Application of High Resolution Direction Finding

Algorithms in Mobile Communications

Umar Hamid Syed Ali Abbas

Institute of Communication Technologies, Muhammad Ali Jinnah University,
Islamabad, Pakistan Islamabad, Pakistan
[email protected] [email protected]

Abstract–Mobile communication systems based on 3G and MATLAB simulations have been developed for MUSIC
4G technologies use adaptive antenna arrays to increase their (Multiple Signals Classification), Root-MUSIC and ESPRIT
coverage and channel capacity. Adaptive antenna arrays (Estimation of Signal Parameter via Rotational Invariance
involve direction finding and beamforming algorithms to Technique) algorithms. Results of these simulations have
localize and track both signals i.e. users and interferers. This been used for characterizing the linear antenna array with
paper presents simulation and analysis of three high resolution optimal performance.
direction finding algorithms namely MUSIC, Root-MUSIC
and ESPRIT. These algorithms provide an estimate about the
number of incoming signal sources and their angles of arrival
on an antenna array. Simulation results have been used to II. HIGH RESOLUTION DIRECTION FINDING
evaluate the performance of these algorithms by varying the
ALGORITHMS
antenna array parameters such as number of mobile users,
number of antenna elements, time samples acquired and
signal-to-noise ratio. A. Data Generation for Antenna Array
Several plane wave signals from P sources are incident
Keywords–direction finding, antenna array, MUSIC, Root-
MUSIC, ESPRIT
on a uniform linear array from different angles θi where i =
1, 2 … P. The array consists of Q antenna elements. At any
time instant t, the received signal vector Rxd(t) at the antenna
array can be written as:
I. INTRODUCTION Rxd(t) = Σ a(θp) x sp(t) (1)
Direction finding algorithms are used to identify the Where s(t) is a vector of Px1 incoming sources, and a(θ)
angular spread of the sources. They work on the signals is a Qx1 array steering vector for a particular direction θ
received at the output of antenna array and compute the relative to the array broadside as given below:
angles of arrival of all the incoming signals. A beamforming
processor uses these angles of arrival for calculating the a(θ) = [ 1 e-jα … e-j(N-1)α ] (2)
complex weight vectors for beam steering. A beamformer Where α is defined as the phase delay across the antenna
steers the radiation in a particular direction and places the array and is given below:
nulls in the interfering directions with the help of known
angles of arrival of the sources [1]. α = (2π/λ) x de x sin(θ) (3)
Conventional direction finding algorithms are quite Where de is the element spacing and λ is the wavelength
vulnerable to the noise of incoming data due to large side of the received signal. The signal vector Rxd(t) can be written
lobe levels in the antenna array beam pattern. In addition as:
these algorithms require a large number of antenna elements Rxd(t) = A x s(t) (4)
to resolve the angles of arrival with precise accuracy [2].
This paper focuses on high resolution direction finding Where A = [a(θ1) … a(θP)] is a PxQ matrix of steering
algorithms. These algorithms exploit properties of Eigen vectors. The antenna array output Oarray(t) consists of both
structures to provide a solution to an underlying estimation signal and noise components and can be written as:
problem for a given application. These algorithms apply
Oarray(t) = Rxd(t) + w(t) (5)
Eigen value decomposition techniques on the input
covariance matrix of received signal to estimate the signal Where Rxd(t) and w(t) are assumed to be uncorrelated
and noise subspaces [3], [4]. and w(t) is modeled as white Gaussian process.

978-1-4673-8753-8/16/$31.00 ©2016 IEEE

 B. Direction Finding with MUSIC D. Direction Finding with ESPRIT
MUSIC is among the most popular algorithms used for The ESPRIT algorithm was proposed by Roy and Kailath
estimating the angles of arrival of the incoming signal [8] and it works by dividing an antenna array of Q elements
sources. It was proposed by Schmidt [5] and involves into two sub-arrays with Q-1 elements. The two sub-arrays
estimating the signal and noise subspaces through Eigen are formed with a fixed displacement vector i.e. a fixed
value decomposition of the input covariance matrix as shown distance in a fixed direction. A concept of sub-arrays for
in fig. 1 below. ESPRIT algorithm is given in fig. 2 below:

