A Machine-Learning-Assisted Antenna Array Isolation Optimization Method Using Intelligent Surfaces for Bistatic In-Band Full-Duplex Communication Systems

In this article, a machine-learning-assisted antenna array isolation optimization method for bistatic in-band full-duplex (IBFD) communication systems based on a novel intelligent surface (IS) is proposed. Both miniaturization and independent dual-polarized transmission phase reconfigurability features are realized simultaneously in this IS. As an illustrative example, a bistatic IBFD array consisting of two transmitting antennas and two receiving antennas working in the 5G-N41 band is designed, where the space coupling wave between transmitter (Tx) and receiver (Rx) is regulated by the IS for neutralizing coupling signals from other paths. With the help of a multilayer perceptron (MLP) network and measured data of the prototype, a surrogate model expressing the mapping between the phase distribution of the IS elements and corresponding Tx-Rx transmission coefficients (TRTC) is learned and called repeatedly in the subsequent genetic algorithm-based optimization. The measured offline-optimized average and worst isolation between two Tx and two Rx ports are 59.9 and 55.7 dB, respectively. Compared with the reference antenna, a 15-dB promotion on both average and worst isolation within the whole 200-MHz operation band is achieved with no significant deterioration on radiation performance. Hence, the proposed method could accomplish relatively wideband self-interference cancellation (SIC) between Tx and Rx while avoiding complicated customized antenna design and relieving the calculation hardware restrictions for system deployment, which can be a promising candidate in future 6G communication systems.