neural net reconstruction

  • DIRC@EIC, ePIC, eicdirc
  • 3-layer spherical lens with sapphire, r=[62,36]
  • 3 mm pixels HIQ400, CE=95%
  • pi/K @ 6 GeV/c
  • 0.5 mrad track smearing at tracking layer
  • phi = 0 (no field); vertex y-shifted by 0.5*barwidth (to hit middle of the bar)
  • 100 ps time precision
  • geometrical reconstruction with averaged LUT from 15M photons
  • GR with per-PMT and chromatic corrections
  • 20k tracks for pdf generation (per each particle species) (generated with 0.5 mrad track resolution); 5k for determining separation

One layer NN, using only pixel information:

After fixing input:

NN eff: 94.3%

GR eff: 96.6%

TI eff: 97.6%

Input data structure. Old:

Input data structure. New:

NN eff for new data structure: 97.8%