Radial basis function network based multilevel channel equalisation techniques for wireless communications

This thesis investigates the employment of Radial Basis Function (RBF) networks in the context of multilevel channel equalisation. The RBF-based Decision Feedback Equaliser (DFE) was found to outperform the conventional DFE at the cost of an increased computational complexity. The RBF DFE was studied in the context of a wideband Burst-by-Burst (BbB) Adaptive Quadrature Amplitude Modulation (AQAM) scheme, where the modulation modes of no transmission (NOTX), Binary Phase Shift Keying (BPSK), 4-QAM, 16-QAM and 64-QAM were invoked by the transmitter, depending on the prevalent channel quality. The 'short-term BER' of the received burst quantifies the channel quality, which was used as the mode switching criterion in order to switch between different modulation modes. The Bit Per Symbol (BPS) throughput improvement for the proposed AQAM scheme designed for a target Bit Error Rate (BER) of 1% was up to a factor of two in comparison to the fixed constituent modulation modes.The logarithmic version of the RBF equaliser referred to as the Jacobian RBF equaliser was derived, which has a reduced computational complexity. Turbo codes were invoked for improving the BER and BPS performance of the BbB AQAM scheme. The Jacobian RBF equaliser provides a logarithmic-domain output, which can be used to provide soft outputs for the channel decoder. We proposed employing the average magnitude of the Log-Likelihood Ratio (LLR) of the bits in the received burst as the channel quality measure for our adaptive scheme. The system exhibited a better BPS performance, when compared with the uncoded AQAM/RBF DFE system at low to medium channel SNRs and also showed an improved coded BER performance at higher channel SNRs.