Bob Camp wrote:

Once you have built the DMTD you need to measure its noise floor.

How do you ensure that the limiters actually achieve a jitter better
than 10ns?
With a < 10Hz beat frequency this is actually quite difficult to do
given, typical mixer and amplifier noise.
Low frequency ground loop noise can be a major problem with low
frequency beat signals.

Some limiting factors for long tau:

1. Mixer phase shift tempco (can be as large as 10ps/C)

2. Limiter phase shift tempco (principally determined by phase shift
   tempco of first stage filter).

3. phase shifter tempco
   If you use coax  the tempco for 100ns delay may be as large as 10ps/C.

4. Delay tempco of RF isolation amplifiers required to prevent injection
   locking.

Another factor not often considered with DMTD systems is the effect of
phase noise aliasing.
The limiter bandwidth of a typical DMTD necessarily exceeds the Nyquist
limit.

RF shielding between the 2 DMTD channels to avoid crosstalk and
injection locking is important.

Bruce

Bob Camp wrote:

Remember that you must measure the actual beat frequency, since you
will need that to calculate the beat-gain. If it is between 5 and 10 Hz
the for a 10 MHz source your gain is 2E6 and 1E6 respectively, which is
a factor of 2 difference or 6 dB. So, your measurements will be
inprecise from that factor alone by +/- 3 dB. The remedy is fairly easy
to come up with, measure the input frequency and beat frequency for each
arm. The best thing is naturally to ensure that the beat frequencies of
both arms is fairly close. EFC steering of either source may work well
enought in open-loop mode during measurement (with the added benefit of
not do spectral interference with the phase noise which locked loop does).

How do you control the input levels to the mixers?

Do you have any isolational amplifiers?

How do you load and pre-filter the mixer outputs?

You haven't convinced me of the expected performance of the limiters.

I'm not sure it will be your biggest problem, but the way you
phase-shift can be of importance for the decorrelation loss.
Phase-shifting such that group-delay moves noise in time will be
problematic, since then the decorrelation gain of having phases coincide
will be partly lost since it is the group-delayed variant of the
transfer oscillator against the current-time transfer-oscillator (both
delayed by each detector arm, but only differnces is important).
Vector-adding phase delays could work around that. The optimum delay
setting for cancelation may not be to fully phase-adjust the leading edge.

That is what just popped up in my head at least.

Cheers,
Magnus

Hi

I agree that the limiters may not actually be "20 ns good". It's certainly well worth checking. I'm also thinking that there may be a compromise in the limiter chain to reduce the phase noise issue. What I'm trying to do here is come up with a "noise floor" number that's adequate before I start.

The isolation amp, mixer, and  phase shifter all operate at RF, so a picosecond change does indeed get me a 1x10^-12 at the output. The limiter operates at audio, so I've already got the down conversion so a picosecond there is less of an issue. Of course the first limiter is going to be a whole lot more time unstable than the RF stuff.

What I'm getting around to is that the counter really does not have to be a SR-620, or even a 5335 to do the job. The problems lie elsewhere. A time tagging FPGA with a 100 MHz clock would do the counting job quite nicely. It also would not be terribly hard to build. I'll grant a 10 or 100 ps/C delay variation with such a gizmo, but it runs at audio, so it's after the 1x10^6 downconversion gain. Running time tags also takes care of issues like measuring the actual beat note frequency.

Here's my guess for temperature stability of the setup;

1-10 seconds < 0.1 C

That would give me:

Mixer : 1ps to 10 sec, 2 ps at 1000 sec.
Phase shifter: half the mixer if I use a switch on the transformer for inversion.
Isolation amps: something to look at

That would give me a limit from mixer and phase shifter of:

1.4x10^-12 at 1 sec
1.4x10^-13 at 10 sec
2.8x10^-14 at 100 sec
better than 1x10^-14 at 1000 sec and beyond

I suppose that if those numbers were 10X worse than that once the rest of it shows up, I would stabilize the temperature of the setup.

So what's still missing?

Bob

On Jan 24, 2010, at 8:43 PM, Bruce Griffiths wrote:

Hi

More or less in order:

The beat frequency is coming out of a rubidium. Hopefully it's fairly stable. It won't be super quiet for 1 or .1 second tau. It looks like the counter will be a FPGA time tagger, so the beat note frequency will drop out for free.

