1. Spectrometer data quality studies
8/02/2016
Straw WG
Spectrometer data quality studies
Dmitry Madigozhin, Dosbol Baigarashev*
* JINR diploma student

2. Outline:
Noises and thresholds
Echo and the time distribution tail

3. Noises and thresholds:
The LHCb-style technique for the thresholds setting is developed
(thanks to Anatoli Kashchuk for the explanations and discussions).
Fit the noise scans (provided by Michal Koval) by the Gaussian with a fixed maximum height of 39.6 MHz (Rice frequency FR, see
A.P.Kashchuk, Instrum.Exp.Tech. 55 (2012) ):
F = FRexp( -(vth — m)2/(2sn2) )
m position of a Gaussian maximum defines a signal baseline for the given straw channel (a place of theoretical maximum of the noise scan, that is invisible due to the inversion in the Carioca scheme).
But if the scan is done in parallel for the full cover (16 straws) we have a lot of cross-talks. So first one need to clean up the scans from the in-cover cross-talks by means of separate straw scans (with a high thresholds in another straws). In such a case the visible cross-talks disappear (done by Michal 23/10/2015).
These cross talks sometimes cause a visible shift of the threshold also for the standard 100Hz procedure, so the separate straw scans should be done for the thrersholds setting.
The LHCb-style technique for the thresholds setting is developed (thanks to Anatoli Kashchuk for explanations and discussions).
Fit the noise scans (provided by Michal Koval) by the Gaussian with a fixed maximum height of 39.6 MHz (Rice frequency FR, see A.P.Kashchuk, Instrum.Exp.Tech. 55 (2012) ):
F =
The position of such a Gaussian defines a signal baseline for the given straw channel (a theoretical maximum of the noise scan, that is invisible directly due to the inversion in the scheme).
But when the scan is done in parallel for the complete cover we have a lot of cross-talks. So first need to clean the scans from the in-cover cross-talks by means of separate straw noise scans with a high thresholds in another straws. In such a case all the visible cross-talks disappear (done by Michal 23/10/2015).
These cross talks sometimes cause a visible shift of the threshold also for the «standard» 100Hz procedure. So periodically the separate straw noise scanning should be done in any case (to prevent the questions about the cross talks effect).

4. Example of the fits and thresholds:
Chamber 3 (from 0-3)
SRB 6 (from 0-7)
Cover 5 (from 0-15)
100 Hz
Crosses : a parralel noise scan with the same current threshold in all the straws of the cover.
Circles : a separate noise scan with a high threshold in the other straws of the cover.
Vertical lines show the thresholds:
Solid: old 100Hz threshold
Dashed: new threshold = baseline + 36 mV
Cross-talk sources
cross-talks

5. For completness: Visually detected in-cover cross-talks:
Orange: receiver.
Blue: source
Yellow: both
Source and receiver may be not a neighboring tubes in the cover => reason in links?
A problem for threshold setting,
but not a problem for data taking, as the source signal always 3-4 orders of magnitude more often than the cross-talk signal.

6. Another visible problem for the thresholds setting (discovered by Michal) :
exists in one SRB only:
Chamber 2 (from 0-3)
SRB 7 (from 0-7)
Here example is shown:
Cover 7 (0-15) ? ?
At some specific thresholds both for parallel and separate noise scans there are point-like noise peaks.
It doesn't looks physical.
But it is able to cheat the standard 100Hz threshold setting algorithm.
For the new threshold procedure we filer out these too sharp peaks.
It should be done also for the old procedure. ? ?

7. The place of strange «noise» at specific thresholds (first found by Michal)

8. Channels
without straws
Noise scan fit helps to distinguish the normally working straws from the empty (or broken) channels.
straws
Noise scan sn(in V)
Baseline is disributed randomly as a Gaussian.
Baseline position m (in V)
If one choose
Threshold = baseline+36 mV
(> 5 average sigma),
the expected noise is reasonably small.
Expected thermal noise distribution

9. Variations of the «standard» thresholds
in the frameworks of Rice model
If one set the threshold at 100 Hz, we have a relation:
100 Hz = FRexp( -(vth — m)2/(2sn2) ),
so for the threshold above the baseline
Vth = vth — m = sqrt(2ln(FR/100Hz)) sn = sn
(standard 100Hz threshold is 5(noise sigma) threshold for straw individual sigmas)
In particular, taking the sn distribution fit from the last slide,
dVth / Vth = dsn / sn = s(sn )/ <sn> = ( V)/( V) = 5%
5% deviation of threshold from the «correct» one (from the Rice model point of view) doesn't look as a big effect.
Even 3s(sn) = 15% is a rather small relative shift of threshold.
(We hope to increase the gain by % in order to improve resolution.)

10. Run 3925 (threshold = +36 mV) and Run 3929 (threshold = + 60 mV)
We pay so much attention to threshold becouse it is the only simple
«free» (not well fixed) parameter in our Monte Carlo models that affect straw time resolution.
This is why we hope to improve time resolution or at least to make it uniform by means of the thresholds control.
But could so small (5%) deviations of thresholds from the «correct» values be responsible for the observed straw leading time resolution variations?
Look into the NA62 data 2015:
Unfortunately our special threshold tests were done at 40% of nominal intensity
Run 3925 (threshold = +36 mV) and
Run 3929 (threshold = + 60 mV)
that causes an additional problems with a big accidental background.
So it is more convenient to start discussion from the low-intensity run
Run 3809 («standard» thresholds Hz) : ~ 1% intensity
As a reference time we use the average time of the HOD hits in the window +/- 25 ns around their peak time (used for 2014 by Giuseppe).
NA62 data from the 2015 RUNS:
Run 3925 (threshold = +36 mV) and
Run 3929 (threshold = + 60 mV)
Run 3932 (threshold Hz) : «standard» setting, smaller statistics
With the same firmware version
Unfortunately with a lot of extra hits — a temporary firmware bug?
As a reference time we use the time of the HOD «clean» candidate — when there is exactly one reconstructed charged particle in the hodoscope.

