Removing harmonic signal nonstationarity by dynamic resampling

Signals produced by rotating machinery often display short term nonstationarity caused by small speed fluctuations. In some cases it is possible to control the data acquisition system sample rate so that Fourier spectral analysis of the signal is not affected by the small variations in signal harmonic frequencies. This is often referred to as order tracking data acquisition. However, electric ground vehicle sounds recorded remotely do not permit use of a local speed reference to control sample rate. This paper demonstrates that dynamic resampling of multi-harmonic acoustic signals in the presence of noise is feasible. A phase tracking algorithm is used to track a principal spectral component and to provide the necessary resampling information. Both synthetic and actual vehicle sounds are processed and improved Fourier power spectral estimates are obtained by simple averaging. If significant noise is present, the signal to noise ratio can be greatly improved by combining individual cycles of the reexampled waveform to approximate coherent detection (an audio demonstration of the improved signal to noise ratio is quite compelling). Also, it is shown that known interharmonic relative amplitude and phase information contained in the original nonstationary signal can be recovered by: tracking; dynamic resampling; coherent averaging; and performing a Fourier series decomposition. Finally, a wavelet approach to improve automatic fundamental frequency component recognition is discussed