Figure 2 Sub-Arrays Concept in ESPRIT Algorithm

This algorithm exploits the rotational invariance property

in the signal subspace by creating two sub-arrays from a
Figure 1 MUSIC Algorithm Flowchart single antenna array with a fixed displacement vector as
shown in fig. 2 above. The calculations for determining the
angles of arrival using ESPRIT algorithm are shown in fig. 3
below:
In this algorithm, the covariance matrix CM consists of P
signal Eigen values that form the signal subspace and Q-P
noise Eigen values that form the noise subspace. This noise
subspace is orthogonal to the steering vectors as shown in
fig. 1 above. The MUSIC spectrum shows peaks appearing at
the angles of arrival of the signals incident on antenna array.

C. Direction Finding with Root-MUSIC

MUSIC algorithm involves a comprehensive search
through all possible steering vectors to provide an estimate
of the incoming angles of arrival. This procedure makes it a
computationally intensive algorithm. On the other hand,
Root-MUSIC is a model based parameter estimation
techniques proposed by Barabell [6]. It is applicable to
uniform linear arrays only and involves rearranging the
denominator of Pmusic as shown in fig. 1 above, to form a
polynomial R(z) on the unit circle. In this way peaks in Pmusic
appear as Roots of R(z) either lying on or close to a unit
circle [7].
For a true θ, ej(2πdsin(θ)/λ) is a root of R(z). The angles of
arrival are calculated by finding the roots, z = z1, z2 … zm,
where m = 1, 2 … P, of this polynomial R(z) as follows:
θm = sin-1 [ (λ/(2 π de)) arg(zm) ] (6) Figure 3 ESPRIT Algorithm Flowchart
 The ESPRIT algorithm provides estimation about the Table. 1 Analysis of Root-MUSIC and ESPRIT with Q = 8
angles of arrival in terms of Eigen values as shown in fig. 3
above and involves less computations and storage as Theta Root-MUSIC ESPRIT
compared to MUSIC. -15o -14.9764o -15.0539o

-8o -7.0388o -6.2230o

III. SIMULATION RESULTS AND ANALYSIS 0o 1.9007o 1.3829o

MATLAB simulations have been developed for
evaluating the performance of MUSIC, Root-MUSIC and 9o 9.3056o 9.9857o
ESPRIT algorithms in terms of providing accurate estimates
about the incoming angles of arrival. In these simulations,
the user can input parameters such as number of antenna Table. 2 Analysis of Root-MUSIC and ESPRIT with Q = 16
elements (Q), number of mobile users (P), angles of arrival
of the mobile users (Theta), number of time samples (Ts) Theta Root-MUSIC ESPRIT
and signal-to-noise ratio (SNR). o o
-15 -14.9634 -15.0261o
Scenario 1: This scenario shows the effects of varying
the number of antenna elements with fixed number of -8o -7.9796o -7.9943o
mobile users, number of time samples and SNR. The values
of parameters used are Q = 8 and 16, P = 4, Theta = {-15o, 0o 0.0200o -0.0027o
-8o, 0o, 9o}, Ts = 128, SNR = 10 dB.
9o 9.0131o 8.9823o

Inference on Scenario 1: The spatial resolution of these

algorithms improved with increase in number of antenna
elements (Q) as shown in fig. 4 and fig. 5 respectively [9],
[10] and [11].
Scenario 2: This scenario shows the effects of varying
the number of time samples with fixed number of antenna
elements, number of mobile users, and SNR. The values of
parameters used are Ts = 128 and 256, Q = 8, P = 3, Theta =
{-10o, 0o, 12o}, SNR = 10 dB.

Figure 4 Number of antenna elements Q = 8

Figure 6 Number of time samples Ts = 128

Figure 5 Number of antenna elements Q = 16

Tables 1 and 2 below show the comparison of Root-

MUSIC and ESPRIT algorithms for the simulation scenario
mentioned above.
 Figure 7 Number of time samples Ts = 256 Figure 8 Number of mobile users P = 5

Tables 3 and 4 below show the comparison of Root-

MUSIC and ESPRIT algorithms for the simulation scenario
mentioned above.

