2017 Progress In Electromagnetics Research Symposium — Fall (PIERS — FALL), Singapore, 19–22 November

Design and Comparative Evaluation of Antenna Array Performance

Using Non Blind LMS Beamforming Algorithms
Rinkee Chopra and Rahul Lakhmani
Indian Institute of Technology, Bombay, India

Abstract— In this paper, 8 element linear antenna array is designed for ISM band at an
optimum operating frequency of 2.48 GHz for conventional microstrip patch antenna using CST
Microwave Studio. Element spacing has been optimized to 0.5λ to remove mutual coupling
which makes focusing of main beam towards unwanted direction and shifts the resonant frequency.
Focusing main beam toward look direction and directing nulls in the jamming signals direction are
obtained by applying excitation weight vectors produced by different variants of Non-Blind LMS
beamforming algorithm. On the basis of these algorithms, antenna array efficiency, normalized
array factor and mean square error (MSE) are examined for mobile communication. Side lobe
levels, null depth and computational complexity have also been analyzed for comparative analysis.

1. INTRODUCTION
With rapid increment in wireless communication network services, the user demands for better
capacity, coverage and higher quality of service within specified electromagnetic spectrum [1, 2].
These challenges motivated researchers to employ better techniques to improve spectrum utilization.
Various techniques like cell sectoring, cell splitting, frequency reuse etc. have been applied so far
to get over these problem. Smart antenna system has been adopted by ITU for third generation
[3G] and fourth generation [4G] wireless networks due to its capability to improve channel capacity,
enlarge bandwidth, multipath and co channel interference suppression. So far from the late 1950s,
there is a gradual improvement in the field of smart antenna. Smart antenna system processes
antenna array elements with digital signal processor to optimize its radiation pattern dynamically
in response to the signal environment. Single antenna can’t be used as smart antenna because
of its inflexibility to modify its radiation pattern so antenna array is to be designed with proper
feeding. Signal received by the antenna array is of continuous analog type and signal processor is
digital in format so data conversion device is needed to make these signals compatible with signal
processor [8].
Smart antenna DSP processor employs two functions: (1) Angle of arrival (AOA) estimation
and (2) Adaptive beam-forming (ABF). AOA algorithm works on the signal received at antenna
array elements and computes the direction of all received signals that comprises desired user signal
i.e., signal of interest (SOI) and all jamming signals, i.e., signal not of interest (SNOI). ABF is a
process in which each user’s signal is multiplied by complex weight vectors to adjust antenna element
current amplitude and phase [8]. Therefore, it steers the main beam toward the look direction and
suppresses the jamming signals of the same carrier frequency from other directions by obtaining
excitation weight vectors W1 , W2 , . . . , WM . Various adaptive algorithms are mentioned in literature
to compute these excitation weights. Most of the algorithms are categorized in two categories-blind
and non-blind algorithms. Non-blind algorithms require information of desired signal but blind
algorithms estimate the direction of arrival of desired signal. Non-blind algorithms are discussed
in this paper.
Least mean square (LMS) algorithm is robust, simple and have slow convergence speed with
computation complexity O(M) [8–10]. Therefore, various modifications to this algorithm are
done including normalized LMS (NLMS) [11], variable step size LMS [12, 13], variable step size
NLMS [14, 15], sign LMS algorithm [16, 17] hybrid LMS [18] and leaky LMS [19]. In this paper,
comparative analysis of non-blind adaptive techniques is done through MATLAB simulation by
varying desired and interfering signal directions and simulated results are also verified through
CST Microwave Studio by using excitation weights as current feeding vector.
This paper is organised in following ways: Section 2 deals with antenna design and optimiza-
tion, Section 3 states the mathematical modelling, Section 4 explains LMS adaptive beam-forming
algorithm along with various modified LMS algorithms such as NLMS, sign Data LMS, sign error
LMS, sign sign LMS, leaky LMS, hybrid LMS etc., Section 4 presents the simulated results and
finally Section 5 concludes the paper.

