Deep Reinforcement Learning Autoencoder with Noisy Feedback

TL;DR

This paper demonstrates that deep neural network autoencoders for communication can be trained effectively with noisy or no feedback channels, maintaining performance comparable to ideal feedback scenarios.

Contribution

It introduces a novel feedback system that enables training of autoencoders without a preexisting feedback link, even over unknown channels.

Findings

Autoencoders perform well with noisy feedback.

The proposed feedback system matches perfect feedback performance.

Effective over AWGN and Rayleigh fading channels.

Abstract

End-to-end learning of communication systems enables joint optimization of transmitter and receiver, implemented as deep neural network-based autoencoders, over any type of channel and for an arbitrary performance metric. Recently, an alternating training procedure was proposed which eliminates the need for an explicit channel model. However, this approach requires feedback of real-valued losses from the receiver to the transmitter during training. In this paper, we first show that alternating training works even with a noisy feedback channel. Then, we design a system that learns to transmit real numbers over an unknown channel without a preexisting feedback link. Once trained, this feedback system can be used to communicate losses during alternating training of autoencoders. Evaluations over additive white Gaussian noise and Rayleigh block-fading channels show that end-to-end communication systems trained using the proposed feedback system achieve the same performance as when trained with a perfect feedback link.