Particle classification capabilities of a silicon dE/dX detector using neural networks

The classification capabilities of light nuclei by a silicon detector measuring up to 32 dE/dX samples of their energy release have been investigated taking into account ionization fluctuations and electronic noise in signal response. The detector has been described as a telescope made of silicon wafers, 6 x 6 cm(2) wide and 380 mu m thick, having tracking capability. It represents the detector used in the NINA space mission, devoted to measure low-energy light nuclei composition in space on a polar orbit at 700 km altitude. The nuclei classification is performed by a neural algorithm (NN) of the Multi Layer Perceptron type, having 32 input neurons. This approach ensures that all measured information is used for event classification. Data sets consisting of several classes of nuclei - from hydrogen to higher Z and mass numbers - have been simulated, both for training and test of the NN. The robustness of this classification approach, against fluctuations and noise, has been investigated using classification efficiency and contamination as merit figures. Results are presented for contained particle, in the energy interval from the lower threshold for triggering up to 30 silicon layers interested; namely, from about 10 MeV for protons to some hundred MeV for higher nuclei. (C) 1998 Published by Elsevier Science B,V. All rights reserved. RI Cafagna, Francesco/A-9299-2010