Equalization Algorithms For Two Dimensional Intersymbol Interference Channels
MetadataShow full item record
This dissertation presents several novel iterative soft decision feedback equalization algorithms for detection of binary-valued two-dimensional images corrupted by 2D intersymbol interference (ISI) and additive white Gaussian noise (AWGN). These algorithms exchange weighted soft extrinsic information between maximum-a-posteriori (MAP) detectors employing different row-column or zigzag scan directions. We first use an independent assumption on the a priori probabilities (APP) used in the computation of the BCJR algorithm, where the extrinsic information transmission (EXIT) charts are used to speed up the optimization of the weight and iteration schemes. Simulation results for the 2 × 2 averaging mask channel show that, at low signal-to- noise ratios (SNR), the new zigzag algorithm gains about 1 dB over previous iterative row-column soft decision feedback algorithm and over a separable-mask algorithm, two of the best previously published schemes. When the zigzag algorithm is serially concatenated with the row-column algorithm, the concatenated system performs better than four of the best previously published algorithms. Based on the definition of the BCJR algorithm, we remove the independence assumption and instead use joint estimation, where we introduce the new two-dimensional extrinsic information flow (TEXIF) chart, which is used to explore the subtraction issue of the extrinsic information for the iterative detection process. For the redesigned joint iterative row-column soft decision feedback algorithm, the block (BLK) algorithm, has the performance within 0.3 dB from the maximum likelihood bound. To address the increased computational and storage complexity introduced by the joint estimation, we develop a simplified block (SBLK) algorithm with almost the same performance as BLK algorithm. The redesigned iterative soft decision feedback zigzag algorithm using joint estimation, the block zigzag (BLKZ) algorithm, provides more than 2 dB SNR gain over the ISDFZA on the 2 × 2 averaging mask channel, and 1 dB for the 3 × 3 averaging mask at high SNRs. When the BLKZ algorithm is concatenated with the joint IRCSDFA, the concatenated system achieves the ML bound at high SNR with the 2 × 2 averaging mask; with the 3 × 3 averaging mask it comes within 0.2 dB of the ML bound.