Conditioning of sonar data

Spectral source characterization is crucial in passive sonar surveillance. The appearance and pattern of spectral lines in the time-frequency domain are important indicators for source identification. The ability of the estimated spectrum to resolve the components of the source is dependent on the quality of the recorded data. The sonar data may be a composition of both narrowband components and broadband transients. Here, we use a stepwise outlier rejection algorithm for removal of non-stationarities. We compare several methods for imputation of the reduced data set. The methods are all based on finding alternative values to the rejected outliers in the bispectral domain. We demonstrate the performance of these methods by means of real sonar data.