LDPC Code Design for ISI Channels via Fitted Detector EXIT Analysis

Abstract

Channels with memory commonly introduce intersymbol interference (ISI), which complicates reliable coded transmission. For such channels, this paper develops a fitted extrinsic-information-transfer (EXIT) framework for LDPC code design with iterative detection and decoding. The soft-input soft-output (SISO) detector is treated as a transfer module: its input-output mutual-information relation is sampled offline, approximated by a smooth surrogate, and incorporated into EXIT chart analysis. The fitted function avoids repeated detector simulations for threshold evaluation and code comparison, and does not rely on detector-specific analytical EXIT derivations. Using the proposed framework, we compare LDPC degree profiles under channels with and without ISI, showing that degree profiles preferred for memoryless channels may not remain favorable for ISI channels, where relatively sparse parity-check matrices tend to perform better. Motivated by this observation, we propose an EXIT-guided masking method that selectively removes admissible nonzero entries from 5G NR LDPC base graphs while preserving their lifting, encoding, and decoding structures. Numerical results show that the enhanced 5G NR LDPC codes achieve lower EXIT thresholds and better finite-length BER performance than the standard 5G NR LDPC codes over the considered ISI channels, with coding gains of about 0.5 dB.