Research on Anti-interference Algorithm for Cab Signal Demodulation Based on Deep Learning

The cab signal, as the movement authority of train operation, plays a vital role in ensuring the safety of operation. The booming high-speed and heavy-haul railways lead to larger traction current, as well as various types of locomotive, the electromagnetic environment is getting more serious than before. The track circuit and traction current take the rails as common transmission channel. It is difficult to physically filter out mixed noise in the received signal and demodulate the information correctly since harmonic component of the interference may occupy the same frequency spectrum as useful signal. Under this background, the paper proposes a cab signal anti-interference algorithm based on PSO-optimized (Particle Swarm Optimization) DBN (Deep Belief Network) aimed at these FSK (Frequency shift keying) modulation signals which were interfered in useful band to improve the accuracy of cab signal demodulation. After verified by on-site and simulated data and compared with commonly used machine learning, this algorithm performs well in the demodulation of in-band noisy cab signals demodulation, the average accuracy of the algorithm reaches 99.87%, in addition, the average data process time is about 0.5s based on the TensorFlow platform with the GPU of GeForce MX250.