Deep Learning Based Channel Estimation for Deep-Space Communications
Lianning Cai, Guanjun Xu, Qinyu Zhang, Zhaohui Song, Wei Zhang
Abstract
During the period of superior solar conjunction, the deep-space channel suffers from solar scintillation and large Doppler shifts, leading to highly time-varying communication links. To guarantee the quality of data transmission, accurate channel estimation is indispensable. In this paper, we use the Gaussian Doppler spectrum to model the time-selective fading channel in deep space communications and propose a data-driven channel estimation framework based on a convolutional neural network-long short-term memory (CNN-LSTM) model. We leverage the strength of CNN for efficient feature extraction and LSTM for modeling temporal dependencies, further enhancing the performance of channel estimation. Simulation results show that the proposed CNN-LSTM method achieves up to 5 dB improvement in normalized mean square error (NMSE) performance over the conventional linear minimum mean square error (LMMSE) method. In addition, it exhibits robustness to solar scintillation in the deep-space channel.