Multi-parameter digital waveform feature extraction method for event position localization

When the detection area increases and the number of SiPMs grows, accurate event localization becomes increasingly critical. This paper proposes a new event localization method for large-area silicon photomultiplier (SiPM) array readout, overcoming the limitations of conventional pulse localization approaches, which rely on amplitude ratios between pulses. Due to the resistor network, signals from SiPMs at different locations undergo distortions. We proposed a simplified model, a tri-exponential model to explain these distortions and leveraged it to design a pulse-shape discrimination (PSD) algorithm that extracts 12 position-dependent features from each signal. These features are then classified using a machine learning approach to generate a model capable of accurately determining the interaction position. The new method achieves over 95% accuracy in identifying interaction positions within an 8 × 8 detector array. • Compared to our previous work, the new method doesn‘t need ASIC like CITIROC 1A or special read-out chips, making a more flexible and cost-effective hardware solution for reconstructing the point of incidence. • Benefiting from the resistor network, signal from each SiPM in the array generates 12 feature values after passing through the method, more feature values lead to better results, the final localization accuracy is better than 95%, even the misidentified positions are adjacent to the correct ones. • The method is based on trapezoidal pulse-shaping algorithm, which is simple enough to implement on FPGA or MCU. • The resistor network reduces the signals from an array of 64 SiPMs into 4. Each of these 4 signals is subsequently processed using a trapezoidal shaping algorithm to extract 3 parameters. Thus, each SiPM position gets 12 feature values after processing through the resistor network and the algorithm.