A Self-Interference Cancellation Method Based on Deep Learning for Beyond 5G Full-Duplex System

While full-duplex wireless communications systems have substantial value to the wireless communications field, such systems have been known to face challenges due to self-interference. Digital cancellation method is an important step to remove the residue self-interference signal in digital domain. In this paper, a self- interference cancellation scheme based on deep learning for full-duplex systems is proposed. We investigate the joint effects of non-linear distortion and linear multi-path channel on the performance of digital cancellation and model those factors by deep neural network which could process arbitrary sequences of inputs with temporal coherence and perfectly handle non-linear problem. The real-time experimental results show that our method has desirable performance than previous research. Its performance has better gain compared with traditional methods since it can solve both linear and non-linear problem. Furthermore, it has good stability to self-interference channel effect.