A hybrid digital self-interference cancellation method with attention-based TCN-GRU for full-duplex systems

Abstract

Full-duplex (FD) technology is expected to be widely used in future communication systems due to its excellent performance in spectral efficiency. Self-interference (SI) is the main problem affecting the transmission quality of the FD systems, thus it is crucial to research the SI cancellation (SIC) techniques. In this paper, we present an innovative hybrid digital SIC method named sliding time window, temporal convolutional network (TCN), gated recurrent unit (GRU) and attention mechanism (AM)-based deep learning network (STGAN). The proposed method can extract complete features and time information from the input sequence, and accurately estimate the SI in FD systems. We first present a preprocessing module that utilizes sliding time windows to handle the original signal sequence. Then, we utilize the TCN and GRU to extract the short-term features and long-term dependencies within the sequence. Finally, the AM is adopted to explore the significance of the information conveyed by the sequence at different time steps. The effectiveness of the STGAN is verified in multiple experimental scenarios. Simulation results indicate that the interference cancellation ratio (ICR) of the STGAN is significantly higher than existing methods.