Conventional and binary time series models of sonar fluctuations

Abstract

The presence of received power fluctuations of underwater acoustic signals and noise is commonplace. This paper addresses the power fluctuations as measured at the output of a narrowband spectrum analyzer. The purpose of this paper is to develop and evaluate models of underwater sound fluctuations that can be used for short-term predictions. Conventional time series models were used, and short-term prediction errors were between 1.3 and 2.5 dB. A new binary time series is developed and results presented. The Binary Model predicts detections (+1) and no detections (-1). The forecast of detections/no detections with the Binary Model resulted in errors as low as eight percent for some of the sonar time series.