Spacing and shape of random peaks in non-parametric spectrum estimates.

ABSTRACT: In this paper, expressions are derived for the expected number of spurious peaks in a spectrum estimate, that is, crossings above a given significance level per frequency unit, as well as the expected width of these peaks. In numerous scientific applications, spectrum estimates are used for the purpose of identifying sinusoidal or modal components, often thinning large sets of candidate frequencies with coincidence detection. Because one always expects numerous false peaks in a spectrum estimate, knowing the expected rate of false peaks helps to decide whether the number observed is abnormal and hence determine the true nature of the process. An example using solar wind data from the Advanced Composition Explorer is given where spectra display pathological numbers of significant peaks, whi

SUBMITTER: Thomson DJ 

PROVIDER: S-EPMC4032558 | biostudies-literature | 2014 Jul

REPOSITORIES: biostudies-literature