Time-frequency Domain Joint Precoding based on the LSTM Network and Repetitive HSR Channel

With the continuous development of high-speed rail (HSR) technology, the HSR communication system demands enhanced precoding schemes to overcome severe Doppler effects and submillisecond coherence time encountered in $350 \mathrm{~km} / \mathrm{h}$ operational environments. Conventional autoregressive methods are insufficient to address channel precoding in high-speed mobility. To resolve this limitation, we devise a time-frequency domain joint precoding method that integrates the long-shortterm memory (LSTM) network with repetitive HSR channel characteristics. The proposed joint time-frequency LSTM network with error correction models takes advantage of historical channel correlations for adaptive single-/multi-step precoding. Furthermore, the paper proposes a novel concept of the HSR network channel state transition map (HSR-CSTM), which uses the similarity of Doppler spectra at the same location caused by the repetitiveness of trains to optimize the accuracy of the joint precoding algorithm and reduce the delay requirement by continuously updating HSR-CSTM. Finally, simulation results show that our method is more stable than the autoregressive algorithm, which only relies on time correlation under different signal-to-noise ratio conditions and lower transmission power conditions, and can meet lower latency requirements.