Reducing the Complexity of the Layer Scheduled LDPC Decoder Based on the Information Bottleneck Method

The complexity of belief propagation algorithms for decoding LDPC codes can be significantly reduced by storing the precomputed sum of messages in variable nodes. This optimization is particularly useful for layered schedule decoders, where it also simplifies the hardware implementation. To reduce the requirement for the amount of information to be processed during decoding, the information bottleneck method is used, which reduces the bit width of all updated messages. However, under this approach, re-calculation of the sum when excluding one of messages becomes complicated. The paper is devoted to development of an algorithm for construction of a discrete binary function corresponding to such subtraction. When using this algorithm, the number of stored and used lookup tables for variable nodes is reduced.