Study Of Multiuser Detection Based On Hopfield Neural Network In CDMA System

CDMA is the basic technology of the three main standards of the third generation mobile communication, with the increasing users, multiple access interfere has become the most important interfere of CDMA.As one of the basic technologies of multiple access interfere elimination of the 3G mobile communication, the main idea is how to extract useful signal from interfere signals starting with the design of the receiver. Using all kinds of information including interfering users (spreading code, time information, amplitude, character parameters) to eliminate or restrain multiple access interfere. This method could succeed in increasing system capacity, and improving system performance.Commonly,the disadvantage of large computation and slow computing velocity consist in most multiuser detection technologies. As to Neural Networks is specially applicable to solve optimization problem for the high speed of optimization and the parallel processing and the essence of the multiuser detection is the method of combination optimization.This article studies the multiuser detection based on Hopfield neural network.The main contents including several aspects are as follows:1. Several multiuser detection methods were detailed analyzed and compared, which are Conventional Detector, Optimal Multiuser Detector, Decorrelation detector and Minimum Mean Square Error detector.2. The multiuser detection based on Hopfield neural networks was deeply studied. That the multiuser detection's optimization is used as the object function, the minimal value of the object function is the optimal solution which is corresponding to the energy function of Hopfield neural networks. Computer simulation results show that the detection based on genetic algorithm and HNN have better performance than the previous several detection methods which relate to the non-linear of neural network.3.we proposed two kinds of modified Hopfield neural network multiuser detection to solve the problem that the Hopfield neural network multiuser detection energy function is easy to run into local minimal value.The first one is to optimized Hopfield neural network multiuser detection, which uses optimization theory to add a penalty factor and updates the continual weight; the other is stochastic Hopfield neural network multiuser detection, which introduces the stochastic perturbation, and uses the control parameters to change the converging effect.