A Reliable and Intelligent Deep Learning Based Demodulator for M-Ary Code Shifted Differential Chaos Shift Keying System With Power Allocation

In high-speed railway systems, channel conditions are dramatically changing, which brings great challenges to information transmissions. Our objective is to improve the reliability performances by utilizing the deep learning (DL) scheme to intelligently extract the features of signals to combat the complex dynamically changing channel conditions. In this paper, we propose a DL-aided demodulation scheme for the M-ary code shifted differential chaos shift keying system with power allocation (PA-GCS-MDCSK) to enhance reliability performance. In this design, the deep neural network (DNN) adopts fully-connected layers (FCLs) to conduct the joint de-spreading of Walsh codes and chaotic demodulations. Meanwhile, the long short-term memory (LSTM) unit with the residual structure is constructed to extract the correlation between the chaotic sequences, thus the interferences induced by real-valued chaotic sequences can be suppressed, thereby improving the reliability performance. Then the computational complexity is analyzed and compared with benchmark schemes. Simulation results over both additive white Gaussian noise (AWGN), fading, and railway channels validate that the proposed design can achieve better bit error rate (BER) and robustness performances than benchmark schemes.

Automated summary

This paper proposes a deep learning-based approach to improve the reliability of information transmission in high-speed railway systems. The proposed scheme utilizes a deep neural network and long short-term memory unit for joint demodulation and feature extraction to handle changing channel conditions. Simulation results demonstrate that the proposed scheme achieves better performance in various channel conditions.