Republic of Iraq Ministry of Higher Education & Scientific Research Mustansiriayah University College of Engineering Electrical Engineering Department Different Algorithms for Implementation of Adaptive Beam Forming in Wireless System A Thesis Submitted to the Electrical Engineering Department, College of Engineering, AL- Mustansiriayah University, in a Partial Fulfillment of the Requirements for the Degree of Master of Science in Electrical Engineering / Electronic and Communications By: Aseel Abdul-Karim Qasim (B.Sc. Electrical Engineering) Supervised By: Prof. Dr. Adheed Hassan Sallomi Rabi' al-Awwal 1442 November 2020 سورة المجادلة :االية ""11 Acknowledgement All praises and thanks to Allah for giving me the strength and wellbeing to complete this project. I would like to thank sincerely my supervisor Prof. Dr. Adheed Hassan Sallomi for his guidance, advice and encouragement throughout the making of this project. A special thanks to my family. Words cannot express how grateful I am to my mother and father for all of the sacrifices that they’ve made on my behalf. Your prayer for me was what sustained me thus far. I would also like to thank my teachers, especially Dr. Raad Hamdan Thaher for their encouragement and advice, and I would like to thank Dr. Bashar Mundher Mansoor for his support throughout this work. Lastly. I would like to thank all my friends who supported me and helped me through my studies, because they were a one hand and they really care about my success. Thank you very much to everyone who helped me. Dedication I dedicate this work to my Mother for her unending love…. Aseel Abdul-Karim Qasim Table of Contents Table of Contents I List of Abbreviations V List of Symbols VIII Abstract X Chapter one General Introduction 1.1 Introduction 1 1.1.1 Smart Antenna System Structure and Principles 1 1.2 Literature Survey 2 1.3Aim of Work 7 1.4 Contribution 7 1.5 Thesis Outline 8 Chapter Two Theoretical Background for Adaptive Antenna Array System 2.1 Introduction 9 2.2 Antenna Fundamental 9 2.2.1 Radiation Pattern 9 2.2.2 Beamwidth 10 2.2.3 Directivity 10 2.2.4 Gain 11 2.3 Isotropic Antenna 11 2.4 Arrays Antennas 11 2.4.1 Introduction 11 2.5 Linear Arrays 12 2.5.1 Array Element 12 2.5.2 Two Element Array 12 2.5.3 N-Element Uniform-Linear Array 13 2.5.4 N-Element Non-Uniform-Linear Array 14 2.5.5 Another Array Geometry 15 I 2.5.5.1 Circular array 15 2.5.5.2 planer array 16 2.6 Smart Antenna 16 2.6.1 Introduction 16 2.6.2 Type of Smart Antenna 17 2.7 Array Weighting 18 2.8 Angle-of-arrival (AOA) estimation 19 2.9 Beam-forming technique 19 2.9.1 Fixed weight beam forming 20 2.9.2 Adaptive Beam Forming 20 2.10 General control algorithm for adaptive antenna array 21 2.10.1 Least Mean Square (LMS) Algorithm 21 2.10.2 Sample Matrix Inversion (SMI) Algorithm 22 2.10.3 Recursive Least Square (RLS) Algorithm 24 2.10.4 Conjugate Gradient Method (CGM) Algorithm 25 2.10.5 Particles Swarm Optimization Algorithm 25 2.11 Benefits and Features of Smart Antennas 26 2.12 Drawbacks of Smart Antenna 27 2.13 Description and Modeling of Wireless Channel 28 2.13.1 Path loss 29 2.13.2 Shadowing 29 2.13.3 Multipath Fading 29 2.13.4 Rayleigh Model 29 2.13.5 Rice model 29 2.13.6 Power Spectral Density 30 2.13.7 Power Delay Profile (PDP) 30 2.13.8 Mobile Station Mobility Model 31 Chapter Three Proposal algorithm analysis 3.1 Introduction 32 II 3.2 The implementation of normalizing LMS algorithm with maximum SIR factor 33 3.3 Variable step size via error controlling algorithm 36 3.4 Improvements for CMA (ICMA&BCM) algorithm Performance 39 3.4.1 Conventional constant modulus algorithm CMA 39 3.4.2 Improvement proposed to the constant modulus algorithm ICMA via variable step size vector 40 3.4.3 Proposed blocking constant modulus algorithm BCMA 41 3.5 Particles Swarm Optimization Algorithm (PSO) 42 3.6 Adaptive antenna array performance via REDS algorithm44 3.7 BER Chapter Four 47 Simulation Results 4.1 Introduction 50 4.2 Characteristic studies of uniform linear arrays 50 4.2.1 Array factor interpretations with variation of array element displacement 50 4.2.2 Array factor interpretations with variation of number of array elements 52 4.2.3 Performance of comparison for smart antenna system using different antenna array algorithms 54 4.2.4 Comparison of Performance for smart antenna system using different antenna array geometry 55 4.3 Performance of smart antenna system via variable Step size aspect 57 4.4 Performance of smart antenna system via SIR-LMS algorithm 59 4.5 Comparison of smart antenna system executions using different improvements for CMA (ICMA&BCM) algorithm 60 III 4.6 Smart antenna performance with particle swarm optimization (PSO) algorithm 64 4.6.1 Weight Vector Optimization 64 4.6.2 Inter Element Spacing Optimization 66 4.6.3 Weight Vector and Inter Element Spacing Optimizatio 68 4.7 Nulling Effect on the Radiation Pattern 69 4.8 Smart antenna array system performance via REDS algorithm 71 4.8.1 (AWGN) Channel Results 72 4.8.2 Rayleigh Fading Channel Results 75 4.9 BER 79 4.10 SNR 81 Chapter Five Conclusions and Future Works 5.1 Conclusion 83 5.2 Future work 84 Refrences 86 IV List of Abbreviation ABF Adaptive beam forming AF Array factor AOA Angle of arrival AWGN Additive white Gaussian noise BCMA Blocking constant modulus algorithm BER Bit error rate bps Bit per second BPSK Binary phase-shift keying BS base station BW Band width CA Circular array CCM Constrained constant modulus CDMA Code division multiple access CGM Conjugate gradient method algorithm CMA Constant modulus algorithm CSLMS Constrained-stability LMS dB Decibel DBF Digital beam-formed DF Direction finding techniques DMI Direct matrix inversion DOA Direction of arrival DS-CDMA Direct sequence code division multiple access ECVSS Error control variable step size LMS algorithm EDS Euclidean direction search EF Element factor FBGA Field programmable gate array FEDS Fast Euclidean direction search V FF Fitness function FM Frequency modulation FNBW First null beam-width FSK Frequency-shift keying GA Genetic algorithm HPBW Half power beam width ICMA Improvement constant modulus algorithm IPSO Improvement Particle swarm optimization ISI Inter-symbol-interference Km/h Kilometer per hour LA Linear array LMS Least mean square algorithm Mbits Mega bits MHz Mega hertz MIMO Multiple-input and multiple-output MRVSS Modified robust variable step size MS Mobile Station MSE Mean square error MUSIC Multiple signal classification MVDR Minimum variance distortion-less response NLMS Normalized least mean square algorithm NULA Non uniform linear array PA Planner array PDP Power delay profile PSO Particle swarm optimization REDS Rabid Euclidean direction search RLS Recursive least square algorithm SDMA Space division multiple access VI SINR Signal to interference plus noise ratio SIR Signal to interference ratio SLL Side lobe level SMI Sample matrix inversion SMI-LMS Simple matrix inversion - least mean square algorithm SNOIs Signals not of interest SNR Signal to noise ratio SOI Signal of interest SSF Step size factor ULA Uniform linear array VSSLMS Variable step size LMS Rnn Noise correlation matrix Rii Interference correlation matrix Rss Signal correlation matrix VII List of Symbols θs Angle of arrival of the desired signal θi Angle of arrival of the interfering signal G Array element gain ̅w ̅̅ Array weight vector φ Azimuth angle L Block length of data fc Carrier frequency (*) Conjugate operator α Constant value less than one for RLS ̅ d Desired signal fd Doppler frequency d Elements spacing of the uniform linear array e Error signal E|•| Expectation operator ∇w Gradient with respect to w. (•) ̅ n H Hermitian of matrix Input noise vector x̅ Input signal vector, [x1 x2 … . xM ] i, i1 , i2 Interfering signals Fj0(k) M x 1 vector, where M is element array λmax maximum eigenvalue of the autocorrelation matrix of the input vector ̅ R nn N x K zero mean Gaussian noise M Number of array elements m count for number of antenna elements array from 1-M K Number of sample Wopt Optimum weights vector of an algorithm VIII y Output signal of an algorithm p p parameter C Speed of light ‖ •‖ 2 Squared Euclidean norm operator ̅ xx R The array correlation matrix r̅ The correlation vector r̂ The estimate of correlation vector ̂ xx R The estimate of the array correlation matrix S(k) The function of desired signal j0 (k) The index of the weight ̅ A The K x M matrix of array snapshots nth The 𝑘𝑡ℎ element of n µ Step size parameter δ The phase shift from element to element a̅(θ) The steering vector σ2 The variance of the noise signal t Time τ Time delay n Time index trace[R] Trace of autocorrelation matrix [R] I̅ Unity matrix µ(n) Variance step size 𝜆 Wavelength of the carrier signal vid Velocity of swarm xid Position of swarm (. )T Transposition. IX Abstract There is a certain variance when comparing different sorts of algorithms like Least Mean Square (LMS) which is largely used because of the low complexity of computational and ease of execution. The least-squares algorithms or (algorithm that content step size factor), such as Recursive Least Squares (RLS), Conjugate Gradient (CG), and Rapid Euclidean Direction Search (REDS), would converge more rapidly and have smaller mean square error (MSE), and another algorithm will be introduced also like blind algorithm such as Constant Modulus Algorithm (CMA) and blocking algorithm such as Sample Matrix Inversion (SMI), also presents an optimization algorithm like the Particle Swarm Optimization algorithm (PSO). This thesis also discussed different tests for different array geometry. Initially, this thesis studies the effects of changing the number of elements, and the spacing distance between elements, and study the effect of changing the array geometry on smart system performance. The first improvement in this thesis will apply on the least square algorithm, by changing the step size factor (LMSSI) that enhance the LMS convergence from more than 80 to about 20 iterations also provided reducing in the MSE. The second development would be the SIR-LMS that gives zero iteration number. The CMA algorithm reinforcement investigates into two-manner, firstly by apply the variable step size algorithm (ICMA) and secondly by blocking the CMA algorithm (BCMA) the result illustrates that the ICMA improve the number of iteration from 50 for traditional CMA to 23 and 6 iterations for BCMA with keeping the directional beam pattern to desired signals and perfectly nulling in the inference directions. PSO is discussed in three-part, firstly to optimization, the weight vector. Then the second part will be the inter-element spacing optimization, most important achievements acquired an economic and effeteness system using the same or less number of elements besides reducing the array size. The improvement PSO (IPSO) that combine the weight and spacing optimization, and third part studied the nulling effect on the radiation pattern with different number of elements and spacing. REDS algorithm utilizing on smart antenna with three array geometry linear, circular and planner gave more stable and acceptable performance, it can be seen that the number of iteration reduce to 5 for circular, 7 for planner and 8 iteration for linear array as compared with other algorithm, also gave improvement in MSE as appear in chapter four early, the effectiveness of REDS algorithm also appear more robust and remain stable when apply X another type of noise like Rayleigh fading noise in single beam and multi beam signals. The BER and the SNR will be studying for linear and circular array and shows that the BER is approximately close for both arrays at the same number of elements, and it is clear to observe that the SNR curve of overall system increase with the number of elements used in the array regardless the geometry of antenna array. The subsequent simulations on arrays are performed using the MATLAB program version R2017a. XI Chapter One General Introduction Chapter One General Introduction 1.1 Introduction The expansion of adaptive arrays initiates in more than 60s years. In the 1940s, the concept of an antenna array was first used in military applications and significantly evolved in wireless communication. In 1959, Van Atta first coined the term adaptive array to describe a self-phased array. Production of arrays plays a significant role in many different fields, most modern radar and sonar systems depend on arrays antenna as an essential part of the system [1]. The rising demand for mobile communications is gradually growing, reaching about one billion mobile phone users globally. Mobile phones have, of course, been one of the most essential components of ordinary life, and a business-critical device across all developed nations. Therefore, further effective use of the radio spectrum is demanded and the need for wider coverage, improved efficiency, and higher quality of transmission increases. Smart antenna array contains several distributed elements of an antenna (dipoles, monopoles, or elements of the directional antenna) arranged in several geometries [3]. Selected control algorithms, with predefined parameters, give adaptive arrays the unique ability to adjust the characteristics of the radiation pattern (nulls, sidelobe level, main beam direction, and beam-width) [4], that relies on the form of an element, the specific positions and the excitation at each element (amplitude and phase). The term smart antenna encompasses all circumstances in which a system uses an antenna array and the antenna pattern is dynamically changed as required by the system, The beam pattern is therefore modified as desired and the interference signals pass [5], there are two basic forms of smart antennas, the first kind is a phased array or multi-beam antenna composed of either a collection of fixed beams with one beam combining the signal output per each element of the array with the required a single beam (established vie phase-only modification) guided to the desired signal by correctly altering the phase between the elements. Another model is the adaptive antenna array with weighted and combining received signals to enhance the desired to interference signal power ratio (SINR) [6]. 1.1.1 Smart Antenna System Structure and Principles The structure diagram of the adaptive array system is shown in Figure (1.1). 1 Chapter One General Introduction Figure (1.1): smart antenna structure. It consists of the following principal elements: A. Array sensors and RF unit: This unit consists of antenna arrays that receive radio frequency (RF) signals from the space, down conversion chains that detach the carrier of the RF signals obtained by the antenna array, and analog to digital converters (A / D) that transform the carrier signals for more processing. B. Adaptive antenna processor: The signal processing unit, depending on the signal received, which subdivided into the signal processing unit and adaptive algorithm. The signal processing unit calculates weight vectors by which the received signal is multiplied of each array element, and this weight vectors calculation can be implemented using various adaptive algorithms and techniques. C. Beamforming unit: It is a process that blends the radiation patterns of each antenna element on the antenna arrays to form a directional and focused energy beam [4]. The beamforming unit is responsible for shaping and directing the main beam in the proper direction whereas generating nulls in the direction of interference signals leading to an increase of SINR [7]. In just the same unit (Digital Signal Processor (DSP)) the beamforming and signal processing units can commonly be integrated. In general, to achieve an adaptive beamforming a necessary sequence of steps must be carried out which is each element of the adaptive array has to be independently weighted the received signal, the weight vector for each of these signals has to be continuously controlled, all signals are then combined to form the output and the decision about the amount of adaptation in the beam unit must be made, based on the output level and then a control signal is to be sent back to the weight control unit to modify the weight vector [8]. 