[time-nuts] Next step up from basic GPS/PPS timekeeping

[Chris Albertson]

My DAC is the ones built into the processor chip.  They are only 8-bits
wide but you can combine two of them by adding First reducing one of them
by a factor of 256 then adding them.  There is a small discontinuity but
software addresses that  Si I get a decent 15-bit DAC for free

As for the counter.  I was wrong.  It runs at full clock speed and it too
is built into the CPU chip.   Using an AVR/Arduino processor

This is actually applicable to a high-end NTP server.   The CPU clock on
the NTP server can be GPS disciplined.  If you do that then NTP has very
little to do as the internal clock will no longer change with the room
temperature.   After all those cheep canned crystals are the reason most
computers need NTP.  So you replace the cheap crystal on you Raspberry Pi
with a GPSDO.   What's the point of a sub uSec NTP server?  None, that is
the nuts part of TN.

On Wed, Feb 24, 2016 at 11:53 AM, Nick Sayer via time-nuts <
time-nuts at febo.com> wrote:

>
> > On Feb 24, 2016, at 8:30 AM, Chris Albertson <albertson.chris at gmail.com>
> wrote:
> >
> > On Wed, Feb 24, 2016 at 6:22 AM, Neil Green <ncguk at hotmail.co.uk> wrote:
> >
> >>
> >> What would be my next step up be, hardware-wise, in terms of improving
> precision, stability, etc? A GPSDO? Budget is limited as far as these
> things go - about £150 UK/$210 US.
> >
> > Given that you already have the GPS and PPS, you can build your own
> > GPSDO for very little cost.  The most expensive part is the crystal
> > and as long as you are not going for top of the line performance a
> > cheaper crystal will do. (You asked about the NEXT step not the LAST
> > step)
> >
> > A simple GPSDO is, well, simple.  All you do is compare the phase of
> > the PPS to the phase of the 10MHz crystal and then adjust the
> > frequency of the crystal to keep the phase difference constant.   Can
> > be done all in software by a $2 micro controller chip.  By Time Nut
> > standards the performance will not be top notch but compared to your
> > NTP server it could literally be 10,000 times better   A while back I
> > tried to build the cheapest and simplest GPSDO possible and except for
> > the price of the crystal and GPS receiver I spent about $5.
>
> The BOM for my GPSDO is similar. It’s not quite $5 until you also exclude
> the DAC ($7), but the GPS module and oscillator are well over 90% of the
> price.
>
> > Performan ce was such that when I put the cheap GPSDO  10MHz output
> > and my Thunderbolt both on a dual trace scope I could see some
> > movement in phase over hours long time but over days in stays locked
> >
> > First divide the 10MHz crystal to say about 1Mhz using a TTL divider.
> > Some (most) micro processor chips have a way that an external signal
> > can trap the current value of a counter.  Let both the PPS and the
> > 1Mhz do this and look at the difference in counts.
>
> For mine, the PPS “captures” a 16 bit timer that is clocked directly from
> the oscillator with no prescale. An overflow interrupt extends that timer
> to 32 bits, which is enough for 10 MHz to run for a few minutes before
> overflowing - plenty of time to capture phase deltas of a 1 PPS signal.
>
> From what I have learned about the whole thing so far, the real limiting
> factor is the phase granularity you can get with just a 1 PPS source. If
> the GPS module output, say, a 1 kHz signal, a GPSDO would be a lot easier
> to keep stable at low tau. Mine has a “hump” at sort of 10^2-10^3 tau that
> I can’t really do a lot about given that the only thing I have to
> discipline against is a PPS rising edge. I’m sampling 10 second periods,
> which gives me a base detection threshold of 10 ppb. I track a window of 10
> of those for the visual feedback. There, I can measure 1 ppb, but it takes
> 1000 seconds to do it. My newer version clocks at 20 MHz in an attempt to
> double the granularity, but from what I’ve been able to measure, it doesn’t
> do a whole lot about the “stumbling around” at low-ish (10^1-10^3) tau. The
> performance lower than that is dominated by the oscillator - in my case
> that’s either around 10^-10 or 10^-11 depending on whether you spend the
> extra $100 or not. :)
>
> My understanding is that better GPSDOs are able to provide for more
> granular phase detection. My system is granular to +/- 1 whole cycle per
> second of sampling. Something clocking a lot faster could come up with a
> system for fractional granularity. I could do that by clocking the logic
> faster and then dividing the output - exactly what I do for the 20 MHz
> variant - except that the 8 bit AVRs I’m using at the moment top out at 20
> MHz (at least, they’re not supported higher).
>
> It also doesn’t help me personally that my 53220A sort of peters out
> around 10^-11 anyway. I can’t really distinguish a Thunderbolt from a
> Connor Winfield OH300 based GPSDO, which clearly isn’t reality.
>
> Getting back to the R.Pi stratum 0 NTP daemon, I’ve got a PI Zero in the
> NTP pool as well. As others have pointed out, there isn’t a whole lot of
> value to be found in attempting to do better over the internet. Attached is
> the current pool monitoring graph for mine. The picture actually is a lot
> worse than the norm - which is represented by the right half of the graph,
> for the most part. Still, you can see that even in the most regular
> periods, it’s -3 or -4 ms +/- 1 ms, and that’s with a monitoring station
> only 300 miles or so away.
> _______________________________________________
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to
> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> and follow the instructions there.
>



-- 

Chris Albertson
Redondo Beach, California