Presentation is loading. Please wait.

Presentation is loading. Please wait.

1 T1-T3 in L1 algorithm  Outlook: I) Summary of L1-confirmation II) About the TrgForwardTracking package III) Confirming (preliminary)  L1-confirmation.

Similar presentations


Presentation on theme: "1 T1-T3 in L1 algorithm  Outlook: I) Summary of L1-confirmation II) About the TrgForwardTracking package III) Confirming (preliminary)  L1-confirmation."— Presentation transcript:

1 1 T1-T3 in L1 algorithm  Outlook: I) Summary of L1-confirmation II) About the TrgForwardTracking package III) Confirming (preliminary)  L1-confirmation  L1-upgrade IV) About Patter Recognition with less stations V) Conclusion and plans L1-confirmation summary T1-T3 in L1 algorithm Status Report Jose A. Hernando (16/2/04)  Quick summary: L1-confirmation (L1 as 1 st step of HLT)  ~5% signal eff lost and ½ reduction of mb L1-upgrade (T1,2,3 in L1)  1.3-1.4 gain in efficiency (40KHz output)  4-6 track candidates to forward track  Questions? How many stations do we need to track? What about the time? Can we “improve” the confirmation, upgrade?

2 2 (I) Summary: L1 confirmation (or) 1 st step of HLT  L1 confirmation : We can reduce the rate to 20 KHz we with a cost ~5% efficiency  Rough time estimations (1 GHz PIII) Redo some L1 calculation  ~ 6 ms Do the full tacking of some candidates  ~9 ms Current HLT time budget  50 ms If we reduce the rate by ½ each L1- confirmed event will have  ~35 ms reconstruction  ~35 ms HLT decision Efficiency (after L1) vs Output rate

3 3 (II) TrgForwardTracking package  The basic: From HltLongTrack package  (O. Callot + N. Arnaud) Use TrgTracks from Velo or Velo-TT  use TrgProviders Separate tracking from hit allocation A Tool to forward track a single track candidate geo Hits Tracks (Forward) ForwardTrackAlg HitCreator IT/OT Clusters Planes Tracks (TT) Tracks ForwardTrackTool Track geo L1Buffer geo RawBuffer Different versions of the algorithm L1-Raw(HLT) buffer, or TDR-DaVinci Common part for TT Use TrgTracks as Input/Output It calls the Forward tracking tool It does clone killing HitCreator Planes Hits Planes  Status: It works! Final study of efficiency needed Plan to incorporate in the repository this week Note: separating hits form tracking allow to “mask” input hits.

4 4 (II) Trg Forward Tracking: resolution & n. of tracks  tracking performance  _p/p ~ 0.5 % Number of input tracks  Velo+TT 41.2  Forward 22.9  A factor 1.8 tracks reduction Pending a full study of efficiency, ghost and clones rate! sigma_p/p and number of tracks B(pi,pi) Signal tracks Velo+TT (TrgTracks) Forward (TrgTracks)

5 5 (II) Trg Forward Tracking: timing  Timming (my laptop: PIII 1GHz) Reference:  L1Decision 8.2 (+13) ms  Scaling factor 5/8.2 = 0.61 Timer decoding Hits 2.8 (+-4.5) ms Timing tracking 31 (+-29) ms (P-info) 19 ms (re-fishing) 11 ms Tracking successfully a track  With previous P Info 0.43(+-0.24) ms  Without no P Info 0.6(+-0.44) ms Tracking in track without P Killing clones Forward Algorithm Traking on track with P info

6 6 (III) Confirming L1-confirmation (L1 global variable)  DaVinci v9r2 L1Decision v2r1 Zurich: DV v8r3, L1 v2r0  Trigger sequence HLT-Velo & Velo-TT TrgForwardTracking L1 from L1Decision My TrgDecision  Compute IP candidates from L1 vertex.  Compute Pt, IPS and distance  Add bonus!! Some changes  No 400 MeV Pt low limit  Results We confirm the confirmation  ~5% lost in signal and reducing ½ mbias  A little bit worse!? B(pi,pi) – L1Global (after L1) % of the sample minbias [cross] B(pipi) [dark start] Bs(DsK) [open start] (after L1) Min-bias : L1 Global (after L1) % signal eff vs % retencion B(pipi) [dark] Bs(DsK) [open] (after L1) From now on.. PRELIMINARY!! 20 KHz

