Time-Frequency Feature Extraction from Nonstationary Multiple Time Series and its Applications
Abstract
A central question of time-frequency analysis is how to decompose a signal with timevarying oscillatory properties into several components with distinct amplitude and frequency behaviors. The time-frequency characteristics of the input signal visualized by the popular spectrogram are often too blurred to decompose the signal reliably into a finite sum of amplitude-phase components and extract the instantaneous amplitude (IA) and instantaneous frequency (IF) of each component. We propose to use the so-called Synchrosqueezing transform (SST) to sharpen the spectrogram for reliable retrieval of IAs and IFs of an input signal, which allows us to analyze and interpret each component separately. However, the current SST theory has several limitations when dealing with real nonstationary signals including sonar signals and music recordings, which often have strongly modulated/discontinuous IAs and/or highly-varying IFs. In order to mitigate them, we will investigate the use of more adaptive time-frequency representations in the SST, e.g., the quilted Gabor transform, the Chirplet transform, and the prolate spheroidal wave functions. As applications of our tools, we propose to analyze sonar recordings and separate out the acoustic waves reflected or emitted by man-made objects from other unwanted sounds (e.g., whale clicks, ambient noise, etc.), which should help the Navy’s effort on Automatic Target Recognition (ATR). In addition, we propose to develop and test a procedure to map each amplitude-phase component in the data to a sequence of musical notes and assign a synthetic musical instrument so that we can hear and characterize the data with our own auditory system, a process known as “data sonification.” For this purpose, we plan to use temperature measurements made at different depths of a lake for aquatic ecosystem studies.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 09, 2016
- Source ID
- N000141612255
Entities
People
- Naoki Saito
Organizations
- Office of Naval Research
- United States Navy
- University of California, Davis