[time-nuts] The amazing $5 timestamper with 6ns, resolution (next stop: 184ps)

[Jeremy Elson]

The sample rate depends on how much work I want to do to optimize the code.
It also depends on if we're talking about maximum continuous sample rate,
i.e. streaming data back to the PC, vs maximum burst sample rate where the
samples are just being buffered in RAM and transferred to the PC after the
experiment is over. (It's interesting to note that in going through the
documentation of my CNT-90 yesterday, it appears only to support the
latter, which surprised me.)

For burst sample rate, the code I've written today has a dead time in
between samples of probably about 500 nanoseconds (i.e. a sample rate of
about 2mhz), which is the time it takes for the interrupt handler to fire
and run a couple lines of code to copy the current counter value from the
capture register into RAM. However, with some tricks such as using DMA to
have the hardware transfer the timestamp value, the dead time between
samples could probably be reduced even further, maybe 3x? The stm32g4
devices have 128K of RAM which suggests about 16kilosamples can be buffered
in burst mode.

The comparatively slow part is transferring the data out to the host. Right
now, the code's fast-path just writes timestamps to a RAM buffer, and then
a relatively slow process comes behind every 50ms and drains that buffer
out to the serial port. This is limited by, at least, the speed of the
serial port, which I'm currently running at 1Mbps. There are various ways
to improve this (e.g. transfer binary samples instead of ASCII text) but
it's all a tradeoff of performance vs ease of use. I like emitting ASCII
because it's trivial for anyone on any host OS to read the output without
needing any special software. But, this limits the transfer rate both by
increasing the amount of serial data transferred and increasing the load on
the microcontroller for doing math conversions and ascii text formatting.

Right now my prototype code is sort of a hybrid; it handles bursts by
buffering, and continuously (but more slowly) drains the buffer out to the
PC over serial. So it means you could, for example, have a burst of 1,000
events in a millisecond, followed by a couple seconds of dead time, then
repeat - and it would stream all those events back to do you all day long.

-Jeremy




On Wed, Feb 17, 2021 at 7:44 AM Dan Kemppainen <dan at irtelemetrics.com>
wrote:

> Hi,
>
> I'm curious how this turns out. Please keep us updated on your progress.
> So, what is the maximum sample rate you can record at?
>
> One thought to look for a lower frequency with less jitter than a signal
> generator is TVB's PicDiv dividers. You could easily take the RB down to
> something lower that you is within your sample rate.
> http://www.leapsecond.com/pic/picdiv.htm
>
> If you ask nice, he has other source code for different divide ratios
> posted on his site. In my case, I needed to take 10Mhz down to 250Khz
> for testing a time stamping counter. It worked very well for this
> purpose, and was cheap and easy.
>
> Dan
>
>
>
>
> On 2/17/2021 4:59 AM, time-nuts-request at lists.febo.com wrote:
> > Date: Tue, 16 Feb 2021 23:48:12 -0800
> > From: Jeremy Elson<jelson at gmail.com>
> > To: Discussion of precise time and frequency measurement
> >       <time-nuts at lists.febo.com>
> > Subject: Re: [time-nuts] The amazing $5 timestamper with 6ns
> >       resolution (next stop: 184ps)
> > Message-ID:
> >       <CAKufN8UdmWh9oTE4pZ=EY57+sxh2JSc4oD7gSw3w7Mt6FS=
> w4w at mail.gmail.com>
> > Content-Type: text/plain; charset="UTF-8"
> >
> > I've used jlcpcb before, but I have to admit I'm more of a pcbway guy.
> I've
> > had a lot of good experiences with them, and their assembly special ($30
> > for qty 10) plus LCSC parts is very cheap, I"ve had good luck with them.
> >
> > You're right - the test I did conflated more than one thing. It was not a
> > pure test of my timestamper so my next attempt once I get some time will
> be
> > to use my Rb xo to run both the clock on my timestamper and a signal
> > generator and then see really how much jitter there is. Of course, there
> > will be some question if the jitter is coming from the timestamper or the
> > signal generator, but I suppose the best case is that there is not jitter
> > in either one! I'll keep the list updated as I do more experiments.
> >
> > Cheers!
> >
> > -Jeremy
>
> _______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com
> To unsubscribe, go to
> http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> and follow the instructions there.
>