A Mixture Density Channel Model for Deep Learning-Based Wireless Physical Layer Design

Abstract

Machine learning is a highly promising tool to design the physical layer of wireless communication systems, but it usually requires that a channel model is known. As data rates increase and wireless transceivers become more complex, the wireless channel, hardware imperfections, and their interactions become more difficult to model and compensate explicitly. New machine learning schemes for the physical layer do not require an explicit model but implicitly learn the end-to-end link including channel characteristics and non-linearities of the system directly from the training data. In this paper, we present a novel neural network architecture that provides an explicit stochastic channel model, by learning the parameters of a Gaussian mixture distribution from real channel samples. We use this channel model in conjunction with an autoencoder for physical layer design to learn a suitable modulation scheme. Since our system learns an explicit model for the channel, we can use transfer learning to adapt more quickly to changes in the environment. We apply our model to millimeter wave communications with its challenges of phased arrays with a large number of antennas, high carrier frequencies, wide bandwidth and complex channel characteristics. We experimentally validate the system using a 60 GHz FPGA-based testbed and show that it is able to reproduce the channel characteristics with good accuracy.