Deep Learning-Aided Coding for the Fading Broadcast Channel with Feedback
Deep Learning-Aided Coding for the Fading Broadcast Channel with Feedback
复制标题
DOI：
10.1109/icc45855.2022.9838631
复制
发表时间：
2022-05
期刊：
影响因子：
--
通讯作者：
Siyao Li;Daniela Tuninetti;N. Devroye
中科院分区：
文献类型：
--
作者：
Siyao Li;Daniela Tuninetti;N. Devroye
研究了带反馈的对称双用户衰落高斯广播信道的实用编码设计。我们在深度神经网络（NNs）的帮助下构建了一个两阶段的编码方案，该方案寻求共同优化编码器和解码器。将通信系统解释为一个自动编码器（用AE表示），我们在各种无噪声反馈信号的情况下训练AE。对瑞利分布式信道状态进行了性能评估，结果表明，在低信噪比条件下，存在一种训练有素的基于神经网络的两相模型，其性能优于最先进的编码。考虑到反馈信号的可用性，我们用不同的输入训练声发射，并观察到在所提出的方案下，由接收信号组成的反馈似乎比信道状态更有利于提高可靠性。我们提供了使用信道状态反馈的编码方案的初步解释。
We consider the design of practical codes for a symmetric two-user fading Gaussian Broadcast Channel (BC) with feedback. We construct a two-phase coding scheme with the help of deep Neural Networks (NNs) that seeks to optimize the encoder and decoders jointly. Interpreting a communication system as an autoencoder (denoted by AE), we train the AE under various scenarios of noiseless feedback signals. Performance evaluation is presented for Rayleigh distributed channel state, which reveals the existence of a trained NN-based two-phase model that outperforms state-of-the-art codes in the low SNR regime. Considering the availability of feedback signals, we train the AE with different inputs, and observe that feedback consisting of received signals appears to be more beneficial than channel states to boost reliability under the proposed scheme. We provide initial interpretations of the encoding scheme which uses channel state feedback.