Table. 3 Analysis of Root-MUSIC and ESPRIT with Ts = 128

Theta Root-MUSIC ESPRIT

o o
-10 -9.3195 -9.2721o

0o 0.0469o 0.0244o

Figure 9 Number of mobile users P = 8

12o 11.6038o 11.5949o

Table. 4 Analysis of Root-MUSIC and ESPRIT with Ts = 256 Inference on Scenario 3: The MUSIC algorithm is
unable to resolve all the signals in the presence of large
Theta Root-MUSIC ESPRIT number of incident signals (P) closer to the number of
o o
antenna elements (Q) as shown in fig. 9.
-10 -9.9961 -10.0316o
Scenario 4: This scenario shows the effects of varying
0o -0.0560o -0.0395o the SNR with fixed number of antenna elements, number of
mobile users, and number of time samples. The values of
12o 11.8737o 12.0146o parameters used are SNR = {-10, 10} dB, Q = 8, P = 3,
Theta = {10o, 20o, 30o}, Ts = 128.

Inference on Scenario 2: The spatial resolution of these

algorithms improved with increase in number of time
samples (Ts) as shown in fig. 6 and fig. 7 respectively [9],
[10] and [11].
Scenario 3: This scenario shows the effects of varying
the number of mobile users with fixed number of antenna
elements, number of time samples and SNR. The values of
parameters used are Q = 12, P = 5, Theta = {-30o, -22o,
-15o, -5o, 3o}, P = 8, Theta = {-40o, -30o, -22o, -15o, -5o, 3o,
11o, 20o}, Ts = 256, SNR = 10 dB.

Figure 10 SNR value -10 dB

 Figure 11 SNR value 10 dB Figure 12 Increased number of mobile users in a low SNR situation

Tables 5 and 6 below show the comparison of Root- Table 4 below shows the comparison of Root-MUSIC
MUSIC and ESPRIT algorithms for the simulation scenario and ESPRIT algorithms for the simulation scenario
mentioned above. mentioned above.

Table. 5 Analysis of Root-MUSIC and ESPRIT with SNR = -10 dB Table. 7 Analysis of Root-MUSIC and ESPRIT with low SNR -10 dB

Theta Root-MUSIC ESPRIT Theta Root-MUSIC ESPRIT

o o
10 o
11.3178 o
10.7428 o
-48 -48.3424 * -48.8873o

20o 21.1347o 20.3730o -40o -40.9461o -40.9438o

30o 30.9762o 29.0924o -30o -31.3881o -31.3054o

-22o -22.2377o * -22.4587o

Table. 6 Analysis of Root-MUSIC and ESPRIT with SNR = 10 dB -15o -14.7468o * -14.8415o

Theta Root-MUSIC ESPRIT -5o -4.1329o -4.2605o

10o 10.2084o 10.3349o
3o 2.7235o * 2.6663o
o o o
20 20.3500 20.5909
11o 11.6717o 11.5122o
o o o
30 30.2524 30.3895
20o 19.9272o * 19.7541o

29o 29.1193o * 29.3672o

Inference on Scenario 4: The spatial resolution of these
algorithms degrades in a low SNR situation as shown in fig.
10 and table 5 respectively.
Scenario 5: This scenario shows the effects of increased Inference on Scenario 5: In this case Root-MUSIC
number of mobile users in a low SNR situation with fixed performs much better than MUSIC and ESPRIT algorithms
number of antenna elements, and number of time samples. as shown in table 7. Root-MUSIC is able to resolve most of
The values of parameters used are SNR = -10 dB, Q = 14, P the incoming signals marked with *.
= 10, Theta = {-48o, -40o, -30o, -22o, -15o, -5o, 3o, 11o, 20o, Scenario 6: This scenario shows the effects of increase
29o}, Ts = 256. in number of antenna elements (Q) along with time samples
(Ts) on the execution time of these algorithms. The values
of parameters used are Q = {4, 8, 16, 32, 64} with Ts = 256
and 512 respectively.
 than MUSIC and ESPRIT algorithms as shown in table 7.
Fig. 13 and fig. 14 show that increased values of Q and Ts
result in increased execution times of all the direction
finding algorithms. The execution time of Root-MUSIC
increases drastically with large values of Q and Ts.

