[time-nuts] Sub Pico Second Phase logger
bruce.griffiths at xtra.co.nz
Fri Dec 5 13:39:49 UTC 2008
> Doesn't this phase detector, like all digital phase detectors, have
> significant non linearity at the ends of its range?
> In the case of an XOR gate phase detector this is caused by the finite
> slew rate of the gate output.
> Thanks for all the ideas and Information.
> True, XORs don't work good at the end of their range for several reasions.
> This is why I had to put in the second pair of detectors.
> K1 and K2 are used in the example as a sort of "smooth linear demulplexer"
> to switch between the phase detectors. As one set of differential
> phase detectors gets near its nonlinear and troublesome range at their zero
> and 180 phase point their output change contribution is COMPLETELY shut down
> by the a "sin" like function product of K and Phase#. Not really a sin
> function and not really done as K's. The function of the sum of the two
> phase detectors will be dependent on the voltages themself.
> That is their modified voltage out will have less effect the farther away
> from zero their phase voltage is. A smooth clipper function.
> For a given detector pair the farther from zero it is the less its change
> will contrubute. When both the zero and 90 degree phase detectors
> are half way between zero and rail, their outputs gets summed together
> with a weighing factor of 1/2 each.
> One disadvantage of the quad phase detectors, is to work good and
> allow the digital processing functions I want to have, there has to be two
> seperate data paths, one from each Pair of phase detectors.
> The data rate is slow enough even with the desired oversampling
> that a multiplexed ADC my be OK.
Multiplexing a sigma delta ADC inputs only works well if the ADC design
Linear technology have some especially designed for use with multiplexed
Sigma delta ADCs are relatively inexpensive so one could always use a
couple of them.
Surely the outputs of both phase detectors are required to sort out the
the sign/direction of phase changes?
One could always log both outputs and sort out the sign of any changes
together with the appropriate output to use for determining the
magnitude in software.
> Is there anything that cares about faster than about 10Hz high resolution
> phase update times, that a simple analog XOR phase detector could handle?
> Also I do not think that the linearity has to be very good. I was thinking if
> it stays within say 90% it would be good enough for what I know of. The important
> thing is Just so it stays monotonic and glitch free with no hysteresis etc which
> it seems to do. It is not 1 million to one accurate as you know, maybe more like
> within 1% or 0.1%. The one I build only has a million to one resolution
> around 90, 270 degs, zero and 180 deg. which is the only place I generally using it.
> At the moment the accuracy and noise performance with the detector away
> from its zero output value is limited by the accuracy of the reference along
> with a lot of other gain error things.
> Is there any use for a truly linear and accurate simple phase detector?
> I supose it could be done, in a similar mater, but may have to add a couple parts.
Only when one uses such a phase detector to measure the frequency offset
between the 2 sources being compared as quickly and accurately as possible.
The principal source of non linearity near the ends of the range is due
to insufficient time available for the transitions to settle accurately
with short pulse widths.
If the value of the first resistor in the RC passive filter is too low
the output resistance mismatch between the 0 and 1 states will
contribute significantly to the nonlinearity.
Modulation of the output resistance of the gate/buffer driving the RC
filter can also be significant.
Both of these effects can easily be reduced to insignificance by using
appropriate feedback circuitry.
> You also said HCMOS buffers have 4ps or so of jitter. Is this the kind of jitter noise
> that can be filtered down into the mud with the present 100ms analog Bandwidth
> and the 250K samples that are effectively being averaging?
> Do you have any knowledge or guess on ICs NON tracking delay change with Temp?
Random jitter will indeed average down to the femtosecond level with
that many samples.
A crude way of estimating the propagation delay mismatch tempco for CMOS
ICs is to assume that the propagation delay mismatch has the same tempco
as the propagation delay itself or about +0.4%/C.
e.g. if the propagation delay is say 10ns and the mismatch is 10% (1ns)
then the delay mismatch tempco will be in the region of 4ps/C.
> I've tried to match all delays in the four loops, by always having a part from the same
> IC in each loop, so mostly, as long as all like parts in any single IC track well then their
> zero errors should cancel.
> I have not tested it much below 1ps except to see what I think was more like 0.1 ps resolution.
> The test I did for that was to move my hand near the center of one channel's shielded
> signal cable the watched the phase output smoothly change like a proximity detector
> as my hand approached the shielded cable, not due to cap to ground but due to the
> cable delay changing. Thoses electrons just don't get very far in 100fs
> I am considering using a faster famly, but I do want to say away from anything that
> produces any heat. It looks like temperture offsets and changes are going to be the
> limiting factor for zero stability.
If the power dissipation is low it would be feasible to use a
bootstrapped oven (similar in thermal design to some of the standard
cell oven designs from the 50's and 60's) like Wenzel has used to
achieve sup-picosecond stability in a frequency multiplier.
> I use the same dual detector design to phase lock now, but I never have checked or
> cared how good it can get.
> Thanks for the review and feedback, I've added several questions, please comment where you can?
> I'm planning to make another pass at it and clean it up for thermos etc. and see how low it will go.
> Concerning other uses of the single differential XOR phase detector.
> If anyone wants to improve the performance of an existing crappy XOR phase detector
> they may want to consider the differential XOR detector. Just be sure to include a good differential
> integrator (with a zero pole) at its output before driving the EFC so that the phase detector's
> output always stays at zero. The differential detector can reduce phase detector errors
> by many orders of magnitude.
More information about the time-nuts