1827
 2017 Progress In Electromagnetics Research Symposium — Fall (PIERS — FALL), Singapore, 19–22 November

2. ANTENNA DESIGN AND OPTIMIZATION

Antenna element shows an important role in shaping and scanning the radiation pattern. Rectan-
gular microstrip antenna is used in this work due to its attractive characteristics as single element
and in arrays. Beside this, they are economical, light weight, conformal, easy to fabricate and ver-
satile [1]. Dimensions of designed microstrip antenna are calculated using conventional microstrip
antenna eqs [1]. The dimensions used for array design in CST Microwave Studio with optimum
inter-element spacing are arranged in the Table 1. An 8-element linear antenna array is designed
as shown in Fig. 1. Parametric optimization is done for inter element spacing of 8 element antenna
array.
Substrate Dielectric
0.5 2.2
Thickness constant
Patch length 40 Patch width 48.4
Inset width 1.9 Inset depth 18
Strip width 1.5 Feed line length 38

Table 1. Parameters of optimized patch antenna. Figure 1. Geometry of eight element array.

Figure 2(a) shows the simulated return loss |S11 | vs frequency plots. The inter element spacing
of 0.5λ gives the proposed frequency 2.48 GHz with very low return loss, i.e., −35 dB and far field
radiation pattern for conventional antenna array directs main beam toward broadside direction with
high side lobe level −9.5 dB as shown in Fig. 2(b). The next section deals with the mathematical
modeling of antenna array which is used for antenna array weight optimization through MATLAB.

(a) (b)

Figure 2. CST simulation results of conventional 8 element array, (a) return loss |S11 | in dB, (b) 3D far-field
radiation pattern.

3. MATHEMATICAL MODELING
Consider an M element uniform linear array receives desired signal s (k) arriving at an angle θ0
and N interference signals in (k) arriving from different angles θn (n = 1, 2, . . . , N ). Each angle is
called angle of arrival (AOA) and defines direction of arrival of signal with respect to a reference
direction normal to the array axis. The signal xm (k) received at the input of every m-th element
(m = 1, 2, . . . , M ). Output of antenna array is
y(k) = wH x(k) (1)
where w = [w1 w2 . . . wM ]T is excitation weight vector and superscript H shows the Hermitian
transpose. The main goal of adaptive beam-forming algorithms is to reduce the error between the
desired signal and antenna array output by applying mean square error to zero. This error can be
expressed [6] as
e(k) = d(k) − wH (k)y(k) (2)
Minimum error is obtained by using different beam-forming algorithms which optimizes weight
vectors in such a way that results in best match of antenna array output with desired signal. The
algorithms are described in the next section.

1828
2017 Progress In Electromagnetics Research Symposium — Fall (PIERS — FALL), Singapore, 19–22 November

4. ADAPTIVE ALGORITHM
To optimize the error between the desired input data and array output, adaptive algorithms are
used. Many algorithms are applied to modify excitation weight coefficient in order to find best
match of array output with desired signal. This section describes adaptive algorithms to gain a
better understanding of adaptive filtering techniques.
4.1. Least Mean Square Algorithm (LMS)
Least mean square is simplest gradient based adaptive beam-forming algorithm that comprises
repetitive process to make successive correction in the negative gradient direction which finally
results in minimum mean square error. LMS algorithm modifies the excitation weights along the
direction of the estimated gradient based on the steepest descent method [8, 9]. The weight update
equation of LMS algorithm is given by

w(k + 1) = w(k) + μe(k)x∗ (k) (3)

1
The convergence of LMS is directly dependent on step size μ. It is given by 0 ≤ μ ≤ 2λmax where,
λmax is the trace of autocorrelation matrix [6]

Rxx = E[x(k)xH (k)]

4.2. Normalized LMS (NLMS) Algorithm

LMS is sensitive to the scaling of its input vector x(k). This results in very difficult selection of
learning rate, i.e., step size μ to assure stability of algorithm. Least mean square is modified to
NLMS which solves this problem by normalizing the input power [11]. Updated equation of array
weight coefficient of NLMS is given by

μe(k)x∗ (k)
w(k + 1) = w(k) + (4)
xH (k)x(k)

where superscript H shows the Hermitian transpose, used for complex conjugate of input x(k).
This algorithm decreases the step size μ to make the changes large. Thus, the step size μ varies
adaptively by following the changes in the input signal level. This avoids weight divergence and
makes the algorithm more stable and faster converging than LMS algorithm where a fixed step size
is used.
4.3. Variable Step Size Algorithm (VSS-LMS)
In conventional LMS low step size leads to extremely large convergence time and large step size leads
to degradation in error performance. Thus, optimum value of step size is necessary to maintain
equivalence. This problem prompted variable step size LMS. In variable step size LMS algorithm,
step size is varied according to square of the prediction error [12, 13]. Large prediction error results
in increased step size with faster tracking while small prediction error leads to decrease in step size
that yields smaller misadjustment [12]. Weight update equation of VSS-LMS can be expressed as

w(k + 1) = w(k) + μ(k)x(k)e(k) (5)

