Study On Blind Multiuser Detection Algorithm Based On Feed-forward Neural Network

In the CDMA communication system, spreading codes of users are not completely orthogonal. Then, multi-user interference and 'far-near' effect in CDMA happen. The capacity of CDMA system is greatly affected by multi-user interference and 'far-near' effect. Multi-user interference and 'far-near' effect can not be avoided, but multi-user detection can be used to restrain multi-user interference and 'far-near' effect. The performance and capacity of the system can be improved. Neural network possesses many merits such as quick operating speed and parallel processing ability and so on. Neural network and blind multi-user detection are combined, and the new blind multi-user detection based on neural network was proposed. Recently, neural network blind multi-user detection has become a hot topic.This article can be summarized as follows:1. The purpose and significance of multi-user detection was briefly discussed; Implement method and its development of blind multi-user detection algorithm was overviewed, and its characteristics were analyzed. The basic principles and many common algorithms of blind multi-user detection algorithm were used to be analyzed. At the same time, the features and structure of the neural network can be analyzed. The Solving method of constraint problems is discussed.2. A blind multi-user detection algorithm based on the constant modulus feed-forward neural network is proposed. Punishment function and augmented Lagrange function are used to optimize the cost function and the optimal solution of feed-forward neural network is obtained. Then a new blind multi-user detection based on feed-forward neural network algorithm was proposed. The simulation shows that the new multi-user detection algorithm improves bit error rate, convergence speed and tracking ability performance.3. A feed-forward neural network blind multi-user detection algorithm based on the minimum kurtosis criteria was proposed. According to the characteristics of higher order cumulants, the cost function based on the minimum kurtosis criteria is founded. Constraint condition ensures that the desired signal can be obtained. The constraint cost function is optimized by the two above optimal methods. The feed-forward neural network blind multi-user detection algorithm based on minimum kurtosis criteria is realized. Simulations show that the new algorithm is superior to the traditional linear constraints algorithm in BER and convergence speed.