Different Adaptive Beamforming Algorithms for

Performance Investigation of Smart Antenna System

Ashraf A. M. Khalaf Abdel-Rahman B. M. El-Daly Hesham F. A. Hamed
Department of Communication and Department of Communication and Department of Communication and
Electronics, Faculty of Engineering, Electronics, Faculty of Engineering, Electronics, Faculty of Engineering,
Minia University, Minia, Egypt Minia University, Minia, Egypt Minia University, Minia, Egypt
[email protected] [email protected] [email protected]

Abstract— Smart antenna systems are of great importance (SIR) and increasing the signal-to-noise ratio (SNR) [3-5].
in wireless communication and radar applications. They Smart antenna is used in many applications such as radar,
effectively enhance the system capacity and reduce the co- sonar, seismology, wireless communications, radio astronomy
channel interference. Smart antenna is an array antenna that and biomedicine.
uses adaptive beamforming algorithms to steer the main beam
toward the desired signal direction and reject the interfering In this paper the performance of five non-blind algorithms
signals of the same frequency from other directions without -LMS, NLMS, SMI, RLS and hybrid LMS/SMI-will be
moving the antenna. This is achieved by continuously updating compared by varying the number of radiating elements and the
the weights of each radiating element (antenna). An algorithm displacement between elements of the array. The performance
with low complexity, low computation costs, high speed of each algorithm is measured in terms of beamforming
convergence rates and better performance is usually preferred. stability, beam width, maximum side-lobe level, nulls depth,
This paper introduces a new performance investigation and and convergence rate. The investigation in this study aims to
comparison between five different beamforming algorithms: determine the optimum algorithm.
Least Mean Square (LMS), Normalized Least Mean Square
(NLMS), Sample Matrix Inversion (SMI), Recursive Least II. SMART ANTENNA MODEL
Square (RLS) and Hybrid Least Mean Square / Sample Matrix
Inversion (LMS/SMI). In this investigation, the number of array
The smart antenna block diagram with non-blind
elements and the displacement among them are changed in each beamforming algorithm is shown in Fig. 1.
algorithm. Simulation experiments are carried out, and the θ0 S 0 (k )
results are compared to verify the effectiveness of this x0 (k )
w *
0
investigation. The performance of each algorithm is optimized
and demonstrated using MATLAB software package. x1 (k ) y (k )
θ1
S1 (k )
w 1* Σ
Keywords—smart antenna; adaptive beamforming; least mean
square; sample matrix inversion; recursive least square.
SM−1(k) xN−1(k)
θM−1 w *N −1
I. INTRODUCTION
A smart antenna system is an integration between array
antenna elements and digital signal processing techniques.
Array signal processing involves the manipulation of signals Adaptive Algorithm
e(k )
Σ
−
induced on the elements of an array antenna. A vital role is +
DOA
played by the smart antenna in the increased desire of various
mobile communications services to achieve the spatial Beamformer d (k )
parameters of a wireless channel under noisy environment [1]. Fig. 1. The adaptive beamforming array antenna (smart antenna) model
The signal processing technique uses two algorithms, a block diagram [3].
direction of arrival (DOA) algorithm and an adaptive
beamforming algorithm. The DOA algorithm calculates the The array factor of spherical angle θ for linear array of N
direction of arrival of all incoming signals. Then, the adaptive radiating elements with elements spacing d is given by the
beamforming algorithm is used to update the system formulas [6]:
parameters (w0, w1, … wN-1 shown in Fig. 1). The weights of N /2
each element change the phase shift and amplitude attenuation
of the received signal so that the main beam is steered toward
AF even (θ ) = w
n =1
n cos(( 2 n − 1) u ) , N = even (1)

