Rapid Emitter Geolocation with Adaptive Learning

Wide-band RF spectrum monitoring and geolocation is a challenging task that is exacerbated by complex collection geometry and the intricate interplay of signals in complex 3D urban environments. This interaction often leads to complex signal behaviors like multipath propagation, reflection, and absorption, significantly reducing the efficacy of traditional geolocation methods. Geolocation, typically treated as a distinct step in signal processing chains, has not yet fully benefited from the AI advancements that have revolutionized other areas like signal detection/identification, as it relies heavily on physical models for signal propagation that are grounded by well-established equations describing physical phenomena. In urban environments, where multi-path and co-channel interference effects are prevalent, these traditional methods falter.Ā Rapid Emitter Geolocation with Adaptive Learning (REGAL)Āpresents the opportunity to revolutionize RF signal processing in urban areas by developing a system that exploits local 3D models of an area to train a SWaP-efficient neural network that is specialized to the local context of the deployment area of interest. The ability to precompute and train a neural network, such as a Convolutional Neural Network (CNN), for rapid edge computation not only aligns with the demands for speed and efficiency in real-world applications, but it also creates numerous opportunities for advanced RF processing at the edge across the DoD, IC, and commercial domain. REGAL is structured to deliver its innovative solutions within a strategically planned timeline and budget. In Phase I, we focus on feasibility studies, system design, and software prototyping for $180,000 over a period of six months. This phase is crucial for establishing the foundation of a novel AI-based geolocation system. Subsequent phases build upon these initial developments, scaling up the technology with payload development and integration into operational scenarios.