In the course of working on BSMD/BPRS speed-up I have noticed why there are these unusual bands in the upper approximately half of the BSMD/BPRS spectra. I guess I should really say, I just have further corroborating evidence that there is a problem and where that problem is. The detailed explanation will have to wait till another day.

Here (figure 1) is the waveform seen by the analog-to-digital converter chip in the readout crate, when a test signal of about 25% full scale amplitude is injected to channel 1 (numbered from 0) of an SCA group on a BSMD FEE board:

The purple trace is the ADC clock, the black trace value at the time of the rising edge of the purple trace is what gets recorded as the ADC value. The channels 0, 2, 3 also seen here have only pedestal signals.

That looks fine (I mean that there is a nice flat top at the moment of sampling). [However, the sample time is much earlier than necessary, i.e., it is wasting time afterwards, i.e. deadtime, but it looks accurate enough. The glitches on the waveform at the ADC clock edges are normal and nothing to worry about. Also I should mention, the above is from the 2nd bucket readout, i.e., good data; 1st bucket is tossed because of known SCA settling issues.]

BUT NOW SEE A BIGGER PULSE: This time about 63% of full scale. (figure 2)

Another one (about 66% full scale): (figure 3)

Another one (about 80%): (figure 4)

You get the idea. Above 50-60% of full scale the waveform becomes wiggly and unstable. The instability probably has an amplitude dependence. The result is that signals in the upper part of the range will get an extra error contribution and some particular values will be favored over others, so "bands" will appear in the spectra. I don't have time right now to quantify that, but I can say from my scope measurement that the unstable wiggle amplitude is of the order of 10% of full scale.

It should therefore be assumed that signals in the upper half of BSMD/BPRS energy range have an extra (independent) uncertainty of about 10% of full scale energy, in addition to what the detector normally delivers, i.e., based on analysis in the lower half of the energy range.

I don't know how much ill effect on analysis this may have. Fixing it will probably be quite difficult at this point. (Needless to say, it would have been trivial to fix at the design stage :(  ). Contact me or Will Jacobs if you have input on this matter. Perhaps other uncertainties such as ionization fluctuations dominate it?

As mentioned, I don't have time to find the actual hardware culprit behind this issue right now, and I hope that perhaps we can just live with it. It probably has to do with the line driver circuit on the FEE board just like the "squashing" problem solved last year. Of course we cannot replace the FEE board, that would be quite an expensive project. In principle we could insert a new "cable buffer" board if it was important to do so and if it could fix the problem.