the desired signal direction and nulls are placed in the ( N +1) / 2

directions of the interference [2]. Smart antennas have many AF odd (θ ) = w

n =1
n cos( 2 ( n − 1) u ) , N = odd (2)
advantages such as increasing the system capacity, reducing where
bit-error rate, compatibility with the various multiple access
technique, enhancing the beam directivity, decreasing the wn = the weight of nth radiating element,
effects of fading, reducing the signal-to-interference ratio
 u = (πd / λ ) sin θ . w ( k + 1) = w ( k ) + μe* ( k ) x ( k ) (7)
Equation (1) and (2) can be written in general form for N where μ is the learning rate and the error signal e(k ) is
even or odd number of radiating elements as: given by:
 − ( N −1 ) 
e(k ) = d (k ) − w H (k ) x (k ) (8 )
N −1 j + n  kd sin (θ )
AF (θ ) = w
n=0
n e  2 
(3) The convergence of the LMS algorithm is directly related
to the step-size (learning-rate) parameter μ. Therefore, it is
where very important to choose a step-size in a range that insures
k = wave number ( 2π / λ ) , the best and the most stable convergence. This happens under
the following condition [10-13]:
λ = wavelength of the incident wave. 1
0 ≤ μ ≤ (9 )
Considering M to be the number of arriving plan waves λ max
that are incident on the linear array from multiple directions
(θ 0 , θ 1 ,......, θ M −1 ) the incident signals on nth radiating λmax = the greatest eigenvalue of the correlation matrix R̂xx .
element are given by [1]: B. Normalized Least Mean square (NLMS)
M −1  − ( N −1 )
j

+ n  kd sin (θ i )
The LMS is a robust algorithm; however, one of its
x n (t ) =  S i (t ) e
i=0
 2 
+ n n (t ) (4) drawbacks is its sensitivity to the scaling of its input. This
makes it very hard to choose a learning rate that insures
where stability of the algorithm. The NLMS is a variant of the LMS
S 0 ( t ) = the desired signal, algorithm that solves this problem by normalizing to the
power of the input signal. The NLMS algorithm updates the
S 1→ ( M −1 ) ( t ) =
the interfering signals, weights as follows [11]:
n n (t ) =
the received noise signal at nth element.
The array output can be given by [6]: μ opt e * ( k ) x ( k )
w ( k + 1) = w ( k ) + (10 )
γ + x H (k ) x (k )
y (k ) = w H ⋅ x (k ) (5) where μopt is the optimal learning rate for the NLMS
where algorithm that is equal to 1 and is independent of the input,
w = [w 0 w N − 1 ] = array weights vector, and γ is small positive value.
T
w1 
x (k ) = vectors of inputs to the array which is equal to C. Sample Matrix Inversion (SMI)
x (k ) = x s (k ) + xi (k ) + n (k ) (6) The SMI algorithm uses a block-adaptive approach,
with which would give better performance than a continuous
xs (k ) = approach; therefore, it has a faster convergence rate than the
vector of desired signal, LMS algorithm. The sample matrix is a time average estimate
xi (k ) = vector of interfering signal, of the array correlation matrix using K-time samples. The
n ( k ) = zero mean Gaussian noise vector, SMI weights can then be calculated for the kth block of length
K-time samples as [2, 7, 12]:
III. ADAPTIVE BEAMFORMING ALGORITHMS w ( k ) = R xx− 1 ( k ) r ( k ) (11 )
The adaptive beamforming algorithms compute the where the estimated correlation matrix and correlation
weights by minimizing the error between the array output and vector of length K-time samples of block number k can be
the desired signal until the weights reach their optimum calculated the time average as:
values. In many application, such as a wireless mobile 1 K

communications system and radar target tracking, it is desired Rˆ xx ( k ) =

K
 x (k ) x
k =1
H
(k ) (12 )
to achieve high speed adaptation of the weights and to reject K
1
the interference and noise. Investigation in this study aims to
determine the optimum algorithm by comparing their
r (k ) =
K
d
k =1
*
(k ) x H (k ) (13 )

performance. Each algorithm is discussed below.

