Abstract: Gaussian approximation models the message density in message passing decoders as Gaussian distribution.In this paper, irregular repeat accumulate (IRA) codes over the mixed white Gaussian noise/erasure channels are designed using Gaussian approximation.Recursive equations of density evolution and analytic expressions of bit error rate are derived.The analytic results not only provide a way to predict asymptotic performance of IRA codes but also help us design good IRA codes over the mixed channel.Comparison between simulation results and thresholds computed from analytic results shows that predication from the theoretical results is accurate enough.The optimized IRA codes designed for the mixed channels have better performance than that of IRA codes designed for AWGN channels when the designed erasure probability is close to the actual probability.

Key words: irregular repeat-accumulate codes, Gaussian noise, random erasure, Gaussian approximation