Two is better than one: Reducing the loss of the window decoder for SC-LDPC codes

In this paper, we consider spatially coupled LDPC codes derived from protographs. In particular, we analyze the performance of the window decoder (WD), which allows reducing the complexity, the memory requirements, and the latency of the flood belief-propagation decoder. We show that the performance degradation of WD is due to the fact that it exploits a single decoding wave instead of two. This has effect both in the ideal case of infinite code length, where it may imply a threshold loss, and in the case of finite length, where it affects the slope of the BER curve in the waterfall region. We show how a forward-backward decoder can reduce such problems at the price of a limited increase of average complexity.