D. Recursive Least Square (RLS)
A. Least Mean Square (LMS) Although the SMI algorithm is faster than the LMS
The least mean square is an adaptive algorithm which algorithm, the computational complexity and potential
produces the least mean squares of the error signal (difference singularities related to computing the correlation matrix
between the desired and the actual signal). It is a gradient inversion can cause many problems. Therefore, we can use
method of steepest decent based on the error at the current the RLS algorithm that calculates the required correlation
time. The LMS algorithm updates the weights as follows [7- matrix and vector iteratively to reduce the computational
10]: complexity with fast conversion rate as [6]:
 Rˆ xx ( k ) = α Rˆ xx ( k − 1) + x ( k ) x H ( k ) (14 ) The pattern beamwidth of the antenna will be narrower
and the directivity will increase by increasing the number of
rˆxx ( k ) = α rˆxx ( k − 1 ) + d * ( k ) x ( k ) (15 ) radiating elements N as shown in Figs. 2 to 6. In LMS
where α is the forgetting factor; that is a positive constant algorithm; the convergence rate speeds up by increasing the
value in rang 0 < α ≤ 1 .The weights of RLS algorithm are number of radiating elements N. At N = 8, LMS converges
updated as follows: after 45 iterations while at N = 51 it converges after only 7
(
w (k ) = w (k − 1) + g (k ) d * (k ) − x H (k ) w (k − 1) ) (16 )
iterations as shown in Fig. 2 and Table I. On the other hand,
in the NLMS algorithm, the convergence rate is insensitive to
where the gain vector g (k ) is defined as: the number of radiating elements N; it converges after 6
α − 1 Rˆ xx− 1 ( k − 1) x ( k ) iteration for all values of N as shown in Fig. 3 and Table I.
g (k ) = (17 ) This convergence independence of NLMS algorithm is due
1 + α − 1 x H ( k ) Rˆ xx− 1 ( k − 1) x ( k ) to the normalization to the power of the inputs.
correlation matrix inverse can be computed iteratively as:
In the SMI algorithm, it is found that the minimum
Rˆ xx−1 ( k ) = α −1 Rˆ xx−1 ( k − 1) − α −1 g ( k ) x H ( k ) Rˆ xx−1 ( k − 1) (18 )
number of time samples K required for acceptable
E. Hybrid Least Mean Square algorithm/Sample Matrix performance is 30, as shown in Fig. 4. Values of K less than
Inversion (LMS/SMI) 30 reduce the computational complexity but the array factor
pattern is distorted. In the RLS algorithm, the convergence
Another way to improve the performance of LMS and
occurs after 3 iterations and the convergence rate is
SMI algorithms is combining their advantages and avoiding
insensitive to the number of antenna elements N as shown in
their drawbacks. To achieve that, the weights of LMS
Fig. 5 and Table I. The hybrid LMS/SMI algorithm
algorithms will be initialized by SMI algorithm of block
convergences much faster than all the previous algorithms.
length (K) calculations as [2, 14]:
w ( 0 ) = R xx− 1 ( 0 ) r ( 0 ) (19 )
Now, not only does the LMS/SMI algorithm take small
time to converge but also the block length (K) need not be
large to ensure that the effect due to the change in the signal
environment during the block acquisition does not affect the
performance of algorithm and computational complexity of
SMI will be decreased.
IV. SIMULATION RESULTS
This simulation is performed by steering the beam in the
desired signal direction and placing nulls in the undesired Fig. 2. Normalized array factor of linear array and mean square error at
directions. Assuming that signals have direction of arrival different radiating elements number using LMS algorithm at d= λ/2.
30°, 0° and -60°; the signal of interest is coming from the
direction of 30° and two random interfering signals are
coming from the direction of 0° and -60°. First, simulation is
carried out using 8, 16, 24, 32, and 51 antenna radiating
elements, then at elements displacement of λ/8, λ/4, λ/2, and
λ. The effect of interfering noise is taken into account by
executing the simulations at 30 dB signal to noise ratio
(SNR), where the noise is represented as additive white
Gaussian noise.

