Full-Duplex Digital Self-Interference Cancellation Based On Deep Learning

Co-frequency co-time full-duplex system can send and receive signals at the same time and frequency.Compared with the time division duplex and frequency division duplex system,it can theoretically double the spectrum utilization efficiency,which is one of the development directions of the next-generation wireless communication technology.One of the critical problems to be solved in the co-frequency co-time full-duplex system is the self-interference signals generated by itself will affect the correct demodulation of useful signals,but it can be suppressed in the antenna domain,analogue domain and digital domain respectively.In this paper,aiming at the problem of insufficient cancellation ability of traditional digital domain self-interference cancellation technology.This paper mainly studies a digital domain self-interference cancellation technique based on deep learning method.The main contents include:Firstly,the self-interference channel model in co-frequency co-time full-duplex system is theoretically derived and modeled,the main sources of interference in the self-interference channel are analyzed.We design a self-interference cancellation scheme based on deep learning according to the theoretical self-interference channel model.The design criteria and functions of each component in the deep learning self-interference cancellation model are studied and analyzed.Secondly,we design a feature extraction method of the digital domain self-interference signal,and combine with the deep learning self-interference cancellation model in the feature layer of deep learning networks.The convergence algorithm,activation function,and the influence of feature dimension on the performance of the deep learning self-interference cancellation model are studied and analyzed.At the same time,the quantitative relationship between the self-interference cancellation ability and the loss function value of the deep learning model is analyzed and established.Thirdly,the code of digital domain self-interference cancellation simulation system is built,and a digital domain self-interference cancellation method based on deep learning is simulated and analyzed.Simulation results show that,under the condition of the 30 dB interference-to-noise ratio of the self-interference signal,the deep learning self-interference cancellation technology can realize the self-interference cancellation 22.70 dB capability without the use of additional reference signal feature extraction algorithm,and the self-interference cancellation capability can be realized 27.33 dB under the use of additional reference signal feature extraction algorithm.In this paper,we studied the digital domain self-interference cancellation technology,which based on deep learning in co-frequency co-time full-duplex system.Additionally,we simulated and analyzed this self-interference cancellation technology,which provides a new method and research direction for further improving the performance of digital domain self-interference cancellation in the full-duplex system.