11. F(t) = A + B exp( -(t -tmax)2/(2 s(t)2 ) ),
Remainder: we use an (asymmetric Gaussian + flat Bg) to fit the leading time distribution peak in order to check its width.
F(t) = A + B exp( -(t -tmax)2/(2 s(t)2 ) ),
where s(t) = s0 + k(t-tmax)
no t0
correction
applied
Run 3809

12. s0(ns) k Red — all hits in a straw Blue — first hit in a straw
(standard in reconstruction)
s0(ns) k

13. still unknown effects causing resolution non-uniformity :
If there would be a considerable sensitivity of resolution to the threshold shift of such a scale, there should be a correlation between the actual sigma and the deviation of the used threshold (100Hz) from the «better» new threshold (extracted in LHCb style).
But the possible correlation of this nature looks negligible in comparison to the
still unknown effects causing resolution non-uniformity :
Visible resolution variations are cover-dependent or place-dependent
s0 in ns
Apart from this (wrong 100Hz),D looks random
D = (100Hz threshold) - (proposed new threshold) in mV
RO channel = (chamber*8+srb)*256 + cover*16 + straw

14. Some regions in details
s0 in ns
Shifted by 1 bin by mistake, sorry
(100Hz threshold) - (proposed new threshold) in mV
s0 in ns
(100Hz threshold) - (proposed new threshold) in mV

15. Some another regions in details
s0 in ns
«Strange noise» cases
(100Hz threshold) - (proposed new threshold) in mV
s0 in ns
(100Hz threshold) - (proposed new threshold) in mV

16. The largest variations of measured leading time resolution seems to be geometrical position-related (angular distribution effects ?) U V X Y
s0 in ns
Chamber 1(0-3)
D = (100Hz threshold) - (proposed new threshold) in mV
Straw position index = (16*chamber + 4*view + 2*half + layer)*122 + straw

17. s0 (ns) (100Hz threshold) — (baseline+36 mV) (in mV)
Globally also there is no visible correlation between the resolution s0 and deviation of the old «100Hz threshold» from the new proposed «baseline+36 mV» one.
So the change of threshold technique itself will not improve essentially the resolution uniformity. Other reasons of resolution variations (place-dependent) are much stronger.

18. Echo and a tail split of the leading time distribution
The standard reconstruction (Giuseppe) takes only the first leading hit in the straw in order to avoid the considerable echo effect that presents in many tubes.
Sometimes echo has two maximums, so in general it can imitate the splitted tail.
Is it possible that also the first-hits distribution has some residual echo effect?
(~ due to some inefficiency for the first hit)
In such a case the effect of straw-wire centers transversal mismatch could be imitated.
With a small background the first peak is almost unaffected by the echo
All hits
All hits
First hits
First hits

19. Oftenly the echo effect is layer-dependent (still not understood why)
All hits
All hits
Layer 1
Layer 0 58 56 54 52 50 48 57 55 53 51 49 48 50 52 54 56 58 49 51 53 55 57
Chamber 0 SRB 0

20. Examples :
1 straw without any visible tail splitting in the first-hit distribution;
2 starws with a large visible tail splitting in the first-hit distributions.
All hits
All hits
All hits
No splitting
split
split

21. Staggered straws may be used for the straw side tagging.
Tagging straw layer 1
The majority of track-related hits are 0-tagged or 1-tagged ones depending on the corresponding hits presence in the another cover of the same cell. The kind of tagging define the side of the hit for the large values of leading time. 1 1
Definition of
0-tagged and 1-tagged hits. 1
First hits 1 1
Leading time for the the straw 2166 (y axis) and the staggered one used for 0-tagging (x axis).

22. Tagged time distribution examples
RO 2166
RO 4727
RO 1145
0-tagged
1-tagged
1-tagged
1-tagged
0-tagged
0-tagged
A real difference between the left and right time distributions happens in straws 1145 and 4727 (they were chosen with a rare large splitting effect).
So shift happens up to +/- 15 ns.
It corresponds approximately to the shift of 250 mm between the straw and wire centers.
If it is a straw transversal shift (rather than wire) it is not a problem for the measurement resolution, and it is still not a problem for the efficiency. With our geometry the shift of up to 400 mm will not change the efficiency. But one need to check it for calibration (say, maximum drift time estimation)

23. Conclusions:
A simple improvements of the threshold setting procedure should be done in any case: 1) separated noise scans and 2) filtering of «strange noise peaks».
An alternative thresholds setting procedure (baseline+) could be implemented. But no big resolution improvements are expected, as the threshold corrections will be ~ 5% of the threshold deviation from baseline.
But now we understand this area much better and we are ready to the expected questions on these problems.
A distortion of the leading time distribution due to echo at least is not the only source of the time distribution tail visible splitting. First-hit distributions (used for reconstruction) seems to be free from echo, but sometimes show the essential tail splittng.
The first-hit-only distributions with the splitted end in the few tested cases reflect the real transversal relative shifts between the straw and the wire centers by up to 250 mm (rarely). We can use it for the geometry control.