17
Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Embed Size (px)

Citation preview

Page 1: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Some mediation on fast track reconstruction of BESIII

Wang DayongMar 10,2004

Page 2: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

– Purpose, status in the system

For background reduction in L3 trigger(see figures below)

For event classification

Real time data quality check and detector monitoring

To enable physicists’ fast analysis in a few hours after data taking

A clear and definite top-level design and decision together with the online system is needed badly

Z0 distrib. of triggered events (Left:Babar

Right: Belle)

Page 3: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Requirements for fast track reconstruction compulsory requirements (whatever the top level design)

System must be completely automated and operable within the Online System

Fast reconstruction system must incur no deadtime

Complete reconstruction for 100% of all events

System must be fault tolerant

Optional requirements: ( according to different top level design)

Detailed quality monitoring data for feedback to detector operators, thus

Low latency (nominally within 2 hours of data acquisition)

Event tagging based upon analysis of fully reconstructed events

Dynamic calculation of traditionally “offline” constants to be used in subsequent reconstruction jobs

Dynamic calculation of alignment corrections

Quality monitoring results must integrate into run summary

Alarms must be recorded and handled properly

Page 4: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Limiting factors , present conditions

it must be performed on online farms

information from hardware triggers could be utilized

the software framework should be identical with at least similar to the final offline reconstruction

clear and flexible event data model accommodating different implementation and later updating

use mature ,fast and stable algorithms

if possible ,make use of the ready-to-wear core codes due to lack of tracking experts (Babar,Belle,BesII, STAR…)

Page 5: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Different track fitting methods(model)

explicit track model (helix tracking)

variational method

spline approximation

multidimensional function parameterization

Predictor-corrector methods

Page 6: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Different track finding methods

A.Local methods Track following

Track roads

Kalman filter

Track element

B. Global methods Global Kalman filter

Hough transform ( histogram)

Hopfield network

Template matching

Minimum spanning tree

Page 7: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Performance of Babar and Belle fast track reconstruction

Babar Algorithms

L1 trigger rates ~1000Hz

L3 trigger rates

CPU time per track:

Belle Algorithms

L1 trigger rates 200-300Hz

CPU time per track:

L4 track finder:5.0 ms/event/CPU of UltraSPAC 167Mhz

compared to offline: 2.7s/event/CPU( total,half from tracking)

Page 8: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

BesIII MDC hardware triggers

1. Track Segment Finding :

Method of look-up table

2. Track Finding:

the same method as TSF

Use SL-5 as pivot to find :

Long track: TSF also found in SL-10

Short track: TSF only found in SL-5

3. Inner layer z trigger(? Still in study)

only for backgrounds suppression without complete z hits information as Belle

Pivot layer

Page 9: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Some possibilities of BesIII fast track recon.

Track fitting scheme: surely the simple helix model!

a. Rewrite the helix fitting of BesII fortran codes with C++ or make it reusable in BOSS as fortran codes(but how to deal with COMMONs?)

b. learn from the related Babar or Belle packages

Track finding scheme: full event data and L1 hits as input

a. r-phi track finding with TSF hits and s-z track finding with full event data from stereo wires,then track fitting with full event data to get better resolution (fast,but maybe online only)

b. using full event data for track finding then fitting with L1 output as a guidance to set regions of interest for track candidates ( could have multiple uses both online and offline, but speed requirements online?)

Page 10: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Possible track finding algorithm algorithms of histogram class in conjugation with

conformal transformation are both simple and fast, which fulfill requirements of BESIII

Possible choices:

I. R-Phi track finding

a) Nose-following with track extension and fitting in conformal plane

b) Phi histogramming in conformal plane and circle fit in the plane

II. S-z track finding and fitting

III. Helix fitting

Page 11: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

About Event Data Model

be general enough to accommodate different implementations

Data should be independent of the algorithms that produce them

Really OO-styled hierarchy of abstract classes for codes update

A well-defined format of hits

“Hits” conversion from the raw data should be provided by framework

Modularization design for more effective debugging

Page 12: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Babar Dch L1 trigger scheme1. Track segment finder:

time information has been considered : Utilize the 267ns sampling over max ~600ns drift time 0.8mm spatial resolution for a segment ~1.5cm σZ for stereo layers.

2. Binary link tracker: Share the algorithm of CLEO-II start from the innermost superlayer, and moves radially outward to link them into complete tracks

3. PT discriminator

Using TSF fine-phi data to find high Pt tracks within Pt envelope.

Page 13: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Babar L3 fast track reconstruction algorithm

L3Dch starts with the track segments from the TSF system and improves the resolution by making use of the actual DCH information.

t0 determination

t0 values for each segment are binned and the mean produced from the values in the most populated bin is used as the estimated event t0. resolution ~1.8-3.8ns

pattern recognition L3Dch searches a look-up-table for matches to segments found by the TSFs,allowing for up to two missing TSF segments per tracktrack fitting algorithm with both the track segments found and the individual hits, adding segments close to the initially fitted track, and dropping hits with large residuals. The final fit does not demand that the track originate from the IP.

Page 14: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Belle CDC L1 trigger scheme

1. r– trigger : from axial superlayers Anode wires grouped into track segment finder (TSF)

hit pattern in each cell is examined by a memory look-up (MLU) to find track segment

TSF-cell outputs to form track finder (TF) wedges

2. z- trigger : axial and stereo superlayers

Page 15: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Belle L3 fast track reconstruction algorithm

• Input: summarized hit signals from hardware trigger

1. R- Track finding : based on the memory lookup method

a. search for fired TSF in the outermost super-layer

b. search for inner hit TSF signals in the limited region

c. logical-AND of the obtained pattern and stored patterns

2. Track fitting

a. conformal transformation

b. fit the line in the conformal space

c. get PT from the distance from the origin to the line

3. R-z Track Reconstruction

a. get z-hit patterns for the r- segment from r-z trigger

b. Fill histogram as a function of the polar angle to get Pz

Page 16: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Belle fast track reconstruction in software trigger• L4 trigger track finding and fitting :

Page 17: Some mediation on fast track reconstruction of BESIII Wang Dayong Mar 10,2004

Algorithms of L4 fast track reconstruction

1.Track segment finding

2. R- phi track building

3.r-phi track fitting

3 track parameters out of 5 determined (dr, phi0, κ )

4.Event time correction

5. 3D track building

6. s-z track fitting

linear fit to determine the other 2 track parameters(dz,tanλ)