Advanced Coding And Modulation For Ultra-wideband And Impulsive Noises

The ever-growing demand for higher quality and faster multimedia content delivery over short distances in home environments drives the quest for higher data rates in wireless personal area networks (WPANs). One of the candidate IEEE 802.15.3a WPAN proposals support data rates up to 480 Mbps by using punctured convolutional codes with quadrature phase shift keying (QPSK) modulation for a multi-band orthogonal frequency-division multiplexing (MB-OFDM) system over ultra wideband (UWB) channels. In the first part of this dissertation, we combine more powerful near-Shannon-limit turbo codes with bandwidth efficient trellis coded modulation, i.e., turbo trellis coded modulation (TTCM), to further improve the data rates up to 1.2 Gbps. A modified iterative decoder for this TTCM coded MB-OFDM system is proposed and its bit error rate performance under various impulsive noises over both Gaussian and UWB channel is extensively investigated, especially in mismatched scenarios. A robust decoder which is immune to noise mismatch is provided based on comparison of impulsive noises in time domain and frequency domain. The accurate estimation of the dynamic noise model could be very difficult or impossible at the receiver, thus a significant performance degradation may occur due to noise mismatch. In the second part of this dissertation, we prove that the minimax decoder in [38], which instead of minimizing the average bit error probability aims at minimizing the worst bit error probability, is optimal and robust to certain noise model with unknown prior probabilities in two and higher dimensions. Besides turbo codes, another kind of error correcting codes which approach the Shannon capacity is low-density parity-check (LDPC) codes. In the last part of this dissertation, iii we extend the density evolution method for sum-product decoding using mismatched noises. We will prove that as long as the true noise type and the estimated noise type used in the decoder are both binary-input memoryless output symmetric channels, the output from mismatched log-likelihood ratio (LLR) computation is also symmetric. We will show the Shannon capacity can be evaluated for mismatched LLR computation and it can be reduced if the mismatched LLR computation is not an one-to-one mapping function. We will derive the Shannon capacity, threshold and stable condition of LDPC codes for mismatched BIAWGN and BIL noise types. The results show that the noise variance estimation errors will not affect the Shannon capacity and stable condition, but the errors do reduce the threshold. The mismatch in noise type will only reduce Shannon capacity when LLR computation is based on BIL.