International Journal of Computer Applications (0975 – 8887)

International Conference on Communication, Computing and Information Technology (ICCCMIT-2014)

A Comparative Study of Different DOA Estimation

Schemes and Adaptive Beam Forming Techniques for
Target Detection and Tracking
Susaritha U.S Ganesh Madhan. M Ragupathi. R
Dept. of Electronics Dept. of Electronics Dept. of Electronics
Engineering, Madras Institute of Engineering, Madras Institute of Engineering, Madras Institute of
Technology Campus, Anna Technology Campus, Anna Technology Campus, Anna
University, Chennai, India University, Chennai, India University, Chennai, India

Ashita Priya Thomas

Dept. of Electronics
Engineering, Madras Institute of
Technology Campus, Anna
University, Chennai, India

ABSTRACT RSSI based location estimation in wireless sensor networks

This paper reports a comparative study of different Direction have been investigated by Raida Al Alawi [1]. The results of
of Arrival (DOA) estimation methods and application of this work reveal the feasibility of RSSI based localization
adaptive beam forming for target detection and tracking. DOA algorithm in designing real-time position monitoring system.
algorithms such as Beamscan, Minimum variance distortion Location estimation has been done using trilateration. The
less response (MVDR), Multiple Signal Classifier (MUSIC), Root Mean Square (RMS) node positioning error was
root-MUSIC, Estimation of signal parameters via rotational computed as the distance between the actual and the estimated
invariance (ESPRIT) have been investigated, with initial data position of the target node. Nwalozie et al [4] have compared
obtained from Received Signal Strength (RSS) measurement various adaptive beam forming algorithms such as Direct
in a Wi-Fi environment. Location estimation using root- matrix inversion, constant modulus algorithm, least square
MUSIC and ESPRIT algorithms have been carried out for a constant modulus algorithm, Least mean square (LMS)
moving target. Optimum combination for best performance is algorithm and Recursive LMS algorithms. Solution for Fast
identified. Simulations have been carried out in MATLAB Mobile Target Tracking based on Kalman Filter and MUSIC
tool. Algorithm is proposed in Ref [5], where, tracking error was
obtained to be less than one degree. Passive location of non
Keywords co-operative emitters can be achieved by triangulation,
Received signal strength indication (RSSI), Direction of direction of arrival measurements or by trilateration, time
arrival (DOA), Root Mean Square Error (RMSE), Signal difference of arrival (DTOA) measurements, which require
Strength Set Identifier (SSID), Signal estimation, Adaptive cross correlation processing to determine the emitter location.
beam forming. Both techniques when used separately result in ambiguities in
a multiple-emitter environment when the various source
1. INTRODUCTION emissions overlap in time/frequency and space. F.J Berle [6]
Location estimation and tracking of objects and vehicles have had described a concept of a mixed triangulation/trilateration
become important in recent days. Significant efforts have emitter location system in order to circumvent the
been made towards the development of efficient approaches ambiguities. He has found that the measurement accuracy is
for target localization and tracking. Location tracking can be strongly dependent on target geometry relative to the baseline
done with RSSI based techniques [1], but the accuracy of between the measuring sensor pair. Based on the approach of
localization highly depends on the path loss model. Ref.[6], this paper focuses on combining trilateration and
Fluctuations in environment conditions affect the localization DOA estimation techniques for detection and tracking.
results. Recently Wi-Fi technology has been widely deployed
for LAN applications and this can be used for position 2. DOA ESTIMATION AND
determination [2]. In our previous report [3], we have BEAMFORMING [7]
estimated the location based on RSSI in a Wi-Fi environment The major classification of DOA estimation schemes includes
and a maximum deviation around 7.6 % was observed. In conventional and adaptive methods. Maximum signal power
order to improve the localization accuracy and to track is detected when the DOA of the signal matches with the
effectively, DOA estimation along with adaptive beam steering direction of the array. Detailed descriptions of
forming is required. Adaptive beam forming ensures that even various algorithms are provided in Refs [7-10]. However, we
when RSSI methods fail to localize target, it tracks the target brief about the schemes which are used in this work, in the
successfully. The accuracy is dependent on correct DOA following paragraphs.
estimation of the required signal and cancelling the
interferences. Hence a combination of RSSI, adaptive beam 2.1 Beamscan Algorithm
forming and DOA estimation techniques can provide The beamscan algorithm develops conventional beam scans
successful localization and tracking. over the region of interest and computes the square of the

