a(θ1 ) a(θ2 )...a(θ D )] is a M*D
a source to a microphone Mi with respect to array steering matrix.
reference microphone M1 will be given by Eq.1 R xx = [s1 (k)s2 (k)....s D (k)] T
[7]: • R xx has D eigenvectors associated with sig-
nals and M âĂŞ D eigenvectors associated
τi = −(di /υ)sin(θ ) (1) with the noise.
• Construction of a family of vectors param-
eterized by the arrival angle with a theo-
retical state of relative phase imposed by
This equation indicates that the sound the geometry of the network. such that[7]:
wave incident on microphone Mi will be VN = [V1 V2 ...VM− D ]
timedelayed version of reference microphone. • Projection of this family of vectors in the
noise subspace and determining the val-
In DOA algorithms we choose in our papier ues of the direction of arrival θ1 θ2 , , θ D for
to working with two algorithms: MUSIC which there is a relative minimum [8].
algorithm and ESPRIT Algorithm. The e MUSIC Pseudospectrum is given as:

PMUSIC(θ ) = 1/abs(( a(θ )θ H VN VNH a(θ )

iii. MUSIC Algorithm (2)

MUSIC Algorithm (MUltiple SIgnal Classi-

fication) is one of the most popular method of iv. ESPRIT Algorithm
The detection methods who use the fact that
signal eigenvalues are larger than the noise ESPRIT (Estimation of signal parameters
eigenvalues. via the rotational invariance technique) is also
MUSIC makes assumption that the a robust and efficient method for estimating
noise in each channel is uncorrelated mak- DOA. It is similar to the MUSIC method. But
ing correlation matrix diagonal. The inci- it uses two tables in that the second element
dent signals are somewhat correlated creat- of each pair is moved the same distance in the
ing non diagonal signal correlation matrix[7]: same direction relative to the first element.[9]

3
 ∆0 is the element separation in terms of wave-
length. Other variations of ESPRIT include
beam-space ESPRIT, resolution-enhanced ES-
PRIT multiple invariance ESPRIT and higher
order ESPRIT.

III. Simulation and Discussion

At first, we implemented the code for several

arrival direction estimation techniques: MU-
Figure 6: Uniform Linear Array with Far Field Source
SIC, ESPRIT and also Capon. MATLAB is used
Let the array signals received by the two to simulate these algorithms.
arrays be denoted by x(t) and y(t) such that[10]:

x (t) = As(t) + n x (t) (3)

i. With three different algorithms:
y(t) = Aθs(t) + ny (t) (4)
Three signals were used with a Direction of Ar-
• A is a Kx M matrix; where M is the num- rival equal to 20 ◦ , 60 ◦ and 110 ◦ , respectively.
ber of steering vectors produced by N ele- And SNR = 5dB, Number of snapshots = 1000
ments of the array. and Number of antennas = 10.
• n x (t) denotes the noise induced at the ele-
ments ofthe y arrays.
• ny (t) the noise induced at the elements of Type of algorithm Mean squared errator
x array MUSIC 0.0023
ESPRIT 0.0067
The numbers of directional sources M are
CAPON 0.0132
estimated using the Akaike information cri-
terion (AIC) and the minimum description
length (MOL).[b] Two matrices Ux and Uy are
formed which designate the M eigenvectors
corresponding to the largest eigenvalues of
the two-matrix correlation matricesR xx and Ryy
(matrix correlation matrices). The eigenvectors
of the following matrix 2M by 2M are obtained
and are designated by:
 
H
Ux
 
V=
 H  Ux Uy (5)
U
y

One eigenvector V is obtained its eignval-

ues λm , m=1,...,M can be computed.Now the
Figure 7: DOA of three signals using MUSIC algo-
DOA is given by :
rithm

ARG (λm )
θm = cos−1 { }, m = 1, . . . , M (6)
2π∆0

4
Figure 8: DOA of three signals using Capon algorithm Figure 10: DOA estimation by varying Number of Di-
rection of Arrival (3)
As can be seen, the CAPON beamformer is
able to identify the signals but the locations are
not represented by sharp peaks (Mean squared
errator = 0.0132). MUSIC and ESPRIT also
identify the three signals, but MUSIC is able to
represent locations with larger peaks.

ii. With a single algorithm (MUSIC) :

In this case we used the algorithm that gives
the most performance results (Music) but by
varying the Direction of Arrival.

