[time-nuts] algorithms and hardware for comparing clock pulses

[Can Altineller]

Hello All,

Another night heroically wasted, and I have achieved a PID lock within 50 /
80 000 000 th of a second between two clocks. 50 is there because it is the
minimum clock ticks for the irq handler to execute.

To summarize:

1. I am using a TI launchpad running at 80mhz. I was using micros() to
measure time, but then I realized I could use SysClockTicks, which is
80,000,000 per second. In Energia, I had to use: time_micros =
HWREG(DWT_BASE + DWT_O_CYCCNT) to get the sysTicks. I am not sure if PID
algorithm is the right one for tuning clocks, but thats what I only knew.

2. The diffuculty arises from the fact that I am using 1pps signal to tune
a DS3231, which also gives out a 1pps signal, (it can be configured to give
faster pulses, but then when I modify the ppms, they wont be reflected in
the pulse output) So in order to observe the reaction after setting aging
register, I have to compare 1pps signal.
If I had a RTC with a tunable 10mhz clock, I could feed that into a PLL /
counter, not the case. Eitherway, I am in the process of ordering a VCTCXO,
and probably will build my own RTC.

3. I have investigated and experimented different interrupt schemes, like
taking measurements on both rising and falling edge, and XOR'ing the clock
signals. Using a 74hc74 and a XOR gate, I have made a mini-circuit so that
if one clock is lagging behind, it will send pulses from one pin, and if
reverse, it will pulse from other pin. However this scheme failed because
same difficulty with interrupts arise again. The pulse is too fast in order
to trigger interrupts on the falling and rising edge, and measure time.

4. I think an analog solution like Bill Hawkins described, would be best
suited for this task. But I have not understood it enough to build it.

Best Regards,
C.A.

On Wed, Sep 23, 2015 at 7:25 AM, Bill Hawkins <bill at iaxs.net> wrote:

> Group,
>
> Seems to me that what's needed here is a current source for linear volts
> vs. time and the cmos switching to control the duration of the capacitor
> charge while the phase flip-flop is on. When it turns off, it interrupts
> the processor and isolates the capacitor so it acts as a sample-and-hold
> device. The processor can take its own sweet time reading the capacitor
> voltage (although this sets minimum limits on the pulse duration). When
> the reading has been captured, the micro toggles a FF that shorts the
> capacitor with a cmos switch. The short is removed when the phase FF
> toggles on.
>
> This is a lot of analog circuitry, but it will operate as fast as the
> parts are capable of switching and not at the whim of whatever the micro
> is doing.
>
> Hope that's useful. Probably already been done.
>
> Bill Hawkins
>
>
> -----Original Message-----
> From: Magnus Danielson
> Sent: Tuesday, September 22, 2015 3:32 PM
>
> Jim,
>
> I had the intent to try this, but never got around doing it. Thanks for
> reminding me. Please share any enhancements.
>
> I did exchange some emails with Lars, but as that project never got off
> the ground, it faded out.
>
> Cheers,
> Magnus
>
>
> On 09/21/2015 10:02 PM, Jim Harman wrote:
> > Hi Can,
> >
> > For a simple analog solution, you might try a 74HC4046 phase detector
> > followed by a diode and RC network as used in Lars Walenius' GPSDO,
> > described here in the archives:
> >
> > https://www.febo.com/pipermail/time-nuts/2014-February/082820.html
> >
> > The phase detector produces a pulse whose width equals the time
> > difference between the two pulses. The RC network converts this to a
> > voltage proportional to the time difference, which you then measure
> > with the MCU's A/D converter. Using the rising edge of the signal at
> > pin 14 as the interrupt source triggers the A/D converter at the end
> > of the pulse, which corresponds to the peak of the analog signal.. The
>
> > 1 meg resistor discharges the capacitor between pulses.
> >
> > Lars' code also includes a filtering algorithm which does a nice job
> > of controlling one of the oscillators to match the 1-PPS generated by
> the GPS.
> > I have enhanced this if you are interested.
> >
>
> _______________________________________________
> 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.
>