End-to-End Learning of Communications Systems Without a Channel Model
This work addresses a key limitation in neural network-based communications for researchers and engineers, enabling learning over diverse channels without prior assumptions, though it is incremental as it builds on existing autoencoder approaches.
The paper tackles the problem of end-to-end learning for communications systems without requiring a differentiable channel model, by proposing an algorithm that combines supervised training of the receiver with reinforcement learning for the transmitter, achieving performance comparable to fully supervised methods on AWGN and RBF channels, with convergence faster on RBF channels.
The idea of end-to-end learning of communications 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 iterates between supervised training of the receiver and reinforcement learning -based training of the transmitter. We demonstrate that this approach works as well as fully supervised methods on additive white Gaussian noise (AWGN) and Rayleigh block-fading (RBF) channels. Surprisingly, while our method converges slower on AWGN channels than supervised training, it converges faster on RBF channels. Our results are a first step towards learning of communications systems over any type of channel without prior assumptions.