A neural online triggering system based on parallel processing

The study of a prototype of the second-level triggering system for operation at LHC conditions is addressed by means of a parallel machine implementation. The 16 node transputer based machine uses a fast digital signal processor acting as a coprocessor for optimizing signal processing applications. A C-language development environment is used for running all applications at ultimate speed. The implementation is based on information supplied by four detectors and includes two phases of system operation: feature extraction and global decision. Feature extraction for calorimeters and global decision processing are performed by means of neural networks. Preprocessing and neural network parameters rest in memory and the activation function is implemented using a look up table. Simulated data for the second-level trigger operation are used for performance evaluation.