Least-squares time-of-arrival estimation for transient signals in a multipath environment

The authors consider the problem of estimating the arrival times of overlapping ocean-acoustic signals from a noise received waveform which consists of scaled and delayed replicas of a deterministic transient signal. They assume that the transmitted signal and the number of paths in the multipath environment are known and consider the algorithms that give least-squares estimates of the amplitude and time delay of each path. The Gauss-Newton algorithm converges quickly to a minimum of the error surface, but if the initial estimates are not sufficiently accurate, only a local minimum will be found. The authors compare several different methods for obtaining initial estimates, including coordinate descent and linear prediction. They demonstrate the performance of the complete algorithm (initial estimates followed by a Gauss-Newton search) on experimental data