Multi-Stage RF Emitter Search and Geolocation With UAV: A Cognitive Learning-Based Method

This paper investigates the issue of searching and geolocating a radio frequency (RF) emitter with an unmanned aerial vehicle (UAV) in a rich scattering environment. The absence of knowledge of the characteristics of the target RF signals, the ground-to-air channels, and the potential geographical distribution of the RF emitter makes efficient trajectory planning (TP) and geolocating intractable. Since the average signal-to-noise ratio (SNR) of the signal captured by the UAV changes across the search area, state-of-the-art TP is inapplicable due to the assumption of a single problem domain, according to the No Free Lunch (NFL) theorem. In contrast, we develop a cognitive multi-stage search and geolocation (CMSG) framework, which adjusts the adopted TP algorithms in various search stages. In the stage with extremely-low SNRs, a random-walk-based TP is proposed to avoid meaningless detours. For stages with higher SNRs, the TP follows policies optimized by a novel search method based on direction of arrival (DOA) estimation and reinforcement learning. To alleviate the impairment of unreliable DOA estimations, empirical emitter position estimations are exploited to find the best flight directions. Simulation results confirm the advantages of the proposed CMSG over various baselines in lowering both failure rate and time consumption.