7 7 (III) Confirming L1-confirmation (log(pt0*pt1))  DaVinci v9r2 L1Decision v2r1  Same trigger sequence Now I do not add the bonus (how?)  Results We confirm the confirmation  ~5% lost in signal for reducing ½ mbias  A little bit better!  Do we open again (toll) the bonus accounts? B(pi,pi) Log(pt0*pt1) (after L1) % signal minbias[cross], B(pipi)[dark start], Bs(DsK) [open start] (after L1) Min-bias Log(pt0*pt1) (after L1) % efficiency vs % retiencion B(pipi)[black] Bs(DsK) [open] (after L1) 20 KHz

8 8 (III) Confirming L1-upgrade (with L1 global)  DaVinci v9r2 L1Decision v2r1 Previous results: DaVinci v8r3, L1 v2r0  Same trigger sequence Compare to L1Decision Forward Tracking tracks with (0.15-3mm) respect L1 prim. vertex I only consider Tracks with Pt (no 400 MeV) My L1 global code.  Results We confirm L1-upgrade B(pi,pi) eff  L1 ~62% eff  L1-upgrade ~82%  Factor ~1.3 B(pi,pi) –L1global (line L1-VeloTT) (dash: forward) % sample minbias[cross], B(pipi)[dark start], Bs(DsK) [open start] Min-bias L1-global (line: L1:Velo-TT) (dash: Forward) % efficiency vs % retiencion B(pipi)[black] Bs(DsK) [open] (L1 and L1upgrade) 40 KHz

9 9 (III) Confirming L1-upgrade (with log(pt0*pt1))  DaVinci v9r2 L1Decision v2r1 Previous results with:  DaVinci v8r3, L1 v2r0  Same trigger sequence L1 from L1Decision Forward Tracking tracks with (0.15-3mm) IP (1 st Vertex) I only consider Tracks with Pt (no 400 MeV) No Bonus!! Compute the sum of the pt of the 2 largest pt track in the IP window  Results We confirm L1-upgrade B(pi,pi) eff  L1 ~68% eff  L1-upgrade ~86%  Factor ~1.26 L1 a little bit better!?,  The bonus? B(pi,pi) –Log(pt0*pt1) (line L1-VeloTT) (dash: forward) % retention Minbias[cross], B(pipi)[dark start], Bs(DsK) [open start] (L1) Min-bias: log(pt0*pt1) (line: L1:Velo-TT) (dash: Forward) % efficiency vs % retiencion B(pipi)[black] Bs(DsK) [open] (L1 and L1upgrade) 40 KHz

10 10 (III) Confirming the confirmation and the upgrade  Confirmation of the confirmation We have ~5% loss in efficiency for reducing the output rate by ½ (20 KHz) A little bit better if we use log(pt0*pt1)  Confirmation of the upgrade For the channel B( ,  )  we go from ~62% to ~84% Similar results with the new forward package and decision We confirm the upgrade  ToDo: More channels I will like to check the intermediate distributions! But It seems that we are in business Efficiency (after L0) vs Output rate L1Upgrade (Zurich)

11 11 (IV) The PR search histogram  Patter Recognition Forward Tracking (~somehow)  Using input direction & q/p  Define a z-plane ref and a x window (depending on p)  Project x-hits onto that plane histogram: Callot’s histo ST (IT) weight 1. OT weight 0.5  Use the histo-peaks as seeds  Plan: Investigate how the PR histogram behaves with less stations Use B(pi,pi), [Bs(Ds,K)] channels  Where are the 2 largest pt tracks from the signal? I can run the tracking finding tool only with the hits of the MCParticle! A minbias event Tracks ordered by Pt (velo-TT) PR histogram

12 12 (IV) The peaks of the PR search histogram  The study: For a given track:  We know the MCParticle from the Velo “hits” We can make different collection of hits:  1) MC: Only the hits associated to the MCParticle  2) FC, fake: Remove the MC hits of this particle (will show up “random” peaks)  3) RC: all the hits Manipulate the PR histogram (to solve the question of the fixed bin-width):  1 st ) Take the highest peak in one bin (1 st row)  2 nd ) Add neighbor bin to highest peak and continue  3 rd ) Add now the other neighbor and continue Study:  What is the highest “peak” of the signal (MC-sample)  What is the highest “peak” of the face sample and how many do we have?  Is then the signal visible?, in witch histogram (1 st,2 nd,3 rd )? Repeat the study removing stations. Note: in TrgForwardTrack we already have a way to “mask” hits and redo the tracking!!

