Comments on fast detection

>Dear John,
>
>    here I am again, with another bunch of questions. Recently I had a discussion
> with one of my customers about fast time series and I would very much appreciate
> your comments:
>
>   Let's assume we want to measure the intensity of a diffraction spot at
> a time resolution of 1 microsecond in a one-shot experiment. We have a
> multichannel scaler with a minimum dwell time of 1 microsecond or better for that.
> We are interested in a dynamic range of about 100 to 1000, i.e. the signal
> should not decay much more than a factor of 1000 during the length of the time
> series which would be something like 16000 dwell times.
>
> Now, what type of detector should we use:
> - a fast one like a plastic scintillator with a shaping time of something
>   like 10-100 nsec?

OK. Don't forget that (for Gaussian shaped pulses) that the effective
pulse full width is about 5x the 'shaping time'. Plastic (use the Pb loaded type!)
exponentially decays with a time constant around 3nsec, YAP:Ce has about
30nec decay. Gaussian shaping is not optimum: gated integration is far better.
Unfortunately, there is nothing available on the market in a 'turnkey'
configuration that I know of.
 

> - a very fast one like an avalanche diode (shaping time around 1-10nsec?) ?

OK. You will find it extremely tough --even with an APD dector and fast
preamp-- to get 'pulse pair resolution' in your counter system to <5nsecs. Still
pretty good! APDs are very different to scintillator detectors: what will be
best for you depends on the nature of your experiment. For e.g., above ~15keV the
direct APDs are lousy due to poor X-absorption...
 

> Here are the more exotic ideas:
> - a diode in current mode with a fast microA-to-frequency converter

Why not?
- diode-preamp signal/noise may kill you unless you have a strong signal;
- the diode-preamp RC time constant may kill you (to go fast you need a
  SMALL pindiode operating under bias);
- and the time response of the I-F convertor
  will likely be a problem. It is hard to find I-F convertors above 10MHz 'full
  scale',and these will not necessarily have a 10 Mhz 'time response'* :
  it may be much smaller!

> - direct detection of photoelectrons from a scatterer with a photomultiplier
    or a channeltron

As above, the questions are experiment related:  mcp's and channeltrons
have good electron detection efficiency, but only a narrow range of good
efficiency for X-ray detection, need good vacuum, etc. etc.
 

>     The other question is: At what point will the time-structure of the
> synchrotron spoil the fun: for a ring like ESRF it takes an electron
> bunch a couple of microseconds to travel around - let's call that the ring period.
> If we want to avoid timing the detectors with the electron bunches, is it
> correct to say, that the minimum time-resolution is given by the ring period? Or do
> we need a higher safety factor (like minimum dwell equals 10 ring periods, if we
> want to treat the source as quasi-continuous)?

Ho ho ho!  the $64 000 question again! The ring bunch structure is a nightmare. I
summarize it thus:

- unless the ring has a symmetric fill pattern, the ring must have a
  'fundamental' frequency that is  c/(ring circumference). For the esrf, this is about
  300kHz
    (note added: nowadys ESRF runs uniform filling at 200mA and 70h lifetime, i.e.
     every bucket is filled. This puts the fundamental at 3usec/900 buckets and thus
     gets one into the MHz regime)
- on top of this, there are probably obvious frequencies associated with
  the diverse fill modes (mult bunch, 1/3, etc. etc.)
  One can always play at Fourier analysis stuff for more.
 

I --pretty naively-- see things thus when it comes to the detectors:

1. 'time response'* of detector >> 1/frequencies of ring.
   In this situation, the detector does not 'see' any structure. When it
   comes to photon being detected, Poissonian statistics will aplly. Situation
   relatively simple.
2. 'time response' of detector << 1/frequencies of ring.
   Now the detector is so fast, it works normally during the 'active'
   spills of X-rays, and just 'turns off' when there is no X spill.

Take example of a detector with time resonse of say 30ns . For the ESRF
in 2/3 fill, this detectector is fast enough to just operate normally for the
2usecs when X spill (normally means it will count according to Poisson deadtime
losses), and just count nothing for 1 usec. It just 'doesn't see' the 3nsec
structure (between successive bunches) in the esrf machine.

3.  'time response' of detector ~ 1/frequencies of ring.
Horribly complicated. Maths is not my bag. There are a couple of
internal reports I've seen, and I'm sure if you looked you'd find some reports for
particular cases, but I doubt they would give usefully accurate predictions for
'real' detectors+electronics for which the 'time response' is never simple to
characterise.

Finally, if you are making repetitive measurements even at frequencies
<< all ring frequencies, be careful of 'beating' effects. The ring will have
some superb quartz clock which stabilizes the machine rf etc., and your data
acquisitions are also probably driven somewhere with quartz clock accuracy. If you

start accumulating good statistics, you can quickly reach a situation where
accumulating n or n+1 buches in a measurement may be a significant error.

*by time response, I mean loosely the 3dB (say) response frequency for a
'continuous' response detector+electronics, or the individual pulse
processing 'dead time' for each X-ray in some counting system.
 

John Morse
Instrument Support Group/Detectors, Experiments Division,
European Synchrotron Radiation Facility
6 rue J.Horowitz, BP 220 , 38043 Grenoble Cedex FRANCE