Short-time spectral estimation of acoustic Doppler current meter data

Abstract

Estimation of water currents by using acoustic Doppler current meters is a well-known method. In these applications, determination of the water current is based on the Doppler shift caused by the movement of many small particles in the water, which are related to the current as an average. These movements cause a Doppler shift with an assumed Gaussian-shaped distribution in time. Furthermore, the Doppler signal is assumed to consist of many reflections with the same frequency, but with random phase. Many bubble sizes are involved in the scattering, and there are other phenomena contributing to the received signal. The Gaussian distribution assumption is thus likely to become inaccurate. In real life, there will be a complex signal in both time and frequency. This signal will consist of several frequency components with a different amplitude and phase. In the study of a new signal model for acoustic Doppler current meters, a thorough investigation of the signal behavior in time-, frequency- as well as amplitude domain has been made. By using both parametric as well as non-parametric estimation methods, the short-time frequency behavior has been analyzed. The estimated amplitude and frequency components are analyzed in terms of bias and variance, and a comparison of different estimators is made. Estimation of Doppler frequency and its variance has been made using a set of data recorded at the Trubaduren light house off the Swedish coast. The results are the basis of the formulation of a novel nonGaussian signal model for acoustic Doppler current meters.