Вы находитесь на странице: 1из 5

2010 3rd International Conference on Computer and Electrical Engineering (ICCEE 2010)

Performance and Quality Analysis of Adaptive Beamforming Algorithms (LMS,CMA, RLS & CGM) for Smart Antennas

Abdul Aziz, M.Ali Qureshi


Assistant Professor, Department of Electronic Engineering, University College of Engineering & Technology, The Islamia University of Bahawalpur, Pakistan. Email: {abdul.aziz, ali.qureshi}@iub.edu.pk
Abstract Due to recent substantial development in the field of wireless communication, there is a need to maximize spectral efficiency so that the extensive increase in traffic can be accommodated efficiently. Smart antenna system is a major source to maximize spectral efficiency and capacity of the wireless networks. It consists of an adaptive antenna array that continuously adjusts its radiation characteristics (beam-width of main lobe, side lobe levels and position of nulls) to produce narrow beam in the direction of arrival (DOA) of desired signal and to place nulls in the DOA of interferer signals so that maximum SINR (Signal to Interference and Noise Ratio) is obtained. Smart antennas are becoming more popular now a days due to extensive advancement in the field of digital signal processing and real time implementation of adaptive signal processing techniques on FPGAs. In this paper we analyze various adaptive beamforming algorithms including LMS (Least Mean Squares), CMA (Constant Modulus Algorithm), RLS (Recursive Least Squares) and CGM (Conjugate Gradient Method) through simulating different parameters like radiation pattern, amplitude response, mean square error and absolute weights of an N-element array for a certain number of iterations. The obtained simulation results are very helpful to evaluate performance and quality of adaptive beamforming algorithms. Keywords- LMS; RLS; CGM; CMA; SINR; Adaptive Beamforming; Smart Antenna; Digital Signal Processing.

M. Junaid Iqbal, S.Zeeshan A. Zaidi Umer Farooq, Usman Ahmad.


Deptt. of Electronic Engineering, UCET, IUB, Pakistan Email: junaid.iqbal19@gmail.com, engineer.sza@hotmail.com,
dabra29@hotmail.com, usman.ahmad@msn.com

I.

INTRODUCTION

Smart antenna is a system of antenna array which uses beamforming algorithms to identify spatial signal signature and is used to compute beamforming vectors to track the antenna beam on receiver. Smart antennas enhance system performance, channel capacity, coverage and spectrum efficiency. Smart antennas use efficient methods to track multiple users and reduce interferences [1]. We can classify smart antennas in two types either switched beam or adaptive array systems. Users in the desired directions are served by the use of multiple fixed beams in switched beam systems. As the mobile user changes its position through the cell, the base station switches between several beams and select the beam that provides the best performance and accuracy in switched beam systems. Generally in a switched beam system, multiple arrays are used with each array covering a certain

area in a specified direction. If we consider a phase shifting network then these multiple beams search for the desired signal in their specific areas and the selection of beam is controlled by the beam logic controller which is governed by an algorithm. This algorithm decides and selects the strongest signal based on the detection of the detector. In high interference regions this technique proved to be less efficient [2]. A process of adapting magnitude and phase of the signal from every antenna element by the use of product of each users signal and weight vectors is called beamforming [3]. Beamforming, as the name depicts, provides narrow beams in desired directions as well as reduces the interferences [4]. If the arrival angles of the desired signal are same then fixed beamforming is used which is a very simple technique but if the arrival angles change with respect to time then there is a need to formulate an optimization scheme that adjusts the weights of arriving signals iteratively. This intelligent technique of adapting the weights with respect to time is called adaptive beamforming. So adaptive beamforming is a technique which separates a desired signal from interfering signal by using technique of adapting the weights with respect to time while maximizing the signal to noise ratio and array output [5]. Adaptive array systems as shown in figure 1 use antenna arrays controlled by strong signal process capability to dynamically vary the radiation pattern in accordance with the varying environment of the signal. Adaptive array systems provide maximum radiation in users direction and also nullify interferences at the same time [2]. Adaptive algorithms are formulated on prescribed performance criteria which are being implemented by a set of iterative equations to meet that criteria. Performance criteria mostly includes MMSE, Maximum signal to interference and noise ratio (SINR), maximum likelihood (ML), Minimum noise variance and Maximum gain [6]. The two major types of adaptive algorithms are blind and non-blind. Non-blind algorithms require a pilot signal to detect the desired signal and update the complex weights. These algorithms include LMS, RLS, SMI and CGM. In contrast to non-blind algorithms, blind algorithms do not

