| The main objective of this study is to use non parametric methods to solve the problem of spectral estimation of missing observations in experiments. In this thesis, six methods are proposed and they are supported by MATLAB simulations that show their results.; The first method used was based on the averaging of nonoverlapping segments to obtain an estimate of the power spectrum. The second method used is based on Welch's periodogram for spectral estimation. In this case, the segments created are allowed to overlap.; The third and fourth methods used are modified versions of the Welch's method as proposed by Poularikas (Course notes EE-748), with the difference occurring during the segmentation of the signals. These methods offer better frequencies resolutions than the first two methods.; The fifth method was the introduction of a set of random data to fill the gap caused by the missing observations. The advantage offered by this method is the reduction of the sidelobes that are usually created by the windows used.; The last method proposed is based on the estimation of the autocorrelation function of the time sequence. Unlike the previous cases where the windows are applied directly on the sequence under study, this approach allows getting the autocorrelation function of the sequence under study instead, and the window is applied to the latter. |
