Speaker
Description
We present a trigger-free method for wavelength-shifting (WLS) fiber localization in a CMOS SPAD pixel array based on hierarchical DBSCAN clustering of dark count rate (DCR) spatial excess maps.
The methodology was evaluated using experimental datasets acquired from a prototype CMOS SPAD matrix readout platform. To test the pipeline under low signal-to-background conditions, spatial DCR and in-pixel counter excess maps were compiled from extended exposure measurements. Even with signal rates per pixel several orders of magnitude below the DCR background, statistically significant spatial excess patterns were isolated, achieving a maximum z-score of 5.93 in the counter channel.
A two-stage DBSCAN pipeline was applied in physical spatial coordinates using the pixel pitch as the spatial metric. A first pass identifies local microclusters, followed by a second pass over their centroids to form superclusters. Scanning the merging radius from 0.28 to 0.38 mm reveals a stable configuration of nine superclusters over a 50 µm range (0.315–0.330 mm), just before a strong merging transition, demonstrating robustness against parameter selection. The dominant supercluster localizes the fiber on the sensor active surface at (1.61 mm, 0.94 mm), with a spatial extent of 0.36×0.58 mm², consistent with the 1 mm physical fiber diameter.
These results demonstrate that unsupervised spatial clustering of DCR excess maps can localize WLS fibers without hardware triggers, offering a complementary calibration tool for SPAD-based particle detectors.