978-1-4244-7224-6 /10/$26.00

2010 IEEE

V6-302

2010 3rd International Conference on Computer and Electrical Engineering (ICCEE 2010)

need a pilot signal to find the complex weights. Blind algorithms include CMA, Spectral Self-Coherence Restoral (SCORE) and Decision Directed algorithm (DD) [3]. A. Least Mean Square Algorithm The Least Mean Square (LMS) algorithm is the famous adaptive filtering algorithm in which Gradient Based approach is employed [7]. It comprises very recent observations and reduces the Mean Square Error iteratively [8], [9], [10]. The direction of gradient vector is opposite to that of steepest descent (SD) [3]. Widrow [10], [11] propounded SD approximation in terms of weights using LMS method mathematically written as [7]. for each k { e (k) = d(k) wH (k) x(k) w(k + 1) = w(k) + e * (k) x(k) } Where, d(k) = reference signal, x(k) = input data vector, w(k) = weight vector, e(k) = error signal, = step size. Stability and convergence rate of LMS algorithm is controlled by the scalar constant [12]. Step size should be chosen in a range in which convergence is insured [8]. 0 < < 2/max Where max is the largest Eigen value of correlation matrix Rxx. The main advantage of LMS is its low computational complexity and disadvantage is slow convergence rate [7]. B. Constant Modulus Algorithm CMA is a well known algorithm of adaptive beamforming of blind adaptation. This algorithm is derived keeping in view the constant complex envelope (amplitude) property of the signal. These signals generally include FM, FSK, PSK, QAM and PAM. If the arriving signal has constant amplitude then this algorithm maintains and restores the amplitude of desired signal [5]. The weights can be calculated using following equations [7]. for each k { y(k) = wH(k) x(k)
e( k ) = y (k) y (k) y (k)

C. Recurssive Least Square Algorithm RLS adaptive algorithm uses method of Least Squares in approximating weight vector w(k) [13]. In LMS, the weight

Figure 1. Block diagram for adaptive beamforming

vector is selected to reduce the averaged error squares while in RLS, it is to reduce the cost function which includes sum of error squares over a time window. The weights can be calculated using following equations [7]. R1(0) = 1I, small positive constant and I the N N identity matrix for each k { k(k) = R1(k 1) x(k) k (k) g (k) = + x H (k )k (k)
-1 k (k)k H (k) R (k 1) + x H (k )g (k) e (k) = d(k) wH(k) x(k) w(k + 1) = w(k) + e(k) g(k) } R -1 (k ) = 1 Where is the forgetting factor, it is also called exponential weighting factor. Its value is such that, 0 1 which shows that it is a positive constant. =1 indicates infinite memory and also ordinary least square algorithm is restored [14]. In comparison to LMS, RLS has faster rate of convergence. This enhanced performance is achieved at the expense of huge computational complexity [7].

w(k + 1) = w(k) + e * (k) x(k) } Where x(k) = input data vector, w(k) = weight vector, e(k) = error signal and = step size. Although CMA does not require a pilot signal but it has a major drawback of slow convergence [7].

V6-303

2010 3rd International Conference on Computer and Electrical Engineering (ICCEE 2010)

D. Conjugate Gradient Method In Conjugate Gradient Method, the gradient is accelerated by choosing conjugate (orthogonal) paths iteratively to get optimal solution for weights that results in faster rate of convergence. It is also called Accelerated gradient Approach (AG) [8]. The weights can be calculated using following equations [15]. r(0) = b Aw(0) g(0) = AH r(0) for each k {
(k) = A r (k) A H g (k)
H 2 2

side lobes and has narrowest beamwidth whereas CGM algorithm has the greater power in the side lobes of radiation pattern as compared to others. CMA algorithm has widest beamwidth as compared to others. In Figure 3, amplitude responses of the four algorithms from -90 degree to +90 degree are shown. It is evident from the figure that LMS, RLS and CGM algorithms reject all the interferences at -30o, -60o and +45o whereas, amplitude response of CMA algorithm is poor as compared to others. CMA algorithm did not reject all the interferences; it suppresses the interferences to some extent. In Figure 4 and 5, mean square error and absolute weights for 50 data samples of four algorithms are shown. It is apparent from the plots that RLS algorithm converges faster among others and LMS has slowest convergence rate. CGM has faster convergence rate than LMS but slower convergence rate than RLS. CMA algorithm depicts unstable behavior.

w(k + 1) = w(k) (k) g(k) r(k+ 1) = r(k) + (k) A g(k)

(k) =

A H r (k + 1) A H r (k)
2

g(k + 1) = AH r (k + 1) (k) g(k) } Where, w(k) denotes array weights, A is the matrix with columns containing consecutive samples from array elements, b is the vector containing consecutive samples of desired signals, r is residual vector and g(k) is the direction vector [15]. II. SIMULATION RESULTS Simulation for an adaptive antenna array is done in MATLAB environment. In simulation we analyze radiation patterns, amplitude response, mean square error and absolute weights for adaptive beamforming algorithms LMS, CMA, RLS and CGM. The parameters used for our simulation are shown in Table 1.
TABLE 1: SIMULATION PARAMETERS Number of antenna elements (N): Element spacing (d): Noise variance (2): Total number of data samples (K): Forgetting factor () (for RLS): DOA of desired signal: DOA of interferer signal (I1) DOA of interferer signal (I2) DOA of interferer signal (I3) 10 0.5 0.001 50 0.9 0o -30o -60o 45o Figure 2. Radiation patterns

In Figure 2, radiation patterns of the four algorithms are shown. It can be seen that LMS algorithm is the best choice among the four algorithms because it has least power in the

V6-304

2010 3rd International Conference on Computer and Electrical Engineering (ICCEE 2010)

Figure 3. Amplitude response

Figure 5. Absolute weights

III.