μ(k + 1) = αμ(k) + γe2 (k) (6)

where, α lies between 0 and 1 and γ > 0. Suitable selection of these parameters assures good
equivalence between convergence time and error performance.
4.4. Variable Step Size NLMS Algorithm (VSS-NLMS)
Variable step size normalized LMS (VSS-NLMS) finds an optimal value of step size through vari-
able step size. This parameter provides a solution to the challenge of fast convergence and low
misadjustment of the NLMS algorithm [14, 15]. VSS-NLMS weight update equation is given as

μx(n)e(n)
w(k + 1) = w(k) + (7)
x(n)xH (n) + ε

where, ε is positive regularization parameter, i.e., small positive number.

1829
 2017 Progress In Electromagnetics Research Symposium — Fall (PIERS — FALL), Singapore, 19–22 November

4.5. Sign Data LMS Algorithm (SDLMS)

High data rate applications of adaptive filters often require maximum speed and minimum hardware
requirement to avoid computational complexity. In this algorithm, computation complexity of LMS
is simplified by applying signum operator to input vector x(k). Weight update equation of sign
data LMS [16] is given by
w(k + 1) = w(k) + μsgn(x(k))e(k) (8)
x(k)
where, sgn(x(k)) stands for signum operator of data and it is given by sgn(x(k)) = |x(k)| . The
drawback of SD-LMS is sometimes the weight vector obtained by this algorithm results in diver-
gence.
4.6. Sign Error LMS Algorithm (SE-LMS)
In this algorithm, weight update equation is modified by applying signum operator to error e(k) in
order to reduce computational complexity. Modified weight update equation is given as
w(k + 1) = w(k) + μx(k)sgn(e(k)) (9)

1 e(k) ≥ 0;
where, sgn(e(k)) is given by sgn(e(k)) = 0 e(k) = 0; SE-LMS can also be seen as two level
−1 e(k) ≤ 0
quantizer to an error e(k).
4.7. Sign Sign LMS Algorithm (SS-LMS)
In this algorithm, signum operator is applied to both data and error so both data and error are
quantized [17]. Thus, the weight update equation is given by
w(k + 1) = w(k) + μsgn(x(k))sgn(e(k)) (10)
Convergence rate of SS-LMS is slower than LMS. This is also known as Zero forcing LMS.
4.8. Hybrid LMS Algorithm (H-LMS)
Hybrid LMS algorithm is designed using conventional LMS and normalized LMS algorithm to im-
prove convergence rate. In terms of robustness, this algorithm is better than conventional LMS [18].
Hybrid LMS weight update equation can be represented as
w(k + 1) = w(k) + 2μ1 e(k)x∗ (k) for 0 < n ≤ p (11)
and w(k + 1) = w(k) + 2μ2 R−1 e(k)x∗ (k) for 0 < n ≤ p (12)
where p: switching point from LMS to NLMS algorithm. R−1 = E[x(k)xH (k)] and μ1 , μ2 represents
step sizes which control the convergence speed and stability.
4.9. Leaky LMS Algorithm
LMS algorithm is modified to leaky LMS to overcome the problem of inadequacy of excitation in
the input data which results in unbounded parameter estimates. This algorithm also gives the best
result in alleviating stalling, where the gradient estimate is very small to adjust the weight vector
due to very low input signal [19]. Leaky LMS weight update equation is given by
w(k + 1) = (1 − μβ)w(k) + μx(k)e(k) (13)
where, β is the leakage factor that is smallest constant greater than zero.
The above weight update equations are simulated using MATLAB R2013a. Array factor pattern
is calculated using these vectors with help of following equations [6]
M
+1
w(k) cos(2(n − 1)u)
n=1
AFodd = (14)
M+1
w(k)
n=1

M
w(k) cos((2n − 1)u)
n=1
AFeven = (15)

M
w(k)
n=1

where, M = Number of antenna elements and u = πd

λ (sin θ − sin θ0 ).