21
 International Journal of Computer Applications (0975 – 8887)
International Conference on Communication, Computing and Information Technology (ICCCMIT-2014)

signal magnitude [8]. The signal power reaches a maximum at

the impinging angle, as illustrated by the following Element spacing 0.5 m
expression.
Propagation speed 3e8 m/s
Pnorm ( )  a H ( ) R xx a( ) / a H ( ) a( )
(1) Wavelength 1m
Where, a(θ) is the steering vector array, Rxx is the input data
Gain at steering angle 10 dB
covariance matrix and H denotes the transpose of the matrix.
This scheme has to determine the maximum power for all 1-D array span 4.5 m
values of “θ” and hence becomes cumbersome. In the case of
linear array, FFT can be used for detecting the power. The
2.3.3 Estimation of Signal Parameters via
resolution of this approach is higher for small array aperture.
But this scheme is restricted to single source cases, as it fails Rotational Invariance technique (ESPRIT)
for multipath and multi source conditions. The property of rotational invariance of the signal space is
utilized to detect the DOA, in this case. The unique feature of
2.2 Minimum variance distortion less this scheme is that it requires less computation and direct
evaluation of DOA. But its use is restricted due to the need of
response beam former (MVDR) more number of sensors.
MVDR method [8] improves the performance by avoiding the
unwanted signals. This scheme minimizes the array output 2.4 Adaptive beamforming techniques
power and the spectrum is given by Adaptive Beam forming [9-10] uses a number of antennas or
sensors to evaluate the DOA in a noisy environment. It also
PMVDR ( )  1/ [a H ( ) R 1xx a( )] (2)
cancels the effect of inference signals. The implementation
This approach does not have array restrictions and leads to schemes include Minimum Variance Distortion less Response
better resolution. However, it needs increased power and does (MVDR) and Linearly Constrained Minimum Variance
not perform well under multiple source environments. (LCMV). MVDR is limited by the self nulling condition;
however this problem is overcome in LCMV approach.
2.3 Subspace based techniques
These techniques determine the signal space based on the 3. ANALYSIS AND RESULTS
received data and evaluate the DOA. For localization, three Wi-Fi access points or base stations
were considered. The device to be tracked accesses these
2.3.1 Multiple signal classification (MUSIC) stations using Wi-Fi, and transmits the information regarding
MUSIC provides a precise evaluation of DOA. It is also based the received signal strength from them, to the stations.
on the concept of orthogonality of signal vector to noise
Table 2. DOA Estimates
vector. The expression for music spectrum is given by

PMUSIC ( )  1/ {a H ( ) Vn Vn H a( )} (3) Impinging

Beam Root
source MVDR MUSIC ESPRIT
where Vn is the noise vector. -scan (°) (°) MUSIC (°)
Broadside
(°) (°)
(°)
The DOA is “(θ)” when PMUSIC(θ) reaches a maximum.
The peaks of the spectrum reveal the DOA. The resolution of
this approach depends on the sensor parameters. 40, 40, 40, 40, 40.0038, -19.9981,
-20 -20 -20 -20 -20.0004 39.9984

25, 25, 25, 25, 25.0005, 25.0009,

40 40 40 40 39.9987 40.0015

30, 35, 30, 30, 29.9977,

30,40
40 71 40 40 40.0041
Fig 1: Uniform linear array
35, 37, 35, 35, 34.9973, 40.0031,
2.3.2 Root-MUSIC algorithm 40 64 40 40 40.0004 34.9918
Root-MUSIC is an approach that provides straight forward
computation of DOA. This scheme is classified under Model-
37, 38, 38, 37, 36.9940, 36.9851,
Based Parameter Estimation (MBPE) technique. This scheme
utilizes the roots of the signal Eigen vector lying in the unit 40 66 14 40 40.0017 40.0008
circle, to determine the DOA of the sources. However this
technique is limited to linear array cases. Using the obtained RSSI values and the path loss model,
Table 1. ULA specifications target position was determined by application of trilateration
algorithm. However, the accuracy of the location estimation
depends on the loss model. This data is used as the initial
No. of elements 10
conditions for the tracking simulation. For analysis of DOA
algorithms, a Uniform Linear Array (ULA) has been used.
Signal frequency 300 MHz Table 1 gives the specifications of the ULA and Figure 1

