Machine classification of code signals in electric train warning systems

Abstract

Automatic train warning systems beeing currently in service on Russian railways use electric track circuits as signal communication media. Electric signals transmitted through a track circuit often get corrupted by the noise produced by electric locomotives and other sources. This, in most cases, causes errors in automatic train warning systems and temporarily disrupts the operation of a railway. To improve the stability of such systems while receiving signals from a track circuit, we propose a machine classification algorithm based on a neural network. In this article, we describe all the stages of this algorithm and discuss the architecture of a neural network for classification of an electric signal received from a track circuit. We also demonstrate the successful application of the algorithm for receiving a noisy electric signal which currently used automatic train warning systems fail to decode.