Curvature-based spatially adaptive sampling of heteroskedastic data with application to Laser Doppler Velocimetry

Spatially varying signals are typically sampled by collecting uniformly spaced samples irrespective of the signal content. For signals with inhomogeneous information content, this leads to unnecessarily dense sampling in regions of low interest or insufficient sample density at important features (or both). A new adaptive sampling technique is presented directing sample collection in proportion to local information content, capturing adequately the short-period features while sparsely sampling less dynamic regions. The proposed method incorporates a data-adapted sampling strategy on the basis of signal curvature, sample space-filling, experimental uncertainty and iterative improvement. Numerical assessment has indicated a reduction in number of samples required to achieve a predefined uncertainty level overall while improving local accuracy for important features. The potential of the proposed method has been further demonstrated on the basis of a Laser Doppler Velocimetry experiment extracting wake velocity profiles.