The Research Of Multiuser Detection Based On Neural Network

In the Code Division Multiple Access (CDMA) system, because of stochastic access of users and the characteristic of real-time change of channel which make that it is nonorthogonal between spreading code, there is MAI (multiple access interference), which limits the performance and the capacity of the system. Among the many methods of multiuser detection, the technique of multiuser detection based on MMSE criteria and MOE criteria has been a hot topic of research in recent years because they have better performance and lower complexity. Unfortunately, the complexity of MMSE and MOE multiuser detection is also higher to practical application, which holds back their real-time realization. In this paper, we get real-time detector by using the ability of high parallel process and looking for optimal point of neural network. So the main work in this paper has as follows:Firstly, the principle and the class of multiuser detection are discussed. The common arithmetic and their improved methods, especially the hot topic in recent are given.Secondly, some neural networks improved from Hopfield neural network, such as wavelet neural network and stochastic neural network, are researched. In addition to this, the common process to resolve problems using Hopfield neural network is also given.Thirdly, a novel MMSE multiuser detection based on neural network in the multipath environment is produced in this paper. The neural network has many merits such as high parallel process and fast process, so the neural network can be used to reduce the complexity of multiuser detection. But, in the multi-path environment, the signal at the receiver is complex number because of the effect of the multipath channel and can not been implemented directly by neural network. To resolve this problem, in this paper the real part and imaginary part of the signal at the receiver is separated, so the cost function based on the MMSE criteria become real number, and by associating with the energy function of Hopfield neural network, a MMSE detector based on neural network is designed. To validate the performance, the simulation of his performance is given. The result of simulation experiment shows that this detector almost has the same performance with the up limitation of the performance of MMSE in theory.Fourthly, the neural network blind multiuser detection based on channel estimation in the multipath channel has been introduced. Firstly the signal is put into the MOE blind multiuser detector, and then the channel coefficient is estimated using the output of detector, at last the information is obtained by combination and judgment, so the signal detection is combined with the technique of channel estimation. The simulation validates the performance of this detector when channel estimation has error and the number of path in channel is estimated wrong. The result of experiment shows that this detector has a better performance under both conditions compared with other detector.