Better narrowband spectral measurements

Abstract

Spectrum analysis based on traditional FFT techniques is a widely available, commonly understood and general purpose tool for use with mechanical systems. In applications such as system identification, modal analysis and rotating machinery signature analysis, relevant information content in signals is greatest near narrowband phenomena in the frequency domain. Effort should be focused on reducing bias error in these regions in preference to reducing variance error. Time aliasing methods of spectrum analysis can result in superior bias error reduction, producing an accurately sampled spectral estimate and permitting window function characteristics to be decoupled from the DFT resolution, or frequency line spacing. Windows created with time aliasing can be optimized by adjusting the amount of time or frequency localization, according to the type of spectrum being measured. Assuming conventional criteria, time aliased windows can achieve better mainlobe and sidelobe characteristics than other windows. An innovative approach, appropriate with continuous spectra, is to use the window as a discrete sampling function in frequency to eliminate bias error inherent in the effective convolution between window and signal transform. Both approaches have been successfully implemented and results are presented using time aliased spectral estimates showing more accurate measurements of narrowband phenomena when used in applications of rotating machinery vibration and system identification.