Multiuser Detection Based On Intelligent Algorithm

Code-division multiple-access is one of several methods of multiplexing that has taken asignificant role in cellular and personal communications. Because the spread spectrum codeof the users is not strictly orthogonal. This is known in CDMA communications asmultiple-access interference (MAI). Although the MAI generated by several users is small, itbecomes the main interference because of the increase in the number of users and the increaseof signal power, thus become a bottleneck restricting the development of CDMAcommunication system. So, how to effectively control this kind of interference has becomethe most important issue that need to be addressed to improve system performance andcapacity. Multiuser detection is an effective way to solve the problem of MAI. It does notregard MAI as interference noise, but as useful information to take advantage of. It can makefull use of the association between users to detect, thus improve the detection performance.This paper is focused on intelligent algorithm in multi-user detection. The optimummultiuser detector (OMD) has good performance in canceling the multi-access interferenceand near-far effect, it is impractical to real word implementation for its computationalcomplexity which is exponentially increasing with the number of users. Since the design ofoptimum multiuser detector can be modeled as a NP-complete optimization problem, theintelligent algorithm can be used to design multiuser detector. Therefore, to explore in depththe combine of intelligent algorithm and multiuser detection, to find the method with lowcomplexity which can restrain the multi-access interference and near-far effect are of greatinterest.In this paper, we start from the CDMA system, study several typical multiuser detectiontechnologies and analyze its strengths and weaknesses. Then we introduce genetic algorithmand Hopfield neural network, and proposed a multiuser detector based on genetic algorithmand Hopfield neural network considering the advantages in solving combinatorialoptimization problems of the two algorithms. The genetic algorithm provides a good initialsolution to neural network, then neural network search for the local optimization by gradientdescent mechanism.The disadvantages of the GA are slow convergence to a good near-optimum solution andeasily to converge at local optima. We combine GA and SA, and design self-adaptation crossover probability and mutation probability according to simulated annealing ideas, inorder to confine diversity of population and speed up the convergence velocity.We present another intelligent algorithm called tabu search, which use the traditionaldetector output as the initial solution in search, use the points with Hamming distance with thecurrent solution is1as neighborhood and searched solutions constitute a contraindicationtable. This method can achieve good performance. Because the choice of initial solution hasgreat impact on the performance of tabu search, we use the best chromosome in the lastgeneration of SAGA as the initial solution of TS. Computer simulations show that theproposed detector’s BER curve is very close with optimum multiuser detector’s and it has ahigher convergence speed.