Nonlinear averagings and overlapping techniques for spectral analysis of noisy signals

The authors propose a new method for accurate spectral estimation of noisy signals. The major drawback caused by trade-off between spectral resolution and variance is examined. The main aim of the paper is to show how the accuracy of periodogram estimation can be improved by adopting nonlinear averaging techniques of partially overlapped time slice of data samples. Also, the effects of the taper functions and the percentage overlap both on the computational cost and the improvements achievable are investigated. The derived expressions for the bias and variance point out the better performance of this method with respect to previous ones. Finally, the experimental validations of the results are shown.