Real time analysis for high bandwidth experiments

Supervisor: Themis Bowcock

High bandwidth experiments are those that produce many TBytes of data per second (current examples are the ones running at the CERN/LHC). Analysis of the data from these experiments takes places in two interlinked phases, first an online phase which receives, processes and refines that data and makes a selection on the information that is retained for further analysis, or that is flagged for immediate attention. The second is a high level refined reanalysis of the data that is performed many months later on aggregate data sets. At this point it is often too late to identify problems, refine or optimize the data collection.  This project is dedicated to the development of a general toolset for the real-time, first stage, analysis. The tools that are used for this process are numerous, ranging from advanced pattern recognition and fitting to physics channel selection algorithms. In all cases the resources are limited and in all cases the final physics performance of the experiment is critically dependent on them. We will develop a set of demonstrator tools based on typical applications (from LHCb and mu3e) These will: a) demonstrate high bandwidth pattern recognition with an adaptive ability (for example to compensate for detector degradation), to ensure the highest possible throughput and purity of the data sample; b) deliver a high level signal “tracker” that provides a monitor of the functioning of the detector and flags and tries to identify (in real time) problems in hardware, firmware or software or changes in the environment; c) enable the  ability to discover regions of interest where the data is evidencing divergence from theoretical and phenomenogical expectation and refine automatically the searches to discover if these are  statistically significant.