Particle filtering for frequency estimation from acoustic time-series in dispersive media

Acoustic signals propagating in the ocean carry information about geometry and environmental parameters within the propagation medium. Accurately retrieving this information leads us to effectively estimate parameters that are of utmost importance in environmental studies, climate monitoring, and defense. This dissertation focuses on the development of sequential Bayesian filtering methods to obtain accurate estimates of instantaneous frequencies using Short Term Fourier Transforms within the acoustic field measured at an array of hydrophones, which can be used in a subsequent step for the estimation of propagation related parameters. We develop a particle filter to estimate these frequencies along with modal amplitudes, variance, model order. In the first part of our work, we consider a Gaussian model for the error in the data measurements, which has been the standard approach in instantaneous frequency estimation to date. We here design a filter that identifies the true structure of the data errors and implement a [chi]^2 model to capture this structure appropriately. We demonstrate both with synthetic and real data that our approach is superior to the conventional method, especially for low Signal-to-Noise-Ratios.

Name thesis:degree_name

Doctor of Philosophy in Mathematical Sciences - (Ph.D.)

Discipline thesis:degree_discipline

Mathematical Sciences

Year

2014

Author dc:creator

• Aunsri, Nattapol

Contributors dc:contributor

• Eliza Zoi-Heleni Michalopoulou
• Ali Abdi
• Sunil Kumar Dhar

• Frequency estimation
• Particle filtering
• Bayesian filtering
• Sequential Monte Carlo
• Underwater acoustics
• Signal processing
• Mathematics

Repository record dc:identifier

https://digitalcommons.njit.edu/dissertations/139

OAI identifier oai:identifier

oai:digitalcommons.njit.edu:dissertations-1194