Hybrid Neural Coded Modulation: Design and Training Methods
This work addresses the problem of enhancing communication efficiency for wireless systems, though it appears incremental as it builds on existing coding methods with neural network integration.
The paper tackles the problem of improving coded modulation schemes by designing a hybrid system with a neural network inner code and standard outer codes, achieving performance gains over conventional QAM-based schemes for modulation orders 16 and 64 using 5G LDPC codes.
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.