Improvements on Accelerating Iterative Decoding Using Eigenmessages

Abstract

The eigenmessage decoder has been shown to reduce the number of decoding iterations required for LDPC and other iteratively-decoded codes by introducing a degree of nonlocality into the decoding algorithm. In this paper, the multiple loop eigenmessage approach is extended using normalized matrices. Performance is examined via EXIT charts, showing that eigenmessage algorithms have wider channels in the chart. Performance as a function of the girth of the graph is examined, showing the performance to be largely invariant to girth. Finally, performance on matrices denser than typical "low density" parity check matrices is examined, showing that eigenmessage methods perform better than message passing, but still break down as the matrix density increases