Deep Thinking Models for Radio Transformer Networks

Abstract

In modern communication systems, message signals are processed with modulation, coding, pulse shaping, etc. for efficient data transmission. Recently, machine learning techniques have been used to replace such signal processing algorithms. Radio Transformer Networks (RTNs) is a technique that can be used to model an entire communication system with neural networks encompassing transmitter, channel, and receiver. These models can then be trained as a whole to generate encoding schemes that are optimized for different channel conditions. In this paper, we incorporate parameter estimation in the receiver trained with the model. Recurrent layers are used to improve parameter estimates whereby the network has the opportunity to “think longer.” Simulation results are presented to illustrate the concepts.