Digital processing of acoustic signals with application to an ASW signal processor

Abstract

There is a growing need within the Na\ry for methods of detecting discrete narrowband signals in a non-stationary background. This paper concerns itself with the application of digital processing and spectral analysis techniques toward that goal. The use of the fast Fourier Transform in estimating the power spectrum of a signal is described. The method involves sectioning the time record, making "raw" estimates of the spectrum from these sections, and averaging these "raw" estimates. It is shown that more stable estimates are available if the segments are overlapped and an optimum amount of overlap for the case of the Manning Window is found. It is shown that the stability of these spectral estimates can be interpreted as processing gain in the case of a discrete narrowband signal in additive noise. And finally, a brief description of signal detection theory applied to a human observer is presented to emphasize the flexibility that a human operator can bring to a signal detection system.