22
 International Journal of Computer Applications (0975 – 8887)
International Conference on Communication, Computing and Information Technology (ICCCMIT-2014)

shows a descriptive picture of the array considered for the For signals separated with a space less than MVDR
study. beamwidth, MVDR does not resolve them and super
resolution techniques are adopted. Root MUSIC, MUSIC and
DOA estimates were obtained using the above mentioned ESPRIT give correct estimates when impinging source angles
algorithms and are tabulated as in Table 2. All simulations are at 37o and 40o whereas MVDR fails. Hence, it is clear that
have been carried out using MATLAB software. beamscan algorithm gives improper estimates for closely
spaced signals. Further, MVDR also fails if difference in
source signal directions is less than its beam width. Better
resolution is achieved by MUSIC, Root-MUSIC and ESPRIT
algorithms.
1

0.9

0.8

0.7

0.6

Amplitude (V)
0.5

0.4

0.3

0.2

0.1
Fig 2: Path traced by the target object
0
When the signals arrived from directions, which are separated 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
less than the beam width of beamscan beam former, improper Time (s)

estimates were obtained. For two signals impinging at 30° and Fig 3: Received pulse at antenna array
40°, the DOA estimates were obtained as 35° and 71°
respectively, using beamscan algorithm. For the same values The scenario considered for the tracking analysis is illustrated
of impinging angle, as taken for beamscan algorithm, i.e. 30° in Figure 2, where the base stations or Access Points used are
and 40°, DOA estimates were obtained correctly using denoted by AP1, AP2 and AP3. The location of AP2 is fixed
MVDR method, as its beamwidth is smaller than beamscan as reference (0, 0). The location of AP1 and AP3 are kept as
beamformer. (198.1, 0) and (106.56, 27.44) respectively. The dots joined
by a line in Figure 2 indicate the path traced by the object.
Table 3. DOA estimates using Root MUSIC and ESPRIT The initial position of the target in the path is given by the
coordinate (60, 2) and was found using trilateration.
Target DOA Estimates Considering the antenna array to be placed at AP 2, for the
Coordinates Elevation Azimuth Broadside (°) path tracked by an object in the Figure 2, continuous DOA
estimates were obtained using root-MUSIC and ESPRIT
angle angle angle ROOT
algorithms. These estimates are tabulated in Table 3.
x(m) y(m) ESPRIT
(°) (°) (°) MUSIC From the results tabulated in Table 3, it is found that the
maximum deviation from the original DOA estimates is 1.41
60 20 51.3247 1.9092 1.1929 1.1887 1.2221
% and 3.06% for root-MUSIC and ESPRIT algorithm
respectively. Thus root-MUSIC algorithm is found to provide
better DOA estimates than ESPRIT algorithm, for the system
90 60 34.7365 33.6901 27.1191 27.1617 27.1139
under study.
120 100 25.6474 39.8056 35.2475 35.2712 35.3434 1

0.9
140 140 20.7470 45.0000 41.3955 41.3514 41.4334
0.8

200 180 15.5750 41.9872 40.1207 40.0895 40.1796 0.7

Magnitude (V)

0.6
250 185 13.5582 36.5014 35.3287 35.3517 35.4247
0.5

300 150 12.6044 26.5651 25.8767 25.9198 25.8544 0.4

0.3
320 100 12.6098 17.3540 16.9227 16.9326 16.8949
0.2

300 45 13.8870 8.5308 8.2796 8.2999 8.2696 0.1

0 0.05 0.1 0.15 0.2 0.25 0.3

250 10 16.6866 2.2906 2.1941 2.2082 2.2247
Time (s)

Fig 4: MVDR output in the presence of interferences

23
 International Journal of Computer Applications (0975 – 8887)
International Conference on Communication, Computing and Information Technology (ICCCMIT-2014)

Adaptive beam forming is implemented for an antenna array Azimuth Cut (elevation angle = 0.0 )
for receiving a narrow band signal in the presence of noise
and interference. Figure 3 shows the received signal, which is 30 LCMV
assumed to impinge on the array at 45° azimuth and 0° MVDR
elevation. We first investigate the performance of MVDR in
20
this scenario. Two interference signals are simulated to arrive
from 30° and 50° in azimuth. MVDR algorithm effectively
cancels the interferences and recovers original signal. This 10
observation is shown in Figure 4.