Nbr of Direction Arrival Mean squared err

2 0.0010
3 0.0023 Figure 11: DOA estimation by varying Number of Di-
4 0.0024 rection of Arrival (4)
5 0.0121

Figure 9: DOA estimation by varying Number of Direc- Figure 12: DOA estimation by varying Number of Di-
tion of Arrival (2) rection of Arrival (5)

5
It is found that the MUSIC algorithm is able Overcomplete Representations IEEE Sig-
to identify several signals with precision com- nal Processing Letters,April 1999.
pared to the other DOA algorithms, with five
sources found that the squared mean error is [2] J.-F. Cardoso, Blind source separation :
still low (0.0121). statistical principles in Proceedings of the
IEEE, vol. 9, no. pp. 2009âĂŞ2025,10 oct.
1998.
iii. With a single algorithm (MUSIC)
but in this case with different values of [3] Angeliki Alexion,Martin Haardt Smart
Number of snapshots : Antenna Technologies for Future Wireless
Systems: Trends and Challenges IEEE
We change the Number of snapshots and communication Magazine October 2004.
we leave the other parameters unchanged.
And also, we used 2 signals with a Direc- [4] V Krishnaveni, PhD. T Kesavamurthy,
tion of Arrival equal respectively: 30 ◦ and 80 ◦ . PhD. Aparna.B Beamforming for
Direction-of-Arrival (DOA) Estimation-A
Nbr of snapshots Mean squared errator Survey International Journal of Computer
50 0.0753 Applications (0975 âĂŞ 8887) Volume 61âĂŞ
600 0.0084 No.11 January 2013.
1000 0.0023
4000 0.0019 [5] Honghao Tang, Sven Nordebo, Pieternella
10000 9.3656e-04 Cijvat DOA estimation based on MUSIC
algorithm Bachelor degree Institutionen fÃűr
This simulation shows the relationship between
Fysik och Elektroteknik 2014-05-16.
the estimate of the direction of arrival and the
number of snapshots.With the increase in num- [6] N. A. Dheringe , Prof. B. N. Bansode Per-
ber of snapshots, DOA estimation accuracy formance Evaluation and Analysis of Di-
increases, but at the same time computational rection of Arrival Estimation Using MU-
load increases. So it has to be selected wisely. SIC, TLS ESPRIT and Pro ESPRIT Algo-
rithms Maharashtra, India , June 2015.
IV. Conclusion:
[7] T.B. LAVATE,V.K. KOKATE, Dr. A. M.
The DOA estimate has a wide range of appli- SAPKAL Performance Analysis of MUSIC
cations in the fields of communications, radar, and ESPRIT DOA Estimation algorithms
biomedicine, etc. This article proposes various for adaptive array smart antenna in mo-
algorithms of the estimation of the DOA. The bile communication. Second International
additional MUSIC algorithm is simulated in de- Conference on Computer and Network Tech-
tail by modifying its various parameters using nology ,2010.
MATLAB. In various DOA estimations based
[8] AOUNALLAH Naceur and BOUZIANI
on a spatial spectrum estimate, the MUSIC
Merahi The use of high resolution algo-
method has a higher resolution, a moderate
rithms for angle of arrival estimation Re-
amount of computation and a better robust-
searchGate ,Algeria ,April 2012.
ness of the matrix structure.
[9] Karan Sharma, Sandeep Santosh Review
References on ESPRIT-Estimation of Signal Parame-
ters via Rotational Invariance Techniques
[1] Te-Won Lee ; M.S. Lewicki ; M. Girolami International Journal of Engineering Research
; T.J. Sejnowski Blind Source Separation (ISSN : 2319-6890) Volume No.2, Issue No3,
Of More Sources Than Mixtures Using pp : 243-252 .01 July 2013.

6
[10] Mohammad Ismat Kadirl , Md. Shadiul
Hoque., Saiful Islam, Direction of Arrival
Algorithms for Adaptive Beamforming in
Next Generation Wireless Systems Proceed-
ings of 11 th International Conference on Com-
puter and Information Technology ,Khulna,
BangladeshDecember, 2008.