Authors
Mehmetcan Gok (Northwestern University), Danijela Cabric (University of California, Los Angeles), and Michale Honig (Northwestern University)
Abstract
We propose radar detection for spectrum sharing by monitoring the a priori estimation errors of uplink Recursive Least Squares (RLS) adaptive beamformers at a cellular base station. Unlike energy or eigenvalue detectors, this approach requires no knowledge of radar waveforms, noise power, or dedicated sensing hardware. The proposed scheme exploits the observation that radar pulses uncorrelated with uplink user pilots produce transient error spikes that the RLS filter attempts to suppress upon converging to steady-state. We employ three detectors: a Quadratic Form (QF) detector for wideband interference, an Adaptive Peak Detector (APD) for spatially localized pulses, and a CUSUM sequential detector for persistent anomalies. Bidirectional processing running the RLS adaptation both forward and backward in time addresses cases where radar pulses coincide with filter initialization transients in one direction but fall within steady-state in the other. Simulations using 3GPP Urban Macro (UMa) channels show that APD achieves probability of missed detection below 0.1 at false alarm rate 0.1 when the radar interference power is 6 dB below OFDM communication signal, enabling detect-and-vacate coexistence in the 3-4 GHz band without incumbent coordination.