Global testing against sparse alternatives in time-frequency analysis (Q309712)

The authors' aims of this paper are to establish a number of results on global testing for periodicity. They construct a new test based on an over-sampled periodogram that adapts to a growing number of sinusoids with general frequences. The focus of this work is to find the asymptotic optimality of this method. Their main result guarantees that as long as frequencies of the complex sinusoids in the mean of \(y=X\beta+z\) obey some minimum separation condition, this over-sampling rate leads to an asymptotically optimal global test where \(X\in \mathbb C^{N\times p}\) with \(p\gg N\) is an extended Fourier transform matrix and \(z\) represents white noise.