The isolation amps are common base buffers. Not much gain, but quite a bit of isolation. Phase shift / C - need to look into that.

Mixer loading likely would be as I've done it before. Resistive termination at RF and fairly high impedance at audio. Resistor here and there to improve the match at RF. LC filtering adequate to suppress the RF stuff on the output of the mixer.  Single pole R-C for audio bandwidth control. Big capacitors and small resistors for low noise.

Until I've measured them I'm not sure of the floor of the limiters. Before I get into that I want to be fairly sure I'm not over spec'ing them. If 100 ns is as good as 3 ns it's not as hard a problem.

The issue of the group delay is an interesting one. I believe that people have been getting good results with coax line for the phase shift. I'm a bit conflicted on the  coax. 15 meters of small diameter stuff will fit in the box (maybe), but it's not super stable.. If I go foam coax then the phase shifter gets pretty big. If I go with some kind of LC setup, temperature stability would likely be an issue.

Crazy Stuff ....

So what did I miss that time?

Bob

On Jan 24, 2010, at 9:01 PM, Magnus Danielson wrote:

Bob

Whilst one can use the limiter design spreadsheets perfected by Roberto:
http://www.ko4bb.com/~bruce/ZeroCrossingDetectors.html
http://www.ko4bb.com/%7Ebruce/ZeroCrossingDetectors.html
(bottom of page)

One needs to know the mixer phase noise spectrum in the vicinity of the
beat frequency (not specified by most manufacturers) and its gain in
order to design an "optimised" limiter.

This can be done using a low noise preamp together with a sound card and
suitable spectrum analyser software.
The noise and gain need to be measured with the proposed mixer IF port
termination.
Using a capacitive IF port termination can significantly reduce the
mixer output phase noise.

Using a mixer that allows the IF port ground to be isolated at low
frequencies from the RF and LO grounds is useful in minimising the
effects of ground loops.

Mixers characterised for use as phase detectors tend to have lower drift
than other mixers.

Your estimate for phase shift tempco of the limiter and associated
filters is wildly optimistic.
If one uses a single pole RC low pass filter using 100ppm/C tempco
components then its equivalent phase shift tempco (at the mixer RF
input) will be around 0.8ps/C if the filter 3dB is equal to the beat
frequency.
If one increases the filter cutoff frequency its output noise increases
as its equivalent phase shift tempco decreases. Finding capacitors with
lower tempcos will be problematic given that the capacitance must be
relatively high to keep the noise from the series resistor down.

Ensuring that the the temperatures of the limiter input filters track
will reduce the differential tempco.

Low frequency isolation of the counter drivers grounds may be necessary
to avoid crosstalk from the large currents used to drive the counter inputs.
This may require using a digital isolator of some kind.
The jitter of the isolator may be significant.
Suitable low jitter optical isolators are available.
The delay tempco of such an isolator may also be an issue.

The phase shift tempco of a well designed RF isolation amp should be
less than 2ps/C.
SpectraDynamics claim about 1.5ps/C.

Bruce

Bob Camp wrote:

Bob Camp wrote:

A cleanup loop may be useful to improve the offset source short term
stability.
The cancellation of offset oscillator noise in a DMTD is imperfect.

You can achieve similar isolation together with lower noise and
distortion with a transformer feedback CE stage.
Transformer feedback CB stages have even lower noise coupled with low
isolation, however they can be useful for amplifying low level signals
ahead of a high isolation amplifier.

That's one of the worst terminations possible from the noise perspective.
To lower the noise its essential to reflect the sum frequency back into
the mixer.
Resistors in series with the mixer LO and RF inputs will then be
required to improve the mixer input VSWR.

You can take the published phase noise for unspecified mixers as a lower
limit.
The noise in the flicker region for the mixers (eg those from
minicircuits) that use integrated diode quads may be somewhat higher.
Initial measurements on a HP10534B (uses discrete diodes) appear
consistent with the typical noise specs for a low level mixer.

NIST's measurements indicate that lowest delay tempco is achieved with a
powdered silica dielectric.
Specialised fibres can have very low delay tempcos.

