Division of Port and Ocean Engineering, University of Trondheim, Norwegian Institute of Technology, N-7034, Trondheim, Norway
Abstract:
In traditional spectral analysis the use of windows is required because the assumption is made that the data outside the sample are zero. A data window is simply equal to unity inside, and zero outside the sample. This is equivalent to the truncation of the autocovariance function beyond some lag after which zeros are added. In Fast Fourier Transform spectral estimation a spectral window with smaller or greater negative side lobes is used. Any use of windows results in a smearing or spectral leakage that limits the spectral resolution. Maximum entropy spectral estimation (MEM) is equivalent to an extrapolation of the autocovariance function being consistent with some model assumptions. The result of the extrapolation is an increased spectral resolution in the frequency domain. In applications of MEM no kind of windows are used. MEM is applied to describe wind wave scalar spectra. The technique can also be used in the plane to estimate directional spectra. Our results show that MEM is a powerful tool for estimation of scalar spectra and simulation of the sea surface. There are also theoretical reasons for assuming that MEM are superior to traditional methods when only short samples are available. Our preliminary results verify this assumption. We propose a data acquisition system based on MEM, and also show that the response of linear systems can be calculated and simulated by MEM.