Related to that issue, let me ask you few questions:

-Of course, fake hits have most of the time a cluster multiplicity of

1, contrary to real hits. So a tracking algorithm could/should

benefit form that fact. Are you going to use that fact in your

tracking alogrithms ?

My understanding is that using a simple cut on 1 pixel clusters would

reduce the efficiency by ~5% based on the smaller cluster sizes for

the high resistivity substrate. This is higher than the cd-4

parameters would allow for sensor efficiency, though if we ended up

having much more noise than we have recently measured, this would be a

good option. This is a simple analysis, finding a clever way to weight

suspected noise hits based on column or position could provide a

larger cut on noise without affecting the efficiency as much, but

figuring this out would take a larger and different data set than we

have. (see noisy pixels answer below)

-Do you want to reproduce a realistic mapping of fake hits or can you

accept just a random fake rate ?

I would say that for a first pass, a random fake rate would be good.

After that we would optimize by looking at the most noisy pixels and

discriminators with the most dispersion and/or threshold variance from

the norm. If the particular column is much noisier than the norm, we

would envision disabling that coulumn. This will depend on just how

noisy this column is. We have a limited amount of frame hit memory, so

we wouldnt like to go into overflow and miss hits just because of a

hot column. These parameters would need to be analysed based on a

large sample of beam data.

-Do you want to tag offline the most noisy pixels ?

See above answer, but I think that the answer is yes.

-Last but not least, fake hits distributions depend mostly on the

discriminator threshold (and also irradiation).

So what is the "efficiency vs fake rate working point" you want to

use ?

This is an very valid observation. We will choose an operating point

by looking at what discriminator setting gives us a high efficiency

with an acceptable noise rate. For the cases where there is an obvious

hot column, we would have to see whether setting the discriminator at

a higher setting for that block (each sensor is divided into 4

separate column blocks with an individual discriminator setting for

each block) is a better solution than just disabling that particular

hot column(s). As a starting point, we should take probably a very

conservative approach which would be something like 98% efficiency and

10^-5 accidental rate. This should (estimate by eye) more than cover

most of the dispersion in the discriminator threshold settings for

most sensors. The measured values give a larger S/N value than that,

but as a starting point this should be conservative. As for

replicating the hot columns and the discriminator dispersion, this

would be valid parts of a complete detector response function.