Transformer-Based Parallel Accurate Prediction in Complex and Time-Varying Underwater Channels

Traditional channel prediction methods are mostly based on autoregressive models, leading to a significant accumulation of errors due to sequential prediction methods and consequently making it challenging to achieve accurate channel prediction results. Therefore, in this paper, we propose a Transformer-based method for underwater acoustic channel prediction. Through rigorous experimental validation conducted across diverse oceanic environments, we substantiate the superior performance of our method when compared to conventional Long Short Term Memory (LSTM) and Recurrent Neural Network (RNN) networks. Our study significantly contributes to meeting the crucial need for precise channel prediction in underwater wireless communication. The average MSE of the prediction results is reduced by approximately 15.5% compared with LSTM.