Machine Learning and Multipath Fingerprints for Emitter Localization in Urban Scenario

Abstract

A hybrid approach is proposed to perform the locate and tracking a non-collaborative radio frequency emitter using ray tracing (RT) simulation tool, channel impulse response (CIR), and machine learning estimation. The technique can enhance communication intelligence (COMINT) systems or even perform the localization using a single sensor in an non-line-of-sight (NLOS) suburban scenario. A multipath fingerprint can identify the target position using the machine learning classification engine to perform the matching. Conventional localization techniques mitigate errors trying to avoid NLOS measurements in processing emitter position, while the multipath fingerprints proposed uses the reflection information to feed the pattern matching engine build on a machine learning classification framework. The method was applied to simulate a tactical scenario, where a navy frigate is in Ipanema and tries to track an RF emitter target in the Rio de Janeiro streets using only one RF sensors fixed in an aerostat in a hypothetical counterinsurgency situation.