IEEE Wireless Communications
AI-Driven Rate-Adaptive LDPC Coding and Decoding
Xiyu Wang, Yukun Cheng, Wei Ren Chen, Lun Li, Bo Ai
Abstract
Artificial intelligence (AI)–driven channel coding has emerged as a promising solution for enhancing error correction performance and reducing decoding complexity, particularly in the context of Low-Density Parity-Check (LDPC) codes. Unlike traditional coding-theory-based encoding and decoding methods, AI-driven approaches optimise codeword design in a data-driven manner under conditions where classical techniques have limitations, e.g., LDPC codes with limited block lengths and finite decoding complexity. AI-designed LDPC codewords and AI-based decoding algorithms are usually optimised for a given block length or code rate, and cannot be directly applied to different block lengths or rates. To this end, this article presents an AI-driven rate-adaptive LDPC coding framework that incorporates a parameter-sharing mechanism within the encoder and decoder, trained via multi-task learning to support multiple code rates. Preliminary evaluations in AWGN channels validate the feasibility of such a design and demonstrate its promising error-correction potential and adaptability. In this article, we present key design principles, emphasise their advantages, and discuss the future research challenges that remain.