1830
2017 Progress In Electromagnetics Research Symposium — Fall (PIERS — FALL), Singapore, 19–22 November

5. SIMULATION RESULTS
LMS, variable step size LMS, NLMS, VSS-NLMS, SD-LMS, SE-LMS, SSLMS, hybrid LMS and
leaky LMS algorithms have been applied on 8 element antenna array. Step size parameter μ = 0.024
and SNR = 20 dB is used in above algorithms. Additional parameters employed by variable step
size LMS, variable step size NLMS, leaky LMS are α = 0.97, γ = 2.8 × 10−4 , ε = 0.001 and
β = 0.001. Various cases have been studied for different signal of interest (SOI) and signal not of
interest (SOI) but only one case is shown here due to brevity of space. All signals are assumed to
be uncorrelated with each other and antenna elements are considered without mutual coupling. All
algorithms are compared in terms of normalized array factor pattern, side lobe level, null depth,
computational complexity and mean square error. These algorithms are run for 100 iterations. The
optimal excitation weights and errors obtained using these algorithms in MATLAB are given in
Table 2. Normalized array pattern, signal tracking, mean square error are plotted using MATLAB
and far field radiation pattern is plotted in CST Mircowave Studio using these excitation weights
as feeding current amplitude and phase.
CASE1: Signal of interest direction 35◦ , interfering signal direction −20◦ .

Table 2. Weights and error of conventional LMS, VSS-LMS, SS-LMS, SE-LMS, SD-LMS, NLMS, VSS-NLMS,
Leaky LMS, Hybrid LMS having desired angle at 35◦ and interfering angle at −20◦ .

Adaptive Beamforming algorithms

weights VSS- SS- VSS- SD- Leaky Hybrid
LMS SE-LMS NLMS
LMS LMS NLMS LMS LMS LMS
w1 1 1 1 1 1 1 1 1 1
w2 0. 341 0.341 0.257 0.470 0.361 0.359 0.7041 0.341 0.341
+1.121i +1.121i +1.276i +0.929i +1.106i +1.086i +1.444i +1.121i +1.122i
w3 -0.7403 −0.740 −0.6247 -0.562 −0.714 −0.715 −1.223 −0.740 −0.740
+1.098i +1.098i +0.811i +0.895i +1.083i +1.061i +1.100i +1.10i +1.098i
w4 −1.0693 −1.069 −1.176 −1.004 −1.060 −1.060 −1.457 −1.069 −1.069
−0.0068i −0.007i +0.693i −0.064i −0.011i −0.018i +0.299i −0.007i −0.007i
w5 − 0.244 −0.244 0.145 −0.384 −0.262 −0.265 −0.277 −0.244 −0.244
−1.041i −1.042i −1.401i −0.947i −1.036i −1.023i −1.539i −1.041i −1.041i
w6 0. 907 0.906 0.293 0.651 0.875 0.867 1.042 0.907 0.906
−0.966i −0.966i −0.399i −0.839i −0.961i − 0.939i −1.087i −0.966i −0.966i
w7 1. 169 1.169 1.319 1.012 1.151 1.144 1.665 1.169 1.169
+0.084i +0.085i −0.497i +0.137i +0.084i +0.096i +0.071i +0.084i + 0.084i
W8 0. 222 0.222 −0.379 0.311 0.232 0.236 −0.066 0.222 0.222
+0.975i +0.975i +1.319i +0.957i +0.976i +0.970i +1.393i +0.975i +0.975i
error 2.6e-07 3.89e-07 0.18901 0.21334 0.0152 0.01435 1.2e-05 2.7e-07 1.6e-11

Figures 3–4 show that the LMS algorithm and its various variants place nulls in the direction of
interfering signals and maximum in the direction of the desired signal. SE-LMS, SS-LMS, SD-LMS
can also locate main beam towards desired user and nulls toward jamming signals while it reduces
computational complexity as shown in Table 3. Analysis of mean square error represents that
the variable step size LMS, normalized LMS, variable step size NLMS, hybrid LMS, leaky LMS
can efficiently converge in less iteration as compare to conventional LMS while LMS has better
capability of directing mean beam toward desired direction and placing nulls toward interferers.
Quantitative comparison of side lobe level, null depth, computational complexity in terms of adder
and multipliers is shown in the Table 3 [21].
Table 3 presents the side lobe suppression is better in case of NLMS, VSS-NLMS, hybrid LMS,
leaky LMS as compare to LMS while adders and multipliers required by these algorithms are more
than the conventional LMS. Hybrid LMS gives best side lobe level suppression along with deep
nulls toward interferers. Convergence of leaky LMS is highly dependable on α and γ. It shows slow
convergence for large value of α and γ while decrease in parameters α and γ decreases stability and
accuracy so optimum value is required to maintain convergence and stability. Table 2 clearly shows
substantial reduction in adders and multipliers for computation in case of SE-LMS, SD-LMS and
SS-LMS but side lobe level and null depth for these algorithms are higher than others.

