next up previous
Next: the level 0 trigger Up: The model Previous: the UPV OR

the CPV OR

Figure 7: CPV OR scaler rate vs. beam current measured at the beginning of the June run period.
\begin{figure}\begin{center}\mbox{\epsfxsize =9.0cm\epsffile{cpvA.eps}}\end{center}\end{figure}
Figure 8: CPV OR scaler rate vs. beam current measured at the end of the June run period.
\begin{figure}\begin{center}\mbox{\epsfxsize =9.0cm\epsffile{cpvC.eps}}\end{center}\end{figure}

The charged particle veto wall is the component of Radphi that encounters the highest rate, and so is the most sensitive part of our apparatus to dead-time effects. It is important, therefore that the model correctly describe the behavior of the logic that generates the CPV OR. Similar to the UPV logic, the CPV OR is formed by discriminating the individual signals from the 29 phototubes, producing for each a train of logic pulses of fixed duration and then OR'ing these signals together. The formulae which embody the model are as follows.

$\displaystyle S_{_C}$ $\textstyle =$ $\displaystyle r_{_C}\,I\,e^{-d_{_C}\,r_{_C}\,I}$ (8)
$\displaystyle {\cal F}_{Chit}$ $\textstyle =$ $\displaystyle 1$ (9)
$\displaystyle {\cal F}_{Cmiss}$ $\textstyle =$ $\displaystyle 1-e^{-t_{_C}\,r_{_C}\,I}$ (10)

where the symbols $S_{_C}$, $r_{_C}$, $d_{_C}$, $t_{_C}$ and ${\cal F}_{Chit}$, ${\cal F}_{Cmiss}$ have the usual meanings. As in the case of the tagger OR, a distinction is made between the coincidence window width $t_{_C}$ and the scaler dead-time $d_{_C}$ although the two are substantially determined by the same quantity, the width of the CPV OR pulse generated by the OR module. The scaler dead-time contains whatever additional dead-time is inherent in the scaler, whereas the coincidence gate width contains corrections due to the minimum overlap required to generate a valid veto, so the two are not exactly the same. However they should both be close to the CPV pulse width measured on the oscilloscope, and this provides an important check that the electronics is working as expected.

The results of a scan taken towards the beginning of the June period is shown in Fig. 7. The tendency of the CPV OR scaler to decrease at high rates is due to the fact that the scaler is counting the transitions between 0 and 1 in the CPV OR signal and not the amount of time it is on or off. At very high rates one expects the scaler to go to zero (the CPV OR is always 1) which is the limit of Eq. 8 at high rates. The two-parameter fit to the data in Fig. 7 gives a fairly precise value of 25.5ns for $d_{_C}$, which disagrees with the value of 10ns that was programmed into the OR module for its gate width. Upon examination, it was discovered that the discriminator modules had been operating in updating mode, which extends the width of the CPV OR signal to however long the pulse remains over threshold. A later scan shown in Fig. 8 gives a more consistent value of 10ns for $d_{_C}$, indicating that the problem was fixed. Once more the model correctly diagnosed a subtle bug in the electronics and produced consistent results after the bug was fixed.

There are still significant deviations from the expected model behavior for the high-rate points in Fig. 8. One possibility is that it is due to CLAS emptying their target during that scan. A slight effect due to this is observable in these data: compare the U-shaped nonlinearity in the low-rate points in Fig. 6 with the corresponding points in Fig. 8. However the suppression at high rates goes beyond what would be expected based upon a comparison between those two figures (the data were taken during the same scan). The observation that the adc spectra from the central CPV counters showed a marked decrease in gain for the high-rate points taken during this scan indicates that some of the CPV tubes are showing signs of saturation.

The most important restriction on the validity of this model arises from its ignoring the effects of discriminator dead-time at the level of the individual counters. All of the observations made above under the tagger section apply here as well. Because the rates in the individual channels are limited to a few MHz, dead-time effects are much less important there than they are once one is working with the OR signal. Nevertheless losses at the level of 5% can be expected from this source for those tubes running at 5MHz under full design intensity. In order to take these losses into account consistently, one must also bring in the two-pulse resolving power of these tubes and electronics, which must be measured. At present the stability of the beam current measurement is not sufficient to measure dead-time effects a level better than 5%. In the present model they are being ignored (set to zero). The restriction $t_{_C}\ge 15$ns discussed under the UPV section above also applies here as a condition for the validity of Eq. 9


next up previous
Next: the level 0 trigger Up: The model Previous: the UPV OR
Richard T. Jones 2003-02-12