AttentionRNN: Novel Propagation Channel Time-Domain State Predictor
Congcong Wang, Pengqi Zhu, José Rodríguez-Piñeiro, Xuefeng Yin · 2024
Real-time communication-based applications are be-coming more popular in the fifth generation (5G) and beyond systems, increasing the demand for accurate near to realtime channel estimation. Channel time-domain state prediction poses as a viable alternative to satisfy these demands. Traditional channel prediction methods usually suffer from high complexity or inaccuracies in their underlying mathematical models, severely limiting their performance. In this paper, a novel channel time-domain state predictor, named AttentionRNN, based on a recurrent neural network with a self-attention mechanism, is proposed. The predictor introduces a self-attention mechanism, allowing the network to focus on the parts of the data that provide more useful information for the channel prediction. The superiority of our proposal with respect to other available solutions in both prediction accuracy and training cost is verified based on measured data from actual vehicular radar scenarios. Our work constitutes a practical framework for realtime channel state prediction for time-varying applications.