Title:A Low Density Lattice Decoder via Non-Parametric Belief Propagation

Abstract: The recent work of Sommer, Feder and Shalvi presented a new family of codes, called low density lattice codes (LDLC), which can be decoded efficiently and approach the capacity of the AWGN channel. A linear time iterative decoding scheme which is based on a message-passing formulation on a factor graph is given.
In this work, we show that the LDLC decoder is closely related to the Gaussian belief propagation algorithm as well as the non-parametric belief propagation algorithm. By bridging those two different domains, we are able to use previous research results about Gaussian belief propagation to extend the original LDLC formulation and convergence analysis to support more general decoding matrices than the original version.
We also show how to apply efficient computational constructs derived for non-parametric belief propagation to speed up the decoding algorithm in practice.
As an application to our new theoretical results, we show that the extended LDLC decoder can be used for solving the multiuser detection problem.