13 13 (IV) Peaks of PR histogram with T 1,2,3  The study: For a given:  1) MC: Only the MC hits  2) fake: Remove the MC hits of this particle, (“random” peaks) Method:  Peak in one bin (1 st row)  Add neighbor bin to highest (2 nd )  Add the other neighbor (3 rd )  Results for B(pi,pi) We clearly see the signal!!. The best is to “add” 2 bins (2nd) There are not many “fake” combinations (~2)  Most likely they will be in evidence in front of the good one 40 KHz Size of the highestpeak Signal[line], fake[dash] # of peaks vs the size [only in fake sample] #of signal peak found [cross] and fakes vs peak height

14 14 (IV) Peaks of the PR histogram removing T2  Removing T2 hits For a given track  1) MC: Only the MC hits  2) fake: Remove the MC hits of this particle. (random peaks) Method:  Peak in one bin (1 st row)  Add neighbor bin to highest (2 nd )  Add the other neighbor (3 rd )  Results for B(pi,pi) We still see the signal !!. The best is to “add” 2 bins (2n row) There are still not many “fake” combinations (~2.5) Size of the highestpeak Signal[line], fake[dash] # of peaks vs the size [only in fake sample] #of signal peak found [cross] and fakes vs peak height

15 15 (IV) Peaks of the PR histogram with only T2  Only T2 hits For a given track  1) MC: Only the MC hits  2) fake: Remove the MC hits of this particle. (random peaks) Method:  Peak in one bin (1 st row)  Add neighbor bin to highest (2 nd )  Add the other neighbor (3 rd )  Results for B(pi,pi) We do not see the signal. But there are still not many “fake” combinations (~2.5) Not everything is lost:  How “close” is the peak to the “expected” (according with P from Velo-TT)  How “good” is the track produced by the “random” peaks? Size of the highestpeak Signal[line], fake[dash] # of peaks vs the size [only in fake sample] #of signal peak found [cross] and fakes vs peak height

16 16 (IV) About PR with less stations  From Studying the PR histogram Combine two bins in one T1,2,3 signal clearly visible Without T2 signal is still visible It seems that there are not many peaks (~2.5) to play. It seems that without T2 we should be able to do the tracking  Next steps Check with Bs(Ds,K) signal. Retune the reconstruction parameters without T2  Feed the reconstruction tool with only the MC of the track, and study the variable distributions.  Digging deeper in tracking code… See if the “distance” of the peak from the “prediction” will help See if the “quality” of the peak will help

17 17 Conclusions and plans:  The Trg (Trigger) Forward Tracking Status:  Is in the “new” Trg code structure  It is working! Planes  Efficiency studies to be done!  Reading from Raw-Buffer pending (no way from L1-Buffer  )  To be incorporated in the repository this week.  L1-confirmation Redo L1-algorithm as 1 st step of HLT ~5% efficiency lost and ½ output rate We have preliminary confirmed with the “new” forward reconstruction. The confirmation works with log(pt-*pt1) [and what about pt0 only?]  L1 upgrade A gain factor 1.3-1.4 in efficiency for same output rate (40 KHz). We only need to forward track: 4 or 6 candidates. We have preliminary confirmed the upgrade with the “new” forward tracking The upgrade works good with the log(pt0*pt1) variable  [what about pt0 alone?]  PR with less station Studying the peaks from PR histogram  Of course, signal clear with T1,2,3  Signal visible after removing T2 Retune the tracking parameters for the case of no T2. Can we ask more?:  What about the “distance” to the prediction (will be P from VeloTT good enough?)


Download ppt "1 T1-T3 in L1 algorithm  Outlook: I) Summary of L1-confirmation II) About the TrgForwardTracking package III) Confirming (preliminary)  L1-confirmation."

Similar presentations


Ads by Google