At LANL, we bonded four channels of a TGV to a silicon pad detector, and left the other four channels floating. We observed the undesirable result that the four bonded channels went immediately into saturation, evidently due to noise coupling, both with the silicon unbiased and biased. The result was the same with a switching, and a linear supply for the silicon bias. Recently, we used a battery array for the biasing and observed marked improvement, that is the output baseline moved out of saturation, but was still very noisy. The situation was improved greatly by raising the threshold on the discriminator high enough to effectively shut it off. In this condition, the bonded channel response is quite close to that of the floating ones. We were able to observe output pulses from an Am-241 source incident on the pads. This is a milestone, and we now anticipate a series of more detailed tests.
b) The current TGV is designed for a front-end dynamic range of 9 MIPs. A resettable feedback design means that the output response of the amplifier will staircase toward it's upper rail for every charge input until a reset is sent. The time between resets envisioned for PHENIX is somewhere in the range of 1-10mS. MVD folks at LANL feel that the 9 MIPs of headroom is not adequate, especially being nervous about noise and unanticipated background rates. The simplest way to extend the dynamic range on the TGV is to reduce the gain. Chuck Britton intends to do this for the next tiny chip submission in May. The gain is currently 40 mv/fC. Chuck does not want to go below 10 mV/fC. Certainly, it would be hard to maintain an efficient 1/4 MIP threshold on the discriminator at a gain lower than this. We would like to have 50 MIP dynamic range. The exact solution to keep enough gain and dynamic range is not yet there.
The current issue concerning the AMU is whether the same capacitor cell can be read twice, preserving enough of the initial information to allow the second read to be useful. Because each level 1 read of the AMU requires 2 samples for the MVD, this becomes an issue when a level 1 accept is generated for each of two adjacent beam crossings. The second sample of the first accept becomes the first sample of the second accept. Based on the Pb-Gl experience, Chuck and Alan Wittenburg have reported that the degree to which the information in a cell is corrupted by a read depends on the feedback capacitance in the read amplifier, and on the amplitude that is stored. The overall effect will appear as a gain shift in the ADC. The effect was 2% of full scale in Pb-Gl, and estimated by Chuck to be ~4% for MVD. If we get the 50 MIP's, this implies as much as a 2 MIP distortion in the second read.
The probability of generating 2 level-1 accepts in adjacent crossings is being looked at by Barbara and Glenn. They each intuit that this must be low. John Sullivan thinks it is not out of the question to consider examining these adjacent events individually, off-line, and perhaps correct for the distortion.
The digital correlator has the advantage that once digitized, the information can be copied and held in memory, so that if the adjacent crossing level-1 situation does occur, one can read the copied sample with no loss of information. The analog correlator does the subtraction automatically, and the individual samples, in normalread, are not preserved. We feel that the overhead on the heap manager to actually implement the digital scheme is not worth it. For the purpose of monitoring the baseline movement of the TGV between resets, we envision a mode in which the correlator is turned off, and we take some integer number of non-subtracted data samples between resets.
Chuck considers a self-adjusting (setting the rails) ADC too much of a complication and proposes setting the ADC endpoints via a DAC.
| 1.2 mW preamp |
| 0.3 mW discriminator |
| 0.75 mW AMU 0.5 mW analog correlator |
| 1-1.5 mW ADC, 10 bit |