CONCLUSION

Figure 4. Mean square error

In adaptive beamforming, the radiation pattern of smart antenna is controlled through various adaptive algorithms. Adaptive algorithm dynamically optimizes the radiation pattern according to the changing electromagnetic environment. Here we analyze four popular adaptive techniques including LMS, CMA, RLS and CGM through simulation of various parameters like radiation pattern, amplitude response, mean square error and absolute weights. Comparative analysis of the four algorithms LMS is the simplest and more suitable choice because it has the narrowest beamwidth in the desired direction, least power in side lobes and complete rejection of interferer signals but it has slow convergence which limits its application in case of quickly varying channel conditions and where quick capturing of the signal is required. While CMA has widest beamwidth in the desired direction, suppress interference to some extent and unstable behavior in case of convergence due to which CMA can be used in applications where complex envelope of the signal should ideally be constant. In case of RLS we have narrowest beamwidth, complete rejection of interference and fastest convergence at the cost of high computational burden but has greater power in side lobes as compared to LMS. RLS is the best choice and has also its application where quick tracking of the signal is required. At the end, CGM has also narrow beamwidth, complete rejection of interference, fast convergence as compared to LMS but has the largest power in side lobes as compared to others. It finds its application in mobile communication where it is used to eliminate multipath fading. ACKNOWLEDGMENT We would like to thank our respected teachers and colleagues for their guidance and ever present support.

V6-305

2010 3rd International Conference on Computer and Electrical Engineering (ICCEE 2010)

We are also very much thankful to Dr. Muhammad Mukhtar, the Vice Chancellor of the university, and Professor Jan Muhammad Keerio, the Principal of UCET, for their efforts to facilitate and provide us most suitable environment for research and development. REFERENCES
[1] [2] M. Chryssomallis, Smart antennas IEEE antennas and propagation magazine Vol 42 No 3 pp 129-138, June 2000 S. Rani, P. V. Subbaiah, K. C. Reddy and S. S. Rani, LMS And RLS Algorithm for Smart Antennas in a W-CDMA Mobile Communication Environment, ARPN Journal of Engineering and Applied Sciences, VOL. 4, NO. 6, AUGUST 2009. M. T. Islam, Z. A. Rashid, MI-NLMS adaptive beamforming algorithm for smart antenna system applications, July 2006. M. Ali Qureshi, Abdul Aziz, Hussain Bakhsh, "CIR Improvement using Antenna Beamforming and Power Control", Proc. 2010 International Conference on Networking and Information Technology, ICNIT 2010, Manila Philippines, pp. 294-297, ISBN: 978-1-42447577-3 F. Gross, Smart Antennas For Wireless Communication, Mcgrawhill, September 14, 2005. Constantine A. Balanis, Panayiotis, Introduction to Smart Antennas (Synthesis Lectures on Antennas) , Series edition 2007 by Morgan & Claypool. S. Werner, Reduced complexity adaptive filtering algorithms with applications to communications systems, Ph.D. dissertation, Helsinki University of Technology, Helsinki, Finland, Oct. 2002. R. A. Monzingo and T.W.Miller, Introduction to Adaptive Arrays, Scitech Publishing Inc., Oct. 2003, Mendham, NJ. ISBN 1-89112124-3. D. H. Johnson and D. E. Dudgeon, Array Signal Processing: Concepts and Techniques, Englewood Cliffs, NJ: Prentice-Hall, 1992. B. Widrow, P. E. Mantey, L. J.Griffiths, and B. B.Goode, Adaptive antenna systems, Proc. IEEE, vol. 55, no. 12, pp. 21432159, Aug. 1967. Widrow,B., and M. Hoff, Adaptive Switch Circuits, IREWescom, Convention Record, Part 4, pp. 96104, 1960. W. Y. Shiu, Noniterative digital beamforming in CDMA cellular communications systems, Masters thesis, Queens University, Kingston, Ontario, Nov. 1998. B. D. Van Veen and K. M. Buckley, Beamforming: A versatile approach to spatial filtering, IEEE ASSP Mag., vol. 5, pp. 424, Apr. 1988. doi:10.1109/53.665 Haykin, S.,Adaptive Filter Theory, 4th Edition., Prentice Hall, New York, 2002. L.C. Godara,Smart.Antennas,CRC.Press.Jan.2004.

[3] [4]

[5] [6]

[7]

[8]

[9]

[10]

[11] [12]

[13]

[14] [15]

V6-306

Вам также может понравиться