James F. Smith
Steven Finette
Acoust. Div., Naval Res. Lab., Washington, DC 20375
A deconvolution algorithm based on simulated annealing has been developed that allows the extraction of the source time series from passively sensed multichannel data measured on an array. The algorithm is general, different cost functions and annealing schedules can be easily implemented allowing various environmental conditions to be treated in an optimal fashion. The algorithm is intrinsically parallel, fast, and requires no assumptions about the initial form of the source time series. The intrinsic parallelism of the algorithm allows an excellent compromise to be reached between stability of the solution and rate of convergence of the simulated annealing process. Tests using simulated data constructed by convolving a known signal with a theoretical Green's function and then adding Gaussian white noise have been conducted. Excellent reconstructions of the source time series have been achieved for cases of high and low signal-to-noise ratio. This methodology is being tested on deep ocean data obtained in a multipath environment. Results from simulations and experimental data will be discussed. [Work supported by ONR and NRL.]