[time-nuts] More results with the ACAM GP22 chip
th.allgeier at gmail.com
Sat Dec 5 07:12:32 EST 2015
I can now add a few more results from "Range 2" of the ACAM GP22 TDC chip.
As before they can be found here: www.stanton-instruments.co.uk/page26.html
The first 2 files are more or less a repeat of what I did for "Range 1" except using a 250 m RG6 coax spool as delay line. In a departure from the original plan I left the whole length intact as one continuous piece. This gives pretty much 1 µs delay, a little less than predicted, and the noise band is around 1 ns wide. I must add that it was somewhat tricky to tune the system to get this as the waveform out of the un-buffered line is not so nice. The output level from the signal generator has to be set just right for the GP22 to trigger, the only reliable setting is a "falling edge" and when the scope is connected at the same time the noise is several times worse.
So I pinned high hopes on Bob Camp's suggestion to use a 74ACT14 hex Schmitt trigger to buffer the line - which I did and results from which are given in the second pair of files.
In a nutshell, this changed very little to the "noise outcome" of 1 ns but made the whole affair much easier to "tune", since of course now I have clean waveforms. It makes little difference now whether "falling" or "rising" is selected in the chip's software.
The Schmitt trigger as expected increases the delay a little, so we have around 1100 ns, and it also seems to add noise if not powered exactly in its sweet spot of around 5.02V. As little as 50 - 100 mV higher or lower will add several ns of extra noise. In some cases having the scope hooked on also added noise, but not as much as with the direct approach of using the coax line only. The 74ACT shows around 20 ns between in and out (double-inverted as suggested by Bob) which may depend a little on Vcc and probably has some jitter of its own even with constant supply.
As before we have a minor frequency dependency, but less so: I have used a 74ACT14 before and after the line, which should have given more consistent rise times as compared to the "raw" output from my Thandar sig-gen.
A long story cut short: With crude and noisy kit you can expect to time a period to within 1ns in range 2 of the GP22, assuming that the noise level does not increase for longer periods. And also allowing for the fact that this is all based on a 32kHz "master clock" on the evaluation board. Which of course could be done differently and with higher precision if needed.
This applies to no averaging being selected in the chip's software. If speed is not essential and you can wait to average 2 or even 10 cycles the whole noise thing goes pretty much away, which of course is pretty clear from the results plots.
I hazard a guess: The fact that the noise level in the "best case" scenario comes out the same with and without the buffers indicates to me that on the inputs of the GP22 are effectively similar Schmitt triggers to what I have used externally.
I hazard a second guess: If you hooked a very precise and stable frequency source as an input onto the GP22 and did it in a way that keeps the signals clean the claims of 50 ps "resolution/stability" could probably be achieved even into the µs range. The 250 m spool sitting on my floor and all the very crude connections don't fill me with confidence. (In the ns range I got the 50 ps pretty much already with the crude setup as per the initial results.)
Which means for what I have in mind the chip is adequate, and what I saw initially (10's to 100's of ns noise) is simply the oscillator on the weighing cell trying to keep itself tuned with the mechanical vibrations which it is meant to lock on to.
More information about the time-nuts