1.2 Literature Survey The literature and the study of adaptive array geometry became the peak topic in last years, which, based on the development of smart antenna array. Here are some of the important researches reviewed as follows: 2 Chapter One General Introduction (Ioannides and Constantine A. Balanis, 2005)[14] examined adaptive beamforming with three different rectangular, circular and concentrated geometries consisting of seven arrays each with different radii and uniformly distrusted elements, and compared the results utilizing LMS and RLS algorithms, it was found that the circular array provides the strongest beamforming capabilities with directivity equal to 14.82 dB with LMS and 14.76 dB with RLS while, the concentrated circular show the deepest nulls against interference. (Mohammad and Zainol, 2006)[15] this paper introduced the “MINLMS”, the MI-NLMS combines the SMI and NLMS algorithms. Simulation results showed that the MI-NLMS algorithm provides remarkable improvements in terms of interference suppression, convergence rate and BER performance over that of LMS algorithms. With respect to LMS, MI-NLMS provides 15 dB improvements in interference suppression, 5 dB gain enhancement, and the reduction of the BER for MI-NLMS is 76% compared to LMS algorithm in the case of 4-element antenna array. ( K. R. Mahmoud, M. El-Adawy, S. M. M. Ibrahem, R. Bansal, K. R. Mahmoud, and S. H. Zainud-Deen, 2007)[16] search many configurations using 18 half-wave dipole elements in free space. The PSO algorithm has been used to optimize the complex excitations, amplitudes, and phases of the beamforming adaptive array components. This took the effects of mutual coupling between the elements totally into consideration. The distinction between circular and hexagonal array showed that hexagonal array geometries had the same beam width as circular array geometries, giving slightly deeper nulls which about -100 dB, while about -60 dB for circular array. (Atrouz, Alimohad, and Aissa, 2009)[17] established an adaptive beamforming algorithm applied a rectangular array to a two-dimensional elevation and azimuth angles. The algorithm developed was based on an algorithm from LMS, produce a new algorithm called the Sequential Block LMS algorithm (SBLMS), the simulation showed that this algorithm's efficiency to accomplish the Jammers suppress tasks was very good. ( C. S. Rani, P. V. Subbaiah, K. C. Reddy, and S. S. Rani, 2009)[18] various adaptive beamforming algorithms such as LMS and RLS algorithms used in smart antennas were addressed. And showing the LMS algorithm convergence rate depends on the array correlation matrix's Eigenvalues. (Zuniga, Erdogan, and Arslan, 2010)[19] assessed the efficacy of PSO in solving adaptive antenna array with rectangular antenna configuration and compared the results with conventional genetic algorithm (GA). The observation that the PSO obtained an average lower side lobe level (SLL) equal to -13 dB suitable to prevent interference, while -8 dB for Ga. 3 Chapter One General Introduction Additionally, the mean minimum power at all the nulls for the PSO algorithm is −35.7 dB whereas the power for the GA is −23.4 dB. These figures suggest that the PSO algorithm tends to perform better than the GA algorithm in terms of radiation power directed towards undesired signals. {Formatting Citation}[20] the performance of various circular array configurations for adaptive antennas was compared. Each design is made up of 19 elements, and the distance between the elements is 0.65. Such configurations include uniform circular arrays, uniform circular centered arrays, planar uniform circular arrays, and planar uniform hexagonal array. The PSO algorithm has been used here to determine the complex weights of the elements to modify the antenna according to the condition. (Amritpal Singh, Bhinder and Arvind Kumar.,2012) [21] presented planar arrays of rectangular micro-strip patches with a spacing of 0.5 between the elements. The outcome of the simulation showed that the main beam of the bigger array elements can get the signal of interest more reliably and rejects signal noise of interest but has a drawback that it increases the cost and complexity of implementation and therefore increases the convergence rate for the adaptive algorithm herewith minimize the bandwidth. (Maina, Langat and Kihato, 2013)[22] developed adaptive beamforming using for the null steering approach in a uniform rectangular array system. PSO algorithm and Simulated Annealing (SA) algorithm had to implement the beamforming procedure, to deliver optimal radiation in desired directions and minimum radiation in undesired directions, the weights of beamformer were synthesized. (Seyed Abolfazl Hosseini, 2013)[23] described a new FEDS algorithm in noise cancellation for speech enhancement, interference nulling, and system identification purposes. The authors showed that the FEDS converges to the true parameters and that its convergence rate is comparable to that of the RLS but, at a much lower computational cost. (Rao and Sarma, 2014)[24] considered three different geometries for the adaptive antenna. Evaluate also the convergence rate for the LMS algorithm, which would rely on the correlation matrix's Eigenvalues. It converges slowly in a dynamic channel setting with major Eigenvalues distributed. RLS Algorithm solves the problem, also a constant mudulos algorithm (CMA) algorithm is used when the reference signal is not available. (Venkatesan, Arunnehru and Umamaheswari, 2014)[25] the author used LMS which is easy to implement with low computation and RLS algorithm which usually converges faster than the LMS algorithm with less than 45 iterations but the price paid is added complexity. And show the rate of CGM was faster than that of the conventional algorithm with less than 20 iterations. 4 Chapter One General Introduction Reliance on these algorithms on the SNR algorithm and the iteration numbers required to obtain the desired signal, which minimizes the error faster than most adaptive algorithms were tested. (Vesa, Alexa, and Baltə, 2015)[26] given a brief overview of antenna array aspects. States that the structures are more clearly in the case of the 2D planar array and 3D array. Even, if the phase shift is applied between the current inserted in the antenna elements the linear array becomes unworkable in terms of the pattern of directivity. The author determines throughout the tests that increasing the number of elements over 7 may not contribute to a significant increase in directivity. (Patel, Makwana and Parmar, 2016)[27] presents adaptive algorithms LMS, SMI, and RLS model and evaluate the radiation pattern. Which demonstrates the LMS convergence rate function as one of the algorithm's weaknesses as it is directly dependent on the step size amount, and also provided the SMI algorithm that overcomes the LMS limits but raises the complex correlation matrix calculation. Finally learning the RLS algorithm where the array weight vectors are updated very fast since the convergence variance is defined by the information of the Eigenvalue of the signal correlation matrix. (Banerjee and Dwivedi, 2016)[28] PSO algorithm was implemented for a uniform linear array of 16 elements, output analyzes are performed using the specific variable tests. To confirm this method, the mean SLL, null depth, and first null beamwidth (FNBW), it has been shown that the PSO based former beam provides better SLL with direct main beam and null placement. {Formatting Citation}[29] established a PSO algorithm for evaluating and optimizing the element positioning and excitation amplitude, which included HPBW, SLL, directivity, and null steering in some spatial locations for the optimization goals output of this thesis. Those goals would be achieved with two fitness functions applying in the PSO algorithm. (S. C. Swati Patidar, Kishor Kumbhare,2017)[30] the genetic algorithm (GA) has been used in smart antenna to locate the beam pattern with optimum signal gain for the given direction. And by using this method the SLL was restricted and the direction of the angle array switched. (P. N. Chuku, Olwal, and Djouani, 2018)[31] examined smart antenna system based on RLS. The RLS algorithm's gain factor has been improved to boost its efficiency in terms of lower MSE resulting in a higher convergence rate. The improved RLS algorithm also holds weights that are identical to the calculated weights as compared to the standard RLS algorithm and the LMS algorithm. (Zhang et al., 2019)[32] this paper focuses on maximizing the performance of an array in transmitting and receiving data on how to optimize the positions 5 Chapter One General Introduction of antenna array elements using a contraction adaptive particle swarm optimization (CAPSO) algorithm. Two functions analyze the convergence rate and compare it with three other methods. The CAPSO shows reasonable performance for the various array element numbers. (Dakulagi and Alagirisamy, 2019)[33] describes the use of an innovative two-dimensional (2D) uniform linear array for adaptive algorithms. This new antenna design has been used to research the common beamforming algorithms, i.e. the least mean square algorithm (2D-LMS) and the least normalized mean square algorithm (2D-NLMS) and also suggested a stepsize variable NLMS (2D-VSSNLMS) algorithm. The suggested 2DVSSNLMS algorithm involves fewer beamforming iterations and high interference resistance. That allowed MIMO-WLAN, WI-MAX, 4 G LTE, low-power, low-cost anti-interference systems, and other advanced communication systems. (Dakulagi and Bakhar, 2020)[34] presented and comprehension many kinds of beamforming and DOA schemes for wireless communications. This work mostly indicates that the RLS algorithm utilizes the array correlation matrix Eigen spread to measure the weights, as the result of which accuracy and convergence rate of this scheme is poor and influenced by the Eigen spread value. The conjugate (orthogonal) for every new iteration for conjugate gradient method (CGM) computes the antenna array, by comparing it with LMS, SMI, and RLS schemes, the convergence rate of this approach is enhanced. Smart antenna using CGM also given improved system capability and can minimize the interference effect by creating very small beams in the looking position, as well as produced the newest LMS algorithm that shows a better convergence rate compared to any LMS algorithm version. ( M. U. Shahid, A. Rehman, M. Mukhtar, and M. Nauman, 2020)[35] a methodical evaluation of the performance of fixed beamforming algorithms for smart antennas such as Multiple Sidelobe Canceller (MSC), Maximum Signal-to interference ratio (MSIR) and minimum variance (MVDR) has been comprehensively presented in this paper. Simulation results show that beamforming is helpful in providing optimized response towards desired directions. And show that the MVDR beamformer provides the most optimal solution and more general application beamformer. 6 Chapter One General Introduction 1.3 Aims of Work The aims of this thesis could be summarized as the following points: o Test the performance of smart antenna system by applying nine different algorithms (least mean square LMS, normalized least mean square NLMS, recursive least mean square RLS, constant modules algorithm CMA, simple matrix inversion SMI, conjugate gradient method CGM, particle swarm optimization PSO, rabid direct Euclidian search REDS), using MATLAB program then looking for suitable one that leads to optimum weight to apply maximum power in the desired user direction and eliminates interference users by steering nulls in that direction. o Analyzing and choosing the most suitable array antenna configuration by made the comparison between three-antenna array configuration (linear array (1D), circular and planner array (2D) and cubic array (3D)) with most proper algorithm to achieve the goal of the work to track the sources of signal in two planes (azimuth and elevation) simultaneously. o Comparing the efficiency of the smart antenna system with traditional antennas by observing the effect on the beam pattern of varying element number and displacement spacing, and showing the relationship between changes on the beamwidth. 1.4 Contribution a. Applied nine algorithms on the antenna array system and comparing the performance result on antenna array between them, and calculate some of the important antenna characteristic like the gain, sidelobe level, beamwidth and finally bit error rate and signal to noise ratio. b. Proposed a new adaptive system design by using a particle swarm algorithm by producing a nonuniform array. c. Produce an enhancement and improvement on the conventional algorithm like least square algorithms or algorithms that has a step size factor by two methods which are variable and hybrid aspects. Also, improve the constant modules algorithm by two different aspects which are the variable and blocking method. d. Introduce a comparison between different array geometry performances on a rapid euclidian direction search algorithm, and apply two types of noise on it which is an AWGN and Rayleigh fading. 7 Chapter One General Introduction 1.5 Thesis Outline This thesis is organized into five chapters, established as follows: Chapter One: Presented a general presentation to the growth of the adaptive antenna array system and how the smart antenna could address some limitations factor for mobile communication systems and afforded a brief literature survey for relevant work in this field. Chapter Two: Provides theoretical history introduction to the smart antenna, beamforming adaptive array techniques, and structure for various geometries that would be used with the proposed algorithm, also produce a system measuring parameters. Chapter Three: where describes the implementation for proposed algorithms are derived and presented their mathematical modal. Chapter Four: provide the simulation results of the system module described in chapter three, which compares the quality of different adaptive algorithms and array geometry concerning convergence errors and resultant beam patterns using the MATLAB program. Chapter Five: require conclusions and recommendations for further work. 8 Chapter Two Theoretical Background for Adaptive Antenna Array System Chapter Two Theoretical Background for Adaptive Antenna System 2.1 Introduction An antenna array is a system consisting of a series of several antennas that are distributed by spacing distance to a specific given point, with phase and amplitude adjustments of the exciting currents in each of the antenna elements, the main beam and position nulls in either location could be scanned electronically [39]. An adaptive array managed by sophisticated signal processing[40], using a combination of signal processing algorithms, allows the adaptive system to efficiently locate and monitor various types of signals to dynamically reduce interference and optimize the intended reception, the different algorithms vary in the determination of certain weights. To get into an antenna and its performance and characteristics some parameters need to be summarized. The section below provides a brief introduction of an array feature. 2.2 Antenna Fundamental To get into an antenna and its performance, characteristics, and some parameters need to be defined. Parameter definitions will be given most of them based on internationally well-known Standards like the IEEE Standard Definitions of Terms for Antennas (IEEE Std 145-1993), and the IEEE Std 145-1983. 2.2.1 Radiation Pattern The pattern might be dependent on the feature representing the electrical or magnetic fields. Thus, when the amplitude or relative amplitude of a specific portion of the electrical field vector is graphically plotted, an amplitude pattern, field pattern, or voltage pattern is identified. When the amplitude square or relative amplitudes are plotted a typical two-dimensional power pattern is labeled, the field pattern plot for rectangular and polar coordinates can be shown in Figures 2.1. 9 Chapter Two Theoretical Background (a) (b) Figure 2.1 (a) Rectangular (b) Polar, Coordinates for Field Patterns [37] 2.2.2 Beamwidth: Which is determined from a radiation pattern's -3dB points that represent the position of the highest radiation-intensity lobe, the normalized pattern is -3dB points that equal to √1/2= 0.707 from the maximum point of the pattern. Figure 2.2 Main lobe of an antenna identifying the HPBW [37]. 2.2.3 Directivity The directivity represents the reflection of the capability of the antenna to systemically locate the energy in some direction and the ratio of radiation intensity in a given direction from the antenna to the average radiation intensity in all directions. The average radiation intensity is equal to the total power radiated by the antenna divided by 4π [38] 𝐷= 4𝜋𝑈𝑚𝑎𝑥 (2.1) 𝑃𝑟𝑎𝑑 2𝜋 𝑃𝑟𝑎𝑑 = ∫0 𝜋 ∫0 𝑈(𝜃, 𝜑) sin(𝜃) 𝑑𝜃 𝑑𝜑 (2.2) Where 𝑃𝑟𝑎𝑑 , is the total radiation power and 𝑈𝑚𝑎𝑥 represents the ultimate radiation intensity. 10 Chapter Two Theoretical Background 2.2.4 Gain Antenna's gain is an adjustment of the directivity which is the potential of the antenna to directing the energy in desired directions, in many other phrases is the ratio of the radiation intensity in a given direction to a radiation intensity that would have been obtained if the power received by the antenna were isotropically radiated, the antenna gain can express as[38] 𝐺(𝜃, 𝜑) = 𝑒𝐷(𝜃, 𝜑) (2.3) Where e presents the complete efficiency of antenna plus loss and mismatch impact. When an antenna has no dissipative loss then its gain is equal to its directivity under any specific direction. 2.3 Isotropic Antenna An isotropic antenna is a theoretical ideal antenna and a hypothetical lossless antenna may be described as an isotropic radiator, which radiates in all directions equally-horizontally and vertically by the same strength. Figure2.3 Isotropic Radiation Pattern The antenna has quite a gain of 0 dB in the sphere around it and a 100 % efficiency. The definition of an isotropic is ideal and yet physically realizable, it is also used as a guideline for presenting the actual antennas' directive properties [38]. 2.4 Arrays Antennas 2.4.1 Introduction The architecture of the array consists of two parts the first is the geometry of the array. Antenna arrays can be one, two, and threedimensional, the array geometry defines the physical locations of the antenna 11 Chapter Two Theoretical Background elements. The second aspect is array pattern in general, all elements of the array are considered to provide isotropic patterns, combined element patterns form the total array pattern which depends on the antenna array geometry and exciting current and radiation pattern thus acquired the array factor [39]. 2.5 Linear Arrays: The linear array is the most simple array geometry. In which the elements are aligned on a straight line and have a uniform or non-uniform inter-element distance space. 2.5.1 Array Element A single radiating element in a linear array or a convenient grouping of radiation elements that have fixed relative excitations, this is a single radiating antenna. 