Neural Networks Applied for Broadcast Channels

Abstract

The famous mathematical definition of the capacity region of broadcast channels assumes that the channel is a degraded. However, not all wireless communication channels can be mathematically represented as degrades ones. Motivated by the artificial intelligence revolution, this paper provides a novel approach for tackling this problem. In particular, this paper utilizes artificial neural networks for providing a solution for the problem of converting a broadcast channel into a degraded one. Creating a degraded channel implies modeling the broadcast channel as a series of cascaded channels connecting the transmitter to the receivers sequentially which in turn implies finding a middle (relay) channel connecting the receivers. This process of finding the middle channel is performed in this paper utilizing different models of neural networks. This approach is applied for binary symmetric channels and Gaussian channel. For the Gaussian case, this paper provides a novel approach for establishing the middle channel by providing an estimation of the distribution of the middle channel itself, not only the crossover probability or the variance of the distribution. As a result, several possible extensions for practical channels (like Rayleigh) is suggested. Furthermore, this paper provides an abundant evidence that artificial intelligent is capable of modernizing classical information theory that faces major problems with the complexity of the analytical solutions.