A. Effect of Varying Number of Antenna Elements Fig. 3. Normalized array factor of linear array and mean square error at
different radiating elements number using NLMS algorithm at d= λ/2.
This subsection shows the simulation results of varying
the number of antenna elements N and its effect on the This is because the weights are initialized by the SMI
convergence, beam width, side-lobes level (SLL), nulls depth algorithm with small block size, K=10, with values that are
and beamforming stability while the spacing of the radiating near the optimum weights solution. The error decreases by
elements is fixed at λ/2. Figs. 2 to 6 show the simulation increasing the number of antenna elements, where the
results of normalized array factor and the mean square error maximum error at N = 8 is 4.2 ×10-4 and at N=51 is 0.45 ×10-
4
for the linear array using LMS, NLMS, SMI, RLS and as shown in Fig. 6. The SMI, RLS and hybrid LMS/SMI
LMS/SMI respectively. Table I show the analytical values of algorithms introduce nulls depth level that are deeper than
simulation results using LMS, NLMS, RLS and LMS/SMI LMS and NLMS. However, LMS and NLMS give better
respectively that are shown in Figs. 2 to 6. side-lobe levels than SMI, RLS and LMS/SMI as Figs. 4 to 6
and Table I.
 Fig. 7 shows the average of 100 simulation results of antenna elements and vice versa, as shown in Fig. 7. In
different cases for each algorithm to measure the stability and general; the side-lobe level is decreased and the nulls depths
compare between the beamforming. Each case has different are increased by increasing the number of elements as shown
conditions, where the characteristics of each of the interfering in Fig. 7.
signals are changed by making them random signals in TABLE I. BEAMFORMING COMPARISON OF N FOR THE ALGORITHMS
addition to the noise. Figs. 2 to 6 and Table I show only one
Null Null
of these 100 different cases. By comparing the results shown Array
HPBW
Beam Max.
Depth Depth
Element Width SLL Convergence
in Fig. 7 and Table I, it is clear that LMS and NLMS are more No.
(Deg)
(Deg) (dB)
at
0°(dB)
at -
60°(dB)
stable than SMI, RLS and LMS/SMI. The beamwidth at the 8 15.12° 35.1° -12.87 -49.22 -48.81 45 iterations
same number of elements is slightly different between 16 7.29° 16.65° -13.07 -50.83 -49.34 18 iterations

LMS
24 4.89° 11.07° -13.16 -50.91 -51.01 16 iterations
different algorithms. However they all decrease by increasing 32 3.66° 8.28° -13.21 -49.32 -48.87 10 iterations
the number of elements as shown in Fig. 7. 51 2.29° 5.22° -13.26 -49.61 -49.39 7 iterations
8 15.15° 35.1° -12.89 -43.91 -44 6 iterations

NLMS
16 7.35° 16.65° -13.06 -46.02 -46.78 6 iterations
24 4.89° 11.16° -13.2 -49.06 -48.85 6 iterations
32 3.69° 8.28° -13.21 -50.26 -49.72 6 iterations
51 2.29° 5.22° -13.24 -51.18 -52.07 6 iterations
8 15.35° 35.58° -9.4 -57.92 -65.35 No iterations
16 7.83° 18.45° -10.83 -60.41 -60.66 No iterations

SMI
24 5.04° 11.34° -12.03 -67.15 -62.25 No iterations
32 3.78° 8.82° -11.74 -61.65 -56.79 No iterations
51 2.24° 4.86° -9.83 -69.76 -59.43 No iterations
8 14.44° 35.1° -10.01 -49.84 -58.55 3 iterations
16 7.06° 16.65° -9.88 -51.91 -59.89 3 iterations

