[Halld-cal] BCAL SiPMs, adding channels, and noise

[Matthew Shepherd]

Hi all,

I wanted to follow up on a detail regarding dark rate in the BCAL and proposed readout scheme.  Dan Bennett has been working to code up the simulation correctly for the new scheme of doing an analog sum of SiPMs into a single FADC channel.  This has got us thinking again about these issues again.

Recall that the readout threshold within a cell (formerly a single SiPM) was set such that for beam off events on average one would have 5% of all BCAL cells above threshold.  The 5% is a my guesstimate of what the DAQ can handle shipping off the crate and onto disk.

When one goes to clusterizing level one achieves a nice background suppression since the clusterizer requires hits on both ends of a cell.  The probability of this is 5%*5% -- small.

If now we gang together three or four SiPMs several things happen:

If we apply the former single SiPM threshold to the sum, as expected, many cells (collections of 3 or 4 SiPMs) are now above threshold.  This two implications:  (1) this is a problem for the DAQ since tons of noise data is transmitted from the crate across the backplane to disk and (2) this also a problem for the reconstruction.  As a simple test, with single 700 MeV photons one now reconstructs on average 18.4 photons per event in the BCAL.  With the old single-cell readout one gets an average of 1.2 photons per event.

The naive solution is to simply increase the cell threshold by a factor of three or four (depending on how many SiPMs are being added).  We did this and, as expected, the average number of reconstructed photons returns to around 1.  However, I'm certain this will have a very negative impact on the physics.  The previous threshold was equivalent to about 7 MeV deposited in a cell -- a factor of 3-4 means a cell energy threshold of 20-30 MeV.  If we start throwing out all cells with this much energy then we can anticipate terrible resolution for low energy photons (which we know are important).

If it is truly the number of cells and not the fraction of the cells that sets the DAQ limit (which is likely) then with a total of a factor of 3 fewer cells we should be able to tolerate a higher percentage of cells on.  For example about 15% of the current planned number of FADC channels is equal to about 5% of the former planned number of FADC channels.  This means we can lower the threshold and tolerate the higher fraction of channels above threshold, but the reconstruction also needs to be able to weed out these noise cells.  Previously we had a fake hit cell rate of 5% of 5% -- that would now go to 15% of 15%.

The ideal solution is that the SiPM noise rate drops by a factor equivalent to the number of SiPMs that have been added together.  This is why I asked this morning if the specification for SiPM noise had been changed in light of the newly proposed readout scheme.

I think the bottom line is that this needs careful study.  We can start by examining the assumptions that went into the estimation of the threshold -- this is documented in GlueX-doc 1231.  We need to verify that all those assumptions are still correct -- if they still suggest a 20-30 MeV cell threshold then I think we have a problem to solve.  One can think of an infinite number of solutions:  time multiplexing if SiPMs, forming an AND between cell ends in the FADC, etc.., but we should verify it is a problem first.

Cheers,

Matt