V. CONCLUSIONS
This paper presented a detailed analysis of high
resolution direction finding algorithms and their application
in mobile communication systems using antenna arrays.
This paper presented MATLAB simulations of MUSIC,
Figure 13 Execution times (msec) with variable Q for Ts = 256 Root-MUSIC and ESPRIT algorithms that allowed the user
to change the design parameters of a linear antenna array i.e.
Q, Ts, SNR, P along with values of Theta. Simulation
results verified that MUSIC algorithm is a preferable choice
for direction finding applications having a large antenna
array with a large number of time samples. Simulation
results also verified that Root-MUSIC algorithm is a
preferable choice for direction finding applications detecting
a large number of mobile users with narrow angular
separation and low SNR environment, keeping the number
of antenna elements and time samples fixed. These results
can provide help to system engineers involved in design and
analysis of linear antenna arrays for mobile
communications.
Figure 14 Execution times (msec) with variable Q for Ts = 512

REFERENCES
Inference on Scenario 6: The execution time of these [1] L.C. Godara,“Application of antenna arrays to mobile communication
algorithms increases with increase in Q and Ts especially II: Beamforming & direction of arrival considerations”, Proc. of
when the value of Ts is high as shown in fig. 13 and fig. 14 IEEE, vol. 85, no. 8, pp., 1195-1245, August 1997
respectively. [2] Hamid Krim and Mats Viberg, “Two Decades of Array Signal
Processing Research”, IEEE SP Magazine, pp., 67-90, July 1996
[3] John Litva and Titus Lo, “Digital Beamforming in Wireless
Communications”, Artech House Mobile Communication, 1996
IV. DISCUSSION ON OBTAINED RESULTS [4] Toby Haynes, “A Primer on Digital Beamforming”, Spectrum Signal
Processing, 1998, Information on http://www.spectrumsignal.com/
Fig. 5 and fig. 7 show that increase in number of antenna [5] R.O. Schmidt, “Multiple Emitter Location and Signal Parameter
elements (Q) and time samples (Ts) improve the Estimation”, IEEE Trans. Antennas and Propagation, vol. 34, no. 2,
performance of all the direction finding algorithms in terms pp., 276-280, March 1986
of accurately resolving the incoming signals with sharp [6] A.J. Barabell, “Improving the Resolution Performance of
peaks and minimal errors in detecting the angles of arrival. Eigenstmchxe-based Direction Finding Algorithms”, Proc. IEEE Int.
Conf Acoustics, Speech and Signal Processing, pp., 336-339, 1983
Fig. 9 shows that increase in number of mobile users (P) [7] Harry Trees, “Optimum Array Processing, Detection, Estimation and
with respect to number of antenna elements (Q) results in Modulation Part IV”, John Wiley and Sons, New York , 2002
poor spatial resolution of MUSIC along with significant [8] R. Roy and T. Kailath, “ESPRIT Estimation of Signal Parameters Via
errors in angle detection of Root-MUISC and ESPRIT Rotational lnvariance Techniques”, IEEE Trans. Acoustics, Speech,
algorithms. This issue can be addressed by increasing the and Signal Processing, vol. 37, no. 7, pp., 984-995, July 1989
values of Q and Ts at the cost of array size and [9] E.M. Al-Ardi, R.M. Shubair and M.E. Al-Mualla, “Performnace
Evaluation of Direction Finding Algorithms for Adaptive Antenna
computational requirements. Fig. 10 shows that as the SNR Arrays”, Proc. IEEE Int. Conf ICECS, 2003
is reduced, the spatial resolution also reduces for almost all
[10] U. Hamid, S. Moin and R.A. Qamar, “Comparative Analysis of DOA
the direction finding algorithms. Estimation Algorithms for Linear Sensor Arrays”, Proc. Int. Conf
Applied Sciences and Technology, IBCAST 2007
Fig. 12 shows that increased number of mobile users (P)
in a low SNR situation degrades the performance of all the [11] T.B. Lavate, V.K. Kokate and A. M. Sapkal, “Performance Analysis
of MUSIC and ESPRIT DOA Estimation algorithms for adaptive
direction finding algorithms. In this case increased values of array smart antenna in mobile communication”, Proc. IEEE Conf
Q and Ts allowed the Root-MUSIC to perform much better Computer and Network Technology, ICCNT 2010