Sampling-ADC based Pile-Up Recovery and Digital Signal Processing in the CBELSA/TAPS-Experiment

Abstract

The CBELSA/TAPS experiment in Bonn investigates the spectrum and properties of baryon resonances by performing photoproduction experiments off the nucleon to reach a better understanding of quantum chromodynamics in the non-perturbative regime. The main calorimeter of the experiment, the Crystal Barrel, consists of 1320 scintillating CsI(Tl) crystals which are read out by avalanche photodiodes and an FPGA-controlled sampling analog-to-digital converter (SADC) readout system. Using a custom firmware enables access to the sampled waveforms and the extraction of different features from this data. Its main advantages over the previous readout lie in increased rate capabilities, an event-based baseline subtraction and the possibility to detect and correct cases of pile-up. <br /> Significant parts of this work were spent on testing and enhancing feature extraction algorithms, with improvements based upon findings from this thesis. The baseline-finding algorithm was rewritten and improved, further enhancing the energy resolution. Timing capabilities of the constant fraction discriminator were enhanced, surpassing the capabilities of the dedicated current time readout for <em>E</em> < 8MeV. In addition, a digital filter was proposed and implemented, which more efficiently suppresses high frequency noise above 1MHz in the sampled data, improving the performance of the subsequent feature extraction. <br /> A central achievement was the development and refinement of pile-up detection and recovery algorithms, addressing previous algorithmic limitations which led to a considerable amount of unrecognized pile-up in the data. A new detection approach was proposed and implemented, which shows significant improvements and does not exhibit systematic weaknesses anymore. Specific attention was paid to the interplay between the pile-up detection and the several feature extraction algorithms. Some specific causes of pile-up, such as <em>&mu;</em>^&plusmn; at rest, were identified and discussed. Furthermore, the pulse shapes recorded by the new readout setup were carefully examined. Since CsI(Tl) is used, they depend on the type of detected particle. The ensuing complications and opportunities with regards to a potential particle identification, specifically a potential discrimination between <em>p</em> and <em>γ</em> pulses, were investigated and implications for the feature extraction and pile-up detection and recovery process were discussed. <br /> A custom deconvolution method was developed and employed to disentangle the pile-up signals, which was found to perform far better for the waveforms at hand than a standard moving window deconvolution. It is followed by a custom fitting and evaluation routine, which takes care of more complex pile-up topologies. The end result is a robust method which achieves a satisfactory performance for over 99% of cases. Finally, the positive effects on the reconstruction of physical particles from corrected data was discussed as well.