AI/ML Interference Cancellation used in STAR Wireless for Radio Astronomy RFI Control

RFI from LEO satellites is a major problem in radio astronomy and cosmology; and self interference from transmitter to receiver in a full-duplex link is an equally hard problem in 6G/NextG simultaneous transmit and receive (STAR) communications. We observe that these two hard problems that occur in passive sensing (radio astronomy) and wireless communications (STAR systems) are related and in fact mathematically identical. Much progress was made in STAR communications aimed at solving self-interference using RF machine learning. This can have new applications when adapted for solving the related problems in radio astronomy RFI control. We show that advancements in STAR communication systems, utilizing learning-based interference cancellation with nonlinear DSP, can be tailored to counteract the low-noise nonlinear effects caused by RFI from LEO constellations in radio telescopes. Our latest experiments for STAR show 30+ dB of self-interference cancellation in the digital domain using real-time RF-MLS. The result is encouraging towards similar levels of RFI reduction in a radio telescope. A straw-man design that employs an auxiliary array, Howells-Applebaum digital adaptive beamforming, and augmented envelope neural network based real-time nonlinear DSP on Xilinx RF-SoC is proposed at a conceptual level to motivate further joint work by the wireless communications and radio science communities.