RLS
24 5.76° 11.07° -11.91 -63.65 -61.64 3 iterations
32 3.54° 8.28° -11.23 -57.01 -66.89 3 iterations
51 2.46° 5.22° -10.89 -63.47 -64.43 3 iterations
8 15.39° 35.64° -11.64 -54.86 -53.68 1 iterations

LMS/SMI
16 7.56° 17.01° -11.54 -61.45 -51.15 1 iterations
24 5.31° 11.34° -12.36 -57.00 -55.33 1 iterations
Fig. 4. Normalized array factor of linear array with different radiating 32 3.64° 8.10° -10.44 -50.32 -57.04 1 iterations
elements number using SMI algorithm at d= λ/2. 51 2.28° 5.13° -11.78 -69.82 -57.76 1 iterations

Half-Power Beamwidth (Deg) Maximum Side Lobe Level (dB)

20 -20
15.19
15.16
15.12

15.12
15.3

-13.24
-13.23
-13.21
-13.21
-13.19
-13.18
-13.09
-13.08
-12.89
-12.88

-12.64
-12.52
-12.49
-12.41

-12.09
-12.05
-11.94
-11.89

-11.79
-11.54

-11.45
-11.31
-11.9
-11.12
15 -15

-10.75
7.38
7.38
7.35
7.29

7.29

10 -10
4.95
4.89

4.88
4.87
4.86

3.71
3.69
3.68
3.66

3.63

2.33
2.29
2.28
2.29
2.21
5 -5

0 0
8 16 24 32 51 8 16 24 32 51
LMS NLMS SMI RLS LMS/SMI LMS NLMS SMI RLS LMS/SMI

Fig. 5. Normalized array factor of linear array and mean square error at Null Depth at 0° (dB) Null Depth at -60° (dB)
different radiating elements number using RLS algorithm at d= λ/2.
-63.71

