TPC parallel tracking

1
TPC parallel tracking

• Marian Ivanov
• CERN ALICE

2
Kalman Filter

• Current tracking is based on Kalman
• two steps
• Cluster finding
• find two dimensional clusters in pad-row -time
  planes
• reconstruct the position of the corresponding
  space point - interpreted as the crossing point
  between the track and the center of pad row
• estimation of the error
• Track finding
• kalman filtering approach
• find a good seed to start a stable filtering
  procedure
• follow track

3
Kalman filter

• Clustering and the track finding are independent
  processes, but
• Cluster shape and error critically depend on the
  track properties (not taken to the account in the
  current cluster finder)
• large fluctuations in the signal shape - even if
  we know estimate of given track properties (deep
  angles)

4
Clusters error estimates - momentum methods

• Main source of the uncertainties diffusion and
  angular effect
• theoretically
• for practical use

5
Clusters error estimates

• Moment 0 - m0 - integral of the charge in the
  cluster
• Ratio - m0/Nell
• coefficients a, b - chosen to estimate product
  of the landau factor in angular effect and m0
  /Nprim ratio
• parameterization work quite well for not
  overlapped clusters - residuals are almost
  gaussian - r.m.s 1.1 s (result from aliroot
  simulation)

6
Overlapped clusters

• When do we have overlapped clusters ?
• Cluster width
• Big fluctuation in cluster shape - r.m.s 14

7
Overlapped clusters

• Fast approach - assumption all clusters have the
  same shape - if the width of the cluster is too
  big - subtract the fraction of the neighbored
  bins
• works and gives better results then current
  official cluster finder

8
Kalman filter

• To get better estimate of error we connect track
  finding with cluster finding
• to define better criteria of the overlapped
  clusters - necessary to track in parallel
• find prolongation of the tracks to the next
  padrow
• for given cluster look around how many tracks
  contributed to him -
• try to deconvolute it - how to do it, to not have
  dependent cluster position estimates ?

9
Problem no 1.

• Kalman filtering starts with a set of seeds
  track candidates
• two times more track candidates then real
  tracks
• not possible to use it for parallel tracking
• necessary to find stronger criteria to remove
  ghost tracks - not decreasing efficiency

10
Hough transform for seeding?

• Try to use Hough transform for finding seeds in
  the outermost padrows
• linear parameterization, parameters (z0,y0, ky) -
  vertex constraint in z direction y0, ky free
  parameters

11
KyY plane fix Z
12
Hough transform for seeding?

• work in progress - number of ghost factor 3 -
  mainly due to the landau fluctuation - additional
  peaks in parametric space - misinterpreted as
  lines
• overall efficiency -99.5

13
Conclusion

• We continue this direction
• It necessary to find whether high number of
  ghosts is a principal problem or it is fault of
  the bad peak finding criteria.
• If it principal problem, than we have to find
  another seeding algorithm