Nolan R. Davis
Code 5160, Naval Res. Lab., Washington, DC 20375-5000
This work considers a systematic approach to data inversion and data fusion for stationary passive sonar in low signal-to-noise situations. Bayesian inversion is applied to the probability distributions that are implicit in conventional signal processing methods. The resulting source location probability distributions are multimodal, reflecting the sidelobe structure of conventional ambiguity functions. Using the probability interpretation of these distributions the secondary sidelobe peaks can be compared quantitatively with the mainlobe. Results of model calculations are presented for an ocean waveguide in order to demonstrate the method, provide a comparison with conventional approaches, and assess the performance under low signal-to-noise conditions. A preliminary discussion of data fusion is given for probability distributions derived from inversion of independent data sets. An application to frequency fusion is made, and a performance improvement is demonstrated with the computational model.