Distortion Element Estimation Technique based on Deep Learning for Self-Interference Cancellation of Full Duplex Communication System

This paper presents the deep learning-based distortion element estimation technique of self-interference signal for DOCSIS 3.1 system with full duplex. In the DOCSIS 3.1 system with full duplex, the self-interference signal estimation and cancellation are the most important issues to enhance the system performance. Proposed technique estimates channel and nonlinear element of self-interference signal by using deep learning method. The practical challenge of application of deep learning technique to communication system is that it is difficult to find proper deep neural network structure for a specific purpose of the system. The proposed estimation technique uses simple deep neural network structure to estimate channel and nonlinear element of self-interference signal in time-domain. The evaluation results of the deep learning-based proposed technique is better than that of existing algorithm-based estimation technique.