2.5.2 Two Element Array The array with two-element is the main basic and easiest array to evaluate. Figure 2.4 illustrates the array consisting of 2 identical polarized vertically aligned placed arrays in the direction of the y-axis with keeping a specific displacement (d) between them. Figure 2.4 Two Infinitesimal Elements [37] The complete radiation field by with two elements , which is the sum of the two, and would be offered within an x-y plane by[37] 𝐸𝑡 = 𝐸1 + 𝐸2 (2.4) 𝐸𝑡 = 𝑗𝑘𝜂𝐼0 𝑒 −𝑗 𝛿 2 𝐿 𝑠𝑖𝑛𝜃 4𝜋𝑟1 𝑒 −𝑗𝜅𝑟1 + 𝑗𝑘𝜂𝐼0 𝑒 +𝑗 𝛿 2 𝐿 𝑠𝑖𝑛𝜃 4𝜋𝑟2 𝑒 +𝑗𝜅𝑟2 (2.5) The field point is placed in r distance away from the reference with r >>d therefore assuming the vectors of the distance 𝑟1 , r, and 𝑟2 are parallel with each other. So the approximations could write as follow: 𝑑 𝑟1 ≈ 𝑟 − 𝑠𝑖𝑛𝜃 (2.6) 2 𝑑 𝑟2 ≈ 𝑟 + 𝑠𝑖𝑛𝜃 (2.7) 2 12 Chapter Two = 𝑗𝑘𝜂𝐼0 𝐿 𝑠𝑖𝑛𝜃 4𝜋𝑟 𝑒 −𝑗𝜅𝑟1 Theoretical Background [𝑒 −𝑗 (𝜅𝑑 𝑠𝑖𝑛𝜃+𝛿) 2 +𝑒 +𝑗 (𝜅𝑑 𝑠𝑖𝑛𝜃+𝛿) 2 ] (2.8) Where δ the electrical phase variation between the two adjacent elements (in radians), L is the dipole length,𝜃 is the angle at the z-axis in spherical coordinate, d is the spacing between elements. By further simplify equation (2.8) would get [37] 𝐸𝑡 = 𝑗𝑘𝜂𝐼0 𝐿 𝑒 −𝑗𝜅𝑟 4𝜋𝑟 𝑠𝑖𝑛 𝜅𝑑 𝑠𝑖𝑛𝜃+𝛿 (2 𝑐𝑜𝑠( 2 )) (2.9) From (2.9) it is obvious that the total array field is equal to the field product of a single element placed at the origin element factor (EF) by the array factor (AF) in general, the AF is a function of the number of elements, their geometrical arrangement, their relative magnitudes, their relative phases and their spacing. A quite accurate representation of the radiation in the array must also include the coupling impact between neighboring elements, whatever this subject is out of the scope of this thesis, more information of element coupling has been produced by Balanis [38]. 2.5.3 N-Element Uniform Linear Array The two-element array has been demonstrated so that the array of Nelements has also been implemented, which is the more general array. Figure 2.5 Indicates a linear N-element array assuming all elements are perfectly aligned and all elements had the same amplitudes but that each corresponding element has quite a progressive phase lead current excitation relative to the previous one (δ represents the phase by which the current in each element leads the current of the preceding element). Figure 2.5 N-Element Linear Array [40] The array factor is given in [40]. If the elements in the array were not isotropic sources, the complete field could be created by multiply the array factor of isotropic sources by the single element field AF = [1 + ej(kdsin θ+δ) + ej2(kdsin θ+δ) … . ej(N−1)(kdsin θ+δ) ]T 13 (2.10) Chapter Two Theoretical Background Where k is the phase constant, θ is arrival angle emitters. Equation (2.10) can be more concisely expressed by j(n−1)(kdsinθ+δ) j(n−1) AF = ∑N = ∑N ψ n=1 e n=1 e (2.11) Where ψ = kd sinθ + δ From equation (2. 10), the array vector can be derived as [40] 1 a̅(θ) = [ j(kdsin θ+δ) e ⋮ ] = [1 ej(kdsin θ+δ) … . ej(N−1)(kdsin θ+δ) ]T (2.12) ej(N−1)(kdsin θ+δ) The array vector was alternatively can be called: array steering vector array propagation vector, array response vector, or array manifold vector. 2.5.4 N-Element Non-Uniform Linear Array The symmetrical radiation pattern can be generated by the unsymmetrical placement of the antenna array. The symmetry situation, therefore, decreases the complexity of the computation. The linear array would consist of N equal radiating elements. The element is positioned asymmetrically around the origin on the x-axis. The schematic of the array structure presented in Figure 2.6, the N elements is assumed to be even number. The elements are subjected to two divides of M elements where N = 2M. The elements numbering are given as −𝑀, −𝑀 + 1 · · · − 1, 1, 2, . . . 𝑀 − 1, 𝑀 .The spaces of elements 1, 2, . . . . 𝑀 from the reference point are named as 𝑑1, 𝑑2, . . . 𝑑𝑀. Figure 2.6 N-Element Non-Uniform Linear Array [37] The symmetrical elements, −1, −2, . . . . −𝑀 have the same values of spacing distance. The array factor is specified by j(n−1)(k𝑑𝑛 sinθ+δ) AF = 2 ∑M (2.13) n=1 e From the schematic diagram, it is clear that the AF's amplitude and phase can be controlled in uniform arrays by carefully choosing the relative phase between the elements; in non-uniform arrays, the distance, amplitude, 14 Chapter Two Theoretical Background and phase could be utilized to monitor the composition and distribution of the total array element. 2.5.5 Another Array Geometry Array antennas can be one, two, and three-dimensional shapes, depending on the dimension of space one wants to access. Figure (2.7) shows a different array of geometries that can be applied in adaptive antennas applications. a)Circular antenna array in x-y c)Cubic antenna array in x-y-z plane b)rectangular antenna array in x-y plane Figure 2.7 Different Array Antenna Geometries [24] Figure (2.7a) shows two-dimensional circular array in the X-Y plane, with uniform angular distribution between elements of a value ∅𝑛 = 2𝜋(𝑛−1) 𝑁 where N represents the number of elements and n= 1,2,…, N . Because of its symmetry, this structure can produce beamforming in any direction and is more suitable for beamforming in one and two-dimensions. A twodimensional rectangular array with horizontal element spacing of ∆X and vertical element spacing of ∆Y is shown in Figure (2.7b). This type is used to perform two-dimensional beamforming (i.e. in both azimuth and elevation angles). Beamforming in an entire space, where there are all angles, requires some sort of cubic, cylindrical, conical, or spherical structure (threedimensional configuration) as shown in Figure (2.7c). This figure shows a 3D cubic structure with elements distributed in X-Y-Z planes. It’s a type of a conformal array antenna used for special applications [40]. 2.5.5.1 Circular Array The array factor or circular array in the x-y plane, with uniform angular distribution between elements of value ∅𝑛 = 15 2𝜋(𝑛−1) 𝑁 . The nth array element Chapter Two Theoretical Background is located at the radius (a) with the phase angle𝜑𝑛 . The AF can be found in a similar procedure as was calculated with the LA as [9] −𝑗(𝑘𝑎𝜌̂ .𝑟̂ +𝛿𝑛 ) −𝑗(𝑘𝑎𝑠𝑖𝑛𝜃cos (𝜑−𝜑𝑛 )+𝛿𝑛 ) 𝐴𝐹 = ∑𝑁 = ∑𝑁 𝑛=1 𝑤𝑛 𝑒 𝑛=1 𝑤𝑛 𝑒 (2.14) 2.5.5.2 Planar Array This type is used to perform 2D beamforming (in both azimuth and elevation angles) with horizontal element spacing of ∆X and vertical element spacing of ∆Y. The AF for PA can be expressed as combining the AF of two LAs [10]. Pattern multiplication can be used to find the pattern of the entire M×N element array. Using pattern multiplication would have 𝐴𝐹 = 𝐴𝐹𝑥 . 𝐴𝐹𝑦 = 𝑗(𝑛−1)(𝑘𝑑𝑦 𝑠𝑖𝑛𝜃sin𝜑+𝛽𝑦 ) 𝑗(𝑚−1)(𝑘𝑑𝑥 𝑠𝑖𝑛𝜃cos𝜑+𝛽𝑥 ) ∑𝑁 ∑𝑀 𝑚=1 𝑎𝑚 𝑒 𝑛=1 𝑏𝑛 𝑒 (2.15) 𝐴𝐹𝑥𝑦 = 𝑗[(𝑚−1)(𝑘𝑑𝑥 𝑠𝑖𝑛𝜃cos𝜑+𝛽𝑥 )+(𝑛−1)(𝑘𝑑𝑦 𝑠𝑖𝑛𝜃sin𝜑+𝛽𝑦 )] 𝑁 ∑𝑀 𝑚=1 ∑𝑛=1 𝑤𝑚𝑛 𝑒 (2.16) The AF for cube array can be expressed as combining the AF of three LAs, Pattern multiplication can be used to find the pattern of the entire M×N×Z elements array as follow 𝐴𝐹𝑥𝑦𝑧 = 𝑂 𝑗[(𝑚−1)(𝑘𝑑𝑥 𝑠𝑖𝑛𝜃cos𝜑+𝛽𝑥 )+(𝑛−1)(𝑘𝑑𝑦 𝑠𝑖𝑛𝜃sin𝜑+𝛽𝑦 )+(𝑜−1)(𝑘𝑑𝑧 𝑠𝑖𝑛𝜃+𝛽𝑧 )] 𝑁 ∑𝑀 𝑚=1 ∑𝑛=1 ∑o=1 𝑤𝑚𝑛𝑜 𝑒 (2.16) 2.6 Smart Antenna 2.6.1 Introduction A smart antenna pattern as mention earlier controlled by algorithms according to certain requirements. These requirements may be maximizing or enhance received the signal to interference ratio (SINR) and may also be considered as forming beams for transmission, reducing the variance, elimination for the mean square error (MSE), tracking the desired signals and interfering signals extinctions. Smart antenna involves the digitizing of the array outputs by using an A / D converter, such digitization would be completed by either at IF frequencies or at baseband [40] Smart antennas offer the potential for advanced radar systems, enhanced mobile wireless network capabilities, and amended wireless communication systems by mean of the enforcement of space division multiple access (SDMA). 16 Chapter Two Theoretical Background 2.6.2 Type of Smart Antenna Smart antenna systems may be classified into the following three types, as describes in Figure 2.8. Figure 2.8 Different Smart Antenna Concepts The following points present a distinction between the three sorts of smart antenna systems [38]. 1. Switched Beam Antennas: Switched beam also can be called a switched lobe antennas which is the extension of the cellular sectorization system that has a collection of several predefined patterns with it the cell is divided into three sectors with macro-sectors of 120 degrees each. The switched beam method more partition macro-sectors into multiple micro-sectors thereby enhancing extent field and efficiency. First, the signal intensity is measured, then one of these established fixed beams is chosen, if the mobile phone travels through the field, the device switches from one beam to another beam [38]. A few more gains are obtained due to the higher directivity comparison with traditional antennas. Such an antenna is simpler to incorporate than the more complex adaptive arrays in existing cell structures but it offers a small device improvement. 2. Dynamically-Phased Arrays: In the previous status of a switched beam antenna the beams are determined and fixed. A consumer perhaps within the zone of one beam at a given time, however as travels the source of the beam away and passes the beam's circumstance, the obtained signal would become weakened and an intra-cell transfer exists. However, by the dynamically phased array, 17 Chapter Two Theoretical Background a direction of arrival (DOA) algorithm monitors the wanted signal as the traverses inside the beam range that follows. 3. Adaptive Antenna Arrays: A going to be lots smartest is adaptive antenna. Adaptive array systems can locate and monitor consumer (SOI) and interferer (SNOI) signals and dynamically change antenna pattern, this structure also can know as the adaptive beamforming or the digital beamforming. Figure 2.10 Shows a functional device block that transforms and digitizes down the received signals to the baseband, then set the SOI used the direction-of-arrival (DOA) algorithm and simultaneously detects the SOI and SNOI[38]. 2.7 Array Weighting The appearance of sidelobes means the array radiates energy in an unwanted direction. Furthermore, the array absorbs energy from unexpected directions due to mutuality. This is the basis on which communications encountered fading. Weighting, shading, or windowing of the array elements would eliminate the side lobes. There are various applications for array weighting in areas like digital signal processing (DSP), radio astronomy, radar, sonar, and communication. Figures 2.9(a) & (b) presents an even and an odd a symmetric linear array elements M. (a) (b) Figure 2.9 (a) Even Array and, (b) Odd Array with Weights [40] The distant field from an array for equivalent elements may be subdivided into the element factor (EF) by the array factor (AF) products [40]. In general, the array factor is found by summing the weighted outputs of every element. The array factor for the weighted even and odd element arrays can be found as: 𝐴𝐹𝑒𝑣𝑒𝑛 = 𝑤𝑀 𝑒 −𝑗 ⋯ + 𝑤𝑀 𝑒 (2𝑀−1) 𝑘𝑑 2 (2𝑀−1) 𝑗 𝑘𝑑 2 𝑠𝑖𝑛𝜃 1 + ⋯ + 𝑤1 𝑒 −𝑗2𝑘𝑑 𝑠𝑖𝑛𝜃 + 𝑤𝑀 𝑒 𝑗 𝑠𝑖𝑛𝜃 (2𝑀−1) 𝑘𝑑 2 𝑠𝑖𝑛𝜃 + (2.17) 18 Chapter Two Theoretical Background Here 2M = N, equal to the total element numbers. The weights 𝑤𝑛 would choose to meet the requirements. Symmetrical scalar weights, therefore, may be employed to form side lobes; to reformulating the even array factor, Euler’s identity for the cosine would utilize such as: 𝐴𝐹𝑒𝑣𝑒𝑛 = 2 ∑𝑀 𝑛=1 𝑤𝑛 cos ( (2𝑛−1) 2 𝑘𝑑 𝑠𝑖𝑛𝜃) (2.18) The 2 was being omitted from the statement in equation (2.18). with no loss of commonality to produce a quasi-normalization 𝐴𝐹𝑒𝑣𝑒𝑛 = 2 ∑𝑀 (2.19) 𝑛=1 𝑤𝑛 cos ((2𝑛 − 1)𝑢) Where 𝑢 = 𝜋𝑑 𝜆 𝑠𝑖𝑛𝜃 If the statement is zero, the array factor is maximum, inclusion θ = 0. Figure 2.11(b) presents the odd array, to have a quasi-normalized odd array factor, it will also summing up each exponential inputs in the array. 𝐴𝐹𝑜𝑑𝑑 = ∑𝑀+1 𝑛=1 𝑤𝑛 cos (2(𝑛 − 1)𝑢) (2.20) Likewise; the array factor could be written in terminology of vector like AF = w ̅ T . a̅(θ) (2.21) Where w ̅ T = [wM , wM−1 … w1… wM−1 , wM ], and a̅(θ) Is the steering vector that was described in equation (2.21) 2.8 Angle of Arrival (AOA) Estimation: The estimation of the angle of arrival (AOA) was also called as spectral estimation, the direction of arrival (DOA) or bearing of the impinging waves was used for the synthesis of beams steered at the desired signal with nulls guided to another signal interference [41]. Through more precisely detecting the arrival angles (AOA) smart antennas have been used to improve direction finding (DF) techniques [2]. 2.9 Beamforming Techniques: Beamforming is the technique used to describe the application of weights to the inputs of an array of antennas to focus the reception of the 19 Chapter Two Theoretical Background antenna array in a certain direction. Beamforming presents several advantages to antenna design:1) Space Division Multiple Access (SDMA) is the main aim in the advancement of cellular radio systems as it ensures that more than one user can be assigned to the same frequency in the same cell [41]. Because the former beam is attained, it can direct its focus toward a certain signal the other signals from multiple directions can reuse the same carrier frequency. 2) Since the beamformer is directed in a specific direction, the sensitivity of the antenna can be increased, particularly when receiving weak signals, for a better signal-to-noise ratio. 3) Signal interference is reduced due to the minimum pattern gain oriented toward undesired signal angles [42]. 2.9.1 Fixed Weight Beamforming: The fixed weight of M element array and N input signals beamformer, as shown in Figure 2.10, is a smart antenna in which fixed weight is used to study the signal arriving from a specific direction. Figure 2.10 Block Diagram of Fixed Weight Beamformer [2] In the fixed weight beamforming approach, the arrival angles do not change with time, so the optimum weight needn’t be adjusted [2]. 2.9.2 Adaptive Beamforming The key benefit of digital beamforming was that phase shifting and array weighting can be achieved on the digitized data instead of the hardware implementation [40]. Different methods of implementing an adaptive algorithm could be used, the following are common adaptive algorithms that are of special interest to this thesis. 20 Chapter Two Theoretical Background 2.10 General control algorithm for adaptive antenna array There are several adaptive algorithms used for the adaptive antenna system, they are typically characterized in terms of their convergence properties and computational complexity [43]. Adaptive algorithms can be classified as non-blind adaptive, blind adaptive and optimized algorithms as shown in Figure 2.11. Figure 2.11 Classification of Adaptive Array Algorithms 1. Non-blind adaptive algorithms - Such algorithms employ a guide signal to change the array weights iteratively, thus that weight output is compared with the target signal at the end of each iteration and the produced error signal are exercised in these algorithms to alter the weights. 2. Blind adaptive algorithms - These algorithms don't use the guide signal, and therefore no array weight adjustment is required. 3. Optimization algorithm - The purpose of optimization is to allows comparison of the different choices to achieve the “best” design relative to a set of prioritized criteria or constraints. These include maximizing some function such as productivity, strength, reliability, longevity, efficiency, and utilization . 2.10.1 Least Mean Square (LMS) Algorithm The weight vector w ̅ = [w1 w2 … . wM ]T must be revised to minify the error while iterating the adaptive weights. The desired signal d(𝑗) and the interferers I1 (j), I2 (j), … IN (i) which is received by each antenna array elements (M), times are noted by the jth time samples [37]. 21 Chapter Two Theoretical Background e(j) = d(j) − w ̅ H (j) x̅(j) (2.22) The squared error is [37] |e(j)|2 = |d(j) − w ̅ H (j) x̅(j)| 2 (2.23) The cost function is given as J(w ̅) = D − 2 w ̅ H r̅ + w ̅ H ̅Rxx w ̅ (2.24) Where 𝐷 = 𝐸[|d|2 ] (2.35) Taken the gradient of equation (2.35) and equate it to zero to reducing the cost function. ̅ xx −1 r̅ w ̅ opt = R (2.26) The prompt estimates of these values are provided as 𝑅𝑥𝑥 ≈ 𝑥̅ (𝑗)𝑥̅ 𝐻 (𝑗) (2.27) 𝑟(𝑗) ≈ 𝑑 ∗ (𝑗)𝑥̅ 𝐻 (𝑗) (2.28) After the cost function gradient, the LMS weight vector would be [37] w ̅ (j + 1) = w ̅ (j) − μ [𝑅𝑥𝑥 𝑤 ̅ − r] =w ̅ (j) + μ e∗ (j) x̅(j) (2.29) As appear in equation (2.29), the convergence of the LMS algorithm is the proportional indirect way with step size factor μ [38]. So ought to select step size within a range which compensates for convergence rate, such as 0≤μ≤ 1 (2.30) 2λmax ̅ xx. If there is only one Where λmax is the largest eigenvalue of R desired signal and the rest interfering signals are noise, equation (2.10) could be approximateas 0≤μ≤ 1 (2.31) 2𝑡𝑟𝑎𝑐𝑒[𝑅̂𝑥𝑥 ] Three main factors guided the response of the LMS algorithm which is, μ, number of weights, and the Eigen-value for the input vector of the correlation matrix. Many iterations ought to before satisfying to achieving the convergence rate, that drawback of the conventional LMS adaptive because of using fixed μ. 22 Chapter Two Theoretical Background 2.10.2 Sample Matrix Inversion (SMI) Algorithm One of the limitations of the LMS adaptive structure will be that the algorithm should have to go through numerous iterations until effective convergence is achieved the eigenvalue spread of the array correlation matrix . The Sample Matrix Inversion (SMI (also known as direct matrix inverse (DMI))) algorithm is the fastest and one of the simplest adaptive algorithms compared to the LMS algorithm and had the minimum MSE algorithms, where the optimal wiener solution offers the optimal array weights. SMI is a time average estimate of an array correlation matrix using K-time samples [49, 50], and the K–length block of data was employ, so this method named a block-adaptive algorithm adapted the weight block by block. The correlation matrix present by determined the time average as [13] ̂ xx(k) = 1 ∑𝐾 ̅ ̅H R (2.32) 1 X K (k) X K (k) K Where K is the observation interval. The correlation vector r̅ can be obtained by 1 ̅∗ ̅ r̂(k) = ∑𝐾 (2.33) 1 d (k) XK (k) K Vector X is given by M×K matrix such as [7] x̅1 (1 + kK) ⋯ x̅1 (K + kK) ̅(k) = [ ⋮ ⋱ ⋮ X ] x̅M (1 + kK) ⋯ x̅m (K + kK) (2.34) Where k is the block number and K is the block-length. Then the estimate of the array correlation matrix is ̂ xx(k) = R 1 K ̅ XK (k) ̅ XKH (k) (2.35) Also, the desired signal vector can be defined as d̅(k) = [d(1 + kK) d(2 + kK) … . d(K + kK)] (2.36) The results of correlation vector is ̂r(k) = 1 K d̅∗ (k) ̅ XK (k) (2.37) Measure the SMI weights for kth block of length K as ̅ xx −1 (k)r̅(k) w ̅ SMI = R ̅ K (k) ̅ = [X XKH (k)]−1 d̅∗ (k) ̅ XK (k) 23 (2.38) Chapter Two Theoretical Background The benefit of the approach is that the convergence rate does not depend on the signal level. However it has several drawbacks and the two principle troubles that related with matrix inversion which is the high computational complexity can't be overcome facilely through using of integrated circuits so the correlation matrix may be ill-conditioned resulting in errors or singularities when inverted, and the exercise of finite-precision arithmetic and the necessity of inverting a large matrix where to invest. 