1831
 2017 Progress In Electromagnetics Research Symposium — Fall (PIERS — FALL), Singapore, 19–22 November
0
Conventional LMS
1 LMS-VSS

Noramilzed Array Factor|AFn|

Sign Sign LMS -10

Noramilzed Array Factor|AFn|

Sign Error LMS
0.8 NLMS -20
VSS-NLMS
SD-LMS -30 Conv LMS
0.6
Leaky LMS LMS-VSS
Hybrid LMS SS-LMS
-40
0.4 SE-LMS
-50 NLMS
VSS-NLMS
0.2 SD-LMS
-60
LeakyLMS
HybridLMS
0 -70
-90 -60 -30 0 30 60 90 -90 -60 -30 0 30 60 90
Angle Of Arrival (deg) Angle Of Arrival (deg)
(a) (b)
1.5 1
Conv LMS
LMS-VSS
1 SS LMS
0.8
SE LMS

Mean square error

0.5 NLMS
0.6 VSS-NLMS
Signals

Desired Signal
Conv LMS SD-LMS
0
LMS-VSS Leaky LMS
SS LMS 0.4 Hybrid LMS
-0.5 SE-LMS
NLMS
VSS-NLMS 0.2
-1 SD-LMS
Leaky LMS
Hybrid LMS 0
-1.5 0 10 20 30 40 50 60 70 80 90
0 10 20 30 40 50 60 70 80 90
No. of Iterations Iteration no.

(c) (d)

Figure 3. MATLAB Simulation results using conventional LMS, VSS-LMS, SS-LMS, SE-LMS, SD-LMS,
NLMS, VSS-NLMS, leaky LMS, hybrid LMS having desired angle at 35◦ and interfering angle at −20◦ ,
(a) normalized array linear pattern, (b) normalized array dB pattern, (c) desired signal tracking, (d) mean
square error.

(a) (b)

Figure 4. CST simulation results of LMS 8 element array desired angle at 35◦ and interfering angle at −20◦
and interfering angle at 3◦ , (a) 3D far-field radiation pattern, (b) polar far-field radiation pattern.

Table 3. Quantitative comparison of beamforming algorithms.

Algorithms SLL (dB) Null Depth (dB) Multiplication Addition

LMS −12 to −20 −18 to −35 8N + 2 8N
Variable Step Size LMS (VSS-LMS) −14 to −23 −20 to −45 8N + 2 8N
NLMS −12 to −25 −20 to −50 10N + 2 10N
Variable Step Size NLMS (VSS-NLMS) −12 to −28 −23 to −55 10N + 2 10N
Sign Error LMS (SE-LMS) −9 to −20 −14 to −28 6N 6N
Sign Data LMS (SD-LMS) −8 to −20 −12 to −30 6N 6N
Sign Sign LMS (SS-LMS) −8 to −18 −12 to −22 4N 4N
Hybrid LMS −13 to −30 −20 to −65 8N + 2 (10N + 2) 8N (10N )
Leaky LMS −15 to −26 −18 to −55 10N + 3 8N + 1

1832
2017 Progress In Electromagnetics Research Symposium — Fall (PIERS — FALL), Singapore, 19–22 November

6. CONCLUSION
This paper talked about eight-element linear array that is designed by optimizing inter-element
spacing at 0.5λ with accurate resonant frequency at 2.48 GHz. The performance has been inves-
tigated and compared with several adaptive beam-forming algorithms such as LMS, VSS-LMS,
NLMS, Hybrid LMS etc.. Antenna array design and optimization for spacing is done using com-
mercial software CST Microwave Studio while analysis and comparison of beam-forming algorithms
for the complex weight calculation is done using MATLAB and these results have been also exam-
ined using CST Microwave Studio. NLMS, VSS-NLMS, Leaky LMS shows faster convergence as
compare to LMS while main beam directing capability of LMS is better than others. Even though
side lobe level suppression and interferers nullifying capability of SD-LMS, SS-LMS, SE-LMS is less
than conventional LMS but it reduces computation complexity at a substantial rate.

