Channel computation based on multi-scale attention residual network

Orthogonal time-frequency space (OTFS) modulation can effectively counter ICI in high-speed mobile scenarios, fully enhance the spectral efficiency of communication systems in high Doppler expansion scenarios, and improve the quality of communication systems. Channel estimation performance serves as a critical evaluation parameter within the OTFS modulation system. In this paper, we propose a multi-scale attention residual neural structure for improved channel estimation of OTFS waveforms in different satellite-ground scenario. Firstly, a multi-scale channel feature extraction module is designed, which applies multi-dimensional feature extraction to the channel matrix, thereby bolstering the capability to capture features at diverse scales. Subsequently, a self-attention mechanism is incorporated to concentrate on subtle yet significant features. The extracted features are then integrated and exploited through a residual convolutional architecture to derive an estimation of the channel matrix. Simulations are conducted using the satellite-ground mobile channel model outlined in 3GPP TR 38.811, with the NTN-TDL-C and NTN-TDL-B channel models representing line of sight (LoS) and non-line of sight (NLoS) conditions, respectively. Results demonstrate that the attention-based approach presented surpasses alternative neural network methodologies in terms of mean squared error (MSE), bit error rate (BER), and complexity, and meets the demands of OTFS channel estimation in satellite-ground scenario.