Fast array-based particle coincidence detection in a TimePix3-based velocity map imaging instrument

Abstract

With the development of high repetition rate laser sources and advanced multi-particle correlation analyses such as covariance mapping, particle detection techniques such as velocity map imaging (VMI) are poised to offer unprecedented views into molecular phenomena. Taking full advantage of the high count rates in these experiments requires the development of detectors with sufficient spatial and temporal resolution that can process data in real time. The TimePix3 camera (TPX3CAM) is an event-based pixel detector capable of spatio-temporally localizing many simultaneous particle hits in an efficient manner. While the sparse nature of the data stream allows for compact representation of particle hits, it also presents algorithmic and computational challenges for clustering individual pixels into hits. Here we present the theory and application of a rapid data processing and centroiding algorithm for ion and electron hits collected in a VMI instrument. The array-based computations that comprise the algorithm take full advantage of the data sparsity of the TimePix3 data stream and localize particle hits on the microchannel plate (MCP) to better than a single pixel on the pixel detector. Centroiding can be parallelized on a commercially available graphics processing unit (GPU) for additional speed. Using these innovations, data processing occurs about 25 times faster than data acquisition, for a 1 kHz repetition rate instrument and tens of particles per shot. In addition to its speed, the TPX3CAM detector outperforms state-of-the-art delay line anode detectors at discriminating multiple simultaneous hits, enabling high-fidelity coincidence and covariance studies in the near future.

Figures (6)

• Figure 1: (a) Illustrated phosphor light curves (solid blue and red curves) along with TimePix3 threshold value (dashed gray line). TimePix3 pixels report their X and Y positions along with the time the signal first went over threshold (Time of Arrival, ToA), and the duration for which the pixel remained over threshold (Time over Threshold, ToT). (b) Example TPX3CAM data stream. Each line represents a pixel event or laser trigger.
• Figure 2: (a) Joint histogram of pixel ToT and ToF values for a typical data collection. The ToT-ToF correlation can be seen at the lowest ToT values, where the center of the distribution is shifted to larger ToF values. (b) Corrected ToT-ToF values after calibration. (c) Example Gaussian fits to the distribution of ToF values for individual ToT slices. The ToF center obtained from the fit moves earlier as the ToT value increases. (d) Plotting the ToF center $\langle \mathrm{ToF}\rangle$ obtained from the Gaussian fit vs. ToT gives the relationship between the two variables. This relationship is fit to the mathematical model shown in the figure and in Equation \ref{['eq:tottof']}, the results of which are plotted in red.
• Figure 3: Example of raw data and centroiding from a single laser shot. (a) X-Y histogram of all pixel data reported on a single shot, with brightness weighted by each pixel's ToT value. Inset: a zoomed particle hit to show distribution of pixel brightness. (b) X-ToF histogram of the same data in the previous panel. Each hit must be centroided in three dimensions, namely X, Y, and ToF. Inset: the same zoomed particle hit as shown in panel (a). Centers of the white circles represent the centroid locations.
• Figure 4: Batching and zero-padding of multiple laser triggers of pixel data for centroiding. Several laser triggers worth of pixel data can be centroided simultaneously on the GPU by forming an array with one extra dimension along the laser trigger axis and applying all of the centroiding operations to the other axes. Zero-padding of the array data must be performed since each laser trigger contains a different number of pixel events.
• Figure 5: Comparison of the X-Y histogram of electron hits from above-threshold ionization of argon at 400 nm wavelength for (a) raw pixel data, and (b) centroids with 0.5-pixel bin width. (c) Underlying photoelectron spectrum obtained via an inverse Abel transform performed on the raw pixel data and the centroided data. The various features in the photoelectron spectrum are sharpened by the centroiding.