Based On Incremental Support And Non-single-point Fuzzy Support Vector Machines, Signal Detection And Its Application In The Mc Cdma System Uplink

Multicarrier code division multiple access(MC CDMA), a fusion of OFDM(orthogonal frequency division multiplexing) and CDMA(code division multiple access), has become a hot topic in wireless communications as soon as it was proposed in the early 1990's. As performance of GSM is more and more limited by system capacity, audio distortion, power radiation and slow transmitting speed, MC CDMA is becoming one of the most competitive techniques in 3G mobile communication system of IMT-2000.However, frequency offset, doppler shift induced by mechanical movement and changes of phase and envelope generated by multipath scattering, introduce inter symbol interference(ISI), inter channel interference(ICr) and multi-address interference(MAI) to received signal. Especially in uplink scenario, users' time offsets, channel's non-linearity and doppler shift destroy the orthogonality of spreading sequence, degrade the uplink performance of MC CDMA system. Linear detection methods used in single carrier spreading system, such as minimum mean square estimation(MMSE), least square estimation(LS), as well as diversity combinater,lose their superiorities in these occasions.Therefore in this thesis, signal detector based on support vector machine(SVM) is employed in uplink MC CDMA as signal detector. First, simulations are conduct to explore influence of RBF kernel's parameters on SVM detector's BER performance, and comparison is made between SVM detector without sample screening and that which selects training samples according to KKT conditions, and assigns exponential damping error penalties to old and new training samples repectively. Second, as the SVM detector with sample screening mechanism has such defects as inaccuracy, inflexibleness, and inefficiency, incremental support vector machine(ISVM) is proposed. Simulations show that this method not only out performs linear detectors, but also more efficiently than the screens policy with KKT condition. Finally, improvement is made in RBF kernel function to improve BER performance when SNR is low. Concept of fuzzy system is introduced and a nonsingleton fuzzy RBF kernel is derived to counteract the negative impact of burst error. Physical meaning is discussed and adaptive updating algorithm of the newly introduced parameter is given. Relation between the old and new RBF kernels is also analyzed. Simulation results show that nonsingleton fuzzy RBF kernel enhances performance when SNR is low.