Channel Estimation for OFDM Systems Over Doubly Selective Channels Based on CEHNet

Abstract

In dynamic scenarios, time-frequency doubly selective channels challenge accurate estimation. Deep learning-based method emerges as a promising way by leveraging temporal correlation and local time-frequency features characterized by wireless channels. To enhance adaptability in dynamic channels with fewer pilots, this letter proposes a novel channel estimation algorithm based on a channel-enhanced deep Horblock network (CEHNet), where the Horblock structure is integrated into the super-resolution convolutional neural network (SRCNN) to capture long-range dependencies effectively. Additionally, the autocorrelation of the channel state information (CSI) matrix, derived from pilot signals, is fed into CEHNet in parallel, thereby emphasizing multipath delay and Doppler frequency shift information therein. Furthermore, the incorporation of Lasso regression accelerates network convergence. Experimental results demonstrate that the proposed algorithm outperforms baseline methods in various scenarios, achieving superior performance with fewer epochs, particularly when pilots are sparse or missing.