Approximate-min* constraint node updating for ldpc code decoding

Abstract

This work introduces a technique for updating messages that originate at the constraint nodes of bi-partite graphs in low-density parity-check codes. The technique computes only two outgoing magnitudes at each constraint node and exhibits no measurable performance loss as compared to exact belief propagation which computes a unique magnitude for each departing edge from a given constraint node. The technique eliminates the need for memory based table look-up in the constraint node processing and has been implemented using only shift, add, and comparison operations. Finite wordlength results for a 'worst-case' code indicate that 6 bit quantization yields performance that is similar to that provided by full floating point computation.