A Machine Learning-Assisted Multi-port Antenna Isolation Enhancement Method Based on a Novel Intelligent Surface for In-band Full-duplex Systems

In this paper, a machine-learning-assisted multi-port antenna isolation enhancement method for in-band full-duplex (IBFD) systems based on a novel intelligent surface (IS) is proposed. This IS realizes both miniaturization and independent dual-polarized transmission phase reconfigurability features simultaneously. Through an artificial neutral network (ANN) trained by measured data of the prototype, a surrogate model expressing the mapping relationship between the phase distribution of the IS and isolations of Tx-Rx port pairs is learned and called repeatedly in off-line genetic optimization. The measured optimal average and worst isolation between two Tx and Rx ports are 59.9 dB and 55.7 dB, respectively. Compared with reference antenna, a 15 dB promotion on both average and worst isolation is achieved in the whole 200 MHz operation band without significant deterioration on radiation pattern. The proposed method accomplishes relatively wideband self-interference cancellation (SIC) in propagation domain between Tx and Rx, while avoiding sophisticated customized antenna design and relieving the calculation hardware restrictions for system deployment, which can be a promising candidate for IBFD solutions in future 6G communication system.