2.10.3 Recursive Least Square (RLS) Algorithm: As noted previously that the SMI procedure does have many disadvantages, the computing time and theoretical singularities can trigger problems, even if SMI faster than the LMS algorithm. The necessary correlation matrix and the correct correlation vector could be recursively calculated the weight vector w ̅ is chosen in the least-squares approach to reduce the cost function consisting of several error squares over some time. No matrix inversion computations are needed for the RLS algorithm, as the inverse correlation matrix would be direct measuring. The following equations will characterize RLS [49, 50] ̂ −1 Through product R ̂(k) canceling the division by K, so the xx (k) r matrix of correlation and the correlation vector removing K has been rewritten as ̂ xx(k) = ∑𝐾 ̅ ̅H R (2.39) 1 X K (k) XK (k) 𝐾 ∗ r̂(k) = ∑1 d̅ (k) ̅ XK (k) (2.40) ̂ −1 The block length presents by k and last time sample k and R ̂(k) xx (k), r is the predictions of the correlation stopping at k sample of time [37] ̂ xx (k) = α R ̂ xx (k − 1) + x̅(k) x̅ H (k) R (2.41) −1 ̂ −1 (k ̂ −1 ̂−1 R R xx − 1) − α−1 g̅(k) x̅ H (k) R (2.42) xx (k) = α xx (k − 1) ̂ −1 g̅(k) = R ̅ (k) xx (k) x (2.43) w ̅ (k) = w ̅ (k − 1) + g̅(k)[d̅H (k) − x̅ H (k)w ̅(k − 1)] (2.44) ̂ xx (k) would be setting to [37] The initial evaluate value of R ̂ xx (0) = x̅(1) x̅ H (1) + R ̅ nn R (2.45) ̂ nn = N x K zero-mean Gaussian noise R 24 Chapter Two Theoretical Background This efficiency enhancement is done at the cost of greater computational complexity [39]. 2.10.4 Conjugate Gradient Method (CGM) Algorithm The Conjugate Gradient Method is an effective method for symmetric positive definite systems. For addressing a system of linear equations, the CGM algorithm aims to search iteratively for the objective functions by selecting conjugate (perpendicular) paths at each new iteration. Conjugacy background is meant to mean orthogonal, in certain terms the CGM method generates orthogonal search directions which lead in the quickest convergence [45]. The CGM is aiming to minimize the quadratic cost function 1 H J(w ̅) = w ̅ ̅A w ̅ − d̅H w ̅ (2.46) 2 x̅1 (1) ⋯ x̅M 1) ̅= [ ⋮ ⋱ ⋮ ] Where, A x̅1 (K) ⋯ x̅M (K) ̅ is the K x M matrix of array snapshots, (M, K equals to numbers of the A element in an array and the number of snapshots, respectively) d̅ = [d(1) d(2) … . d(K)]T represents the desired vector of K snapshots Mathematically, simple CGM algorithm can be summarized as [40] ̅ (k) = The general weight update w ̅ (k + 1) = w ̅ (k) + µCGM (k)D µCGM (k) = r̅H (k) ̅ A̅ AH r̅(k) = ̅D ̅H ̅ ̅ (k) D AH A step-size of CGM (2.47) (2.48) ̅ (k) = the updates for the residuals r̅(k + 1) = r̅(k) + µ(k)A̅ D (2.49) ̅ (k + 1) = ̅AH r̅(k + 1) − α(k)D ̅ (k) = the direction vector update D (2.50) The linear search is used to determine α (n) which reduces 𝑗(w ̅ (n)) [40] α(k) = r̅H (k+1) ̅ A̅ AH r̅(k+1) r̅H (k) ̅ A̅ AH r̅(k) (2.51) Thus, locating the residual and the corresponding weights and updating till achieved the convergence rate is the technique to use CGM. 2.10.5 Particles Swarm Optimization Algorithm Particles swarm optimization PSO is often used to solve prosperous complex multidimensional optimization problems in a different domain, such 25 Chapter Two Theoretical Background as antenna design and device modeling, 𝑒𝑡𝑐 [46].PSO algorithm can be applied to enhance the adaptive antenna radiation pattern in every iteration. The converged value can be achieved by a different parameter such as controlling the amplitude, phase, position and complex weights (Includes amplitude and phase control, this paper deals with complex weight (amplitude and phase) and also position control. The representation of (PSO) algorithm is based on the following steps below to find the optimal radiation pattern of an adaptive antenna system [47]: Stage 1: Parameters initializations that require to be optimized and give them an appropriate range like initialize population, number of iterations, regulating parameters (φ1andφ2) and weights (w) Stage 2: Initialize random swarm position and velocities of the 𝑘𝑡ℎ variable which becomes each particle's respective individual best and then selected, the first global best among these initial positions. Stage 3: Compute particle fitness 𝐹𝐹(𝑖, 𝑘)which is assigned to the present locations. Stage 4: Update the personage best and global best by comparing their fitness value at the current position with the best fitness value which has ever gating at each time 𝑝𝑖𝑑 (𝑖, 𝑘) = 𝐹𝐹(𝑖, 𝑘)and also, the global best can be defined by𝑝𝑔𝑑 (𝑖)= max (𝑝𝑖𝑑 (𝑖, 𝑘)) which it's the best position through all of the personage best positions. Stage 5: Update velocity and position using 𝑣𝑖𝑑 (𝑖 + 1, 𝑘) = 𝑤 ∗ 𝑣𝑖𝑑 (𝑖, 𝑘) + 𝐶1 ∗ 𝑅𝑎𝑛𝑑( . ) ∗ (𝑝𝑖𝑑 (𝑖, 𝑘) − 𝑥𝑖𝑑 (𝑖, 𝑘)) + 𝐶2 ∗ 𝑅𝑎𝑛𝑑( . ) ∗ (𝑝𝑔𝑑 − 𝑥𝑖𝑑 (𝑖, 𝑘)) (2.52) 𝑥𝑖𝑑 (𝑖 + 1, 𝑘) = 𝑥𝑖𝑑 (𝑖, 1) + 𝑣𝑖𝑑 (𝑖 + 1, 𝑘) (2.53) Stage 6: Update fitness function to the𝐵𝐹(𝑖 + 1, 𝑘) Stage 7: If fitness function for 𝐵𝐹(𝑖 + 1, 𝑘) > 𝑓𝑖𝑡𝑛𝑒𝑠𝑠 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 for 𝐵𝐹(𝑖, 𝑘)then 𝑝𝑖𝑑 (𝑖, 𝑘) = fitness function for 𝐵𝐹(𝑖 + 1, 𝑘) Stage 8: Update 𝑝𝑔𝑑 (𝑖, 𝑘) = max (𝑝𝑖𝑑 (𝑖 + 1, 𝑘)). Stage 9: Termination criteria. If 𝑖 < 𝑖𝑚𝑎𝑥 then boost𝑖 and go to step 5. The number of iterations reaches the maximum permissible limit, else stop. 2.11 Benefits and Features of Smart Antennas The operating benefits for the network operator can be summarized as follows by accessing the spatial domain through smart antenna systems [44]: 26 Chapter Two Theoretical Background • Capacity improvement: The central cause for developing an interest in smart systems was the increased capability. Smart antennas allow multiple cell users to use the same frequency without interfering with one another because the smart antenna beams of base station are split to hold different users in separate beams at the same frequency. Coverage extension. Smart antennas should concentrate its energies on expected users, in-state wasting energy in unnecessary directions like a traditional antenna, which is as well as indicate that fewer base stations may be required to serve a wide range of areas and longer battery life in mobile stations and high data rates could be enabled. • Locating users and due to the nature of spatial detection of the smart system, the DOA algorithms are also used to locate human beings in an emergency for any location-specific operation. • Security is another advantage of smart-antenna systems. Security is a significant problem in a society that is more focused on doing business and forwarding on personal information and made it intractable to tap the connection, as the attacker should be located in the same orientation as the consumer from a base station to tap the connection efficiently. The features of a smart antenna are [2] • Signal gain: Multiple antenna inputs are coupled to maximize the available power necessary to create a given coverage rate. • Interference rejection: Antenna pattern can be produced towards points of interference, improving the signal-to-interference ratio of the received signals. This eliminates interference said by the base station on the inverting link or uplink. That also would limit the amount of downlink or forward link interference distributed in the network. These carrier enhancements to the interference ratio provide an increase in efficiency. • Spatial diversity: complex array information was utilized to reduce fading and other unwanted multipath propagation impacts. • Power efficiency: join inputs with many elements to maximize the processing gain available in the downlink (against customers). 2.12 Drawbacks of Smart Antenna Even though the advantages of using smart antennas are considered quite so many, some significant limitations still occur [39]. 27 Chapter Two Theoretical Background • A smart transceiver antenna is more complicated than a conventional transceiver to the base station. For each of the array antenna elements, separate transceiver chains are required and correct real-time calibration of each is necessary. • Adaptive beamforming and DOA is a computationally demanding process therefore, quite powerful numeric processors and control systems should be included in the smart antenna base station. • Smart antenna base stations would probably cost a lot more than traditional base stations. 2.13 Description and Modeling of Wireless Channel To analyze the efficiency of a smart antenna system, specific knowledge of the channels and the parameters of channel parameters is substantial, since the propagation channel is the main benefactor of difficulties and drawbacks which are an effect on the mobile radio systems. In many forms, the physical channel influences the transmission of radio signals on both the straight-ahead (from the base station into mobile) and inverted (from the mobile into the base station) connections. The direct path with the multipath channel is indicated in Figure 2.12 [40]. Figure 2.12 Multipath with a Direct Path Channel. Every propagation path has its own power, time delay, and angle of arrival. The signal received is abundant poorer than the signal transmitted on account of factors like the loss of mean propagation, gradual fading, and fast fading. The mean loss of propagation derives from square-law spreading, water and foliage absorption, and ground-reflections impact, mean loss of propagation depends on the range and variations quietly slow also for quick mobiles. Fading may also be described as flat or selective fading of frequencies [40]. The following two central categories could be classified 28 Chapter Two Theoretical Background into parameterized of the physical model applied to describe the wireless channel between the mobile station and base station [13]:Firstly: Temporal parameters that include the path loss, Shadowing, Multipath fading, Power spectral density, and Power delay profile. Secondly: Spatial parameters that include the angle of arrival and mobile station mobility. Such parameters are modeled in detail in the coming sub-sections. 2.13.1 Path loss: It is such an average reduction in signal strength of path length between certain transmitters and the receiver. Studies illustrate that the average signal power received reduces exponentially together with distance [39]. 2.13.2 Shadowing: Shadowing or slow fading due to changes in propagation factors related to buildings, pathways, trees, and other obstacles in a specific region, which reflects the difference in the average path loss in a local area [39]. 2.13.3 Multipath Fading This form of fading is comparatively quick and thus accountable for shortage signal fluctuations [13]. That's induced due to the constructive and destructive combination which arbitrarily slowed, reflected, scattered, and the diffracted subpath signal sources through distances of the range for the little wavelengths. Many signal replicas arrived at the receiver, having traversed several propagation paths, combining constructively and destructively. 2.14.4 Rayleigh Model The Rayleigh distribution had to paradigm the multi-path fading in non-line of sight (NLOS) fading conditions without direct line of sight (LOS) direction between the mobile station and base station [13] 2.13.5 Rice model The Rice distribution concepts the propagation paths composed of one direct LOS consequence and several randomly weaker NLOS consequences [13]. The Rayleigh (non-line-of-sight) fading models are considered in this study, which is widely was using to define the statistical time-varying form of the personage multipath envelope. 29 Chapter Two Theoretical Background 2.13.6 Power Spectral Density The shift in the frequency of the received signal can be apparent when there is proportional movements occur between the transmitter and receiver, because of the Doppler shift [13]. The Doppler shift is different for every sub-path as it depends on the Angle of Departure of the subpath related to the direction of movement of the mobile station. The Power Spectral Density (PSD) produced by the sum of all the scattered and reflected sub-paths, which cause a continuous spectrum of Doppler frequencies. The inverse Fourier transform of the PSD is the autocorrelation function of the fading signal [2]. Depending on the particular propagation environment and the underlying communication scenario different PSDs have been proposed, [48]. A popular selection for land mobile communications that indicates in this research is the Jake PSD [2]. For the Rayleigh fading case, the normalized Jakes PSD and the coincide autocorrelation function would be defined by [13] S(f) = α2 f π fD √1−(f )2 |f| ≤ fd (2.54) d R(τ) = j0 (2π fD τ) (2.55) Where j0 (. ) is the zero-order Bessel function of the first kind and fD is the maximum Doppler shift that the signal undergoes, given by[48] v v fc λ c fD = = (2.56) Where v is the user speed (in m/s), λ is the wavelength of the transmitted signal, c is the velocity of light and fc is the carrier frequency. In this thesis, the maximum Doppler frequency used is 117 Hz. This corresponds to a speed of v = 140 km/hr for fc = 900 MHZ. 2.13.7 Power Delay Profile (PDP) There is more than one propagation path for each transmitter and receiver in several mobile channels and a received signal composed of two or more different elements and each moves a separate path from the transmitter. Each multipath aspect arrives with a delay depending on the length of the path [41]. Based on the sort of environment (indoor or outdoor) as well as the common conditions for propagation, the PDP model can take various forms. In this work, for simple, the time delays relay with the variable 30 Chapter Two Theoretical Background resolvable multipath would be independent of the Angles of Arrival (AOA’s) [13]. 2.13.8 Mobile Station Mobility Model Modeling the mobility of mobile stations is a central topic of spacetime channel simulations. MS motion affects the properties of both spatial and temporal channels. The smart antenna should monitor and steer its beam towards the desired target in the situation of adaptive beamforming. Hence, the efficiency of a smart antenna could not be realistically measured without mobile station mobility simulations. The subsequent criteria have been used to define the impact of MS mobility on execution efficiency [13]:• With mobile moving for a particular Monte Carlo simulation run, numbers of sub-path and multipath components should behold constant. • As mobile motions, the sub-path parameters (departure angles, random phases, and angular spread) are held constant. 31 Chapter Three Proposal Algorithm Analysis Chapter Three Proposed Algorithm Analysis 3.1 Introduction Adaptive algorithms have been used either in block mode or in iterative mode to adjust the weight vector 𝑤(𝑛). Block processing techniques measure a new solution using the statistical estimates obtained out of the latest data block. In iterative algorithms, the current weight vector is adjusted by an incremental amount to form a new weight vector that approximates the optimal solution. There is a certain variance when comparing as an example the algorithm Least Mean Square(LMS) which is largely used by many researcher because of the low complexity of computational and ease of execution. The least-squares algorithms (an algorithm that content step size factor), like Recursive Least Squares (RLS), Conjugate Gradient (CG), and Rabid Euclidean Direction Search (REDS), would converge more rapidly and also have smaller mean square error (MSE) versus LMS. Nevertheless, their complexity enables them to intrinsically unfit to real-time applications. In this chapter, different tests of common antenna array geometry like linear, circular, planar, and cube antenna array comparing for various antenna array performance also, overall system performance had been inspected along with changing in some system parameters. Figure 3.1 illustrates the flowchart of implementation for different antenna array geometry and different algorithms, it should be noted that the flowchart bellow gave a prime vision of smart antenna algorithm function. Throughout this chapter, several developments on adaptive beamforming algorithms will be presented to enhance the achievement of the smart antenna in terms of rate convergence, interference suppression capabilities, and the tracking capabilities of the desired signal. Such techniques use both the block adaptive and sampled methods. 32 Chapter Three Proposal Algorithm Analysis Figure3.1. Flowchart for a conventional smart antenna. 3.2 The Implementation of Normalizing LMS Algorithm with Maximum SIR Factor The benefit of this process is converges better than the traditional process of adaptation. The proposed adaptive beamforming algorithm is LMS normalize by maximum SIR would be introduced here, the maximization of the SIR is based on one principle that could apply to maximize the received signal and eliminate the signals of interferences. It is obvious that by putting nulls at their angles of arrival, the cancelation of all interference would automatically optimize the SIR, to boost the performance of adaptive 33 Chapter Three Proposal Algorithm Analysis algorithms along with convergence speed, weight stability, and interference suppression. The SIR strategies were used to calculate the optimum vectors of weights allocated for each element in an array by considering a fixed beam pattern instead of random value before computing the ultimate weight vectors by the least square algorithm. The weight coefficients extracted from standardization were set as initial coefficients and then updated by the LMS algorithm to improve device stability and convergence speed. The proposed optimum updated weight vectors of combining the two aspects of 𝑆𝐼𝑅𝑚𝑎𝑥 and LMS algorithm introduces according to the following equations: The desired signal weighted array output power is provided as [37] 𝜎𝑠 2 = 𝐸[|𝑤 ̅ 𝐻 . 