A Study On Blind Equalization For Wireless Communication Channel

In digital communication system, the intersymbol interference(ISI) caused by multipathtransmission and fading causes that the received signals suffer from severe distortion. In order toimprove the communication quality, it is necessary to equalize the received signals to alleviate oreliminate ISI. The traditional adapitive equalization needs periodically sending training sequencesto trace the channel variation, which enhances the reliability of data but reduces the efficiency ofcommunication. Blind equalization technique doesn’t need any training sequences, it improves thebandwidth availability and lows the demand of cooperativity between the transmitter and receiver.Blind equalization technique has been one of the research focus in the field of telecommunication.This paper systematically studies the blind equalization technique applied to the wirelessfading channel. The several improved blind equalization algorithms are proposed. The simulationresults and theoretical analysis are given to demonstrate the availability of new algorithms. Themain works of this paper can be summarized as follows.(1) The dynamic characteristics of wireless fading channel are introduced and the channelmodel is given. Meanwhile, the basic principles of blind equalization are discussed, including thedefinition, ideal equalization conditions, equalization criterion and algorithm performanceevaluating criteria.(2) Some classical blind equalization algorithms are summarized. The fractional spaceequalizer (FSE) is discussed and its good equalization performance is shown. And from the aspectof the error function, we propose improved modified constant modulus algorithm(MCMA) based onconstellation match error(CME-MCMA), improved MCMA based on mean square error decisiondirection(DD-MCMA) and the Stop-and-Go MCMA(SAG-MCMA). In order to overcome thedefect of fixed step size, we give a new varying step size rule replacing fixed step and propose animproved MCMA algorithm based on varying step size(VSS-MCMA). On this basis, we furtherpresent several new hybrid blind equalization algorithms, including DD-CME-MCMA,ME-VSS-MCMA, DD-VSS-MCMA and CME-DD-VSS-MCMA. Besides, we also study the superexponential iteration(SEI) algorithm which converges at a very fast nearly super-exponential rate.Both theoretical analyses and simulation results show that these proposed new improved algorithmscan effectively reduce the residual ISI and accelerate the convergence speed.(3) Blind equalization algorithms using support vector machines(SVM) for wirelesscommunication channel models are analyzed and studied. According to the fact that SVM has a fastconvergence behavior but its calculated amount soars dramatically with the growth of data lengthand the conventional blind equalization algorithm has simple update rule but poor convergence performance and is sensitive to the initialization of weight vector, this paper proposes a novel blindequalization algorithm called SVM-SCA with the organic combination of SVM and square contouralgorithm(SCA), which initializes the weight taps of equalizer by SVM. The simulation resultsshow that SVM equalization has good performance, even when only a small number of datasamples is available, such as in burst data transmission over fast fading channels and theequalization performance of the SVM-SCA has been remarkably improved compared with SCA.(4) The performance of blind equalization algorithms are analyzed from the aspect of thegeometrical structure of cost function. Due to the nonconvex topological structure of its costfunction of the constant-modulus-type algorithms, the existence of undesired local minima andsaddle points in cost function causes ill convergence or slow convergence problems. Therefore,better convergence behavior can be obtained by the use of convex cost functions as they have onlyglobal minima and no saddle points. This paper studies a novel blind equalization algorithm viasubgradient projections which use the convex infinite norm of the equalizer output as the costfunction. Simulation results show that the blind equalization using subgradient projections is lesssensitive to initial point selection and the amount of data required for the training of the equalizer issignificantly lower than most of the existing schemes and a fast convergence speed can be achievedwith a judicious selection of step sizes, thus, this algorithm has desirable equalization performance.