[time-nuts] Help with Novatel OEM6 external oscillator

[Magnus Danielson]

Hi Skip,

On 2019-01-21 22:51, Skip Withrow wrote:
> Hello Time-Nuts,
> Hopefully, this doesn't get too long.  What I desire to do is hook up
> an external oscillator to a Novatel OEM6 series dual-band GPS
> receiver.
>
> Sounds easy. But, Novatel has four fixed (TCXO, OCXO, rubidium,
> cesium) plus a USER specifications to model the external clock
> performance.  Each model specifies three coefficients of the power law
> spectral density (h-2, h-1, h0).  If you google 'novatel
> externalclock' you can see the command. My question is, how do I
> translate an oscillator ADEV plot into the three coefficients?

First of all, an excellent question!

> I'm not an oscillator noise guru by any means (in fact, I'm probably a
> noise dummy).  It took me a while to figure out that the number range
> that Novatel is looking for is Allen Variance.  So taking the square
> root of their model numbers would be Allen Deviation (hope this is
> correct).

Actually, these numbers is really power-levels for the various noise-types.

You can however use Allan Deviation or Allan Variance to figure these 
out, so you got that part right.

>    I believe the h-2 coefficient represents Random Walk FM
> (Would this be slope=-2 on ADEV plot?), the h-1 would be Flicker FM
> (slope =-1?), and the h0 coefficient would be White FM (slope =1 (or
> is it 0)?).

If you look at this part of the Allan Deviation Wikipedia page, I 
prepared a handy table for you:

https://en.wikipedia.org/wiki/Allan_variance#Power-law_noise

The White Frequency Modulation (WFM) is the h0 value and has the -1/2 
slope in the ADEV, i.e. 1/sqrt(tau).

The Flicker Frequency Modulation (FFM) is the h-1 value and has the flat 
slope in the ADEV.

The Random Walk Frequency Modulation (RWFM) is the h-2 value and has the 
+1/2 slope in the ADEV, i.e. sqrt(tau).

As you see there is a scaling factor from each h-value to the ADEV 
beyond the tau-one, and this is important.

So, if you estimate the line-slope, you need to compensate for tau and 
scale factor to estimate the power-level of that noise-type. You do this 
by dividing your ADEV reading with the scale-factor for that noise-type, 
and for the tau-value of the ADEV reading. Naturally you can do that on 
the AVAR values.

> If I look at oscillator ADEV plots I often don't see anything as steep
> as -2 slope.  Is this because it is below the lowest tau of the plot
> and/or below the noise floor of the measurement system?
For the shortest tau you see the -1 slope of ADEV which is either the 
instruments resolution or the white phase modulation (WPM) noise. This 
represents the h2 level. The flicker phase (FPM) is the h1 level, but in 
ADEV we can't tell them apart in any practical manor, for that we need 
to go to MDEV which is what Dave Allan wants us to use, for that 
specific reason.
> The real question is how to I translate a typical ADEV plot into the
> three coefficients that the Novatel receiver wants to see.  An
> illustrative example would be most helpful.  From looking at the
> numbers in the Novatel documentation it appears that the h-2 number
> represents short taus, h-1 intermediate taus, and h0 long taus.

The other way around. h-2 is the long-tau vs. h-1 and h0 which is 
shorter taus.

I'd love to show it more elaborative, but emailing of the night-train 
has it's limits.

> For extra credit - Google ' Novatel clockmodel', and explain (simply)
> what the covariance matrix is.  Again, if an illustrative example
> could be shown it would be most helpful.
>
> That's all for now before my brain explodes.  Thanks in advance for any help.

We can't have you do that. I hope we can help you to figure this out.

Maybe I can do better on my way back tomorrow.

Cheers,
Magnus