𝑥̅𝑠 |2 ]= 𝑤 ̅ 𝐻 . 𝑅̅𝑠𝑠 . 𝑤 ̅ (3.1) 𝑅̅𝑠𝑠 = 𝐸[𝑥̅𝑠 𝑥̅𝑠 𝐻 ] presents the signal correlation matrix The power of undesired signal weighted array output is provided as [37] 𝜎𝑢 2 = 𝐸[|𝑤 ̅ 𝐻 . 𝑥̅𝑢 |2 ]= 𝑤 ̅ 𝐻 . 𝑅̅𝑢𝑢 . 𝑤 ̅ (3.2) 𝑅̅𝑢𝑢 = 𝑅̅𝑖𝑖 + 𝑅̅𝑛𝑛 (3.3) Where 𝑅̅𝑖𝑖 , 𝑅̅𝑛𝑛 presents the correlation matrix to the interferers and noise. The SIR described the proportion of the desired signal power split by undesired signal power. 𝑆𝐼𝑅 = 𝜎𝑠 2 𝜎𝑢 2 = ̅ 𝐻 .𝑅̅𝑠𝑠 .𝑤 ̅ 𝑤 𝐻 ̅ ̅ .𝑅𝑢𝑢 .𝑤 ̅ 𝑤 (3.4) With applying the derivative relative to 𝑤 ̅ and returns the impact to zero. The procedure of optimization was given in Harrington. Rearranging terms, the relationship derives as follow −1 𝑅̅𝑢𝑢 𝑅̅𝑠𝑠 . 𝑤 ̅ = 𝑆𝐼𝑅. 𝑤 ̅ (3.5) 𝑆𝐼𝑅𝑚𝑎𝑥 is equal to the largest eigenvalue 𝜆𝑚𝑎𝑥 for the 𝐻𝑒𝑟𝑚𝑖𝑡𝑖𝑎𝑛 matrix 𝑅̅𝑢𝑢 , 𝑅̅𝑠𝑠 . The eigenvector associated with the largest eigenvalue is the optimum weight vector̅̅̅ 𝑤𝑜𝑝𝑡 .Thus −1 𝑅̅𝑢𝑢 𝑅̅𝑠𝑠 . 𝑤 ̅𝑆𝐼𝑅 = 𝜆𝑚𝑎𝑥 . 𝑤 ̅ 𝑜𝑝𝑡 = 𝑆𝐼𝑅𝑚𝑎𝑥 . 𝑤 ̅𝑆𝐼𝑅 (3.6) The weight vector can pose in terms of the optimum Wiener solution [37] −1 𝑤 ̅𝑆𝐼𝑅 = 𝛽. 𝑅̅𝑢𝑢 . 𝑎̅0 (3.7) The correlation matrix is defined as 34 Chapter Three Proposal Algorithm Analysis 𝑅̅𝑠𝑠 = 𝐸[|𝑠|2 ]𝑎̅0 . 𝑎̅0 𝐻 Where, 𝛽 = 𝐸[|𝑠|2 ] 𝑆𝐼𝑅𝑚𝑎𝑥 (3.8) . 𝑎̅0 𝐻 . 𝑤 ̅𝑆𝐼𝑅 (3.9) The following part means with LMS algorithm, the error square gave as [37] |e(k)|2 = |d(k) − w ̅ H (k) x̅(k)| 2 (3.10) The LMS weight vectors would give as w ̅ (k + 1) = w ̅ (k) + μ e∗ (k) x̅(k) (3.11) The proposed algorithm can be drawn in the flowchart as indicated in Figure 3.2. 35 Chapter Three Proposal Algorithm Analysis Figure3.2. Normalized LMS algorithm 3.3 Variable Step Size via Error Controlling Algorithm Many algorithms evolved from the regular LMS algorithm to enhance the convergence rate, the LMS algorithm is implemented by the normalization known as normalized LMS (NLMS). To improve the LMS algorithm's convergence the Variable Step Size LMS (VSSLMS) algorithms is produced while maintaining stable achievements. The NLMS algorithm is 36 Chapter Three Proposal Algorithm Analysis an application of the LMS step size improvement (LMSSI) algorithm that selects for each iteration of the algorithm a different step size value 𝜇(𝑛) this factor of step size is proportional to the opposite of the cumulative predicted energy from the array instant value of the inputs signal. The NLMS minimizes the step size μNLMS (k) to allow major improvements to the weight vectors of the updates. This avoids the update weight vectors from diverging and allows the convergence algorithm more stable and quicker than when using a fixed step size [15] µ 0 μNLMS (k) = ‖x̅(k)‖ 2 (3.12) Where µ0 is a constant small positive value between (0-1). The final weight vector could be updated by [15] µ 0 w ̅ (k + 1) = w ̅ (k) + ‖x̅(k)‖ e∗ (k)x̅(k) 2 (3.13) Where e(k) = d(k) − w ̅ H (k)x̅(k) present the error signal w ̅ (k + 1) = w ̅ (k) + μ0 q+‖x̅(k)‖2 e∗ (k)x̅(k) (3.14) A small constant q must be added to the denominator to avoid denominator being zero when the data at any instant is zero. A new way of choosing the step size for the LMSSI algorithm which is the error driving algorithm to improve the performance of smart antenna by modifying the step size of the NLMS algorithm. In the NLMS algorithm, the small constant q has a fixed effect in the step size factor and can lead to a decrease in its value. This decrease in step size has an impact on the NLMS algorithm's convergence rate and weight stability. The error signal can be utilized to prevent the denominator from going to zero for every iteration and to control the step size. Under this approach, the q parameter could put as σ n q = ‖e(k)‖ 2 (3.15) Where σn is the square root of the noise variance σ2n . Thus, the step size would be μSS (k) = μ0 σn +‖x̅(k)‖2 ‖e(k)‖2 (3.16) Therefore, the weight vector of LMSSI algorithm is w ̅ (k + 1) = w ̅ (k) + μ0 σn +‖x̅(k)‖2 ‖e(k)‖2 e∗ (k)x̅(k) 37 (3.17) Chapter Three Proposal Algorithm Analysis Figure3.3. Flowchart of LMSSI The step size μ(n), change its value with respect to the array signals of error, and updating the weight vector to detect any updates in the smart antenna environments. Also can say, at the beginning of the adaptive cycle, when the error signal is high, q is low, and step size is high for rapidly down the error. Nevertheless, in steady-state the signal of error is small, q is high and the step size is low for a low degree of mismatch. This avoids divergent update weights and made the step size more constant and converges quicker than the least square algorithms. 38 Chapter Three Proposal Algorithm Analysis 3.4 Improvements for CMA (ICMA&BCM) algorithm Performance 3.4.1 Conventional Constant Modulus Algorithm CMA The CMA algorithm is useful because it requires no carrier synchronization and can be successfully applied to a non-constant modulus signal. Many wireless communication and radar signals are frequency or phase-modulated signals. Some examples of phase and frequency modulated signals are FM, PSK, FSK, QAM, and polyphase. This being the case, ideally, the signal amplitude should be a constant. Accordingly, the signal has a fixed amplitude or modulus. However, the received signal in fading channels, since there are multipath terms, is the combination of all multipath aspects. So the channel incorporates a difference in amplitude on its magnitude signal. Dominique Godard [39] primarily rely on the concept of a constant modulus (CM) to establish a group of blind equalization algorithms used in 2D wireless systems. Godard was using a cost function named order p dispersion function and the optimum weights are allocate after reduction [39] 𝐽(𝑘) = 𝐸[(|𝑦(𝑘)|𝑝 − 𝑅𝑝 ) 𝑞 ] (3.18) Where p and q is the positive integer equal to 1. Godard suggested that the cost function gradient is zero when 𝑅𝑝 is described by 𝑅𝑝 = 𝐸[|𝑠(𝑘)| 2𝑝 ] (3.19) 𝐸[|𝑠(𝑘)| 𝑝 ] Where 𝑠(𝑘) is calculated Zero memory of 𝑦(𝑘) and the error signal is produced as 𝑒(𝑘) = 𝑦(𝑘)|𝑦(𝑘)𝑝−2 (𝑅𝑝 − |𝑦(𝑘)|𝑝 )| (3.20) When p = 1 case reduces 𝐽(𝑘) = 𝐸[((|𝑦(𝑘)|) − 𝑅1 ) 2 ] the cost function would be (3.21) Where 𝑅1 = 𝐸[|𝑠(𝑘)| 2 ] (3.22) 𝐸[|𝑠(𝑘)|] If the output calculated 𝑠(𝑘) set to unity, the signal of error can produce as 𝑦(𝑘) 𝑒(𝑘) = (𝑦(𝑘) − |𝑦(𝑘)|) (3.23) Here when the p = 1 case, the vector of weight, would be [40] 39 Chapter Three Proposal Algorithm Analysis 1 𝑤(𝑘 + 1) = 𝑤(𝑘) + 𝜇(1 − |𝑦(𝑘)|)𝑦 ∗ (𝑘)𝑥(𝑘) (3.24) And when p = 2 case decreases the cost function to the formula 𝐽(𝑘) = 𝐸[(𝑦(𝑘)2 − 𝑅2 ) 2 ] Where, 𝑅2 = (3.25) 𝐸[|𝑠(𝑘)| 4 ] (3.26) 𝐸[|𝑠(𝑘)| 2 ] The weight vector with p = 2 cases would be as [40] 𝑤(𝑘 + 1) = 𝑤(𝑘) + 𝜇(1 − 𝑦(𝑘)2 )𝑦 ∗ (𝑘)𝑥(𝑘) (3.27) The case p= 1 was found to converge somewhat more fast than the case p = 2 [40] 3.4.2 Improvement Proposed to the Constant Modulus Algorithm ICMA via Variable Step Size vector Improvement that applies in this section replaced the step size factor (𝜇) of CMA algorithm by variable step size value that produced in previous section 3.3.1 above so, the new update weight vector equation would be as follow: 1 𝑤(𝑘 + 1) = 𝑤(𝑘) + 𝜇(1 − |𝑦(𝑘)|)𝑦 ∗ (𝑘)𝑥(𝑘) Where, 𝜇 = μSS (k) = μ0 σn +‖x̅(k)‖2 ‖e(k)‖2 (3.28) (3.29) Where the error equation gave as 𝑦(𝑘) 𝑒(𝑘) = (𝑦(𝑘) − |𝑦(𝑘)|) (3.30) So, the weight vectors equation gave as 𝑤(𝑘 + 1) = 𝑤(𝑘) + μ0 σn +‖x̅(k)‖2 ‖e(k)‖2 1 (1 − |𝑦(𝑘)|)𝑦 ∗ (𝑘)𝑥(𝑘) (3.31) The advantage of this step would enhance the coverage speed, the number of iteration, MSE value, and more stability for output signal amplitude. The result of this part will be shown in chapter 4. 40 Chapter Three Proposal Algorithm Analysis 3.4.3 Proposed Blocking Constant Modulus Algorithm BCMA (Static & Dynamic) One important drawback of Godard CMA was its weak convergence rate, which restricted the efficiency in dynamic environments of the algorithm when the signal should captivate rapidly. Therefore this limits CMA's quality if channel conditions happen fast. The BCMA is a resemblance to the SMI that discussed previously in chapter Two. The static BCMA was incorporated with only one block of data, the algorithm is iterated through n values before it converges. The input array block of K length vectors give as 𝑋̅ = [𝑥(1) 𝑥(2) 𝑥(3) . . . 𝑥(𝐾)] (3.32) The newly updated weight vector would produce as 1 𝑤(𝑛 + 1) = 𝑤(𝑛) + 𝜇(1 − |𝑦(𝑛)|)𝑦 ∗ (𝑛)𝑋 (3.33) So, the initial weights w ̅ (1) were select the complex to restrict output data vector r̅(1) would name as y̅(n) = [𝑤 𝐻 (𝑛)𝑥(1) 𝑤 𝐻 (𝑛)𝑥(2) . .. 𝑤 𝐻 (𝑛)𝑥(𝐾)] (3.34) After that the following weight w ̅ (2)was found, then the iteration will continue until sufficient convergence is achieved. This will be termed a static BCMA algorithm as this iteration method involves just one static block, of the length K. The main achievements of the static BCMA is could converge faster than the conventional CMA algorithm to converge after just a few iterations. The static BCMA easily calculated the weights based upon a sampled data fixed block. To ensure the adaptation in a dynamic environment for the data, it should update the blocks of data for every iteration. Let us describe a dynamic data block as the output of the array before weights apply to nth iteration, the nth block of length K will be 𝑋(𝑛) = [𝑥(1 + 𝑛𝐾) 𝑥(2 + 𝑛𝐾) 𝑥(3 + 𝑛𝐾) . . . 𝑥(𝐾 + 𝑛𝐾) ] (3.35) The output of weighted array to the nth iteration describes as y̅(n) = [𝑤 𝐻 (𝑛)𝑥(1 + 𝑛𝐾) 𝑤 𝐻 (𝑛)𝑥(2 + 𝑛𝐾) . .. (3.36) 𝑤 𝐻 (𝑛)𝑥(𝐾 + 𝑛𝐾)] Then the vector of the complex restricted output data is produced as 𝑦(𝑛) = [𝑦(1 + 𝑛𝐾) 𝑦(2 + 𝑛𝐾) 𝑦(3 + 𝑛𝐾) . . . 𝑦(𝐾 + 𝑛𝐾) ] 𝑇 1 𝑤(𝑛 + 1) = 𝑤(𝑛) + 𝜇(1 − |𝑦(𝑛)|)𝑦 ∗ (𝑛)𝑋(𝑛) 41 (3.37) (3.38) Chapter Three Proposal Algorithm Analysis So the dynamic BCMA algorithm is more suitable due to the constancy and coverage rate and speeds iterative improvement. 3.5 Particles Swarm Optimization Algorithm (PSO) Particles swarm optimization PSO is often used to solve prosperous complex multidimensional optimization problems in a different domain, such as antenna design and device modeling 𝑒𝑡𝑐. PSO algorithm which could be used to obtain the enhancement of recent standards in various parameters to improve the pattern of adaptive antenna radiation in each iteration, PSO flowchart can be shown in Figure 3.4. The converged value can be achieved by a different parameter such as: Controlling the Amplitude, Phase, Position and Complex Weights (Includes amplitude and phase control), this thesis deals with complex weight (amplitude and phase) and also position control. 42 Chapter Three Proposal Algorithm Analysis Figure3.4. Particle Swarm Optimization Flowchart. The PSO algorithm needs appropriate fitness function for optimizing so, two fitness function that uses and should have the possibility to directivity improvement, SLL reduction, and minify the HPBW, as follow 𝐹1 = 𝑀𝑎𝑥{𝐴𝐹(𝜃𝑑 )} 𝑤1𝑓1+𝑤2𝑓2 𝐹2 = 𝑀𝑖𝑛{ |𝐴𝐹𝑚𝑎𝑥 | (3.39) } (3.40) 𝑓1 = |𝐴𝐹(𝜃1 )|2 + |𝐴𝐹(𝜃2 )|2 (3.41) 𝑓2 = 𝑀𝑎𝑥{|𝐴𝐹(𝜃𝑆1 )|2 + |𝐴𝐹(𝜃𝑆2 )|2 } (3.42) 43 Chapter Three Proposal Algorithm Analysis Where AF is the array factor equation, which taking from equation (2.11) in chapter two. 𝜃𝑑 is the arrival angle of desired, 𝜃1 &𝜃2 are the two interference angle and 𝜃𝑆1 & 𝜃𝑆2 are the angles where the maximum sidelobe level is acquired during the optimization process. Thus, 𝑓1 function is to force the null at a specific angle and 𝑓2 used to keep the side lobe minimum as possible. The specification of each function on antenna array would be introduced later in chapter 4. 3.6 Adaptive Antenna Array Performance via REDS Algorithm Figure 3.5 illustrated the block diagram of the smart antenna beamforming system. 𝑆(𝑘) 𝑦(𝑘) I(k) 𝑒(𝑘) d(k) Figure3.5. Block Diagram of Smart Antenna System. The weight vector w ̅ = [w1 w2 … . wM ]T as shown in Figure 3.5 Antenna array adaptive weights should be adjusted in this manner to reduce the error when iterating. The desired signal S̅(k) and interferers I1 (k), I2 (k), … IN (k) are received with M potential weights by M elements array. The noise also includes in each received signal at m element. The k th represents the time samples for the time. The weighted array output is given by y= w ̅ H (k). x̅(k) (3.43) Where I1 (k) I (k) x̅(k) = a̅0 S(k) + [a̅1 a̅2 … . a̅N ] . [ 2 ] + n̅(k) ⋮ I𝑁 (k) = x̅S (k) + x̅I (k) + n̅(k) = input signal (3.44) x̅S (k) is the desired signal vector,x̅I (k) is the interfering signals vector. n̅(k) = zero-mean Gaussian noise for each channel. 44 Chapter Three Proposal Algorithm Analysis a̅i = M-element array steering vector for θ𝐼 the direction of arrival. In this thesis three type of array factor would be apply with this algorithm. An error signal is, e(k) = d(k) − w ̅ H (k) x̅(k) Utilizing the cost function gradient, the weight vector becomes as follow w ̅ (k + 1) = w ̅ (k) + μ e∗ (k) x̅(k) (3.45) The reformulation will be written with the use of the REDS algorithm, the error signal and the weight vector can give as e(k) = d(k) − w ̅ H (k)x̅(k) w ̅ (k + 1) = w ̅ (k) + μ e∗ (k)x̅(k) Where, (3.46) (3.47) x̅(k) = [x1 x2 … . xM ]T = input signal d̅ = [d(1) d(2) … . d(K)]T = desired signal The error signal e(k) can be write as e(k) = d(k) − ∑M i=1 wi (k)xi (k) (3.48) Considering the samples k-L, k-L+1, k-L+2 …... k, where L>M, equation (3.48) can be written as e̅(k) = d̅(k) − U(k)w ̅(k) Where (3.49) d̅(k) = [d(k), d(k − 1), d(k − 2) … d(k − L + 1)]T (3.50) ̅ (k) = [u̅1 (k), u̅2 (k) … . u̅M (k)] U (3.51) u̅j (k) = [xj (k), xj (k − 1) … xj (k − L + 1)]T The new approaches error at k is provided as (3.52) ̅ (k − 1)w e̅0 (k) = d̅(k) − U ̅(k − 1) (3.53) By update just one weight in w ̅ (k − 1) the new error can be written as update e̅1 (k) = d̅(k) − [U(k)w ̅ (k − 1) + U(k) w ̅ j0(k) (k) Fj0(k) ] (3.54) j0 (k) Is the index of the weight to be updated in the zeroth P-iteration iteration at time k and Fj0(k) is M x 1 vector with 1 in position j and 0 in all other positions.By choosing index j0 (k), the updated weight of the element is expressed as update w ̅ j0(k) (k) = w ̅ j0(k) (k − 1) + w ̅ j0(k) (k) (3.55) update Where w ̅ j0(k) (k) a projection value of the error e̅0 (k) and can be givens as 45 Chapter Three update w ̅ j0(k) (k) = Proposal Algorithm Analysis ̅j0 (k) (k)> <e̅0 (k)u ̅j0 (k) (k)‖ ‖u (3.56) 2 When P-iteration is equal to zero with step-size the updates the array weight would be introduced as follow update w ̅ o (k) = w ̅ (k − 1) + μREDS w ̅ j0(k) (k) Fj0(k) (3.57) The optimize value of μREDS would be established by applying the same step size parameter (μ) that used in Equation (2.42). (LMS). Figure 3.6 demonstrates the flowchart for the proposed algorithms. 46 Chapter Three Proposal Algorithm Analysis Figure 3.6. REDS algorithm flowchart 3.7 BER By modeling a communication system, it includes a smart antenna system to display the enhancement in received power (SNR and BER), so creates array factor and guides the pattern for both Linear, Circular and Planar 47 Chapter Three Proposal Algorithm Analysis Arrays, generates the signal, modulates it and transmits it with added Noise through the channel. Figure 3.7 presented the flowchart of simulate for a communication system module that included transmitter of signal message, modulator for the signal, beamformer, transmission channel, receiver, demodulator, and sampler. It transmits and create a pattern in the direction of DOA and receive message signal as user/target and measure received signal quality and BER. Figure 3.7 Processing Flowchart 48 Chapter Three Proposal Algorithm Analysis Different tests of common antenna arrays like a linear array and circular array antenna are comparing between performances also overall system performance had been inspected along with changing some parameters in the system. Testing BER and impact of an increasing number of elements in three types of the antenna array, linear array, circular and planar array besides the effect of increasing the noise value added to the system. So the BER had been measured by sending and generating a random message of 1000 Bits using MATLAB software package and comparing the received single with the transmitted one. 49 Chapter Four Simulation Results Chapter Four Simulation Results 4.1 Introduction In this chapter, the proposed system simulation results had been obtained and discussed, the simulation results were described to compare the potential of different algorithms to form beams in the direction of the desired signal and to locate position null in the direction of interference signal with the expectation of the desired signal angle and interference angle, and their production was evaluated by the beam distance, high sidelobe level, null depth and convergence rate. The algorithm that studied is LMS, RLS, NLMS, SMI, CMA, CGM, REDS, GA and PSO algorithms, and also improvement in CMA algorithm and algorithm with step size factor and normalized SIR enhancement algorithms. Another aspect deliberated in this section would be exploring the radiation patterns of different antenna array geometry which can be classified like one dimension (1D) linear array, tow dimension (2D) planar and circular array and three dimensions (3D) cubic array by varying various parameters, likewise the displacement spacing and the number of elements in array system. The subsequent simulations on arrays are performed using the MATLAB software package version R2017a. 4.2 Characteristic Studies of Uniform Linear Arrays This simulations are presented using a uniform linear array configuration. All simulations are assumed to have the signal input (s(k) = sin (2πft)) with f=900 MHz, the signal of desired is (d(k) = s(k)) and the number of sample intervals K is set at 100. The interfering signals and noise signals are negligible. In this chapter (M) would represented the total number of ements. 