Machine Learning for Channel Coding: A Paradigm Shift from FEC Codes

The design of optimal channel codes with computationally efficient Forward Error Correction (FEC) codes remains an open research problem. In this paper, we explore optimal channel codes with computationally efficient FEC codes, focusing on turbo and Low-Density Parity-Check (LDPC) codes as near-capacity approaching solutions. We highlight the significance of accurate channel estimation in reliable communication technology design. We further note that the stringent requirements of contemporary communication systems have pushed conventional FEC codes to their limits. To address this, we advocate for a paradigm shift towards emerging Machine Learning (ML) applications in communication. Our review highlights ML's potential to solve current channel coding and estimation challenges by replacing traditional communication algorithms with adaptable deep neural network architectures. This approach provides competitive performance, flexibility, reduced complexity and latency, heralding the era of ML-based communication applications as the future of end-to-end efficient communication systems.