-63.73

-63.01
-65.3
-62.27

-61.65
-61.43
-60.85

-60.72
-60.52
-62.7
-59.38

-59.18
-58.99

-59.02
-58.31

-57.35
-75 -75

-59.6
-55.39
-55.31

-55.09
-53.98
-53.74

-53.17

-52.78
-52.35
-52.12

-51.57
-51.43

-50.89
-50.21

-50.15

-50.14
-50.11
-49.65

-49.33
-49.01

-48.86

-48.79
-48.75

-48.55

-48.46
-48.36
-50.2

-47.35
-49.7

-46.93
-46.84
-43.88

-43.57

-55 -55

-35 -35

-15 -15

8 16 24 32 51 8 16 24 32 51
LMS NLMS SMI RLS LMS/SMI LMS NLMS SMI RLS LMS/SMI

Fig. 7. The average results of 100 simulation results at different cases of

interfering input signals and shows HPBW (deg), Max. SLL (dB), Null
Depth at 0° (dB) and Null Depth at -60° (dB) of different adaptive
beamforming algorithms under different number of antenna elements.
Fig. 6. Normalized array factor of linear array with different radiating
elements number using LMS/SMI algorithm at d= λ/2.
B. Effect of Varying Antenna Elements Spacing
The SMI algorithm introduces the maximum side-lobe level This subsection shows the effect of varying the spacing d
while the LMS and NLMS give the lowest side-lobe level as between the antenna radiating elements on convergence,
shown in Fig. 7. The deepest nulls are given by the SMI beamwidth, (SLL), nulls depth and beamforming stability
algorithm followed by the RLS algorithm, while the LMS and while the number of radiating elements is fixed at 16
NLMS give the lowest nulls depth. The LMS algorithm has elements. Figs. 8 to 11 show the simulation results of
deeper nulls than the NLMS algorithm at low number of normalized array factor and mean square error for the linear
array using LMS, NLMS, RLS and LMS/SMI, respectively. The directivity will be improved and the pattern beam-
Table II shows the analytical values of the simulation results width of antenna will be narrower by increasing the spacing
that are shown in Figs. 8 to 11. The convergence rate is between antenna elements d as shown in Figs. 8 to 11. The
insensitive to the spacing between antenna elements d. Also error level increases by increasing the displacement between
the SMI, RLS and LMS/SMI algorithms give deep nulls the antenna elements as shown in Fig. 11.
levels and bad side-lobe levels compared to the LMS and Fig. 12 shows the average result of 100 simulation results
NLMS algorithms as shown in Fig. 8 to 11 and Table II. of different cases for each of the previous algorithms to meas-
ure the stability and compare between the beamforming
characteristics while changing the spacing between the
antenna elements d. Each case has different conditions, where
the characteristics of each of the interfering signals are
changed randomly along with the noise. Figs. 8 to 11 and
Table II show only one case from the 100 different cases. By
comparing the results shown in Fig. 12 and Table II, the LMS
and NLM algorithms introduce stable results. The HPBW
changes slightly between the algorithms, but the HPBW
decreases by increasing the displacement between antenna
elements as shown in Fig. 12. The LMS and the NLMS
Fig. 8. Normalized array factor of linear array and mean square error at algorithms give the lowest side-lobe level while the SMI
different radiating elements spacing using LMS algorithm at N=16. algorithm gives the highest side-lobe level, as shown in Fig.
12. The SMI and RLS have deeper nulls than the other
algorithms. The LMS and NLMS algorithms have the lowest
nulls depth as shown in Fig. 12.
Fig. 12 illustrates that the best results occurred at d = λ/2.
Another problem appears when the spacing between antenna
elements d becomes larger than λ/2. A side-lobe begins to
grow until it becomes another main-lobe at d = λ, as shown
in Figs. 8 to 11. This new main-beam causes errors in the
received signal due to interferences and wasted power. By
tracking the growth of this side-lobe with increasing the
Fig. 9. Normalized array factor of linear array and mean square error at displacement between elements, it is found that it reaches -
different radiating elements spacing using NLMS algorithm at N=16. 6.9 dB at d = 0.64 λ which is an undesirable value as shown
in Fig. 13. Therefore the displacement between elements
should not exceed the value of 0.63 λ. The best displacement
between antenna elements d = 0.6 λ that gives the narrowest
beamwidth and directivity without generating high side-lobe
level. Also, the best performance of the smart antenna system
is achieved at d = 0.6 λ and it is better than the performance
at d = λ/2 for each algorithm as shown in Fig. 12 and 14.
TABLE II. BEAMFORMING COMPARISON OF D FOR THE ALGORITHMS
Null Null
Array Beam Max.
HPBW Depth Depth at
Element Width SLL Convergence
(Deg) at -
Spacing (Deg) (dB)
0°(dB) 60°(dB)
Fig. 10. Normalized array factor of linear array and mean square error at
/8 30.24° 90° -13.44 -45.35 -45.25 18 iterations
different radiating elements spacing using RLS algorithm at N=16.
LMS

/4 14.75° 34.28° -13.13 -46.39 -46.70 18 iterations

/2 7.36° 16.67° -13.08 -47.88 -47.05 18 iterations
3.67° 8.26° -13.07 -47.89 -47.51 18 iterations
/8 30.24° 90° -13.46 -47.03 -46.04 6 iterations
NLMS

/4 14.84° 34.27° -13.14 -47.46 -46.19 6 iterations

/2 7.36° 16.67° -13.17 -47.06 -47.08 6 iterations
3.68° 8.25° -13.19 -47.74 -48.06 6 iterations
/8 33.57° 90° -10.95 -64.56 -58.87 No iterations
SMI

/4 14.75° 33.3° -9.9 -57.68 -63.76 No iterations

/2 7.11° 15.75° -9.85 -54.35 -62.37 No iterations
4.14° 10.17° -9.88 -62.75 -65.85 No iterations
/8 33.93° 90° -15.04 -53.38 -53.49 3 iterations
RLS