4.2.1 Array Factor Interpretations with Difference Element Displacement The impact of varying element spacing on the beamwidth had been simulated with a range from 0.1λ to1λ. Figure 4.1 represents the radiation pattern for a linear array of 10 elements for different element spacing. It can be observed that smaller element spacing produces a wider beamwidth in comparison to a half-wavelength spaced linear array. Also, for an element spacing of 1λ, a grating lobe (undesirable radiation pattern) that is equal in the magnitude of the main radiation lobe is generated. This is the large spacing between elements such that having full wavelength spacing will generate secondary lobes with large magnitudes. 50 Chapter Four Simulation Results Radiation pattern 10-element array for DOA = 30, d=0.1λ Radiation pattern 10-element array for DOA = 30, d=0.25λ Radiation pattern 10-element array for DOA = 30, d=0.5λ Radiation pattern 10-element array for DOA = 30, d=1λ Figure 4.1 Radiation Pattern for Different Element Spacing 𝟎. 𝟏𝛌, 𝟎. 𝟐𝟓𝛌, 𝟎. 𝟓𝛌, 𝟏𝛌 51 Chapter Four Simulation Results As it is observed that element spacing below 0.5λ produces wide beamwidth and would not be as useful in this instance. This is because the increase in beamwidth increases both fading and interference levels. Also, when element spacing is increased to 1λ, the grating lobe begins to appear. The beamwidth of the radiation pattern is measured at the -3dB point where the radiated signal magnitude is 0.707 of the maximum power radiated. Typical values for -3dB beamwidth and SLL for varying number of elements can be seen in Table 4.1. This table shows the results obtained from a constrained iterative process of conventional code. Table 4.1 Summary result for different inter-element spacing values Displacement space d 0.1 0.25 0.5 1 SLL(dB) HPBW(degree) -13 -13 -13 0 67.3 24 11.7 5.8 The standard spacing between elements is half-wavelength to minimize mutual coupling and avert grating lobes. 4.2.2 Array Factor Interpretations with Variation of Number of Array Elements Array for 300 DOA desired user and elements spacing of 0.5λ. The simulation was produced to examine the behavior of the beamwidth as the number of elements is increased from 2 to 25. Figure 4.2 illustrated both linear and polar plots of radiation pattern for 2, 6, 12, and 25 array elements, where the relationship of beamwidth and the increasing number of radiating elements can be shown for a various number of elements. Radiation Pattern 2-Element Linear Array for DOA=30, d=0.5λ 52 Chapter Four Simulation Results Radiation Pattern 6-Element Linear Array for DOA=30, d=0.5λ Radiation Pattern 12-Element Linear Array for DOA=30, d=0.5λ Radiation Pattern 25-Element Linear Array for DOA=30, d=0.5. Figure 4.2 Linear and polar plots of radiation pattern for 2, 6, 12 and 25 array elements As seen in Figure 4.2, as the number of elements increases, the beamwidth will decrease. This increase in accuracy and control is obtained at an increase in cost and complexity. Typical values for -3dB beamwidth and SLL for varying number of elements can be seen in Table 4.2. This table shows the results obtained from a constrained iterative process of conventional code. The table includes the beamwidth values for different element number of radiation patterns. 53 Chapter Four Simulation Results Table 4.2 Summary result for a different number of element values Number of elements M 2 6 12 25 SLL(dB) HPBW(degree) -3 -12.45 -13.1 -13.25 86.014 19 8.9 4.5 4.2.3 Performance of Comparison for Smart Antenna System Using Different Antenna Array Algorithms This section would be concerned with studying or comparing several types of algorithms and the comparison would be relying on the same parameter that examined in recent subjects above. Figure 4.3 shows linear and polar array pattern plot sequentially for LMS, RLS, NLMS, SMI, CMA, CGM, REDS, GA, and PSO algorithms. Figure 4.3 linear & polar plot, for radiation pattern for linear array using a different type of algorithms with N=5, d=0.5λ and DOA user at 𝟎° . 54 Chapter Four Simulation Results All algorithm has a DOA 0° and also an interference DOA at 45° , for linear array using a different type of algorithms with 5 elements and d=0.5λ, except the PSO and GA algorithm which would be discussed there interference effect on radiation pattern in the following section. Table 4.3 Summary result for applying a different type of algorithms Algorithm type LMS RLS NLMS SMI CMA CGM PSO GA REDS SLL(dB) -11.1 -10.9 -9.8 -9.5 -8.2 -9 -9 -6.5 -10 HPBW(degree) 20.76 25.89 20.6 20.1 31.6 30.4 22.2 26.5 20.9 The results of simulated algorithms are compared to various parameters like computation complexity, convergence rate, SLL, and HPBW. Accordingly, the analysis showed that, given computational complexity, RLS was the best algorithm followed by LMS, CMA, and SMI based on their magnitude of error. The LMS algorithm is a common solution used for the technique of beamforming. Generally, the RLS algorithm converges with magnitude level faster than the LMS algorithm but complexity is applied to the price paid. The Conjugate Gradient Method is expected to converge with less iterations but still has a wide HPBW compared with the other type of algorithm, the previous disadvantages of each algorithm could be solved with REDS algorithm. As can appear in Table 4.3 the LMS, NLMS and the REDS algorithms incorporate lower degree of sidelobe. Also, REDS, LMS, RLMS and SMI algorithms give the deepest nulls at the directions of interfering signals. PSO produce the most stable beam pattern and also gave initial adaption aspect. 4.2.4 Comparison of the Performance of Smart Antenna System Using Different Antenna Array Geometry The comparison in this part was based on the changing in antenna array geometry, where the shapes that would be used are linear, rectangular, circular and cube arrays, this test done with standard LMS algorithm by N=5.and d=0.5. Figure 4.4 illustrates the polar plot for radiation pattern using different array shapes. 55 Chapter Four Simulation Results Figure 4.4 Polar plot for comparison between different antenna array geometry Figure 4.5 shows the mean square error curve for each antenna geometry, and also to compare and differentiate between each shape Table 4.4 below clarify the summary results for each antenna array radiation pattern. Figure 4.5 Mean square error curves for LA, CA, PA and CUA arrays a circular array can be considered as the fastest convergence array, which is about 10 iteration, while in linear array iteration reaches 60 as shown in Table 4.4. Table 4.4 Summary result for applying linear and circular array geometry Antenna type SLL HPBW(dB) MSE Linear (M=10) Circular (M=10) -12.9 -7.8 10.11 21.23 0.0029 1.5679e-04 Iteration numbers 60 11 From the results of the Table 4.4 and Table 4.5, it could be illustrated that the circular array and cube array had the lowest percentage of error, and under the same number of iterations. 56 Chapter Four Simulation Results Table 4.5 Summary result for planner and cube array geometry Antenna type SLL HPBW(dB) MSE Planar (M=5*5) Cube (M=5*5*5) -23.8 -54 14.65 9.72 0.0024 1.0036e-04 Iteration numbers 55 45 The beamwidth (HPBW) and side lobe level (SLL) per every configuration had shown in Table 4.4 and Table 4.5 and the result clarifies that the linear array has a suitable result but it takes a lot of time to start with adaption process comparing with the other geometry 4.3 Performance of Smart Antenna System via Variable Step Size Aspect This slit gives the achievements of proposed μ(n) improvement algorithm. In this simulation, the type of algorithm that chose to improve is the LMS algorithm, and it is worth mentioning that this proposed improvement could apply in any algorithm had step size factor (μ), the number of sample intervals applied equally to (K=100). Figure 4.6 produced the linear and polar plots of the array pattern in dB for the LMSSI algorithm, the main beam pattern of the system here steers the direction at 00 . Figure 4.6 Linear & polar radiation pattern plots of LMSSI algorithm at the number of element M=10, DOA at 0, and DOA interference at 30,-10. The proposed step size algorithm in Figure 4.7 converges fast and without fluctuation compared with the LMS algorithms. On the other hand, the variation of weight values using the SSF algorithm in the steady-state is more stable than LMS in converging weights. From Figure 4.7, it can be observed that, the number of samples needed for variable μ(n) algorithms to converge are much lower than those for the LMS algorithm. 57 Chapter Four Simulation Results (a) (b) Figure 4.7 Tracking of the desired signal for (a) LMS (b) LMSSI algorithms Figure 4.7 display the obtaining and tracking of the array output for the desire signals after about 80 iterations for LMS and about 20 iterations for variable step size algorithm. Figure 4.8 Magnitude of linear array weights for LMSSI algorithm The proposed algorithm LMSSI reduces and increases the step size μSS (k) to track any changes in the smart antenna environment. Figure 4.9 Mean square error curves for LMSSI 58 Chapter Four Simulation Results Moreover, from the MATLAB software, at the steady-state, the LMS error is almost 4.5028e-04 whereas the LMSSI error is almost 2.7274e-04 at around 100 iterations. 4.4 Performance of Smart Antenna System via SIR-LMS Algorithm The SIR-LMS algorithm is developed for SA application which combined the individual pretty side of maximum SIR and LMS algorithms. This algorithm presented when the number of samples applied is (K=100). The linear and polar radiation pattern plots for the SIRLMS algorithms are shown in Figure 4.10 This figure shows the null depth provided by the SIR-LMS at −61.5dB. As shown in this Figure 4.10 the main beam pattern of the system steers the direction at 00 . Figure 4.10 Linear & polar Radiation pattern plots of SIR- LMS algorithm at the number of element M=10, DOA=𝟎° , and DOA interference at−𝟒𝟓° . Figure 4.11 illustrates the MSE curve for the SIR-LMS algorithm. It can be shown that the SIR-LMS algorithm can achieve the fastest convergence than any previous algorithms. Figure 4.11 Mean square error curves for LMSIR algorithm Moreover, comparison with some algorithms like conventional LMS and NLMS algorithms start converging, simultaneously, from iteration numbers 80 and 25, while the converge in the SIR-LMS algorithm starts to 59 Chapter Four Simulation Results from the primer or initial stat of iterations. In this case, the SIR-LMS error is 7.2038e-28 with zero iteration number. Figure 4.12 Tracking of desired signal Figure 4.13 Magnitude weights for LMSIR Figures 4.12&4.13 display that, the SIR-LMS start adjusting from the initial weight vector values to optimal weights. On the opposite, an algorithm such as LMS, NLMS, ... etc, is beginning to converge from arbitrary weight towards optimum weight values the simulation results illustrated that the proposed SIR-LMS algorithm delivers noticeable improvements in the elimination of interference, the convergence rate and weight stabilizing which quickly reach to their optimum values and without fluctuation compared with the other algorithms like LMS and NLMS algorithms. 4.5 Comparison of Smart Antenna System Executions Using Different Improvements for CMA (ICMA&BCM) Algorithm A fastness of converging CMA can be improved with many sides, one of them that examined in this section was using variable step size factor (μSS (k)), that was introduced in section 4.3 above, Figures 4.19&4.20 shows the beam pattern and the effect of the ICMA algorithm on the received multipath combined signal to achieve the desired signal path1. The second aspect to improve the converging speed of CMA is the Least Square CMA (LS-CMA) which is a block update iterative algorithm that is guaranteed to be stable and easily implemented. Typically LS-CMA converges in 5 to 10 iteration regardless of the block size and as seen the static LS-CMA can converge 100 times faster than the conventional CMA. However, the computational load makes the LSCMA impractical for a realtime application, but this algorithm gave a disadvantage on the effect of the algorithm on the received multipath combined signal for getting the desired signal, so the proposal refinement is to blocking the conventional CMA algorithm that leads to convergence rate at 6 iterations only and also obtained obvious desire signal (path 1) 60 Chapter Four Simulation Results The simulation was used on a uniform linear array of 8 elements. Described the direct path for which the signal arrives as a 32-bit binary chipping series with a value of ±1, at 45 °, the first multi-path signal reaches at 0 ° and the second multi-path signal reaches at -30 °. Figure 4.14 shows the arriving signal paths of the direct path, the first multipath, the second multipath, and the combined path as seen by the receiver. Figure 4.15 indicates the plot of the array factor as well as how the CMA algorithm minimized the multipath signals whereas guiding the maximum signal to a direct path. Figure 4.16 shows the effect of the algorithm on the received combined signal. a b c d Figure 4.14 Arriving signal amplitude ((a) Direct path (desire signal), (b) path 2, (c) path 3, and (d) combined signals) Figure 4.15 Linear & polar Radiation pattern plots for conventional CMA algorithm at 50 iterations number of element M=8, DOA=𝟒𝟓° , and DOAs interference at𝟎° &−𝟑𝟎° . 61 Chapter Four Simulation Results a Figure 4.16 Arriving signal and Output signal for CMA Figure 4.19Linear & polar Radiation pattern plots for ICMA algorithm at 23 iterations number of element M=8, DOA=𝟒𝟓° , and DOAs interference at𝟎° &−𝟑𝟎° . Figure 4.20 Arriving signal and Output signal for ICMA Figure 4.21show the comparison between two error curve (a), for conventional CMA and (b) for ICMA, and as could be obvious the number of iteration for each algorithm. 62 Chapter Four Simulation Results Figure 4.21 Error signal amplitude for CMA and ICMA algorithms Figure 4.20 shows that the performance of the CMA algorithm on output signal improvement when the number of iteration n is decreasing from 50 to 23. Figure 4.22 Linear & polar Radiation pattern plots for BCMA algorithm at 6 iterations number of element M=8, DOA=𝟒𝟓° , and DOAs interference at𝟎° &−𝟑𝟎° . The combination of three algorithms associated with the CMA algorithm shown below. Figure 4.23 Linear & polar radiation pattern plots for CMA, ICMA and BCMA algorithm at number of element M=8, DOA=𝟒𝟓° , and DOAs interference at𝟎° &−𝟑𝟎° . As can be seeing in Figure 4.23 the two proposal algorithm achieved more accentuate result in terms of depth nulling beside the fastest convergence. 63 Chapter Four Simulation Results 4.6 Smart Antenna Performance with Particle Swarm Optimization (PSO) Algorithm For an adaptive array antenna simulation, the training sequence's 200 input signals having values marked as ±1to represent a transmitted dispatching binary quantities as shown in Figure 4.24. Figure 4.24 Binary Signal. Initialize the parameters that will be used for the system development in MATLAB shown in Table 4.5 below. Table 4.5 Initialize simulation parameter Parameter Frequency of operation spacing between element Value 900 MHz 𝑁𝑜.Of element Type of antenna The phase between two element Variable(8-12-18-25) Isotropic 0 Radian 0.5𝜆 For PSO algorithm design, the necessary parameters were described as supervening: Size of population P equal to 10, Maximum number of generation equals 50, Maximum and minimum value of inertia weight range are (0.80.3), Acceleration constantsC1 and C2 are 2. The previous parameters should be given before algorithm applications. Their values affect the process of optimal solution search and results. 4.6.1 Weight Vector Optimization The radiation pattern results according to the equation for uniform element spacing for the optimized weighting vector (phase and amplitude) of ULA could be shown in Figure 4.25. 64 Chapter Four Simulation Results Figure 4.25 Linear and polar for PSO algorithm with uniform array (ULA) Radiation Patterns at 8, 12, 18, and 25 elements and DOA user at𝟎°. Table 4.6 indicates that increasing the element numbers lead to an increase in the numbers of sidelobe also that companion with reducing in SLL and HPBW leading to increasing in directivity calculation were performed using fitness function equation. Table 4.6 (ULA) Parameter result No.elements Directivity Sidelobe level reduction (𝑑𝐵) Beamwidth (𝑑𝑒𝑔) 8 21.043 -10 12 22.983 -12.1 18 24.888 -15 25 26.27 -14 12.8 9-8.8 5.3 3.97 As observed in Table 4.6 that the directivity is increased to 24.888, HPBW is decreased to 5.3 °and the minimal SLL up to -15 dB with N increased to 18. 65 Chapter Four Simulation Results Figure 4.26 Linear and polar for PSO algorithm with Uniform Antenna Array (ULA) Radiation Patterns at λ/4, λ /2, 1.5 λ and 3 λ displacement spacing element and DOA user at𝟎°. Figure 4.26 shows that with an increase in displacement distance between element the sidelobe level (SLL) would decrease with the same number of elements, as can visualizing at λ /2 the SSL equal to -13.2 dB, while at 3 λ the SLL equal to -8.6 dB. 4.6.2 Inter Element Spacing Optimization The displacement distance between neighboring two elements is the one of influential LA performances. PSO algorithm used to find the optimal elements distance value for the element with the equal excitation amplitude 𝑎𝑛 = 1 and excitation phase 𝜃𝑖 = 0° employing the equation of fitness function. The simulation result of the radiation pattern of the optimized weighting vector is viewed in Figure 4.27. The array elements number suggested by the author is equal to 4, 8, 12, 18 and 25. 