Digital SI Cancellation Based on LSTM for Co-frequency Co-time Full Duplex Communication Systems

Abstract

To address the high self-interference (SI) in co-frequency co-time full duplex (CCFD) communication systems, a digital SI suppression method based on deep learning is proposed. A memory polynomial is firstly utilized to simulate the SI channel. Then a long short-term memory network (LSTM) is used to reconstruct the SI signal. The clean signal after interference suppression can be achieved by subtracting the SI signal at the receiver. To better reflect the changes of the SI signals after being transmitted through the multiple channels and reduce dependence on data volume, feature preprocessing and feature reconstruction are performed at the input of the LSTM. We use the multi-segment delayed transmitted signal as the multiple features and utilize the brainstorm optimization (BSO) technique to estimate the ideal time delay. We verify the proposed SI suppression scheme in the OFDM transmission system. The experimental results show that compared with the traditional least squares estimation (LS) and adaptive filter least mean square (LMS) algorithms, the proposed scheme has a significant performance boost. A SI signal suppression capability of 47.19 dB is achieved.