Digital Self-Interference Cancellation for Full Duplex Wireless Communication Based on Neural Networks

Abstract

How to cancel self-interference (SI) signals has always been a challenge for full duplex (FD) wireless communication. SI signals are difficult to cancel given that they contain complex nonlinear distortions introduced by analog components in the FD chains, such as the power amplifier (PA) and analog-to- digital converter (ADC). Even after passive cancellation and active analog cancellation, the residual signal still contains nonlinear components that cannot be ignored. In order to ensure effective demodulation and channel decoding, it is necessary to perform further SI cancellation (SIC) in the digital domain. In this paper, a feedforward neural network is applied to the digital SIC stage for SI signal reconstruction. We design an adaptive linear filter to extract features from the baseband transmitted signal or feedback signal as input of the neural network. Experiments are performed on 20-MHz QPSK-modulated OFDM signals to demonstrate that the neural network can reconstruct the SI signals accurately and outperforms the polynomial based cancellation approach. With the interference- to-noise ratio (INR) being 30 dB, the SIC capability of our proposed method is 29 dB, which is 10 dB better than the polynomial based cancellation approach.