1 M. Tiemens The Latest Developments on the Online Cluster Finding Algorithms ?
2 M. Tiemens Hit Creation Cluster
3 M. Tiemens Topology of the Data Stream ∆ τ ∆ t Datastream: ∆ , τ defined by time difference between Two time-scales: consecutive hits ∆ t , the time difference between any pair of hits
4 M. Tiemens Topology of the Data Stream ∆ τ ∆ t Datastream: because t ( E ), timestamps Hit belonging to the same cluster can be spread out drastically 4D position needed to disentangle hits
5 M. Tiemens Scenarios Cluster Hit Spatially fully No gain Make ∆ t large to separated clusters when using make sure no hits ∆ t are excluded
6 M. Tiemens Scenarios Cluster Spatially Use ∆ t to May not help overlapping clusters separate within events
7 M. Tiemens Scenarios Cluster Spatially Use ∆ t to overlapping clusters separate
8 M. Tiemens Time is Important! However, there are problems:
9 M. Tiemens Time is Important! The evil-doer:
10 M. Tiemens Time is Important! The evil-doer: The prize to pay: No more pile-up recovery
11 M. Tiemens Time is Important! The prize to pay: No more pile-up recovery Possible solution: Shorten length of waveforms HOWEVER Severely compromises reconstruction effeciency! WHY? Unknown... Some bug in the tasks?
12 M. Tiemens 'Default' Cluster Finding ( PndEmcMakeCluster ) Datastream: ∆τ … Yes Yes Is neighbour? Is neighbour? Hit No No … Hit Hit
13 M. Tiemens 'Default' Cluster Finding ( PndEmcMakeCluster ) Datastream: ∆τ … Grow clusters from “seeds” Clusters Issues: ● For each new hit, check if neighbour to cluster ≡ check if neighbour to member hits ● Large number of loops → clock cycles on a computing chip → latency
14 M. Tiemens 'Default' Cluster Finding ( PndEmcMakeCluster ) Datastream: ∆τ … Grow clusters from “seeds” Clusters r Get 4-Momenta: IP EMC
15 M. Tiemens Online Cluster Finding ( PndEmcMakeClusterOnline ) Datastream: ∆τ … Build hit neighbour relations, make clusters Clusters
16 M. Tiemens Distributed Cluster Finding ( PndEmcDistributedClustering ) Datastream: ∆τ … Assign to virtual Data Concentrators, build hit neighbour relations, make preclusters Preclusters DC1 DC2
17 M. Tiemens Distributed Cluster Finding ( PndEmcDistributedClustering ) Datastream: ∆τ … Assign to virtual Data Concentrators, build hit neighbour relations, make preclusters Preclusters DC1 DC2 Build precluster neighbour relations, make clusters Clusters
18 M. Tiemens 2-Step Cluster Finding ( PndEmcMakePreclusters ) Datastream: ∆τ … Assign to DCs, build hit neighbour relations, make preclusters DC1 DC2 Preclusters
19 M. Tiemens 2-Step Cluster Finding ( PndEmcMakePreclusters ) Datastream: ∆τ … Assign to DCs, build hit neighbour relations, … make preclusters Repeat for all timebunches, build precluster datastream DC1 DC2 Preclusters Precluster datastream
20 M. Tiemens 2-Step Cluster Finding ( PndEmcMergePreclusters ) Datastream: ∆τ … Assign to DCs, build hit neighbour relations, … make preclusters Repeat for all timebunches, build precluster datastream ∆τ DC1 DC2 Preclusters Precluster datastream
21 M. Tiemens 2-Step Cluster Finding ( PndEmcMergePreclusters ) Datastream: ∆τ … Assign to DCs, build hit neighbour relations, … make preclusters Repeat for all timebunches, build precluster datastream ∆τ DC1 DC2 Build precluster Preclusters Precluster neighbour relations, make clusters datastream Clusters
22 M. Tiemens Testing the Algorithms
23 M. Tiemens A More Challenging Channel
24 M. Tiemens Finding Optima for ∆ τ and ∆t Low rates <2 MHz If (Integral ≈ #events): Lower bound = ∆τ ∆t? Integrate
25 M. Tiemens Finding Optima for ∆ τ and ∆t Low rates <2 MHz
26 M. Tiemens Finding Optima for ∆ τ and ∆t High rate: 20 MHz
27 M. Tiemens Finding Optima for ∆ τ and ∆t High rate: 20 MHz Integrating doesn’t work because of event mixing and pile-up
28 M. Tiemens Finding optima for ∆ τ and ∆t High rate: 20 MHz
29 M. Tiemens Conclusions for ∆ τ and ∆t ∆ τ can be determined from time-difference spectra ∆ t should be larger than ∆ τ , but not too much larger, + 25 ns Exception: high rate, but ∆ t is still ∆ τ + 25 ns
Recommend
More recommend