Bruce

Hi

Again more or less in order:

I'm trying to keep things as simple as I can at least to start. That rules out the clean up loop oscillator at least in the beginning. It is a good idea, and eventually I'll probably put one in.

I guess I'm going to need to do some looking on transformer feedback high isolation amps. Everything I've seen so far on hight isolation has been straight / no feedback stuff.

The loading at RF on the mixer does reduce the audio output, but it improves the isolation / match on the mixer. You trade one for the other.

Looks like some kind of local temperature stabilization might be a good idea for the audio band limiting stuff. It's after the down convert, but some of the time constants are indeed very long.

I suspect that silica dielectric cable is outside the budget constraints on this project. Cheap foam coax in a spool on the floor or tiny stuff in the box, possibly with better temperature control are about the only two choices.

Another very real choice is to simply move the goal post a bit. Pushing the 1x10*-12 point to 10 seconds from 1 second could turn out to be the only economical basement alternative.

Off to bed ....

Bob

On Jan 24, 2010, at 10:26 PM, Bruce Griffiths wrote:

Bob Camp wrote:

John Miles did some phase noise measurements
(http://www.thegleam.com/ke5fx/norton.htm) on designs like those at:

http://www.ko4bb.com/~bruce/CE_TransformerFeedback_BufferAmplifier.html
http://www.ko4bb.com/%7Ebruce/CE_TransformerFeedback_BufferAmplifier.html

John and I actually used 2N5109s and 2N5943's respectively.
I have LTSpice simulation files for some of these designs (however I
don't have a spice model for either transistor).
If you need more isolation just cascade a few such stages.
Simulation indicates that such stages can easily produce an output of
+23dBm or more should you need it.
The reverse isolation of a single stage is about 40dB which is easily
measured with a scope.
The collector current of such a CE stage is significantly lower than
that of a CB stage with similar output and distortion.

The input distortion of a CB stage limits the distortion performance
unless one augments the circuit with another transistor or a transformer.
Then collector output capacitance modulation limits the distortion
performance especially with high collector impedances.

The matching requirement is a red herring (there are various HP/Agilent
and Watkins-Johnson application notes on reducing mixer noise and loss
by reflecting all the unwanted mixer products back into the mixer). NIST
also did some work on the advantages of a capacitive mixer IF port
termination.
Resistors in series with the input will largely fix the matching and
locating the output stage of the isolation amplifier close to the mixer
also helps.
One way of reducing the mixer phase shift tempco (NIST claim a factor of
about 10) is to use it with the RF port unsaturated, however this
increases the noise.
The noise disadvantage can be offset by using a high level mixer.
Choose a mixer with high isolation as this usually indicates good diode
match and low transformer imbalance.
Usually DMTDs saturate both the IF and RF mixer ports.

One can easily obtain 0.22uF NPO/C0G caps so one could parallel a few of
these and use low tempco resistors.

Yes, NIST found it was outside their budget as well.

Bruce

Hi

I will be running with both of the input ports on the mixer saturated. In that case, it's not clear that I'm worried about harmonic distortion in the isolation amplifiers. I get plenty of that when I saturate the mixer ports. There are indeed other kinds of distortion in an amplifier and they may be an issue.

The mixers will be RPD-1's or the SMT equivalent. They have done sub 1x10^-12 under the same conditions in the past. They are cheap and easy to get ahold of.

There is a pretty old model for the 5109, I don't know of one for the 5943. Last time I used the 5109 model it was not very useful for sub 200 MHz work. I wound up tossing one on a network analyzer to get some better data.

I've used 0.1 uf NPO's for some stuff in the past. They are out there. They aren't cheap. Of course they may be cheaper than ovenizing an active filter that uses four inch long plastic capacitors.

To get back to the original question - what's good enough? The objective is not  state of the art. I'm quite happy with 1x10^-12 at 1 second. I'll probably be willing to accept that at 10 seconds if that's what makes more economic sense.  If I hit anything at or below 1x10^-13 it will be out beyond 1,000 seconds. I'm guessing that anything that does the 1 or 10 second number will also be quite adequate from there on out.

So many things to consider ....

Bob

On Jan 24, 2010, at 11:30 PM, Bruce Griffiths wrote: