Acoustic multipath identification with expectation-maximization

Deep-sea autonomous positioning is susceptible to unmodeled measurement errors. Acoustic range measurements for long baseline navigation are not normally distributed, a situation that erodes the robustness and precision of estimation techniques. A mixed-distribution model captures the combination of direct-path, multipath, and spurious returns common for acoustic localization in limited-range survey. Expectation-Maximization concurrently identifies the parameters of the model and characterizes the observations, converging on a model for homogeneous instrumented environments. The resulting representation aids autonomous navigation for precision applications.