Premium
Detection of hypervelocity impact radio frequency pulses through prior constrained source separation
Author(s) -
Nuttall Andrew,
Kochenderfer Mykel,
Close Sigrid
Publication year - 2016
Publication title -
radio science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.371
H-Index - 84
eISSN - 1944-799X
pISSN - 0048-6604
DOI - 10.1002/2016rs006108
Subject(s) - hypervelocity , blind signal separation , computer science , noise (video) , source separation , underdetermined system , signal (programming language) , radio frequency , probabilistic logic , synthetic data , acceleration , algorithm , acoustics , remote sensing , physics , artificial intelligence , telecommunications , geology , channel (broadcasting) , classical mechanics , image (mathematics) , thermodynamics , programming language
Abstract Hypervelocity dust impacts produce electromagnetic pulses in the radio frequency (RF) spectrum that scales with impactor mass and velocity. Due to the mass acceleration limitations of ground‐based facilities, detecting emissions from impacts in a laboratory setup is difficult due to their low output power. This paper presents a general probabilistic technique to perform signal excision, which was applied to synthetic and hypervelocity impact data sets. The task of excising multiple signals from a single observation of their mixtures is referred to as underdetermined blind source separation (BSS). This paper introduces a framework for solving underdetermined BSS problems when there is only one observation signal by leveraging often overlooked prior information. The most probable solutions for the source signals are computed by solving an iterative constrained optimization problem that seeks to maximize the posterior probability of the system model. In the hypervelocity impact data set, the goal was to reduce the noise floor on an RF antenna by modeling and extracting exterior sources of noise. It was found that the algorithm described in this paper was able to model signals in the observation and subtract them while still maintaining the spectral and temporal content of the remaining signals. Through the use of this methodology, previously hidden impact emissions were able to be isolated and identified for further characterization.