Model-free Training of End-to-end Communication Systems
This addresses a key bottleneck for deploying neural network-based communication systems in real-world scenarios with unknown or non-differentiable channels, offering a practical solution with incremental improvements over existing methods.
The paper tackles the problem of training end-to-end communication systems without requiring a differentiable channel model, presenting a novel algorithm that iterates between training the receiver with true gradients and the transmitter with approximated gradients, achieving performance comparable to model-based training and demonstrating state-of-the-art results in hardware implementations.
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.