Evaluation of real-time event selection mechanisms for the CP-LEAR experiment

CP-LEAR is an experiment dedicated to the measurement, with high accuracy, of several CP violation parameters. Its trigger system, for the on-line selection of interesting events, is an important part of the detector. The implementation, of the part of the trigger, that calculates the z component of the momentum of each primary track, is presented. In addition, the method, of choosing the best on-line kinematical selection, is described. The method finally selected reduces the number of events by a factor 1.67, by, keeping at the same time, all good events. A model of the function of the multi-stage trigger system was developed. It allows the comprehension of the behavior of the system, given the reduction ratio of each individual stage and its delay time. The optimum beam intensity is found to be 800KHz. Then 450 events per second would be written on tape. The performance of a new trigger decision, based on the results of the on-line calculations, was also investigated. Two possibilities were studied, a traditional selection method, implementing off-line cuts on-line, and a neural network algorithm. The performance of both systems is comparable, leading to a rate reduction of a factor of two, by accepting 80% of the interesting events. Finally, a new pattern recognition algorithm has been developed. It is based on a neural network architecture, and it is implementable in hardware. Its performance compares well with that of classical algorithms. It is indeed better than the current system used in the trigger, as it recognizes all the existing tracks of an event, and finds just 6% more tracks than what it should.