Real-time complex-valued RF blind source separation via in-physics computing.

An integrated photonic vector processor performs real-time complex-valued blind source separation on raw RF signals up to 24 GHz, reducing EVM to 6.18% and 5.3% for two recovered outputs.

• Why it matters: Edge-centric 6G needs ultralow-latency processing of overlapping RF signals, and blind extraction of source components is a physical-layer capability that must adapt in real time.
• What they did: The IPVP implements CV-BSS via in-physics computation on raw RF, using in situ learning for blind adaptation to unknown sources and channels, tested in MIMO communications and multi-band radar interference suppression.
• The result: In experiments, EVM drops from over 50% to 6.18% and 5.3% for the two recovered signals, and SIR improves by up to 26 dB, enabling effective interference suppression.

AI summary of the abstract · may contain errors · not medical advice · read the paper.