[time-nuts] Experience with THS788 from TI?

SAIDJACK at aol.com SAIDJACK at aol.com
Wed Mar 21 22:52:33 UTC 2012


Two suggestions:
 
How about using an off-the-shelf Wavecrest counter?
 
These have a 0.8ps resolution, and typically have a noise floor of  around 
3ps when averaging. They certainly can do 10ps single shot. There is a  
DTS-2050 on Ebay now for $700: item number 120647180882
 
You can't get much lower than that for the resolution and accuracy that  
system provides in a working off-the-shelf solution.
 
Alternatively, if you want to design your own, you could use a  
time-expander. It only requires a small micro with counter/timer, and a  little bit of 
external circuitry for charging/discharging a precision cap. You  charge 
fast gated by the signal to measure, then you discharge slowly (expanded  time) 
and measure the amount of charge deposited on the cap. The Linear Appnote  
I mentioned earlier already has most of the capacitor charge pump circuitry 
in  it that you would need for this.

The idea is to design the cap discharge and charge cycles at  different 
time scales, say 1000x to 1, so that the capture time get's expanded  out to 
intervals that the micro can measure. If the micro has say 60MHz counter  
resolution (16.66ns) then a 1000x to 1 expansion would allow a 0.016ns (16ps)  
resolution. Using say 2000x expansion and a 100MHz counter in the  micro 
would get you to 5ps resolution.
 
This setup works very well without having to buy a $170 TI chip and  
designing with an FPGA, and only requires a little bit of software in the micro,  
and a small number of analog components. I think the operating  principle 
and circuitry is explained in great detail in the service manuals for  the HP 
5334A counters, and the PRS-10 rubidium service manual as well.
 
bye,
Said
 
 
In a message dated 3/21/2012 14:58:57 Pacific Daylight Time,  
bgamari at physics.umass.edu writes:

Bob Camp  <lists at rtty.us> writes:

> Hi
>
> Ok, for a legit  12 ps with 0.1 ps drift and 200 mega samples per second -
> not to many  alternatives. The FPGA stuff will get you to 50 to 100 ps on 
the
> same  basis this gets you to 12 ps. They will get you to 20 to 40 ps on a
>  good day - sort of the way this chip gets 8 ps. The FPGA will do it at  a
> much lower data rate. 
>
In our experiments, we are  typically observing very low count rates
(100kHz at absolute most). I've  occassionally stumbled upon a paper
which claims to get 10ps on a standard  FPGA, but naturally they never
show the code. Given that I'm a relative  novice at high-speed
electronics and FPGAing in general, I'll consider  myself lucky if I get
the 50ps advertised by the CERN core.

In  particular, one issue I've been struggling with is the
discriminator. Our  fast detectors produce a NIM negative-current pulse
which will ultimately  need to become suitable input for the FPGA. Of
course, the most precise  time measurement in the world is useless if the
discriminator front-end has  a nanosecond jitter. Unfortunately, I have
yet to find any open, high  precision discriminator designs. In principle
a constant fraction  discriminator doesn't seem to difficult to
implement, but when it comes to  preserving the high-speed signal
integrity, it seems like it could get  pretty hairy. Comments?

> If you average over many samples, all of  these will get you a better
> estimate. How much better depends on a  bunch of things. The TI part 
*could*
> do very well if you have a 200  MHz signal to look at.
>
For time-correlated single photon counting  (our primary use for
precision timing), having high temporal resolution is  quite
important. That being said, all of those arrival times all get  combined
into a correlation function so shot-per-shot jitter will be in  large
part averaged out.

Cheers,

-  Ben




More information about the time-nuts mailing list