An Interpretable Synthesis Method of Intelligent Surface for Dynamic Antenna Isolation Control Based on SHAP Analysis

In this letter, an interpretable synthesis method of intelligent surface (IS) for dynamically generating desired antenna isolation enhancement in a given sub-band is proposed. Though an optimization method based on artificial neural network has been demonstrated to effectively enhance antenna isolation, the lack of interpretability has become a main obstacle to analyzing the decoupling mechanism of the IS. Based on Shapley additive explanations (SHAP) analysis, each element's influence ability on different kinds of antenna isolation between Tx and Rx ports in a 2-transmitting-2-receiving array is given, and an interpretable method which generates IS phase distribution for isolation enhancement at desired frequency subband is extracted with the help of the SHAP feature distribution in a visualization form with a dataset of the scale of 22,000, which is relatively small. Experimental results reveal the effectiveness and robustness of the synthesis method for generating a 100 MHz high isolation band where the worst and average isolation are 55.4 dB and 65.1 dB, with a 16.1 dB promotion of the worst isolation and a 21.6 dB promotion of the average isolation compared to the initial isolation.