Power (dB)
0

0.06 -10

0.05 -20
Magnitude (V)

-30
0.04

-40
0.03 0 10 20 30 40 50 60 70 80 90
Azimuth Angle (degrees)

0.02
Fig 7: LCMV and MVDR beamformer response pattern
From the analysis, it is inferred that, LCMV is a better choice
0.01
as it places nulls only at the interferences and does not
suppress the required signal at 45°, despite a signal direction
0 0.05 0.1 0.15 0.2 0.25 0.3 mismatch. Thus, LCMV method is a better choice for receiver
Time (s) array signal processing. The optimal DOA technique is root-
MUSIC algorithm as it is more accurate than ESPRIT
Fig 5: Output of MVDR with signal direction mismatch algorithm.

4. CONCLUSIONS
1 Different DOA and adaptive beamforming techniques for
0.9
target tracking in a Wi-Fi environment were analyzed in this
work. The merits and demerits of various DOA schemes are
0.8 identified. An RSS based position determination is first
carried out, which is followed by DOA and adaptive beam
0.7
forming techniques. Root-MUSIC algorithm provides better
Magnitude (V)

0.6 DOA estimates than ESPRIT algorithm, for the system

considered in the study. It is observed that LCMV performs
0.5 better as it avoids self nulling and recovers the required signal
even under the presence of interfering signals.
0.4

0.3 5. REFERENCES
[1] Raida Al Alawi. 2011. RSSI Based Location Estimation
0.2 in Wireless Sensors Networks. Proc. ICON '11, 17th
IEEE International conf. on Networks. Singapore. 118 –
0.1
122.
0 0.05 0.1 0.15 0.2 0.25 0.3 [2] Xiang, Z. Song, S. chen, J. Wang, H. Huang, J. and Gao,
Time (s) X. 2004. A Wireless LAN based indoor positioning
technology. IBM Journal of Research and Technology.
Fig 6: LCMV beamformer output with signal direction Vol.48, Iss5.6. (2004), 617-626.
mismatch
[3] Ganesh Madhan, M. Susaritha, U.S. Raghupathi, R. and
However, MVDR fails in case of self-nulling. To implement Ashita priya Thomas. 2014. RSSI based location
this case, the real signal is considered to arrive from a estimation in a WiFi environment: an experimental
direction of 45° azimuth. Due to sensor position error, the study. ICTACT Journal of Communication Technology,
impinging angle is estimated as 43°. Also it suppresses the Vol.5, No.4,(Dec 2014), 1015-1018.
required signal and provides a distorted version of the original
received signal. This response is shown in Figure 5. [4] G.C Nwalozie, V.N Okorogu, S.S Maduadichie, A.
Adenola 2013, A Simple Comparative Evaluation of
In the case of LCMV method, it provides better output for the Adaptive Beam forming Algorithms International
same conditions. Figure 6 shows the LCMV output, which Journal of Engineering and Innovative Technology
clearly indicates the recovery of the original signal even in the (IJEIT) Vol. 2, No. 7, January 2013,417-424.
case of signal direction mismatch. A comparison of beam
former responses is shown in Figure 7. [5] Guo Yan, Qian Zuping, Yao Zeqing and Li Ning. 2008.
Target Tracking Based on Kalman Filter and MUSIC

24
 International Journal of Computer Applications (0975 – 8887)
International Conference on Communication, Computing and Information Technology (ICCCMIT-2014)

Algorithm. Proc. International Symposium on Computer [8] David Munoz, Frantz Bouchereau Lara, Cesar Vargas,
Science and Computational Technology. Vol. 1, (20-22 Rogerio Enriquez-Caldera. 2011 Position Location
Dec. 2008), 243-246. Techniques and Applications. Academic press, Newyork.
[6] Berle, F. J. 1986. Mixed triangulation/trilateration [9] Zhizhang Chen, Gopal Gokeda, Yiqiang Yu. 2010
technique for emitter location. Proc. IEE. Vol. 133, Pt. F, Introduction to Direction-of-Arrival Estimation. Artech
No. 7 (1986), 638 – 641. house, Boston.
[7] Prabhakar.S.Naidu 2001 Sensor Array Signal Processing, [10] Johnson, D. H. and Dudgeon, D. E. 1992 Array Signal
CRC press, Newyork. Processing: Concepts and Techniques. Prentice-Hall.
Englewood Cliffs, New Jersey.

25