Optimizing time-pickup algorithms in radiation detectors with a genetic algorithm.

Abstract Pulse digitization at high rates has paved the way for the use of complex, non-standard algorithms, beyond the conventional constant fraction discriminator (CFD) option to obtain the time-stamps of gamma photons arriving to nuclear detectors. Detector pulses can be shaped with digital filters in digital signal processing (DSP) hardware or software, and timestamps can be derived by interpolating or extrapolating the acquired samples. Digitized pulses can be stored in a computer for off-line analysis allowing for many different algorithm variations to be tried on the same data. One difficulty is that the number of filters and parameters to be tuned in order to optimize the timing performance may become too large to be done by hand. For this, in this work we use a genetic algorithm to tune in-silico digital filters, digital time pickup algorithms and their associated parameters. The data set consisted of 500,000 pulses obtained from the last dynode of two ultrafast photomultiplier tubes (model R9779 by Hamamatsu) coupled to monolithic fast inorganic LaBr 3 (Ce) scintillators, in a truncated cone geometry 1.5 × 1 × 1.5". With finely tuned conventional (CFD+TAC) timing electronics, a CRT of 155 ps (FWHM) for a 60Co source has been measured by our group, within the state of the art for detectors of this size. Our alternate digital, in-silico implementation of CFD+TAC, fine-tuned with the same parameters obtained during manual adjustment of the conventional DAQ and applied to pulses digitized at 5 GS/s and 14 bits of vertical resolution, reproduces the performance of the conventional DAQ. On the other hand, a genetic algorithm (GA) is employed to optimize digital recursive filters with up to 8 parameters, and different time-pickup strategies (upper level thresholding, CFD, time extrapolation). The same 500,000 pulses are employed for which the GA finds several families of filters outperforming the manually tuned algorithm by more than 10%. [ABSTRACT FROM AUTHOR]