Impact of windowing and subsampling algorithms on acoustic scattering strength databases

Abstract

This paper describes the results of a study of potential artifacts in seafloor acoustic scattering strength databases through algorithmic pathologies. Specifically, several methods of "windowing" and subsampling timeseries reverberation data for the estimation of statistical parameters and the resulting georeferenced scattering strength database are evaluated. Data for this study were collected using an ANISQS-53C sonar system and processed with the Sonar Active Acoustic Boundary Loss Estimation (SABLE) [ l ] system. Traditional methods of estimating scattering strength employ overlapping, fixed-length, periodic, windows (block processing) of the time series from which estimators of energy and other parameters are computed. In scientific experiments with limited objectives, viz., the calculation of a single parameter such as Scattering strength at a specific location, computational efficiency, and the size of data sets are not important concerns. When objectives are to compute super-resolution multipurpose databases over large geographic areas, then computational efficiency, data rates, and the size of the resulting database ate important issues. In this study, a number of georeferenced scattering strength databases were created from the same data using different windowing and subsampling schemes. Two nontraditional methods used sampling based on a combination of thresholding and grouping of samples in time.The third, more traditional method, uses overlapping window sampling. An analysis of scattering strength deduced from databases obtained using 11 different sampling parameterizations is reported in this paper.The approach compares the algorithms on a pair-wise basis in terms of estimated bottom scatter strength. The comparison includes the scatter strength estimate and a measure of its statistical reliability. Reverberation time series from the SQS-53C sonar were processed with several windowing and subsampling methods and sorted into georeferenced grid cells. The standard deviation of scatter strength was computed for each geocell, and those with standard deviation greater than 8 dB were removed from the analysis. An ensemble of cell-totell scatter strength differences was defined for each pair of algorithms. For each ensemble a mean difference, rootmean-square differences, and an estimation error were computed. These results were displayed in a twodimensional pseudotolor matrix.The mean scatter strength over all geocells was computed for each database. The mean differences between the various sampling methods were all within 2 dB.The nontraditional algorithms produce smaller databases than did block processing. For example, with a threshold of 8 dB. the database size was 23 MB compared to 134 MB for periodic window processing with a window size half the transmitted pulse length and with 50-percent overlap.

Variability of observed reverberation and estimated sea-floor scattering strength

“…Intervals are usually selected to be of uniform length and can be overlapped or independent. Other methods of selecting intervals are also available [2] but will not be discussed here. Once statistical measures of each interval are determined, each interval is geo-referenced using the sensor position, beam steering information, mean receive time for the interval and the estimated mean sound speed.…”