Geolocation: A Fusion of Measure and Zero-Shot Classification of 1-D Signals

Dr. Ryan Casey

Southwest Research Institute

 

Abstract

The proposed talk is focused on two distinct, and somewhat disjoint, topics providing a breadth of information at the sacrifice of depth. The first half of the talk will focus on radio frequency (RF) geolocation, different forms of geolocation measurements, and the fusion of different geolocation measurement forms. Geolocation is the process of isolating the location of an RF emission, e.g., SOS transmission, EMI emissions. The accuracy of the geolocation is highly coupled with the number of sensors receiving the emission, the placement of these sensors, and the form of measurements such as angle of arrival (AoA), time of arrival (TOA), time difference of arrival (TDOA), and frequency different of arrival (FDOA). The fusion of different forms of measurements can help improve geolocation accuracy. We will present the basics behind the forms of geolocation measurement along with their impact on geolocation. We will also discuss fusing these measurements together in a single geolocation measurement. The second half of the talk will focus on the use of zero-shot classification of 1-D signals. Machine learning signal classifiers (i.e., algorithms that recognize signal modulation types, or classes) are typically trained to approximate a target given a set of labeled training data that includes all possible input values requiring extensive labeled data sets. In some cases, a rich training set is not always available for the desired input types. However, analogous inputs may have rich representations in the data sets. Zero-shot classification is the ability of a classifier to identify an instance of a previously unseen class based on only a description of the class. We implemented a framework capable of learning a semantic mapping function and performed modulation classification on the high-level semantic description.

 

Speaker Biography

Dr. Casey's areas of expertise include signal processing, multi-rate signal processing, adaptive filtering, communication theory, detection theory, and classification theory. In addition, he has extensive experience in software defined radios, compressive sampling, array theory, and machine learning algorithms for signal processing applications. He has a strong understanding of high frequency (HF) ionospheric propagation, ionospheric sounding techniques, and over-the-horizon radio (OTHR). Dr. Casey has 15 years of experience in automated signal detection and classification focusing on signal exploitation algorithm research and development. He has been a technical leader in development of algorithms to detect and classify waveforms ranging from tradition communications waveforms to spread spectrum technics as well as RADAR and ionospheric sounding waveforms of various varieties. He has developed, fielded, and maintained multiple wideband RF automated monitoring system. He has been the principal investigator in multiple IR&D efforts and has been co-principal investigator in both IR&D and eternally funded research efforts. Additionally, Dr. Casey has also taught graduate level classes in the areas of digital signal processing, communications theory, and machine learning.