Hybrid Neural Coded Modulation: Design and Training Methods

TL;DR

This paper introduces a hybrid coded modulation scheme combining neural network-designed inner codes with traditional outer codes, demonstrating improved performance over standard QAM schemes for 5G LDPC codes at higher modulation orders.

Contribution

It presents a novel neural network-based inner coding method integrated with standard outer codes, optimized using a mutual information-inspired loss function.

Findings

Outperforms conventional QAM modulation for 16 and 64 symbols

Uses deep neural networks for inner code design

Achieves better error performance with 5G LDPC codes

Abstract

We propose a hybrid coded modulation scheme which composes of inner and outer codes. The outer-code can be any standard binary linear code with efficient soft decoding capability (e.g. low-density parity-check (LDPC) codes). The inner code is designed using a deep neural network (DNN) which takes the channel coded bits and outputs modulated symbols. For training the DNN, we propose to use a loss function that is inspired by the generalized mutual information. The resulting constellations are shown to outperform the conventional quadrature amplitude modulation (QAM) based coding scheme for modulation order 16 and 64 with 5G standard LDPC codes.