Minimum Variance Power Spectral Estimation of Noisy Signals With Improved SNR

Power spectral analysis of a signal is nothing but analyzing the signal power as a function of frequency components in the signal. Periodogram, the fundamental power spectral estimation technique has the limitation that it is not a consistent estimate, but good at frequency resolution. Hence a number of modifications have been suggested in literature like Bartlett and Welch etc., to improve the statistical properties, especially to increase the consistency by reducing the variance of the estimate at the expense of frequency resolution. One important task in spectral estimation is to estimate frequencies in noisy background with high resolution. This requirement can be achieved in two steps. One is to develop a good spectral estimate in the sense it should be a consistent estimate and also provide high resolution. Second part is to enhance the signal detection and/or estimation by reducing the background noise. Hence, in this Paper, a new modification for estimation of Periodogram, which provides minimum variance and good resolution and a new denoising algorithm to estimate the spectral frequencies of narrow band signals in unknown background noise with minimal information about signal characteristics, has been proposed. This technique performs well in case of signals with and without noise compared to the previous algorithms.