/4 14.94° 34.47° -14.12 -58.12 -64.8 3 iterations

/2 8.01° 18.45° -11.20 -59.32 -58.27 3 iterations
3.87° 9.27° -11.12 -73.88 -58.71 3 iterations
/8 31.23° 90° -11.40 -52.99 -62.98 1 iterations
LMS/
SMI

/4 14.58° 32.85° -10.70 -51.35 -70.12 1 iterations

Fig. 11. Normalized array factor of linear array and mean square error at /2 8.05° 18.63° -12.51 -51.08 -60.61 1 iterations
different radiating elements spacing using LMS/SMI algorithm at N=16. 3.87° 9.18° -13.34 -60.18 -45.76 1 iterations
 Half-Power Beamwidth (Deg) Maximum Side-Lobe Level (dB) 0.6 λ is the optimum spacing between the antenna elements.
-20
40
The LMS algorithm has good performance and stable pattern
30.42
30.24
30.18
29.97
29.97

-13.45
-13.44

-13.14

-13.12
-13.11

-13.11

-13.09
-12.74

-12.69

-12.62

-13.2
-12.46
-12.39

-12.21
-12.16
-11.89

-11.82

-11.77
beamforming except that its convergence is slow and depends

-11.53

-11.21
30 -15

on the number of antenna elements. The LMS and the NLMS

14.85
14.86
14.84
14.87
14.76
-10
20
algorithms introduce the lowest side-lobe level. The SMI and
7.37
7.35
7.36
7.28

7.27
the RLS algorithms improve the convergence at the expense

3.69
3.69
3.69
3.67
3.67
10 -5

of high side-lobe level and instability of pattern

0 0
λ/8 λ/4 λ/2 λ λ/8 λ/4 λ/2 λ beamforming. The SMI and the RLS algorithms give the
LMS NLMS SMI RLS LMS/SMI LMS NLMS SMI RLS LMS/SMI
deepest nulls at the directions of interfering signals. The
Null Depth at 0° (dB) Null Depth at -60° (dB)
hybrid LMS/SMI algorithm also improves the convergence,
-62.02
-59.37
-59.11

-59.34
-58.77

-58.54
-57.89

-57.99
-57.63

-57.89
-57.82

-57.71
-57.31
-57.13

decreases the mean square error and minimizes

-56.25
-75 -75

-58.3
-52.93

-53.05
-52.68

-52.08

-52.09

-52.29
-52.14

-52.01
-47.66

-47.43
-47.25

-46.59

-46.99
-46.39

-46.31
-46.28

-46.51

-46.41
-46.19

-46.19
-46.04
-45.26

-47.6
computational complexity at the cost of high side-lobe level
-45.6

-55 -55

and beamforming instability. The NLMS algorithm achieves

-35 -35
the best performance with respect to speed convergence,
-15 -15 beamforming stability, low side-lobe level, computational
simplicity and appropriate nulls depth.
λ/8 λ/4 λ/2 λ λ/8 λ/4 λ/2 λ
LMS NLMS SMI RLS LMS/SMI LMS NLMS SMI RLS LMS/SMI REFERENCES
Fig. 12. The average results of 100 simulation results at different cases of [1] L. C. Godara, "Application of antenna arrays to mobile
interfering input signals and shows HPBW (deg), Max. SLL (dB), Null communications. II. Beam-forming and direction-of-arrival
Depth at 0° (dB) and Null Depth at -60° (dB) of different adaptive beam- considerations," Proceedings of the IEEE, vol. 85, pp. 1195-1245,
forming algorithms under different of antenna elements Spacing. 1997.
[2] W. Ali, D. Mohamed, and A. H. Hassan, "Performance analysis of least
mean square sample matrix inversion algorithm for smart antenna
system," in Antennas and Propagation Conference (LAPC)
Loughborough, UK, 2013, pp. 624-629.
[3] C. A. Balanis, Antenna theory: analysis and design, 3rd ed. vol. 1: John
Wiley & Sons, 2005.
[4] C. A. Balanis and P. I. Ioannides, "Introduction to smart antennas,"
Morgan & Claypool Publishers’ series Synthesis Lectures On
Antennas, vol. 2, pp. 1-175, 2007.
[5] L. C. Godara, "Applications of antenna arrays to mobile
Fig. 13. Normalized array factor of linear array with different radiating communications. Part I. Performance improvement, feasibility, and
elements spacing using LMS algorithm with d = 0.6 λ, 0.61 λ, 0.62 λ, 0.63 λ system considerations," Proceedings of the IEEE, vol. 85, pp. 1031-
and 0.64 λ at N=16. 1060, 1997.
[6] F. B. Gross, Smart antennas for wireless communications with
13.98

