A New Multi-User Detection Method Based On Random Neural Network

In the Code-Division-Multi-Access (CDMA) communication system, all consumers share the same physical channel. Because of the multi-path delay , the signals from different consumers can not reach the receiver at the same time, so that the received signals have relative time delay. As a result, the spread codes between different consumers are not orthogonal at the receiver, so there comes the Multi-Access Interference (MAI). On the other hand, different signals have different power due to their different distances from the base station. The stronger signals nearby restrain the weaker signals in the distance, which is called Near-Far Effect (NFE). The existence of these two problems makes the performance of the traditional Match Filter (MF) Receiver very bad, and so does the performance of the CDMA system.In order to solve the problem, Multi-User Detection (MUD) technology is proposed. This technology is a signal detection method which simultaneously considers the signals from all consumers or part of the consumers in the same channel, eliminates the influence from other consumers, and detects the signals from all consumers or part of the consumers at the same time. In a short word, MUD fully considers the structure of spread codes and interference information which causes the MAI to eliminate the MAI. At present, MUD has become one of the key technologies which have anti-MAI ability and anti-NFE ability. As the development of MUD, it has become more and more mature. But all types of MUD have some defects such as complicated computation and slow convergence speed, which make MUD can not be used in the practical CDMA system.In recent years, new MUD algorithms combined with the science in other fields have showed their unique advantages and have become new focuses in MUD research field, which include the MUD algorithm based on the artificial neural networks (ANN).On the basis of MUD research, this thesis proposes a new MUD method based on the random neural networks (RNN), which can realize the MUD with slow bit error rate, easy computation and fast convergence speed.The main work done by this thesis:1. The model of CDMA communication system is given, and the cause of MAI is analyzed by mathematical formula. It can help to know necessity of MUD and the environment of this thesis.2. The principles and classification of MUD are introduced. Several typical MUD algorithms are analyzed and compared, and their disadvantages are given to figure out the new research directions.3. The MUD methods based on ANN are analyzed. The merits and defects of BP algorithm, RBF network and Hopfield network are discussed.4. A new MUD method based on the RNN is proposed in this thesis. It not only uses the RNN to realize MUD, but also introduces the Mean Field Theory Annealing (MFTA) into the new method which owns the merits of all above. The new method's performance simulations are presented at the end of this thesis.