[time-nuts] yet another GPSDO design, or so

Attila Kinali attila at kinali.ch
Tue Jun 29 08:39:29 UTC 2010


On Sun, 27 Jun 2010 02:24:31 -0700
Said Jackson <saidjack at aol.com> wrote:

> The 4002 expects a tight phase lock on the two inputs to properly stay
> locked,

Why does the ADF4002 need that? Or do you mean by "locked" that the
locked output of the ADF4002 does reflect the actual locked state?

If you mean that, i dont think i will use that at oput. Maybe as
a debugging help, but not for the control loop itself.
As for the design, i wanted to rely on the ADF4002 haveing a
linear phase difference response within [-2Pi,2Pi] and integrate
over the phase error to average out the jitter introduced by GPS. 

> and your adc/dac will likely introduce too much phase lag and
> cause oscillation. In fact when using the Analog Devices PLL simulator
> one has to closely follow the component values of the loop filter that
> the software suggests otherwise the system won't be stable, unless you do
> a lot of manual math. The 4002 works well with loop bandwidths of 30 Hz
> or more.

The exact control of the loop parameters are exactly the reason why
i want to use a uC implementing the control loop. Implementing such
low frequency control loops is a lot easier in software (or rather
digitally) than it is doing with analog electronics. And yes,
i'm quite aware that i'd have to tweak the loop parameter for stability.
But i dont expect this to be difficult as long as the analog part
is operated in its linear range. Unless of course, there is
something in the electronics i have not taken into account. Which
is why i'm asking here for advice :-)

> For a GPSDO you are looking for loop bandwidths of 0.001Hz or less, a
> totally different world. Even if you use the 10 Mhz output rather than
> the 1 pps. This is because your Isotemp Ocxo will be much more stable
> than the gps for short time intervals, say 0.1s to 500 s

Yes, i'm thinking about a 1/f in the range of 100s to several 1000s.
> What would be easier to try is to replace the gps internal Tcxo with
> an external ocxo, but you have to generate the frequency the gps is
> using, such as 26 MHZ and do some soldering on the gps itself.

Yes, that would be an idea. But it's not that easy. I dont know
how the control loop in the LEA-6 works and whether they are actively
changing the TCXO frequency. If they are, i would need to provide
a VCO that is stabilized, either in form of an seperate OCXO or by
locking/syncinc the VCO to my 10MHz OCXO.

I have one open LEA4-S infront of me. That has crystal mounted
marked with R4130 and that looks from the configuration (two
of the 4 pads on GND, two connected to other components) that
it is a simple quartz without any voltage control.

The Atmel uC (probably a SAM7 OEM variant) has only one small
crystal connected to it that looks like a 32kHz quartz. No
other frequency source is on the uC side of the module.
Although the LEA4 is two generations older than the LEA6,
i expect it to be at least similar by design.

> It would be very interesting to see just how good the uBlox lea-6t can
> work with an ultra stable frequency source rather than a $5 tcxo..

Yes. But then i'd have to build two devices, one that uses a stock
LEA6-T and one that uses a modified LEA6-T and measure both to have
an indication how much improvement this brings :-)

			Attila Kinali

If you want to walk fast, walk alone.
If you want to walk far, walk together.
		-- African proverb

More information about the time-nuts mailing list