Detecting Nonstationary Signals in Noisy Data Using Several Tapers

We present a multiple taper method for detecting nonstationaty signals in a time series based on a statistical test that measures the confidence one can assign to a signal's existence at any given frequency. We consider the problem of detecting decaying sinusoids immersed in additive white noise. This type of data is common in the analysis of mechanical vibrations and nuclear magnetic resonance experiments. The data are multipled by K tapers derived by a variational principle, creating K independent time series.