Coherently averaged power spectral estimate for signal detection

Abstract

A common approach to detect sinusoidal signals buried in noise is based on spectral analysis, such as the periodogram. The fluctuations of the spectral components associated with the noise can be alleviated via incoherent averaging of the power spectral estimates of each segment, which is the basis of Welch’s method. However, Welch’s method only utilizes the incoherent information between segments of signals. In this paper, we propose a method of coherent averaging between segments, which enhances ratio of time-invariant sinusoidal signals relative to the level of the noise background. The gain of coherent averaged power spectral estimate has been derived in terms of time duration of the signal. The proposed method provides a flexible, computationally efficient implementation of signal detection, which can be formulated to allow for various integration times to be realised in different frequency bands. Simulation and experimental results show that the proposed method outperforms the Welch’s method and the periodogram method.