在3G移动通信CDMA系统中，主要的干扰有多址干扰、多径衰落、“远-近”效应、噪声和窄带干扰。多用户检测技术是CDMA通信系统的一个关键技术，由于CDMA系统是一个干扰受限系统，随着干扰的消除，系统的容量和性能都将会得到提高。多用户检测技术将各用户发送的信号做联合检测，此方式缓解了远近效应问题，有效地消除了多址干扰，改善了系统性能，提高了系统容量。但是，其实现复杂度高，一般只用于基站，而对于移动台一般采用盲多用户检测。在神经网络理论迅速发展的同时，人们也加强了其在通信领域应用的研究。近年来，基于神经网络的盲多用户检测结合了盲多用户检测对已知信息量需求少和神经网络运算速度快，并行处理能力强的优点，而成为研究的热点。 本文对无线多径衰落信道条件下的盲多用户检测的两种算法进行了研究和分析。第一种是将BP神经网络和MMSE准则结合起来的BP-MMSE算法。第二种是将改进的遗传算法与BP神经网络相结合并应用于盲多用户检测。仿真结果表明第一种算法相对于单纯的MMSE准则的盲多用户检测算法具有更好的收敛特性，但是其抗远近效应能力却有待进一步提高。而第二种是将遗传算法全局搜索最优和传统的BP神经网络模型局部寻优结合起来，取长补短，既可以减小遗传算法的搜索空间、提高搜索效率，又可以较容易地收敛到最优解，使算法具有一定的实用性。且仿真结果表明：基于遗传算法的BP神经网络盲多用户检测算法其收敛性相对于第一种算法收敛性更好，系统误码率更低，并且在某种程度上能够有效地抑制多址干扰，抗远近效应的能力也更强。

In the third generation(3G) mobile communication Code Division Multiple Access(CDMA) system,the main interference are multiple access interference (MAI),multi-path fading,”near-far” effect(NFE),the noise in the system and some narrow-band interference.Multiuser detection is a key technique in CDMA communications system.Because CDMA system is a kind of system whose performance is limited by the interference.The capability and performance of the system will be improved with the elimination of the interference.The multiuser detection can jointly detect the signals transmitted by each user,this method decrease the “near-far” effect,cancel the MAI and increase the system capacity.However, the realization is high complexity, it is generally used in the base station, the detection for mobile stations generally uses blind multiuser detection into. With the rapid development of neural networks, pepole also strengthenes research of its applications in communications area. In recent years, blind multiuser detection based neural networks combined with blind multiuser detection of small amount of known information and neural networks of rapidly computing speed, the advantages of parallel processing ability, become a research hotspot. In this paper, the two algorithms of blind multiuser detection in infinite multi-path fading channel are studied and analyzed. The first is the BP-MMSE algorithm which combined the BP neural network and the MMSE criterion. The second is the combination of the improved genetic algorithm and BP neural networks and applies blind multiuser detection. Simulation results show that the first algorithm has better convergence properties relative to the blind multiuser detection algorithm of simple MMSE criteria, but its NFE capacity need to be further improved. The second combine genetic algorithms globally searching optimization and traditional BP neural network model locally searching optimization, each other, not only can reduce the search space of genetic algorithms to improve search efficiency, but also easily converge to the optimal Solutions to make the algorithm is practical. And simulation results show that: Based on genetic algorithm, BP neural network blind multiuser detection algorithm whose convergence compared to the first convergence is better has lower bit error rate. To some extent, it is able to effectively suppress the multiple access interference and ability of resisting near-far is more stong.