Two asymptotic approaches for the exponential signal and harmonic noise in Singular Spectrum Analysis
Provides theoretical insights into SSA limitations for unbounded signals, relevant to time series analysts using SSA.
The paper analyzes asymptotic behavior of Singular Spectrum Analysis (SSA) for exponential signals with sinusoidal noise, showing that reconstruction errors do not uniformly vanish when the signal grows unbounded, but do vanish under a bounded discretization scheme.
The general theoretical approach to the asymptotic extraction of the signal series from the perturbed signal with the help of Singular Spectrum Analysis (briefly, SSA) was already outlined in Nekrutkin 2010, SII, v. 3, 297--319. In this paper we consider the example of such an analysis applied to the increasing exponential signal and the sinusoidal noise. It is proved that if the signal rapidly tends to infinity, then the so-called reconstruction errors of SSA do not uniformly tend to zero as the series length tends to infinity. More precisely, in this case any finite number of last terms of the error series do not tend to any finite or infinite values. On the contrary, for the "discretization" scheme with the bounded from above exponential signal, all elements of the error series tend to zero. This effect shows that the discretization model can be an effective tool in the theoretical SSA considerations with increasing signals.