Accurate automatic PDV jumpoff detection with CUSUM

Photonic Doppler Velocimetry (PDV) measures the speed of a moving surface by shining a laser on it and observing the Doppler shift of the reflection by heterodyne detection. PDV's high time resolution (< 1 ns) and dynamic range (m/s to km/s) make it a natural choice in the study of explosions. In this application it is often necessary to determine when the explosion has reached the area covered by the PDV probe (the jumpoff time) by inspecting the PDV signal. Currently, this is a bothersome and slow manual process, because the noisy, complicated PDV signal has defied attempts at automatic change detection. We present a robust automatic jumpoff detection method for PDV signals that is accurate to better than 1 ns. Our method is adaptable to varied experiment types and can be trained to new data by interactively annotating examples.