Abstract
When the iterative Belief Propagation (BP) decoding algorithm is applied to low-density parity-check (LDPC) codes, the convergence speed in the waterfall region and the error floor in the high SNR region are two of the most important metrics of performance measurement. Both can be significantly improved using the scheduling techniques proposed in this thesis. Fast convergence can be achieved using informed dynamic scheduling (IDS) since the important decoding messages have more opportunities of being updated. However, greedy groups and silent variable nodes can be observed in many IDS decoders, and these obstruct the decoders from providing a satisfactory convergence error-rate performance. In this thesis, Q-RBP (Quota-based residual BP) and SVNF-BP (Silent-Variable-Node-Free BP) are proposed in order to suppress greedy groups and silent variable nodes, respectively. Since the number of updates for each message is limited by the proposed Q-RBP schedule, the message updates that would potentially form a greedy group are forced to release the occupied computation resources. On the other hand, following the SVNF-RBP schedule, the messages associated with all variable nodes are arranged to have an equal chance of contributing their intrinsic messages, and hence the silent variable nodes are totally avoided. Both the Q-RBP and SVNF-RBP schedules can provide a significant improvement in decoding performance when compared to other IDS decoders presented in the previous literature. Additional pre-computations are required in most of IDS decoders, including Q-RBP and the SVNF-RBP schedules, so as to order customized decoding sequences for individually received codewords. However, rather than arranging the decoding schedule based on each received codeword, the proposed maximum mutual information increase (M^2I^2)-based algorithm determines the schedule based on maximizing the increase in mutual information. A pre-determined and fixed decoding schedule can be applied to all codewords, and the decoding convergence can be accelerated without increasing the decoding complexity. Moreover, when multiple distinct schedules are applied to a single codeword to create schedule diversity, the error floor can be significantly lowered without requiring any knowledge of trapping sets. When the proposed decoding schedules are applied to punctured LDPC codes, the benefit in increasing convergence speed can be more significant compared to dedicated codes. If rate-compatible (RC)-LDPC codes constructed based on puncturing are considered, the $\mathrm{M^2I^2}$-based algorithm can be used to arrange fixed schedules for incremental decoding, and further reduce the required number of iterations. With the assistance of the proposed decoding schedules, the puncture-based RC-LDPC codes can be a potential solution for delay-sensitive HARQ (Hybrid-Automatic Repeat reQuest) applications.