A Machine Learning-Empowered Adaptive Self-Interference Cancellation for In-Band-Full-Duplex Base Station Antennas

In this paper, an adaptive antenna self-interference cancellation (SIC) technique based on a machine learning method is presented. Two collocated dual-polarized antenna arrays, one of which is transmitting array (Tx), while the other is receiving array (Rx), are used as the reference antenna. Then, a RIS between them with miniaturized unit cells are designed and with hardware configuration, we are able to capture the scattering parameters from all ports, with respect to different phases of the cells. Consequently, an ANN is trained from measurement data and optimizations on all the isolations with and without metal obstacle are carried. It is shown that isolations can be improved from around 40dB to more than 52.5dB in both cases, demonstrating great potential of the proposed method to be used in future full duplex applications.