Digital self-interference cancellation for full-duplex systems based on deep learning

Abstract

Full-duplex (FD) systems can double the spectrum efficiency by allowing users to transmit and receive signals at the same time and in the same frequency band. However, in FD systems the receiver suffers severe self-interference (SI) caused by the transmitter and eliminating the SI is challenging. In this paper, we propose a novel digital self-interference cancellation (SIC) scheme based on deep learning, which combines the power of sliding window, gated recurrent unit (GRU) network and attention mechanism (AM). The proposed scheme can accurately reconstruct the SI through two steps. It first processes the sample signals utilizing the sliding window and then extracts the dependence in preprocessed signals by applying the GRU with AM. We adopt the orthogonal frequency division multiplexing (OFDM) signal as the baseband transmitted signal to verify the proposed scheme. Extensive experiments demonstrate that the SIC capability of the developed scheme is at least 14.46 dB higher than the existing polynomial and neural network (NN) schemes. Furtherly, the effectiveness of the scheme is verified in different interference-to-noise ratio (INR), multipath channel, training sample and OFDM symbol situations.