REFERENCES

1. Balanis, C. A., Antenna Theory: Analysis and Design, John Wiley & Sons, 2012.
2. Godara, L. C., “Application of antenna arrays to mobile communications, Part II: Beamform-
ing and direction-of arrival considerations,” Proceedings of IEEE, Vol. 85, No. 8, 1195–1245,
1997.
3. Monzingo, R. A. and T. W. Miller, Introduction to Adaptive Arrays, SciTech Publishing, 1980.
4. Gross, F. B., Smart Antennas for Wireless Communications: with MATLAB, McGraw-Hill,
New York, 2005.
5. Godara, L. C., Smart Antennas, CRC Press, 2004.
6. Frank, B. G., “Smart antennas for wireless communications with MATLAB,” 2005.
7. Haji, I. A., M. R. Islam, A. Z. Alam, O. O. Khalifa, S. Khan, K. A. Abdullah, and A. A. Yussuf,
“Design and optimization of linear array antenna based on the analysis of direction of arrival
(DOA) estimation and beamforming algorithms,” International Conference on Computer and
Communication Engineering (ICCCE), 1–4, IEEE, May 2010.
8. Gondal, M. A. and A. Anees, “Analysis of optimized signal processing algorithms for smart
antenna system,” Neural Computing and Applications, Vol. 23, Nos. 3–4, 1083–1087, 2013.
9. Upadhyay, S. C. and P. M. Mainkar, “Adaptive array beamforming using LMS algorithm,”
International Journal of Engineering Research and Technology, Vol. 2. No. 1, ESRSA Publi-
cations, January 2013.
10. Kawitkar, R. S. and D. G. Wakde, “Smart antenna array analysis using lms algorithm,” IEEE
International Symposium on Microwave, Antenna, Propagation and EMC Technologies for
Wireless Communications, MAPE 2005, Vol. 1, 370–374, 2005.
11. Mohammad, T. I. and A. A. R. Zainol, “MI-NLMS adaptive beamforming algorithm for smart
antenna system applications,” Journal of Zhejiang University Science, Vol. 7, No. 10, 1709–
1716, 2006.
12. Kwong, R. H. and E. W. Johnston, “A variable step size LMS algorithm,” IEEE Transactions
on Signal Processing, Vol. 40, No. 7, 1633–1642, 1992.
13. Aboulnasr, T. and K. Mayyas, “A robust variable step-size LMS-type algorithm: Analysis and
simulations,” IEEE Transactions on Signal Processing, Vol. 45, No. 3, 631–639, 1997.
14. Huang, H.-C. and J. Lee, “A new variable step-size NLMS algorithm and its performance
analysis,” IEEE Transactions on Signal Processing, Vol. 60, No. 4, 2055–2060, 2012.
15. Shin, H.-C., A. H. Sayed, and W.-J. Song, “Variable step-size NLMS and affine projection
algorithms,” IEEE Signal Processing Letters, Vol. 11, No. 2, 132–135, 2004.
16. Sethares, W. A., et al., “Excitation conditions for signed regressor least mean squares adapta-
tion,” IEEE Transactions on Circuits and Systems, Vol. 35, No. 6, 613–624, 1988.
17. Rahman, M. Z. U., R. A. Shaik, and D. V. Reddy, “An efficient noise cancellation technique to
remove noise from the ECG signal using normalized signed regressor LMS algorithm,” IEEE
International Conference on Bioinformatics and Biomedicine, BIBM’09, IEEE, 2009.
18. Chern, S.-J., C. H. Jyh, and K. M. Wong, “The performance of the hybrid LMS adaptive
algorithm,” Signal Processing, Vol. 44, No. 1, 67–88, 1995.
19. Rigling, B. D. and P. Schniter, “Subspace leaky LMS,” IEEE Signal Processing Letters, Vol. 11,
No. 2, 136–139, 2004.

1833
 2017 Progress In Electromagnetics Research Symposium — Fall (PIERS — FALL), Singapore, 19–22 November

20. Khan, S. A. and S. A. Malik, “Adaptive beamforming algorithms for anti-jamming,” Interna-
tional Journal of Signal Processing, Image Processing and Pattern Recognition, Vol. 4, No. 1,
95–106.
21. Yasin, M., P. Akhtar, and Valiuddin, “Performance analysis of LMS and NLMS algorithms
for a smart antenna system,” International Journal of Computer Applications, Vol. 4, No. 9,
25–32, 2010.

1834