-13.16
-13.18

-59.59
-59.42

-59.42
-58.54

MATLAB: McGraw-Hill, New York, 2005.

-53.27

-52.86
-12.82
6.14

-49.74

-49.62
-47.54

[7] S. Imtiaj, I. S. Misra, and R. Biswas, "A comparative study of

-47.07
13.89

LMS
-12.49

beamforming techniques using LMS and SMI algorithms in smart

6.13

13.86
13.86

NLMS antennas," in International Conference on Communications, Devices

6.12
6.12

-11.95

and Intelligent Systems (CODIS), 2012 Kolkata, 2012, pp. 246-249.

13.77

SMI
[8] G. Nwalozie, V. Okorogu, S. Maduadichie, and A. Adenola, "A Simple
6.11

RLS Comparative Evaluation of Adaptive Beam forming Algorithms,"

International Journal of Engineering and Innovative Technology
LMS/SMI
(IJEIT), vol. 2, pp. 417-424, January 2013.
HPBW (Deg) Beam Width Max. SLL (dB) Null Depth at Null Depth at - [9] A. P. RAO and N. SARMA, "Adaptive Beamforming Algorithms for
(Deg) 0°(dB) 60°(dB) Smart Antenna Systems," WSEAS TRANSACTIONS on
Fig. 14. The average results of linear array with different beamforming COMMUNICATIONS, vol. 13, pp. 44-50, 2014.
algorithms at d = 0.6 λ and N=16. [10] P. Saxena and A. Kothari, "Performance analysis of adaptive
beamforming algorithms for smart antennas," IERI Procedia, vol. 10,
V. CONCLUSION pp. 131-137, 2014.
[11] H. Simon, Adaptive filter theory, 4th ed., 2002.
In this paper five adaptive beamforming algorithms; [12] L. Surendra, S. Shameem, and D. H. Khan, "Performance Comparison
LMS, NLMS, SMI, RLS and hybrid LMS/SMI have been of LMS, SMI and RLS Adaptive Beamforming Algorithms for Smart
applied and their performance have been compared according Antennas," International Journal Of Computer Science And
Technology (IJCST), vol. 3, pp. 973-977, April - June 2012.
to the beamwidth, maximum side-lobe level, nulls depth and [13] Veerendra and M. Bakhar, "Design and Performance Analysis of
the convergence rate. The comparison has been performed Adaptive Beamforming Algorithm for Smart Antenna Systems,"
under changing the number of radiating element N and the International Journal of Advanced Research in Electronics and
spacing between elements d. It is clear from the simulation Communication Engineering (IJARECE), vol. 3, pp. 704-707, July
2014.
results that by increasing the number of antenna radiating [14] W. Ali and A. H. Hassan, "A hybrid least mean square/sample matrix
elements, the overall performance of the smart antenna inversion algorithm using microstrip antenna array," in Science and
improves in all algorithms. The convergence of adaptive Information Conference (SAI), London, UK, 2014, pp. 871-876.
beamforming algorithms is insensitive to the change in the
displacement among the antenna elements. It is clear that d =