Curving Around Obstacles via NN-Enabled Wavefront Shaping in Sub-THz Wireless Networks

The sub-THz band offers an attractive solution to future wireless networks, thanks to its ultra-low latency as well as its large available bandwidth. However, link blockage remains a major setback towards reliable sub-THz end-to-end communication systems, due to narrow beamwidth and inherently high penetration losses. To achieve blockage mitigation in sub-THz communication, this paper takes advantage of unique near-field properties and manipulates curved wavefront trajectories. Unfortunately, finding the best curved beam configuration is non-trivial due to the lack of a closed-form equation for received power calculation under blockage scenarios, even if the wireless environment is precisely known. To address this, we present a physics-informed learning-based framework that optimizes the phase profile of the transmitting array, such that the resulting wavefront could curve around obstacles and adapt to dynamic environments in real time. Through extensive near-field simulations, we evaluate the performance of our AI-generated curved beams as opposed to optimal Airy beams achieved via impractical exhaustive scans with prohibitively large time and complexity overheads. Importantly, simulated results show that our AI-generated curved wavefront provides an average SNR gain of 19.83 dB compared with conventional beam steering and 2.13 dB compared with near-field beam focusing, across ~400 random and independent test scenarios.