Authors
Mehmetcan Gok, Danijela Cabric, and Michael Honig
Abstract
As part of the U.S. National Spectrum Strategy, the 3.1‚Äì3.45 GHz band is considered for coexistence with commercial broadband. This band supports defense radar systems with strict covertness, ruling out cooperative methods. This work proposes an ML-assisted approach for detecting radar chirps in a non-cooperative setting using beamforming errors from uplink training. We utilize supervised and unsupervised algorithms without explicit channel estimation or sensing. Simulations show significant improvements, with supervised models achieving AUC below 1% and OCSVM around 3%, demonstrating ML’s potential for radar-communication coexistence in dynamic environments.