66 Chapter Four Simulation Results Figure 4.27 Linear and polar for PSO algorithm with non-uniform array (NLA) Radiation Patterns at 4, 8, 12, 18, and 25 elements and DOA user at𝟎°. Table 4.7 (NLA) Parameter result No.elements Directivity Side lobe level reduction (dB) Beam-width (𝑑𝑒𝑔) 8 21.716 -10.3 12 23.542 -12.2 18 25.845 -10.52 25 26.02 -13.5 9.5 6.87 4.19 3.75 As shown above, there is a clear and good improvement in the difference between patterns of the two types of optimization variables, so by changing the dimensions of antenna array system the enhancement appears in both directivity and beamwidth, and the number of sidelobe also. Table 4.8, shows the resulting displacement position of each element only by using the (PSO) optimization algorithm process. 67 Chapter Four Simulation Results Table 4.8 (NLA) antenna array position on x-axis normalize to 𝒍𝒂𝒎𝒅𝒂(λ) [𝑥0 , 𝑥1 , 𝑥2 , . . . . . . . . , 𝑥𝑛−1 ] 8element 12element 18element [0.1983,0.3274,0.0958,0.1994,0. 1782 ,0.1571 ,0.0922 ,0.0374] [0.2342,0.1929,0.1822,0.1925,0. 2207,0.1992,0.1980,0.0995 ,0.1326,0.284,0.2366,0.1774] [0.2715,0.3019,0.04232,0.3044, 0.2107,0.03251,0.0928,0.1822,0 .3191,0.3216,0.0525,0.3235,0.3 190,0.16179,0.2667,0.04729,0.1 4058,0.3052] ∑ [𝑥0 , 𝑥1 , 𝑥2 , . . . . . . . . . , 𝑥𝑛−1 ]= (total length of NLA) (1.2858) ∑ [𝑥0 , 𝑥1 , 𝑥2 , . . . . . . . . , 𝑥𝑛−1 ]= (total length with 0.5 spacing ) (4) (2.3498) (6) ( 3.14) (9) 4.6.3 Weight Vector and Inter Element Spacing Optimization PSO algorithm workout even to study the possibility of developing weights and spacing distance between elements together and that will be called improved particle swarm optimization (IPSO). The algorithm will be operated by two parts first one uses to calculate the distances and then combined with the original program to update the weights. The fitness function are given in chapter 3, design for this purpose, observations are made for N = 8 and 20 in Table 4.9. 68 Chapter Four Simulation Results Figure 4.28 Linear and polar for IPSO Radiation Patterns at 4, 8, 20 elements and DOA at𝟎° Table 4.9 IPSO parameter results No .element Directivity 8 20 22.48 24.39 Sidelobe level reduction (dB) -15.7 -15 Beamwidth (𝑑𝑒𝑔) 10 4.77 As observed in Table 4.9 that at 8 elements the directivity is increased to 22.48, HPBW is reduced to 10° and the SLL is minimizing up to -15.7 dB So, from the comparison with the old test, it's clear that an enhance of -5.7 dB in SLL and also in the number of the side lobe, 2.8° in the HPBW by using IPSO algorithm shown in Figure 4.28. 4.7 Nulling Effect on the Radiation Pattern One of the most important applications of a smart antenna is the nulling effect. The null control was designed to transmit minimal power in locations where the eavesdropper is present. Controlling the parameters, the excitation amplitude, the phase, the array distance spacing, and the number of the elements will achieve the null control, but this thesis deal with a number of elements for uniform LA and non-uniform LA by using the fitness function equation. Two cases are considered using a different number of antenna elements and nulls at specified directions. Figures 4.29 & 4.30 showed an improvement in the main beam direction and reduction in the other direction which contains the interference according to Table 4.10 results. 69 Chapter Four Simulation Results Figure 4.29 Radiation Pattern for 8 elements at DOA at 𝟎° and 𝟓𝟎°&−𝟓𝟎° and 𝟑𝟎°&−𝟑𝟎° Interferences Respectively Figure 4.30 Radiation Pattern for 8, 12 and 20 elements at DOA at 𝟎° and 𝟑𝟎°&−𝟑𝟎° Interference Respectively Table 4.10 SLL, HPBW AND depth null results 𝑁𝑜.element HPBW SLL Null depth at 30 Null depth at 30 8 12 12.4 4.15 -11.8 -12.8 -40 -46.1 -43.4 -43 Figure 4.31 shows that with an increase in displacement distance between elements, the sidelobe level would decrease with the same number of elements, as can visualizing at λ /2 the SLL equal to -13 dB, while at 1 λ the SLL equal to 0 dB. 70 Chapter Four Simulation Results Figure 4.31 Radiation Pattern at λ /2, 0.75 λ and 1 λ displacement spacing element at 8 elements and DOA user at 𝟎° and 𝟓𝟎°&−𝟓𝟎° DOA Interference Respectively 4.8 Smart Antenna Array System Performance via REDS Algorithm REDS algorithm is applied to different geometries of the smart antenna array to investigate its potentials in beamforming convergence. The essential benefit of this smart system is that it appears to have the quickest rate of convergence. The results of the simulation are comparing with three classical geometries of the adaptive array, which is linear, circular, and rectangular antenna array. In each shape, the weight vector of the array, mean square error, and radiation pattern performance are presented. The number of elements in all discussion is presented in Table 4.11 for comparison purposes. Table 4.11 Initialization parameters 𝑁𝑜. 𝑒𝑙𝑒𝑚𝑒𝑛𝑡𝑠 Displacement distance Linear array 10 0.5𝜆(x-axis) Circular array 10 2π/𝑁 Planar array 5*5 0.5𝜆 ∗ 0.5𝜆(x-y plane) A linear array made up of isotropic elements that applied according to the below requirements: • Input signal S(k) = sin(2πft(k)) with f = (1: K) ∗ T K 1 T = 900 MHZ and t = Where K represents the sample interval numbers, and T is the period and that will be the desired signal d(k). (d(k) = S(k)). • Desired DOA is θ0 = 00 and interfering signal, I1 each of the rndn(1, K) with DOA, θ1 = 450 . 71 Chapter Four Simulation Results • Step size initial parameter µ0 = 1 • Every element in the array combined with zero mean Gaussian noise with variance σ2n = 0. 001 added to the input signal in the array. • • Signal to Noise Ratio (SNR) and Signal to Interference Ratio (SIR) are placed at 30 dB and 10 dB • The step size parameter is calculated according to the equation. (2.42) and within the MATLAB software package as follows μ = 1/(4 ∗ real(trace(R xx ))), where R xx = X ∗ X′ • In the beginning, every weight vector is set to zero. Two types of noise are discussed here to clarify the performance of the REDS algorithm under the effect of different noise channel, which is, Additive White Gaussian Noise channel (AWGN) and Rayleigh fading channel, respectively. 4.8.1 (AWGN) Channel Results The basic noise model used in information theory to minimize the effect of many random processes that occur in nature. To measure the efficiency of the smart antenna system, for each received signal at element M an (AWGN) channel model is authorized Figure 4.32 contains just a zero mean AWGN. The curves of mean square error are presents in Figures 4.33& 4.35& 4.37 (c) provide that the convergence speed utilize PA geometry is quicker than LA and CA configuration. Moreover, the array output using PA was good predicts of the input signal comparing with LA and CA but, at the same time, it can see that the error magnitude of CA is less than both LA and PA. Figure 4.32 Linear and polar radiation patterns for linear array (LA), all geometry worked out with 50 sample/AWGN. 72 Chapter Four (a) Simulation Results (b) (c) Figure 4.33, (a) Signal tracking of desired and output curves (b)Weight magnitudes (c)MSE for linear array (LA) worked out with 50 sample/AWGN. Figure 4.34 Linear and polar radiation patterns for circular array (CA), all geometry worked out with 50 sample/AWGN. (a) (b) (c) Figure 4.35, (a) Signal tracking of desired and output curves (b) Weight magnitudes (c) MSE for circular array (CA) worked out with 50 sample/AWGN. 73 Chapter Four Simulation Results Figure 4.36 Linear and polar radiation patterns for planar array (PA), all geometry worked out with 50 sample/AWGN. (a) (b) (c) Figure 4.37, (a) Signal tracking of desired and output curves (b) Weight magnitudes (c) MSE for planar array (PA) worked out with 50 sample/AWGN. The studies indicate that the REDS algorithm has a more fast rate of convergence by comparison with the previous algorithm like LMS which needs about more than 80 iteration to start adaptation processes, NLMS algorithms that need more than 25 iterations, CMA algorithm that need 50 iteration and so on. Moreover, Figures 4.33&4.35 &4.37 (a) can illustrate the array output of PA and CA is a good estimate of the desired signal. Also Figures 4.33&4.35&4.37 (b) shows the magnitudes of the estimated complex weights of the, LA, CA and PA array configurations, which is the magnitude of the weights for each element in array .the results also show that the performance of the REDS have very good stability and it can notice the adaptation processes starts approximately after 5 samples only. As apparent in Figures 4.32&4.34&4.36 above that LA and PA antenna configuration have more side lobe than CA array but at the same time the null depth of LA and PA is larger than CA array, also the sidelobe amplitude PA radiation pattern is less than LA and CA array configurations. The numerical result can be shown in Table 4.12 below. 74 Chapter Four Simulation Results Table 4.12 AWGN results LA CA PA Depth null(dB) SLL(dB) Weight stability -61 -60 -96 -13 -10 -23.6 8 iterations 5 iterations 7 iterations Maximum mean square error 5.9119e-04 0.0141 5.1338e-04 4.8.2 Rayleigh Fading Channel Results Rayleigh fading is a statistical model for the effect of a propagation environment on a radio signal, is most applicable when many objects in the environment scatter the radio signal and that lead to no dominant propagation along a line of sight between the transmitter and receiver, such as that used by wireless devices The magnitude of a signal that has passed through such a transmission medium (also called a communication channel) will vary randomly is assumed. To evaluate the performance of an adaptive antenna with an analytical model operating with AWGN and Rayleigh fading environment. In the assessment of results the following criteria are considered: • The signals arriving at every element undergo independent Rayleigh fading, both for the desired and interfering signal. • Both desired and interference signals have the same amplitude as is a release from 0° and45°. • Every simulation relates to a run (data rate) of 500 Kbits. • Doppler frequency fd of 117 Hz, corresponding to the speed of 140 km/h at 900 MHz will be used in this simulation to present a worstcase scenario. • All antenna geometry is implemented when K=60 is set as the sample number. Figure 4.38 Linear and polar radiation patterns for linear array (LA), all geometry worked out with 60 sample/ Rayleigh channel. 75 Chapter Four (a) Simulation Results (b) (c) Figure 4.39, (a)Signal tracking of desired and output curves (b)Weight magnitudes (c)MSE for linear array (LA),all geometry worked out with L=6,K=60 sample/ Rayleigh channel. Figure 4.40 Linear and polar radiation patterns for circular array (CA), all geometry worked out with 60 sample/ Rayleigh channel. (a) (b) (c) Figure 4.41, (a)signal tracking of desired and output curves (b)Weight magnitudes (c)MSE for circular array (CA),all geometry worked out with L=6,K=60 sample/ Rayleigh channel. 76 Chapter Four Simulation Results Figure 4.42 Linear and polar radiation patterns for planar array (PA), all geometry worked out with 60 sample/ Rayleigh channel. (a) (b) (c) Figure 4.43, (a)Signal tracking of desired and output curves (b)Weight magnitudes (c)MSE for planar array (PA), all geometry worked out with L=6, K=60 sample/ Rayleigh channel. In Figures 4.38&4.40&4.42 the results for the REDS algorithm are shown at block length set to 6. The results show that the performance of the REDS is working better when compared with planar array than the LA&CA array shapes. As shown in Figures 4.39&4.41&4.43 the convergence speed using LA antennas is slower than the CA antenna configuration. Also, the output of the array using CA is a good predictor of the input signal comparing with the LA configuration. Figure 4.43 indicates the faster convergence speed and the strong estimate of the output array in the PA configuration. From Figures 4.39&4.41&4.43 (c), the LA antenna converges after more than 30 iterations, CA antenna converges after 20 iterations, whereas the PL antenna convergence can occur at about less than 10 iterations. Figures 4.39&4.41&4.43 (b) shows the magnitudes of the complex weights of LA, CA, and PA configurations. These figures show the magnitude of the weights for each element in the array. And as illustrated in the Figures 4.39&4.41&4.43 above, the data show that PA efficiency is better and faster than other types of configuration. Numerical simulation results of array factor with Rayleigh channel are shown in Table 4.13. 77 Chapter Four Simulation Results Table 4.13 Rayleigh Fading results LA CA PA Depth null(dB) SLL(dB) Weight stability -62 -48 -90 -13 -6 -24 40 20 10 Maximum mean square error 8.7810e-04 0.0033 0.0014 Figures 4.38&4.40&4.42 show that the PA array introduced a deep null which about −90 dB in the direction of the interferer, However, the simulations show, the LA and CA array configurations provide a deeper null of about -62 dB and -48 dB at K=60. It is apparent from this array pattern in Figures 4.38&4.40&4.42 the CA array gives the minimal sidelobe number while PL array gives the lowest array SLLs compared with the other array configurations. Figures 4.38&4.40&4.42 presents that the amplitude of the side lobes of the radiation pattern from the PA array is greater than the amplitude of the side lobes of the radiation pattern from LA and CA array. Multipath and their effects on antenna array pattern with using REDS algorithm also have been introduced, Figure 4.44 illustrates a linear and polar plot for antenna array pattern curve under multipath signals. 78 Chapter Four Simulation Results Figure 4.44 Linear & polar radiation pattern plots with 8 element linear array (LA)where a-desire(S)=0/Interference(I)=45,-20,b-S=0,35/I=60,-30,c-S=0,30/I=50,-30,-70,d-S=0,20,30/I=50,-45,-70, worked out with 50 sample/AWGN. Table 4.14 Analyze the multipath effect a b c d Desired DOA(S) 0° 0° &35° 0° &30° 0° &30° & − 20° Interference DOA(I) 45° & − 20° 60° & − 30° ° 50 & − 30° & − 70° 50° & − 45° & − 70° Depth null (dB) -48.5&-59.2 -61&-41.7 -38.1&-42.3&-62.7 -36.7&-51.2&-55.1 As illustrated in Table 4.14 above, the effectiveness of the REDS algorithm appears to guidance or steering the mean beam in the direction of users and steering the depth extinct to interference directions. 4.9 BER In this section, the estimated BER had been measured, by sending a message of 1000 Bit in random way using MATLAB program and comparing the received single with the transmitted one. With AWGN as a source of noise in the channel. This section studied the effect of the number of elements on the overall system performance. Figure 4.45 shows at lower noise values there are no errors when receiving a signal message. However, errors occur when the noise starts to reach high levels and start to increase by increasing the values of noise giving much higher error values. This figure shows the comparison of the different number of elements (m=2,4,8,16) and shows also that with 79 Chapter Four Simulation Results two elements give a lot of errors than using 8 elements in the array as so on when the number of elements increased, better results are given. Figure 4.45. BER curve for different number of the linear array The circular array represents 2D antenna arrays. The circular array is also very resistive to noise and the BER is approximately close to that of the linear array approximately with the same number of elements (M=4, 8, 16) as shown in Fig.46 Figure 4.46. BER curve for different number of the circular array Table 4.15 and Figure 4.47 shows the result value of the arrays with 8 elements for linear array and circular array. Table 4.15. Number of errors concerning noise Noise (10−16) 1 5 10 50 100 500 No. error of linear array 0 0 0 0 0 0 80 No. error of circular array 0 0 0 0 0 0 Chapter Four Simulation Results 1000 5000 10000 30000 50000 100000 200000 300000 400000 500000 600000 800000 1000000 1500000 0 0 0 0 0 1 13 39 68 103 121 162 204 274 0 0 0 0 1 9 62 111 142 188 216 266 310 386 Figure 4.47. BER curve for linear and circular antenna array geometry with 8 elements 4.10 SNR The output Results of the signal to noise ratio with applying Noise of 1*10 for 1000 Bits message signal to test the benefit of changing the element numbers on SNR values, are given in Table 4.16 −16 Figure 4.48 Relation between number of elements and SNR (dB) for a linear and circular array 81 Chapter Four Simulation Results Table 4.16 SNR to a different number of elements Number of elements SNR(dB) Linear array SNR(dB) Circular array 2 4 8 12 16 20 -13.8067 -7.7861 -1.7655 1.7563 4.2551 6.1933 -22.7126 -16.6925 -4.3701 0.5706 3.5825 5.7610 Figure 4.48 and Table 4.16 shows that there is a small difference in the SNR curve for linear and circular array and it is clear to observe that the SNR curve of overall system increase with the number of the elements of the array regardless of the geometry of antenna array. 82 Chapter Five Conclusions and Future Work Chapter Five Conclusion and Future Works 5.1 Conclusion From the work presented in this thesis some of the points could be concluded as follow: 1-It was noticed that by increasing the number of array elements the main lobe will be (smaller beamwidth) and more director, but the sidelobe numbers will increases. And shows that lead to an increase in the distance between elements leads to produce a grating lobe which is typical to the main lobe but in the reverse direction. 2-The analysis showed that RLS was the best algorithm despite the computational difficulty followed by LMS, CMA, and SMI based on their error magnitude. LMS algorithms are well known for their simplicity and robustness. The Conjugate Gradient Method is expected to converge with fewer iterations but has a high HPBW relative to the other algorithms, the previous disadvantages of each algorithm could be solved with the REDS algorithm. PSO produce the most stable beam pattern and also gave an initial adaption aspect. 3- Through using different antenna arrays geometries which are linear, rectangular, circular and cube with LMS algorithm, it could be found that the circular array and cube array had the lowest percentage of error, and under the same number of iterations circular array can be considered as the fastest convergence array, which is about 10 iteration while in linear array reach to 60 iterations. Despite the result, it clarifies that the linear array has the desired outcome in terms of SLL and HPBW, but it takes a lot of time to begin with the phase of adaptation compared to the other geometries. 