Monte Carlo uncertainty analysis of an ANN-based spectral analysis method

Abstract

This work presents the uncertainty analysis of an artificial neural network (ANN)-based method, called multiharmonic ANN fitting method (MANNFM), which is able to obtain, at a metrological level, the spectrum of asynchronously sampled periodical signals. For sinusoidal and harmonic content signals, jitter and quantization noise contributions to uncertainty are considered in order to obtain amplitude and phase uncertainties using Monte Carlo method. The analysis performed identifies also both contributions to uncertainty for different parameters laboratory configurations. The analysis is performed simultaneously with our method and two others: discrete Fourier transform (DFT), for synchronously sampled signals, and multiharmonic sine-fitting method (MSFM), for asynchronously sampled signals, in order to compare them in terms of uncertainty. Regarding asynchronous methods, results show that MANNFM provides the same uncertainties than MSFM, with the advantage of a simpler implementation. Regarding asynchronous and synchronous methods comparison, results for sinusoidal signals show that MANNFM has the same uncertainty as DFT for amplitude and higher uncertainty for phase values; for signals with harmonic content, amplitude conclusions maintain but, regarding phase, both MANNFM and DFT uncertainties become closer as the frequency increases, which implies, in fact, that when synchronous sampling is not possible, spectrum analysis can be performed with asynchronous methods without incurring in uncertainty increases.