Some properties of the present treatment of time in reconstruction An Svt Digi has a time stamp that is the time when the pulse went over threshold. It is measured in units of 16/238MHz = 67.23 ns. The time that is measured in this way is delayed for small pulse heights. We recognized early on that we need to correct for this, and a chip-by-chip correction is made based on pulse shape measurements that were made by Amedeo Perazzo. The figures in http://costard.lbl.gov/~grl/SvtTime/mar23time.ps have the distributions of raw and corrected digi times, showing how much the time correction improves the distributions. In these plots, and all others in this note the time that is used is the difference between the SVT time and the DCH time, and it is assumed that the accuracy of the DCH time is much better than what we get in the SVT. In addition in this plot I combine the phi-side and the z-side, and also put layers 1 to 3 together and layers 4 and 5 together. The layer 1-3 and 4-5 groups are not combined with each other because they are intrinsically different: they use different wafer types and they are set with different values of peaking time and "skip control". For layers 1-3 the time-over-threshold (ToT) unit is the same and the time stamp time unit, whereas it is twice the time stamp unit for layers 4 and 5.
The time corrections that are made as a function of ToT are illustrated in http://costard.lbl.gov/~grl/SvtTime/mar23timevstot.ps , which has the measured values on the left, the correction that we make in the middle, and the corrected times are on the right. Clearly the corrections make a big improvement, but they have problems at large ToT. I have not tried to see what improvement we can get by modifying this. These plots were made with 20000 calibEvents in in 14222.
We presently get poor time resolutions
The resolutions that we see here of 0.47 and 0.66 ticks are much larger than we might expect. One might hope that the resolution would be the reciprocal of the square root of 12, or 0.29. Actually, I estimate that the granular nature of the TOT time corrections increases this to 0.31. Still our time resolutions are much larger than we might expect, and what we see in Monte Carlo. The time resolution on the digis is important because we use a time cut in SvtHitReco to eliminate background digis. The better our resolution, the more background we can eliminate, saving time and improving track finding. In the reconstruction we average the corrected time stamps of the digis in a cluster to get the cluster time or hit time, and in the stand-alone SVT track finders (and also in the most recent version of TrkSvtPeelOff) we average the cluster times to get the track time. We have seen that in real data the resolution of the track times has been 14 to 16 ns, compared with 6 ns for Monte Carlo data.
Calibration by Readout Section
We never understood why this time resolution for the real data is so poor. For some reason we never attempted to calibrate these times. The top plot in http://costard.lbl.gov/~grl/SvtTime/mar23timevsros.ps has the average corrected time for each of the 208 readout sections (ROS) in the SVT, made with a sample of 10000 events. We see that the values are sizable. The middle plot shows the sigmas for the Gaussian fits that were done to each ROS. I extracted these average times and used them as corrections to the corrected time. The bottom plot in this page shows the average times for a different sample of 10000 events in the same run when the corrections are added. Clearly this correction works for nearby events. I have not tried them on other runs.
Chip-by-chip calibration
The plots in http://costard.lbl.gov/~grl/SvtTime/mar23timebychip.ps show the average time for all chips in the SVT after the ROS time corrections are used. In most cases these averages are small for layers 1-3, but larger for layers 4 and 5. When these averages are put into the corrections, the time resolutions improve, as shown in http://costard.lbl.gov/~grl/SvtTime/mar23correctedtime.ps . These plots have the time distributions seen before the calibration, after the ROS calibration, and after the chip calibration. The sigmas for Gaussian fits go to 78% and 82% of their pre-calibration values when the ROS calibration is done, and to 77% and 78% after the chip calibration is used.
The place where the effects of this calibration show up most significantly is in the resolutions of track times. The figures in http://costard.lbl.gov/~grl/SvtTime/mar22tracktimes.ps , which was done with only 1000 events. The track time resolution goes from 13.9 ns to 8.2 ns to 7.7 ns after the ROS and chip calibrations are employed.
Summary
A simple calibration can considerably improve the time resolution of Svt Digs and Svt tracks. The digi time resolution can be improved by about 22% and the track time resolution by nearly a factor of two. This makes the resolutions much closer to the "expected" values, but we are not there yet.
We can reduce the time cuts on digis that are now in the program. We now cut the digis at 3 ticks in order to cut out background. It looks like we can safely reduce the 3.0 digi time cut to 2.5, and perhaps more. We can also do something that was possible before, which is to make the cut smaller in layers 1-3, where a cut at 2.0, which is what is used for Monte Carlo data, looks safe. This may be premature - a preferable solution would be to make layers 4 and 5 better. Before we reduce the digi time cut we need to understand the ToT consequences of this change. We do not want to cut out a large fraction of the large pulse height digis.
We can reduce the time cut on SVT tracks, which are presently at 60 or 67 ns, to about 30 ns. However, before we do this we need to be sure that the DCH time will be more stable than it has been sometimes in the past.
It is possible to do channel-by-channel time calibrations, but this would require much larger statistics than the 20000 events that I have used.