4-Step size one of most effective factor on some of the algorithm that dependent on step size factor through adaption process such as LMS, NLMS, RLS, CMA, as appear in get a more stable and fasten convergence rate and display how the output of the antenna array acquires and chase the desired signal with about 80 iterations for LMS and about 20 iterations for the algorithm with a variable step size. 5-Hybridtation between the good aspect of fixed beam with maximum SIR algorithm and LMS algorithm produce (SIR-LMS) algorithm with the initial adaptation process and give an enhancement in MSE. 6-The improvement would also expiate on CMA algorithm with two aspects firstly by applying variable step size (ICMA) which produce an enhancement on the amplitude of array output and also on convergence rate so the number of iteration change from 50 by basic CMA to 23 with (ICMA), secondly by 83 Chapter Five Conclusions and Future Work applying the blocking idea (BCMA) on basic CMA and that lead to enhance the convergence rate to 6 iterations only with lower MSE value as appear earlier. 7- Applying PSO on smart antenna came in two parts, at the beginning PSO used to find the adaptive weight vector. The second part of applying PSO on smart antenna which designs non-uniform linear array which presented an effective array with appearance improvement in SLL -6 dB, BW 3˚and directivity in 2 dB as compared with standard linear array and the most important thing have created an array with less antenna number and antenna with less length with the same number of antennas would economically benefit us. 8-REDS algorithm utilizing on the smart antenna with three array geometry linear, circular and planner gave a more stable and acceptable performance, it can be seen that the number of iteration reduce to 5 for circular, 7 for a planner and 8 iterations for the linear array as compare with other algorithms that reach to 85 and 25 for LMS and NLMS resistivity, also gave an improvement in MSE as appearing in chapter four early, the effectiveness of REDS algorithm also appear more robust and remain stable when applying another type of noise like Rayleigh fading noise as comparing with another algorithm in the single beam and multi-beam signals. 9-Calculating BER by different antenna array geometry shows the effectiveness of smart antenna to receive the data with great resistivity to the noise and fading signal. It also demonstrates BER's progress about a growing number of array components. 10- SNR estimated in terms of different number of elements with linear and circular array configuration, which presents an increasing in SNR with increase in number of elements and that gives a great performance to a smart antenna to receive the transmitted signal with the lower number of error. 5.2 Future Work 1. Currently, proposed algorithms have been implemented using MATLAB software alone. Executing an algorithm using either a Digital Signal Processor (DSP) chip or Field Programmable Gate Arrays (FPGA) would make it useful to check the actual complexity of the computation. 2. Using the PSO algorithm to design a more non-uniform antenna array with an acceptable performance like a circular array. 3. Design different array geometry like hexagonal and planner circular, cylindrical...etc., and examine their performance through various array algorithms. 84 Chapter Five Conclusions and Future Work 4. Analysis of the impact of mutual coupling between the array's components. 5. Study the effect of each algorithm on power reduction property, by improve the base station sensitivity or reduce their radiated power, this would translate into cell phone battery life extension and reduced the base stations density at the required coverage area to reduce RF pollution levels. 85 References References References [1] Hema Singh and Rakesh Mohan Jha, “Trends in adaptive array processing”, Hindawi Publishing Corporation, International Journal of Antennas and Propagation, vol.2012, no. 361768, 2012. [2] Sharma, P., “Neural network based robust adaptive beamforming for smart antenna systems”, M.Sc. Thesis, Department of Electrical Engineering, National Institute of Technology, Rourkela, 2009. [3] D. B. Salunke and R. S. Kawitkar, “Analysis of LMS, NLMS and MUSIC algorithms for adaptive array antenna system”, International Journal of Engineering and Advanced Technology (IJEAT), vol.2, no.3, pp.130-133, 2013. [4] Balanis, C. A., and Panayiotis, I. I., “Introduction to smart antennas”, Morgan & Claypool Publishers, 1st ed., 2007. [5] L. C. Godara, “The electrical engineering handbook,” Electr. Eng. Handb, Boca Raton London New York Washington, D.C., 2005. [6] Pradeep Kumar Nayak, Sachidananda Padhi and Subrat Sethi, “ Performance analysis of smart Antenna using beam forming technique”, Advance in Electronic and Electric Engineering, vol.4, no.2,pp.201-206, 2014. [7] Zhizhang Chen, Gopal Gokeda and Yiqiang Yu, “Introduction to direction-of-arrival estimation “, Artech House, ch.1, pp.13-14, 2010. [8] Suraya Mubeen, Dr.A.M.Prasad, and Dr.A.Jhansirani, “An overview of smart antennas and its techniques beam-forming and diversity”, International Journal of Computational Engineering Research (IJCER), Vol.2, No.3, pp.732-736, 2012. [9] Lakshmi, T. S. J., and Sivvam, S. “Smart antennas for wireless applications”, International Journal of Applied Engineering Research, 86 References 12(Special Issue 1), pp. 100–105. doi: 10.1109/aps.2001.959392, 2000. [10] J. Homer, “Adaptive equalisers and smart antenna”, Thesis, the University of Queensland, , Brisbane, Australia, october, 2002. [11] Spencer, M. G. J. “Smart antennas in handsets for improving 3G performance”. Multiple Access Communications Limited, UK, IET,PP.53-57, 2003. [12] I. Stevanovic, A. Skrivervik, and J. R. Mosig, “Smart antenna systems for mobile communications, relatório técnico,” Fed. Lausanne, no. January, p. 120, 2003. [13] Durrani, S. “Investigations into smart antennas for CDMA wireless systems”. Ph.D. Thesis, the University of Queensland, Brisbane, Australia, August, 2004. [14] Ioannides, P. and Balanis, Constantine A. “Uniform circular and rectangular arrays for adaptive beamforming applications”, IEEE Antennas and Wireless Propagation Letters, 4(1), pp. 351–354. doi: 10.1109/LAWP.2005.857039, 2005. [15] T. I. Mohammad and A. A. R. Zainol, “MI-NLMS adaptive beamforming algorithm for smart antenna system applications,” J. Zhejiang Univ. Sci., vol. 7, no. 10, pp. 1709–1716, 2006. [16] K. R. Mahmoud, M. El-Adawy, S. M. M. Ibrahem, R. Bansal, K. R. Mahmoud, and S. H. Zainud-Deen, “A comparison between circular and hexagonal array geometries for smart antenna systems using particle swarm optimization algorithm,” Prog. Electromagn. Res., vol. 72, pp. 75–90, 2007. [17] B. Atrouz, A. Alimohad, and B. Aïssa, “An effective jammers cancellation by means of a rectangular array antenna and a sequential block lms algorithm: Case of mobile sources,” Prog. Electromagn. Res. C, vol. 7, pp. 193–207, 2009. 87 References [18] C. S. Rani, P. V. Subbaiah, K. C. Reddy, and S. S. Rani, “LMS and RLS algorithms for smart antennas in a W-CDMA mobile communication environment,” J. Eng. Appl. Sci., vol. 4, no. 6, pp. 78– 88, 2009. [19] Zuniga, V., Erdogan, A. and Arslan, T. “Control of adaptive rectangular antenna arrays using particle sSwarm optimization’, Loughborough Antennas & Propagation Conference. IEEE, pp. 385– 388, 2010. [20] N. H. Noordin, V. Zuniga, A. O. El-Rayis, N. Haridas, A. T. Erdogan, and T. Arslan, “Uniform circular arrays for phased array antenna’, Loughborough Antennas & Propagation Conference. IEEE, pp. 1–4, 2010. [21] Amritpal Singh Bhinder and Arvind Kumar, “Synthesis & analysis of adaptive beamforming smart antenna for advanced communication systems”, International Journal of Emerging Technologies in Computational and Applied Sciences (IJETCAS), Vol.3, No.3, pp. 290-293, 2012. [22] Maina, R. M., Langat, K. and Kihato, P. K. “Comparative analysis between use of particle swarm optimization and simulated aAnnealing algorithms in beam/null steering in a rectangular array system”, International Journal of Innovative Research and Development (ISSN 2278–0211), vol.2, no.7, 2013. [23] Hosseini, Seyed A et al. “Fast euclidean direction search algorithm in adaptive noise cancellation and system identification”, International Journal of Innovative Computing, Information and Control, vol.9, no.1, pp. 191–206, 2013. [24] A. P. Rao and N. V. S. N. Sarma, “Adaptive beamforming algorithms for smart antenna systems,” WSEAS Trans. Commun., vol. 13, pp. 44– 50, 2014. 88 References [25] K. Venkatesan, T. Arunnehru, and G. Umamaheswari, “Implementation of an Adaptive Antenna Array Algorithm for AntiJamming Techniques,” vol. 5, no. 3, pp. 1–6, 2014. [26] Vesa, A., Alexa, F. and Baltə, H. “Comparisons between 2D and 3D uniform array antennas”, Proceedings of the 2015 Federated Conference on Computer Science and Information Systems, FedCSIS 2015, 5, pp. 1285–1290. doi: 10.15439/2015F266, 2015. [27] Patel, D. N., Makwana, B. J. and Parmar, P. B. “Comparative analysis of adaptive beamforming algorithm LMS, SMI and RLS for ULA smart antenna’, International Conference on Communication and Signal Processing, ICCSP 2016, pp. 1029–1033, 2016. [28] Banerjee, S. and Dwivedi, V. “Performance analysis of adaptive beamforming using particle swarm optimization”, in 2016 11th International Conference on Industrial and Information Systems (ICIIS). IEEE, pp. 242–246, 2016. [29] S. U. Rahman, Q. Cao, M. M. Ahmed, and H. Khalil, “Analysis of linear antenna array for minimum side lobe level, half power beamwidth, and nulls control using PSO,” J. Microwaves, Optoelectron. Electromagn. Appl., vol. 16, no. 2, pp. 577–591, 2017. [30] S. C. Swati Patidar, Kishor Kumbhare “Improvement of spectral effecinecy and power control of smart antenna”, vol.6, no.5, pp. 736– 741, 2017. [31] P. N. Chuku, T. O. Olwal, and K. Djouani, “Enhanced RLS in Smart Antennas for Long Range Communication Networks,” Procedia Comput. Sci., vol. 130, pp. 196–205, 2018. [32] X. Zhang, D. Lu, X. Zhang, and Y. Wang, “Antenna array design by a contraction adaptive particle swarm optimization algorithm,” Eurasip J. Wirel. Commun. Netw., vol. 2019, no. 1, 2019. 89 References [33] Dakulagi, V. and Alagirisamy, M. “Adaptive beamformers using 2Dnovel ULA for cellular communication”, SN Applied Sciences. Springer International Publishing, vol.1, no.9, 2019. [34] Dakulagi, V. and Bakhar, M. “Advances in smart antenna systems for wireless communication”, Wireless Personal Communications. Springer US, vol.1, no.2, pp. 931–957, 2020. [35] M. U. Shahid, A. Rehman, M. Mukhtar, and M. Nauman, “Analysis of fixed beamforming algorithms for smart antenna systems”, vol. 14, no. 5, pp. 110–116, 2020. [36] Hailu, D. H., “Adaptive antennas array algorithms and their impact on code division multiple access system (CDMA)”, M.Sc. Thesis, Addis Ababa University, March 2003. [37] Gross, F. B., “Smart antenna for wireless communication”, McGrawHill, Inc, USA, 2005. [38] Balanis, C. A., “Antenna theory: analysis and design” John Wiley & Sons,INC., USA, 3rd Edition, 2005. [39] Zhigang Rong, “Simulation of adaptive array algorithms for CDMA systems”, M.Sc. Thesis, Virginia Polytechnic Institute and State University, Virginia, P.143, September 1996. [40] Balaem Salem, S.K. Tiong, S.P. Koh, et al, “Avoiding self nulling by using Linear constraint minimum variance beamforming in smart antenna”, Research Journal of Applied Sciences, Engineering and Technology, Vol. 5, No.12, pp. 3435-3443, 2013. [41] Allen, B., and Ghavami, M., “Adaptive array systems: fundamentals and applications”, John Wiley & Sons, Ltd, 2005. [42] Savitri Katariya, “A survey on smart antenna system”, International Journal of Electronics & Communication Technology (IJECT), Vol. 2, No.3, pp.123-126, 2011. 90 References [43] G.C Nwalozie, V.N Okorogu, and S.S Maduadichie, et al, “A simple comparative evaluation of adaptive beam forming algorithms ”, International Journal of Engineering and Innovative Technology (IJEIT), Vol.2, No. 7, pp.417-424, 2013.. [44] Schreiber, R., “Implementation of adaptive array algorithms”, IEEE Transactions on Acoustics, Speech, And Signal Processing, Vol. Assp34, No. 5, pp.1038-1046, October 1986. [45] Anurag Shivam Prasad, Sandeep Vasudevan, Selvalakshmi R., Sree Ram K., Subhashini G., Sujitha S, and Sabarish Narayanan B., “Analysis of adaptive algorithms for digital beamforming in smart antennas”, IEEE-International Conference on Recent Trends in Information Technology, MIT, Anna University, Chennai. June 3-5, pp.64-71, 2011. [46] S. Banerjee and V. V. Dwivedi, “Performance analysis of adaptive beamforming using particle swarm optimization,” 11th Int. Conf. Ind. Inf. Syst. ICIIS 2016 - Conf. Proc., vol. 2018-Janua, no. April 2017, pp. 242–246, 2018. [47] M. A. Panduro, A. L. Mendez, R. Dominguez, and G. Romero, “Design of non-uniform circular antenna arrays for side lobe reduction using the method of genetic algorithms,” AEU - Int. J. Electron. Commun., vol. 60, no. 10, pp. 713–717, 2006. [48] Petrus, P., “Novel adaptive array algorithms and their impact on cellular system capacity”, Ph.D. Thesis, Virginia Polytechnic Institute and State University, Virginia, March 1997. 91 List of Publication • A. A. K. Qasim and A. H. Sallomi, “Optimisation of Adaptive Antenna Array Performance Using Particle Swarm Algorithm,” 1st Int. Sci. Conf. Comput. Appl. Sci. CAS 2019, pp. 137–142, 2019. • A. A. Qasim, “Performance improvement for smart antenna system least square beamforming algorithm,” pp. 155–165, 2020. • A. A.-K. Qasim, “Analysis of Adaptive Antenna System using Rabid Euclidean Direction Search (REDS) Method for Different Antenna Geometry,” Bull. Electr. Eng. Informatics, vol. 10, no. 2, 2020. • Adheed Hassan Sallomi , Aseel Abdul-karim Qasim ‘’Constant modulus algorithm (CMA) improvement by variable step size and blocking technique”, International Journal of Interactive Mobile Technologies (IJIM)(ISSN: 1865-7923, SCOPUS Indexed).(Acceptance) • Aseel Abdul-Karim Qasim, Adheed Hassan Sallomi ‘’ Design and Analysis of Phased Array System by MATLAB Toolbox”, published in the Al Kitab Journal for pure sciences Vol.4, Issue 1, June 2020. (Acceptance) جمهورية العراق وزارة التعليم العالي و البحث العلمي الجامعة المستنصرية كلية الهندسة قسم الهندسة الكهربائية تطبيق خوارزميات مختلفة في تشكيل الحزمة التكيفية في نظام الالسلكي رسالة مقدمة الى قسم الهندسة الكهربائية،كلية الهندسة،الجامعة المستنصرية كجزء من متطلبات نيل درجة ماجستير علوم في الهندسة الكهربائية/ االلكترونيك واالتصاالت من قبل: اسيل عبد الكريم قاسم (بكالوريوس هندسة كهربائية) باشراف أ.د .عضيد حسن سلومي ربيع أول 1442ه تشرين ثاني 2020م الخالصة هناك اختالف معين عند مقارنة أنواع مختلفة من الخوارزميات مثل Least Mean Square ) (LMSوالتي تستخدم إلى حد كبير بسبب التعقيد المنخفض للحسابات وسهولة التنفيذ .وغيرها من الخوارزميات التي تعتمد على مربع الخطأ ( ، )algorithm that content step size factorمثل (( ، ) Recursive Least Squares)RLSو ( ، )Gradient)CG(Conjugateو())REDS ، )Rapid Euclidean Direction Searchوتكون أسرع ولها مربع خطأ صغير مقارنا" ضا مثل الخوارزمية العمياء مثل بخوارزمية .LMSسيتم تقديم بعض الخوارزميات األخرى أي ً خوارزمية () ) Constant Modulus Algorithm)CMAوخوارزمية (((Sample )SMI ، Matrix Inversionكما وتوجد خوارزميات التحسين مثل خوارزمية (( Particle Swarm ضا اختبارات مختلفة لعدد من .)Optimization algorithm (PSOسوف تناقش هذه الرسالة أي ً مصفوفات الهوائيات ذات الترتيب الهندسي مثل )Planner ،Circular (CA) ،Linear (LA ) ، (PAوصفيف هوائي مكعب ) )Cubeومقارنة األداء هذه االشكال عن طريق تغيير بعض خصائص النظام التكيفي. في البداية ،درست هذه الرسالة تأثيرات تغيير عدد العناصر ،كما تدرس مسافة التباعد بين العناصر تأثير تغيير هندسة المصفوفة على أداء النظام الذكي .سيتم تطبيق التحسين األول في هذه الرسالة على الخوارزمية المربعة ،حيث تُظهر خوارزمية LMSSIأن سرعة تقارب LMSتقل من 80إلى تكرارا ،وتحقق ايضا" انخفاضا" ملحوضا" في قيمة .MSEسيكون التطوير الثاني هو حوالي20 ً SIR-LMSوالذي حقق التكيف للمصفوفة الهوائية عند نقطة البداية للنظام .يدرس تعزيز خوارزمية CMAفي اتجاهين ،أوالً عن طريق تطبيق خوارزمية حجم الخطوة المتغيرة ICMAوثانيًا خوارزمية ، BCMAتوضح النتيجة أن ICMAيحسن عدد التكرار من 50لـ CMAالتقليدية إلى 23و 6تكرارات لـ BCMAمع الحفاظ على نمط الحزمة االتجاهية لإلشارات المطلوبة وإبطالها تما ًما في اتجاهات االستدالل. وتم مناقشة خوارزمية PSOفي ثالثة أجزاء ،أوالً للتحسين متجه الوزن الذي تظهره النتيجة بالمقارنة مع Gaعند 8عناصر يتحسن الوضع مع 2.3ديسيبل في SLLو ° 3.77في HPBWبأستخدم .PSOكان الجزء الثاني هو تحسين التباعد بين العناصر ،وهذا يمثل أهم منجز لتوفير نظا ًما اقتصاديًا بفاعلية استخدام نفس العدد أو أقل من العناصر إلى جانب تقليل حجم المصفوفة ،المرحلة التالية ستكون في تحسين ) PSO (IPSOالذي يجمع بين تحسين الوزن ,سيكون الجزء الثالث هو دراسة تأثير االتداخل على نمط اإلشعاع لعدد مختلف من العناصر ولمسافات التباعد المختلفة. وستتم تطبيق خوارزمية REDSمع هندسيات مصفوفة مختلفة وستتم المحاكاة باستخدام ضوضاء ( )Additive White Gaussian Noise (AWGNو( Rayleigh fading multipath .)channelsيمتلك REDSمعدل تقارب بين 8إلى 5تكرارات وهو أسرع مقارنة بخوارزمية تكرارا تكرارا لبدء عمليات التكيف وأكثر من 25و 50 LMSالتي تحتاج إلى أكثر من 50 ً ً لخوارزمية NLMSو CMAوما إلى ذلك .إخراج الصفيف PAو CAيظهر جيدا" لإلشارة المطلوبة ،وكذلك MSEمن PAحول ( )1.2625e-04وهو األفضل على اإلطالق ,تُظهر نتيجة تكرارا ،يتقارب هوائي CAو PA محاكاة قناة Rayleighأن هوائي LAيتقارب بعد أكثر من 30 ً بعد 20وحوالي أقل من 10تكرارات ،كما يقدم صفيف PAصفرا ً أعمق والذي يبلغ حوالي 90 ديسيبل تجاه المسبب للتداخل مع توفير 62-ديسيبل و 48-ديسيبل قيمة خالية أعمق بأستخدام مصفوفة CAو LAعلى التوالي. وتمت ايظا" دراسة معدل الخطأ ( )BERونسبة اإلشارة إلى الضوضاء ( )SNRللمصفوفة الخطي والدائري ويوضح أن معدل الخطأ في البتات ( )BERمتقارب تقريبًا لكال المصفوفتين في نفس العدد من العناصر ،في حين يُظهر المعدل النشط ( )SNRوجود اختالف بسيط في منحنى SNRفي المصفوفة الخطي والدائري و من الواضح أن منحنى SNRللنظام اإلجمالي يزيد مع عدد العناصر المستخدمة في الصفيف بغض النظر عن هندسة مصفوفة الهوائي. و قد تم استخدام الحقيبه البرمجية ( )MATLABطراز( )2017aفي تنفيذ عمليات المحاكات في هذه الرسالة.