DnCNN-SRCNN: A Two-Stage Channel Estimation Method for Deep-Space Communication Systems

Abstract

During superior solar conjunction, deep-space communication links are susceptible to solar scintillation, Doppler shifts, and low signal-to-noise ratio (SNR), which make accurate estimation of the complete channel response challenging. To address this issue, this work proposes a two-stage channel estimation method based on a denoising convolutional neural network (DnCNN) and a super-resolution convolutional neural network (SRCNN). Specifically, the least squares (LS) method is initially employed to obtain noisy preliminary estimates at pilot positions, followed by the utilization of DnCNN to mitigate the noise. Subsequently, SRCNN reconstructs the denoised pilot-channel features to recover the full channel response. Simulation results demonstrate that the proposed method achieves a lower mean squared error (MSE) compared to conventional LS and minimum mean squared error (MMSE) methods under complex deep-space conditions. Notably, at an SNR of 10 dB, the proposed approach reduces the MSE by approximately 82.1% and 68.3% relative to conventional LS and MMSE methods, respectively.