Signal-processing techniques for low signal-to-noise ratio laser Doppler velocimetry signals

Abstract

A variety of methods have been developed to obtain acurate frequency estimates from laser Doppler velocimetry (LDV) signals. Rapid scanning and fiber optic LDV systems require robust methods for extracting accurate frequency estimates with computational efficiency from data with poor signal-to-noise ratios. These methods typically fall into two general categories, time domain parametric techniques and frequency domain techniques. The frequency domain approach is initiated by transforming the Doppler bursts into the frequency domain using the fast Fourier transform (FFT). From this basic transformation a variety of interpolation procedures (parabolic, Gaussian, and centroid fits) have been developed to optimize the frequency estimation accuracy. The time domain approaches are derived from the parametric form of a sinusoid. The estimation of constants in this relationship is performed to satisfy specific constraints, typically a minimization of a variance expression. A comparison of these techniques is presented using simulated signals and additive Gaussian and Poisson white noise. The statistical bias and random errors for each method are presented from 200 signal simulations at each condition. Frequency estimation via the FFT with zero-padding and a Gaussian interpolation scheme was found to produce the lowest bias and random errors.