eff of clusters

Overview

The cluster finder of the ssd proceeds as well :

Scan a list of strips and find the central strip by the mean of a signal to noise cut : s/n > 5
Add the neighborings strips if their id are consecutives (positive or negative) to the central strips

When looking as the results, i found that some big clusters could be reconstructed, just because there is no another cut on the neighborings strips.

Example

The next plots show the size of the clusters (called nstrip) without cut on the neighbourings strips, and by applying cuts.


Figure 1 : No cut	Figure 2 : s/n > 1	Figure 3 : s/n > 2	Figure 4 : s/n > 3

As you can see, without any cut, the cluster finder can reconstruct unrealistic clusters, which is bigger the size set up in the simulation (4 strips).
With cuts on the neighborings strips, the size of the clusters decrease.

One event


Figure 5 : Signal of the clusters as a function of their size : side P	Figure 6 : Signal of the clusters as a function of their size : side N

One wafer particularly


Figure 7 : Signal of the strips vs their id : side P	Figure 8 : Signal of the strips vs their id : side N

log

Here is a log a the values of the strips found, then the clusters built , then the point reconstruct for one wafer where I found clusters which size is > 10 strips :
wafer 7917

another way of cluster finder

The steps of this method are :

find the mean of the signal : per wafer and per side (signal_mean)
find the mean number of strips that are actually fired by the passage of particles through each wafer
for each peak, find the mean number of strips above the signal_mean, this will define the number of strips in clusters.
In fact, the selection of these strips will be :
- signal (strip) > signal_mean + 3*&sigma (from the landau fit)

Step 1

For 1 ladder I plot here for the first ladder the distribution of the signal of the strips.
The fit is done by a landau distribution.
Most of them are ~ 9 ADC and is not the results of passage of particles.


Figure 9 : side P	Figure 10 : side N

The values of the signal_mean and &sigma can be found here : log.txt
From these values, i find :

signal_mean = 8.853789 ADC
&sigma = 0.442536

signal_mean = 8.845757 ADC
&sigma = 0.490469

For all ladders


Figure 11 : side P	Figure 12 : side N

When I loop over all the wafers, i have :

Side P

Ntotwafers = 284
signal_mean = 8.809391 adc
&sigma = 0.674177

Clusters size correction

I found a bug in StSsdClusterList.cc that was one of the reason why the code found huge clusters.
When it found first a cluster, it try to split it, ie to find mininmas and maximas in the signal shape.
If the difference between the signal of on the strip and his neighboring is bigger than a value : tolerance = (signal_strip - signal_neighboring)/(signal_strip)
,the splitting is done.
The cause was that the current strip after the splitting wa never updated so one loop was still beginning with the first strip.

Comparison

I made the comparison by running over the same number of events.

Clusters size

These 2 plots represent the clusters size (number of strips) for the 20 ladders (P and N side).(log scale)
One can see clearly that the size of clusters get back a reasonnable level with the correction.


Figure 13 : without the correction	Figure 14 : with the correction

For all the ladders : here

Recent update

one bug in the cluster finder was fix : see the pdf
differents studies were done : they consisted in studying how evolve the cluster size when I applyied cuts :
1. on the noise of the seed strip
2. when introducing a cut on the neighboring strips and after on their noise
3. when removing strips for which signal is below a cut
4. when removing strips for which noise = 0
5. when fixing a noise (not from the db) but a constant
comparison with the simulation
comparison with the pp data (new : 01/22/07)

Cut on the neighboring strips

Same as the ALICE NOTE related to the beam test of SSD wafers.
(Signal to noise)_seed = 10
(Signal to noise)_neighbouring = 2

Figure 15 :

Removing strips with signal is below a cut

I applied a cut on the signal of strips in the primary list of strips.


Figure 16 :	Figure 17 :	Figure 18 :

Removing strips with noise = 0

I removed strips for which the noise is 0 in the primary list of strips.

comparison with simu

Figure 19 : simulation data

comparison with pp data

Figure 20 : pp data

Determination of the ground

I determine the ground level of signal looking at the raw data (signal of the strips). The mean value is the result of a landau fit.
It wad done wafer per wafer for each ladder but one can see that sometimes the fit does not converge so that ultimately, I have to find the mean signal chip by chip.
But as first try, it is still a good approximation.

Side P


Figure 21 : ladder9 side P	Figure 22 : ladder 10 side P	Figure 23 : ladder 11 side P	Figure 24 : ladder12 side P	Figure 25 : ladder 13 side P	Figure 26 : ladder 14 side P


Figure 27 : ladder15 side P	Figure 28 : ladder 16 side P	Figure 29 : ladder 17 side P	Figure 30 : ladder18 side P	Figure 31 : ladder 19 side P

Side N


Figure 32 : ladder9 side N	Figure 33 : ladder 10 side N	Figure 34 : ladder 11 side N	Figure 35 : ladder12 side N	Figure 36 : ladder 13 side N	Figure 37 : ladder 14 side N


Figure 38 : ladder15 side N	Figure 39 : ladder 16 side N	Figure 40 : ladder 17 side N	Figure 41 : ladder18 side N	Figure 42 : ladder 19 side N

Log text

The fit results are here :

Comparison cut neighborings / Correct hits

Clusters Size

Size of clusters :


Figure 43 : Number of strips in clusters with the cuts on right hits	Figure 44 : Number of strips in clusters with cut on the neighboring strips

The size of the clusters when applying cut on the neighborings strips is decreasing.
Of course removing fake hits do not affect the size of the clusters as it is a previous step to the reconstruction of the hit.
The details (all wafers) can be found here : correct hits
and here : cut on the neighborings strips

Efficiency

Efficiency :


Figure 45 : Efficiency (all ladders) with the cuts on right hits	Figure 46 : Efficiency (all ladders) with cut on the neighboring strips

It is hard to said which cut has an impact on the efficiency (even if the cut of removing fake hits seems to increase a little bit the efficiency) but I will increase the stats soon.

Jonathan BOUCHET

Last modified: Thu Feb 15 19:17:40 CET 2007