# Model-free Training of End-to-end Communication Systems

## TL;DR

This paper introduces a novel training algorithm for end-to-end communication systems that does not require a differentiable channel model, enabling effective training with unknown or non-differentiable channels, demonstrated through hardware implementation.

## Contribution

The paper presents a new learning algorithm that trains communication systems without relying on a differentiable channel model, applicable to real-world unknown or complex channels.

## Key findings

- Works as well as model-based training across various channels
- Effective in hardware implementations with SDRs
- Achieves state-of-the-art performance over cable and wireless channels

## Abstract

The idea of end-to-end learning of communication systems through neural network-based autoencoders has the shortcoming that it requires a differentiable channel model. We present in this paper a novel learning algorithm which alleviates this problem. The algorithm enables training of communication systems with an unknown channel model or with non-differentiable components. It iterates between training of the receiver using the true gradient, and training of the transmitter using an approximation of the gradient. We show that this approach works as well as model-based training for a variety of channels and tasks. Moreover, we demonstrate the algorithm's practical viability through hardware implementation on software-defined radios where it achieves state-of-the-art performance over a coaxial cable and wireless channel.