[time-nuts] No State Of The Art Counter

Magnus Danielson magnus at rubidium.dyndns.org
Thu Jan 6 22:05:24 UTC 2011

On 01/06/2011 08:02 PM, Tijd Dingen wrote:
> To whom it may concerns,
> Currently I am building a DIY frequency counter. Since this is my first serious counter project I am trying to keep things simple, hence It Will Not Be State Of The Art. Maybe a not-too-difficult hobby level counter will be of interest to some, so I'd thought I'd post here...
> The architecture in a couple of bulletpoints:
> - fpga based as much as possible to keep the parts count down
> - coarse counters running at max 200 MHz for now
> - interpolation is done using TDC's. The TDC's look suspiciously much like tapped delay lines and are implemented inside the fpga, using mainly the carry chains.
> - 10000 continuous time stamps per second
> - 500 ps timestamp resolution. And with resolution I mean the smallest resolvable thingy (related to bin size), not precision nor accuracy.

500 ps single-shot resolution is what you probably want to say.

How will the input side work? How will you handle input signals of 
various kinds? In particular sine of various amplitudes and frequencies. 
Slew-rate can be a limiting factor as white noise will convert into 
jitter if hitting a straight comparator. Choice of trigger point can be 
done to achieve lowest jitter, so just AC-blocking and trigger on the 0V 
may not be the best solution if shape is not well known. Noise of input 
stage comes into play.

> That 500 ps is conservative. I intend to do better than that, but just to give some idea of the starting point... The main thing for now is to get a handle on the DNL of the TDC.
> I am currently playing around with some designs in a spartan-6 and a spartan-3e. The spartan-6 design has more promise resolution wise, but it is also not very hobby friendly (bga package). So if this is ever to be built using hobby level soldering skills it will be spartan-3e. The 250k gates devices are available in tqfp-100 package which is doable IMO.

BGAs solder with hot-air gun and correct use of fluss.

> The timestamps are transmitted over usb to the pc for number crunching. The idea is to do some curve fitting to get a frequency estimate, computate Allan Deviation, and do the obligatory plots. With regard to Allan Deviation, as long as I make sure the measurements have zero dead time, I can compute Allan Deviation using the raw time stamps, right?

Yes. Make sure time-stamps has a format such that software can do 
time-wrapping extension.


More information about the time-nuts mailing list