Study On Multi-user Detection Based On Support Vector Machine

CDMA system is superior to other multiple access methods because of greater efficiency capability, soft handover, anti-multipath fading, higher spectrum efficiency, supporting multiple-rate communication, etc. it has become the most important technology in the third generation. But its inherent multiple access interference (MAI) and "near-far effects" problem have severely restrained the improvement of system performance. Multi-user detection (MUD) is an efficient method suppressed MAI and "near-far effects", it views the MAI as a useful resource and makes full use of the relationship among users to increase the detection performance. So the MUD is one of the key technology in 3G mobile communication system. Support vector machine (SVM) is a novel machine learning method developed from the middle of 1990s. With limited samples, SVM has stronger ability of generalization in comparison with existing machine learning algorithm. The SVM is applied to the MUD in this paper, main work of the dissertation can be summarized as follows: After an introduction of the CDMA system model, the performance of several representative multi-user detectors is analyzed through MATLAB simulation, the received signal- and system parameters that have to be estimated or known by the receiver and the computational complexity of these detectors are discussed. The SVM MUD is studied more deeply, we analyze system models of MUD based on SVM in the AWGN channels and multipath channels, and simulate their performance, the experimental results show that their performance is superior to that of linear MMSE MUD. A kind of multi-user detector based on LS-SVM is proposed to solve the complexity of SVM MUD and improve the speed of SVM, It is shown that the method is effective in theoretical analysis and computer simulations.