Non-linear graph-based codes for joint source-channel coding

We study the behavior of a new family of nonlinear graph-based codes, previously introduced for compression of asymmetric binary memoryless sources, for the joint source-channel coding scenario in which the codewords are transmitted through an additive white Gaussian noise channel. We focus on low entropy sources (with high redundancy) and compression rates. Monte Carlo simulation and density evolution results show that the proposed family, with a regular and simple parametrization of the degree profiles, outperforms linear codes.