Charles posted:
but the locked frequency will be different from both oscillators'
free-running frequency and
the EFC will not correctly indicate the test oscillator deviation
because it isn't the only control input in the system.

Good point and No argument  (except for the deviation part)
Because the EFC is the only control input THAT IS VARYING.

Also what I have said in the past, maybe unclearly:
NOW, Frequency offset due to loading effects, that is another issue and a
possible problem, so I'm surprised it has not yet been brought up.

The effects you are talking about are there and can be significant and are
easy to measure (with a second TPLL tester) but the BIG IMPORTANT point is
they are a fixed type of offset error and constant and do not effect the
Delta EFC with delta freq change.
The effect you state is a limit for the absolute accuracy of the DUT's
frequency measurment that the tester can make but it does not limit the
delta freq accuracy.(which is all that is really required for ADEV.)

BTW the way I reduce that effect you are talking about so that it also
becomes significant, compared to the accuracy of the Ref osc, is to add
attenuator pads in both osc paths. This does raise the noise floor some, but
the noise floor is so low that it can be raised and still not be significant
compared to the Ref osc, so it allows a good compromise to be made.

simple Example:
Take an osc that has some IL, buffer up its output real good at 90 deg and
now couple just a little of that buffered signal into the osc output. The
1e-11 to e-12 freq shift that will cause does not cause the osc to become
unstable or have some other significant different EFC shape, It just causes
the freq to change a little (offset) about the same as any other fixed and
constant load would have.

In any case there are lots of little subtle things that are going on that I
can not address in a single email, so that is why all I clam is that the
simple TPLL is better than the reference Osc, so it is good enough.
And YES it can be made MUCH better, if one uses more than a single active
part, but so what? for most things it is good enough as is.

Charles posted:

Warren wrote:

Actually, this is not true.  If either or both oscillators are
affected by injection locking (and they pretty much all are, to some
degree -- in this connection, note that you want to make measurements
down to E-12 or better [I thought you mentioned E-14 somewhere early
on], so even the least bit of IL will affect the results), what you
have is two control inputs to the controlled oscillator (the EFC and
the reference oscillator) and one control input to the "reference"
oscillator (the oscillator under test, which is itself controlled by
both EFC and the reference oscillator).  They will reach equilibrium
(unless the recursive feedback is unstable), but the locked frequency
will be different from both oscillators' free-running frequency and
the EFC will not correctly indicate the test oscillator deviation
because it isn't the only control input in the system.

Best regards,

Charles

I promised myself I would not get into this any more, but here we go again...

---- WarrenS warrensjmail-one@yahoo.com wrote:

Any parasitic control input is a problem in that system, like any other system.
I thought the point of all this was to measure the noise of an oscillator?
If it is noisy (and they all are, to some level, otherwise you would not need to measure it), then its frequency (or phase) is varying.

If the test oscillator is coupled (via injection locking) to the reference oscillator, the test oscillator will force the ref oscillator to follow its noise without the need to move the EFC. The EFC voltage will be stable (because the oscillators move together), while you have two synchronously noisy oscillators. If you measure the EFC, you will be left to believe your oscillator is better than it is.

Please note that the effect is not simply a scaling factor, because injection locking is a non-linear effect, or rather it is a mostly linear effect over a typically very limited dynamic range. Small variations will be totally coupled, where larger ones could possibly unlock the oscillators, producing steps in the EFC voltage. Said another way, you cannot eliminate the effects of injection locking by post-processing the data.

Injection locking is a parasitic control input and it is a problem with ANY method that purports to measure noise. Ignore it at your own risk, but don't say it does not matter, unless you want to prove something we already know.

Didier

Warren wrote:

No, it's not.  The strength with which each oscillator pulls on the
other also varies as the equilibrium frequency (the result of all
three recursive control inputs) moves around relative to the two
instantaneous free-running frequencies.  How much EFC is required
depends, in part, on the strength of the pulling.  There are three
varying inputs.

Magnus suggested that the effect of injection locking may be enough
smaller than the EFC input that it has little practical
significance.  That may be so, but when dealing with measurement
accuracy in the hundreds or tens ot ppt, this needs to be verified by
the results of carefully constructed experiments and hopefully also
supported by mathematical analysis.

Best regards,

Charles

On 06/16/2010 05:45 AM, Charles P. Steinmetz wrote:

What you get is a scale error. Consider that you have an amplifier gain
of 1000 and the injection locking provide a gain of 1, that will result
in actual gain of 1001 and the gain error on the EFC will become
1000/1001. Considering that Allan deviation estimation has problem of
its own, this scale error is not significant. What you do need to check
is that the relationship between intended gain and injection gain is
sufficiently different. Since oscillator frequency from EFC may not be
completely correct, we already want calibration of that scale factor
(K_O) and the gain error due to injection locking would be included into
that correction factor.

So, sufficiently small amount of injection locking gain will change the
apparent EFC coefficient K_O [Rad/sV] on which the scale of TPLL
frequency measurements depends. The fractional frequency observed is

y(t) = 2pif_0 / K_O,eff EFC(t)

Cheers,
Magnus

Hi

The gotcha is that the injection gain is phase angle dependent.

Bob

On Jun 16, 2010, at 1:57 AM, Magnus Danielson wrote:

On 06/16/2010 04:30 PM, Bill Hawkins wrote:

They are parallel.

Cheers,
Magnus