On sparse graph coding for coherent and noncoherent demodulation

Abstract

In this paper, we consider a bit-interleaved coded modulation scheme (BICM) composed of an error correcting code serially concatenated with a M-ary non linear modulation with memory. We first compare demodulation strategies for both the coherent and the non coherent cases. Then, we perform an asymptotic analysis and try to show that the design of coding schemes performing well for both the coherent and the non coherent regimes should be done carefully when considering sparse graph based codes such as low-density parity-check (LDPC) codes. It will be shown that optimized coding schemes for the non coherent setting can perform fairly well when using coherent demodulation, while on the contrary, optimized coding schemes for the coherent setting may lead to "non stable" coding schemes in the non coherent setting.

Short Non-Binary Low-Density Parity-Check Codes for Phase Noise Channels

“…The detector is concatenated with the decoder of various binary codes from the literature to form a turbo detection scheme. Binary low-density parity-check (LDPC) code design for continuous phase frequency shift keying modulation and a blockwise non-coherent additive white Gaussian noise (AWGN) channel was performed in [13] for a bit-interleaved coded modulation scheme. In [14], a code design for binary codes using differential modulation was considered.…”

Precoding for Non-Coherent Detection of Continuous Phase Modulations

“…When BICM CPM schemes are considered, this convergence enables the use of efficient turbo-detection schemes involving simple outer coding schemes based on convolutional codes such as in the DVB-RCS2 standard [8] or based on sparse graph based codes [11], [12]. For the non-coherent case, it has been pointed out by [9] that non-coherent EXIT curves do not converge to point (1, 1) contrary to the coherent case. As a result, non-coherent iterative detection associated with a classical convolutional code or sparse graph based codes that have been optimized for the coherent case would tend to have error floors and/or capacity penalty.…”