Deep learning-based high-speed railway communication systems.

High-speed railway wireless communication systems are characterized by severe Doppler shifts and fast time-varying multipath, which challenge reliable connectivity in Long-Term Evolution for Railways (LTE-R). This paper studies the two data-driven frameworks to enhance the communication quality so as to reduce the error probability under such highly dynamic propagation conditions. The first framework investigates a deep neural network (DNN) trained to learn the channel behavior and avoid the restrictive assumptions and predefined interpolation models used in conventional time-domain channel estimation (TDCE) schemes (Nearest, Linear, Cubic Hermite, and Cubic Spline). Besides, the second framework utilizes an autoencoder/decoder architecture to replace the traditional processing units and gain prior information from the dataset. Numerical results show that the proposed DNN-based frameworks more effectively track rapid channel fluctuations and mitigate Doppler-induced distortions, thereby improving detection performance in high-mobility LTE-R scenarios. These findings highlight the potential of learning-based signal processing to enhance the reliability and efficiency of high-speed railway communications.