Refined Density Evolution Analysis of LDPC Codes for Successive Interference Cancellation

Successive interference cancellation (SIC) is a fundamental decoding technique for Gaussian multiple access channels (GMAC). In SIC, transmit signals of each user are separately decoded. In this paper, we analyze an asymptotic decoding threshold of SIC decoding for practical low-density parity-check (LDPC) codes over
$N$
-user GMAC. Conventionally, the decoding thresholds are evaluated based on the so-called channel approximation (CA) in which the channel model of each decoding stage of a SIC decoder is approximated by simple Gaussian noise channels resulting in errors of decoding thresholds. To avoid this, we propose a refined density evolution (DE) analysis called DE-SIC which uses mixed-Gaussian noise channels corresponding to each decoding stage. We demonstrate DE-SIC by solving a received power optimization problem in GMAC and comparing it to the conventional DE analysis with CA. The results show that DE-SIC accurately evaluates the decoding thresholds whereas the conventional analysis underestimates the thresholds.