1. 7/1/2016IEEE RT 2009 Beijing A Topological Muon Trigger for IceCube Christian Bohm, Sebastian Euler 1, Per-Olof Hulth, Hossein Kavianipour, David Nygren 2, Clyde Robson, Seon-Hee Seo, Carl Wernhoff and Gustav Wikström Stockholm University 1 University of Aachen 2 Lawrence Berkeley National Laboratory Overview IceCube with deep core The algorithm Implementations Conclusion 1

2. 7/1/2016IEEE RT 2009 Beijing IceCube A km 3 neutrino detector in the ice at the South Pole 80 strings with 60 Digital Optical Modules each 2

3. 7/1/2016IEEE RT 2009 Beijing IceCube A km 3 neutrino detector in the ice at the South Pole 80 strings with 60 Digital Optical Modules each IceCube now consisting of 60 strings replaces the older Amanda detector developed and operated during 1992-2009 IceCube can detect neutrinos from the northern hemisphere The neutrinos interact with some nuclei below IceCube producing muons. These activate the optical modules as they pass through the ice. Hit energy and time stamps reported to the counting room Neutrinos from above are more difficult to detect due to a large amount of muons from cosmic rays that also activate the detector 3 *

4. 7/1/2016IEEE RT 2009 Beijing IceCube A km 3 neutrino detector in the ice at the South Pole 80 strings with 60 Digital Optical Modules each IceCube now consisting of 60 strings replaces the older Amanda detector developed and operated during 1992-2009 IceCube can detect neutrinos from the northern hemisphere The neutrinos interact with some nuclei below IceCube producing muons. These activate the optical modules as they pass through the ice. Hit energy and time stamps reported to the counting room Neutrinos from above are more difficult to detect due to a large amount of muons from cosmic rays that also activate the detector A recent development : 6 additional strings around the center increase the module density and the sensitivity to low energy muons 4

5. 7/1/2016IEEE RT 2009 Beijing IceCube 5 86 strings x 60 modules x 500 Hz = 2.6 MHz average random in entire system 5  s window contains 2.6 x 5 = 13 hits Majority trigger > 13 + 4  = 28 hits in 5  s window In current operation, local coincidences between neighboring modules on the same string reduce the rate Leads to loss of hits without neighbors Need a way to trigger on full flow of hits

6. 7/1/2016IEEE RT 2009 Beijing Track Engine Motivation Majority triggers are not efficient for low energy muons Hard local coincidences limit efficiency Topological information essential for low energy track detection In the track engine approach pairs formed by hits from a real muon track tend to cluster in a v, , r and z space (~Hough transform), where v~c. This will also resolve multiple tracks if they are well separated It is a minimal approach without additional geometric constraints than those given array geometry * * * * 6 * *

7. 7/1/2016IEEE RT 2009 Beijing Track Engine Algorithm For each  s select all hits in 5  s windows Form all hit pairs in window Remove all hit pairs outside a reasonable speed range around c Represent and store tracklets (pairs) in a , r and z space Clusters represent possible tracks 7

8. 7/1/2016IEEE RT 2009 Beijing Track Engine Simulations Initial simulations show an efficiency improvement of 10 times below 50 GeV More simulations in progress to study efficiency and to compare with alternative approaches 8

9. 7/1/2016IEEE RT 2009 Beijing9

10. 7/1/2016IEEE RT 2009 Beijing10 FPGA-board XILINX ML507? Server HP Proliant DL380 G6 GbE Switch GbE 10MbE <87 Hardware implementation Being assembled. Remote firmware re-configuration essential PCIe

11. Implementation 7/1/2016IEEE RT 2009 Beijing11 5  s buffer memory Sort Module Soft TE 2 HLC Sort Module 5  s buffer 2d-Histo Sort Module 5  s buffer 4d-Histo Ambition level 1 – Soft TE Ambition level 2 – 2d histograming FPGA -TE Ambition level 3 – 4d histograming FPGA -TE Several alternatives pursued in parallel and more to come Data collector

12. 7/1/2016IEEE RT 2009 Beijing 2-d histogram implementation 2-3/  s8-12/4  s~30/5  s ML-507 Virtex V FX70 12 Firmware built and tested - integration with PC-software remains

13. 7/1/2016IEEE RT 2009 Beijing 4-d histogram implementation 13 A histogram with overlapping bins with a width corresponding to the cluster size can be used for cluster recognition - demanding a certain bin occupancy for detection Must be homogeneous and overlapping to make sure that a cluster will fall into a bin. The direction part of such a histogram can be constructed from a football structure, representing each face with its center point: Or with finer granularity;

14. Unequal bin sizes 7/1/2016IEEE RT 2009 Beijing14 The problem with unequal bin sizes can be cured by moving all hexagonals away from the pentagons, but reducing the effect with distance

15. Storing direction data in multiple bins 7/1/2016IEEE RT 2009 Beijing15 Overlapping bins can be achieved by allocating the three closest bins to each data points. If all three bins are incremented to store the event we can be reasonable sure that a cluster will be recognized as a maximum even if it is located close to a bin border. An alternative would be to store the data in the 4 closest bins

16. Managing the impact parameters 7/1/2016IEEE RT 2009 Beijing16 Two impact parameter dimensions can be included by associating a 2d- subspace to each bin. The overlap can be taken into consideration by checking the sum of the occupancy in the closest 4 bins each time a bin is updated.

17. Managing the impact parameters 7/1/2016IEEE RT 2009 Beijing17 There is a problem of allocating the x and y-directions in the subspaces. If the y-direction coincides with up we get problems to process vertical incidences. One solution is to defer those events to second level processing and instead of up align to one side, e.g. north.

18. The histogram 7/1/2016IEEE RT 2009 Beijing18 The histogram is large and difficult to analyse. However, incrementing bins give info about possible peaks. No need to search histogram Large histogram - minimum probability for random multiple bin occupancy. Duplication or triplication good indications for data clusters. With sparse data list storage may be better than full memory. This list must be kept up to date, removing obsolete data CAM-memories useful modules

19. Firmware block diagram 7/1/2016IEEE RT 2009 Beijing19 Under development

20. 7/1/2016IEEE RT 2009 Beijing Present status of PC software String hub processing: Initiated Data collection and sorting: Data collection and a simple sorting software has been tested using test data transmitted from another PC up to 20 MB/s (TE needs ~10MB/s) A more sophisticated software will improve functionality with a new sorting algorithm and improved buffer handling. 20

21. 7/1/2016IEEE RT 2009 Beijing Summary & conclusion A standard PC with FPGA extension is sufficient to process the entire data flow from the complete IceCube + deep core detector Implementation is in progress and hardware installation at the South Pole is planned for early 2010 Several solutions under development. The software solution is given priority. More simulations needed to fix parameters. 21

22. 7/1/2016IEEE RT 2009 Beijing Thank you for your attention 22