Pulse identification and shape analysis by derivative-based peak detection using a Convolutional Neural Network

The shape analysis of electrical pulses is a well-established technique adopted in the experiments of nuclear physics, in particular to identify the particles revealed by the detectors. The advent of the signal digitization has led to easier and more accurate Pulse Shape Analysis to measure the physical quantities embedded in each pulse (rise time, widths, slopes, etc.) that allow to trace back to the properties of the detected particles. In case of significant noise overlapping the pulses, a strategy aimed at its reduction is usually achieved in the frequency domain by appropriate Fourier and Wavelet transforms. The drawback of this approach is represented by the noise harmonics overlapping the signals, the filtering of which can alter the embedded physical contents in the pulses and affect the particle discrimination performances. An alternative technique is represented by the neural network algorithms that can perform pulse shape analysis even in the presence of a significant amount of noise, thus ensuring high recognition efficiency and high spectroscopy performances without the need of signal filtering. In this paper we describe a new technique based on Convolutional Neural Networks to recognize and analyze the output signals of charge preamplifiers fed by silicon detectors.