Results

The important rates that enter the Radphi trigger are shown in Table 1. The first column shows the number that is read from the instrumentation, for example a scaler. The value in the second column, if any, is the corrected rate in the case that the scaler is affected by an internal dead time. This is only important for rates above several MHz. The predictions in the last column are copied from Table 3 of Ref. [1].

Table 1: The rates in all levels of the Radphi trigger from the front end to the tape. The measured values were taken during 1999 running with 5.5GeV electrons at 125nA and a double radiator each of nominal $1\cdot 10^{-4}$ radiation lengths. This corresponds to $5\cdot 10^7$ tagged photons per second, to within 10%. The predicted values are taken from Table 3 of Ref. [1].

signal

scaler

corrected

predicted

tagger or

30

MHz

50

MHz

48

MHz

CPV or

33

MHz

40

MHz

67

MHz

UPV or

550

KHz

750

KHz

BSD layer 1

4.1

MHz

4.7

MHz

3.9

MHz

BSD layer 2

2.4

MHz

2.6

MHz

2.0

MHz

BSD layer 3

1.5

MHz

1.6

MHz

1.2

MHz

BSD $1\cdot 2\cdot 3$

900

KHz

440

KHz

$BSD \cdot \overline{UPV}$

800

KHz

390

KHz

$BSD \cdot \overline{UPV} \cdot TAG$

750

KHz

150

KHz

The comparison between predicted and measured rate for the tagger OR simply reflects the choice of beam current. All of the other comparisons are a check of our understanding of the apparatus. There is excellent agreement for the cpv (charged particle veto) and upv (upstream veto) in spite of the fact that these counters are sensitive to the tune of the machine. The predicted values were calculated assuming an ideal (zero-emittance) electron beam. The calculation assumed that the CLAS liquid hydrogen target is present and full, whereas during the measurement it was present but empty. The difference between the calculation and the measurement can be taken as an estimate of the effect of filling the CLAS target on the charged-particle rates in Radphi.

There is an overall scale factor in the predicted BSD (barrel scintillator detector) rates that depends on the setting of the BSD discriminator. In Ref. [1] the discriminator setting was taken to fall at the maximum of the minimum-ionizing peak in the scintillators. Figures in that report show the difference in the energy loss spectra for recoil protons and the minimum-ionizing background from electromagnetic showers. Those figures can be compared with Figs. 1-2 below. The Monte Carlo did not include the effects of attenuation in the scintillator paddles. The fact of attenuation implies that one must chose a position along the length of the paddles where to set the threshold to the minimum-ionzing maximum, and that where that peak falls relative to the threshold will depend on polar angle, or $z$. The overlaps of the barrel scintillators allow them to be divided longitudinally into pixels of about 10cm in length. Pixel number increases with $z$ from 0 (upstream) to 7 (downstream). Pixels 0 and 7 are half-pixels. To be safe, we chose to set the threshold in the middle of the MIP peak for pixel 6 near the downstream end, as shown in Fig. 1. The upper plot includes all triple-coincidences between the ORs of the three BSD layers, with the BSD threshold set low to reveal the minimum-biased spectrum. Most of these hits are MIP background. The lower plot shows the same spectrum taken with the threshold set at the operating point (indicated by the red line) and a trigger that requires a large energy deposit in the lead glass. Most of these hits are presumably recoil protons.

Figure 1: Pulse height spectrum for pixel (triple-coincidence between three overlapping counters) hits in the BSD, for different trigger conditions. Pixel 6 is located near the downstream end of the barrel, in the vicinity of $30^o$ polar angle.The upper plot is mainly electromagnetic background, whereas the lower plot is mainly recoil protons. The red line indicates where the threshold was set for normal running.

Figure 2: Pulse height spectrum for pixel (triple-coincidence between three overlapping counters) hits in the BSD, for different trigger conditions. Pixel 2 is located near the upstream end of the barrel, in the vicinity of $60^o$ polar angle.The upper plot is mainly electromagnetic background, whereas the lower plot is mainly recoil protons. The red line indicates where the threshold was set for normal running.

Fig. 2 shows the same pair of spectra for hits in pixel 2, towards the upstream end of the counter. Here one can see the increased pulse height relative to pixel 6, which results in a decreased discrimination against background for upstream pixels, over what was expected without attenuation in the Monte Carlo. As can be seen in Fig. 3, the background rates are strongly peaked downstream, which means that the sacrifice in background rejection at the upstream end of the counter does not have too large an effect on our overall level 1 trigger rate. The three independent layers are hence 25% over the predicted rates, which can be roughly cubed to obtain the factor of 2 in the measured triple-coincidence rate over the prediction. We were careful during the test to ere on the side of a low threshold, knowing that for the test we could raise it offline. With the threshold taken into account, the agreement with Monte Carlo is very good. At this point a more careful study can be made using the t-slope for particular reactions and varying the target diameter, to determine the threshold that optimizes between dead-time and acceptance.

Figure 3: Pixel hit distribution for various triggers, ranging from minimum-biased at low threshold (top) to the physics trigger (bottom). The distribution of minimum biased hits background is peaked towards forward angles, as expected for electromagnetic background. The bottom profile agrees well with the recoil proton angular distribution from diffractive photoproduction. The middle plot shows what comes in at trigger level 1, which is intermediate between the two.

The tagging coincidence window was set to 30ns during the test run. This contrasts with the 10ns gate that was assumed in the Monte Carlo calculation. At present the alignment between the different tagging channels and BSD elements does not allow an efficient tag with a gate as narrow as 10ns. For the purposes of accidentals subtraction it may prove valuable to keep the online gate window somewhat larger than the actual coincidence peak. With some effort cutting cables, we expect to be able to reduce this gate width to 20ns.

The comparison between measured and predicted values for the higher levels of the Radphi trigger are shown in Table 2. Here the measurements are much better than the predictions, which compensates for the excess at level 1 over expectation. The reason for the large errors in the calculation is that this is the first time we have run the Radphi trigger without a charged-particle veto at level 1, and it turns out that estimates of leakage from charged-particle events through the upper-level trigger decisions were far too conservative. The net effect is that the rate of events to tape is an order of magnitude lower than predicted.

Table 2: Measured and predicted rates in the upper levels of the Radphi trigger, under the same conditions as shown in Fig. 1. The dead-time per event are average values under high-intensity conditions.

signal		measured		predicted
level 1 trigger		300		KHz		73		KHz
level 2 trigger		6.2		KHz		8.3		KHz
level 3 trigger		80		Hz		1.0		KHz
level 2 dead time		1.7		$\mu$s/event		1.2		$\mu$s/event
level 3 dead time		11		$\mu$s/event		14		$\mu$s/event
daq dead time		300		$\mu$s/event		300		$\mu$s/event
total live time		40%		@ 80Hz		50%		@ 1KHz

The majority of the experimental dead-time comes from the level-2 logic. The excess dead-time at level-2 per event over the 1.2$\mu$s that was foreseen was inserted to give adequate time for the high-resolution TDC to fast-clear. A simple rearrangement of our electronics is discussed below that will allow us to reduce this dead-time by a factor of 2.