[time-nuts] Sub Pico Second Phase logger

[Joseph M Gwinn]

Bruce,


time-nuts-bounces at febo.com wrote on 12/09/2008 06:24:22 PM:

> Joseph M Gwinn wrote:
> > Bruce
> >
> >
> > time-nuts-bounces at febo.com wrote on 12/08/2008 07:12:22 PM:
> >
> > 
[snip]
> >
> > 
> >>> Many mixers have IF response down to DC.
> >>>
> >>> 
> >> Oops, I meant "single diode type double balanced mixer style phase
> >> detector".
> >> 
> >
> > Ah.  Four single diodes in a ratrace ring.  Max drive +13 dBm or so. 
> > Called Class I or Type I.
> >
> > MiniCircuits ZRPD-1 being one example.
> >
> > By the way, despite the circuit diagram in the datasheet, the 
> > corresponding phase-detector module MPD-1 can be wired to have the IF 
> > output ground isolated from the common RF, LO and case ground.  A 
> > little work with an ohmmeter will tell the tale.  This can help to 
contain the 
> > low frequency beatnote.
> 
> Yes, that's usually the case for the Minicircuits PCB mount phase
> detectors and mixers except for some surface mount versions (usually the
> very high frequency models).

So it was already known.  It looks to me that MiniCircuit's intent is to 
support automated testing of modules.


> A PCB mount mixer package is also preferable as its then much easier to
> use a capacitive IF port termination (for lower noise) in conjunction
> with series resistors at the RF and LO ports (for lower VSWR) than if a
> mixer with SMA or other coax connectors were used.

I've been using 3 and 8 dB coaxial attenuators at the LO and RF inputs 
respectively, and it makes a big difference.

But I don't understand the part about capacitive loading of the IF port. I 
would think that the low pass filter would need to present a matched 
impedance at the sum frequency, so the emerging high-level 20 MHz signal 
is not reflected back into the mixer.

MiniCircuits AN-41-001 "FAQ about Phase Detectors" has on page 2 a 500 ohm 
resistor to ground and a 5000 ohm resistor to the first filter capacitor, 
so the capacitor is isolated from the IF port by the resistors.

I just got your posting about paper 112, so more later.

 
[snip]
> >>>> 
> >>> I've read many or most of the classical DMTD papers, and have seen 
> >>> various passing estimates that diode-ring mixers have a temperature 
> >>> sensitivity of 8 to 10 pS per degree C.  (I recall your figure was 
10 pS/K.) 
> >>> I assume that the DC offset also varies with temperature and drive 
signal 
> >>> amplitude.
> >>>
> >>> 
> >> The only reference I have on the offset tempco is a miniciruits
> >> application note from which one can deduce that the equivalent phase
> >> shift tempco associated with the offset tempco is a few hundred
> >> femtosec/C (@ 10MHz +7dBm) at some temperatures for the particular 
mixer
> >> used. The graph also indicated (if you are lucky) that the offset 
tempco
> >> may be zero at around 20C.
> >> 
> >
> > Do you recall the part number?
> >
> >
> > 
> Supposedly an SRA-1, but some caution is in order as some statements as
> to the effect of the input offset of an opamp based IF preamp in the
> same application note were of dubious veracity unless one were to use an
> inverting opamp input stage.

This issue was mentioned in another app note, but their main issue 
appeared to be that the opamp bias currents could cause an offset.


> >> A NIST paper indicated that mixer phase shift tempco was around 10x
> >> lower if the RF port was unsaturated. It also indicated that the 
mixer
> >> phase shift tempco is much lower if the input frequency is 100MHz 
rather
> >> than 10MHz. This was one reason given for shifting to 100MHz 
> >> DMTD systems.
> >> 
> >
> > Do you recall which paper?
> >
> >
> > 
> http://tf.nist.gov/timefreq/general/pdf/971.pdf
> <http://tf.nist.gov/timefreq/general/pdf/971.pdf>
> Has some measurement data on mixer phase shift tempco and power
> sensitivity and their frequency dependence etc.

I do know this paper.  At the bottom of page 834, to the right, is the 
estimate 3.5 pS/K.

Another reason to go to 100 MHz is that the temperature coefficient of 
electrical length of polymer-insulated coax is far lower at 100 MHz 
compared to 10 MHz.


> I'll search for the paper that stated that the phase shift tempco was
> lower if the RF port was unsaturated.

I think I have seen this too, but don't recall where.  But it's why I use 
an 8 dB pad on the RF input.


> AFAIK there was no accompanying measurement data
> 
> > What I've seen that seems useful is the Watkins-Johnson application 
note 
> > from 1978 on use of mixers as phase detectors: "Mixers as Phase 
> > Detectors", Stephan R. Kurtz, 8 pages.  This may be the source of the 
NIST 
> > article's information.  The electrons are available on the web from WJ 

> > Communications (now owned by TriQuint), filename 
<
http://www.wj.com/archive/documents/Tech_Notes_Archived/Mixers_phase_detectors.pdf
>. 
> > Don't know how long this URL will work, as WJ is assimilated into 
> > TriQuint.

This app note is reference 7 of paper 971 above.


[Soundcards]
> >>> 
> >> Preliminary (non optimum) tests by Ulrich indicate that picosecond
> >> stability for times up to 100sec is very easy to achieve.
> >> Beyond that mixer phase shift tempco mismatch may be significant.
> >> 
> >
> > It would not be that hard to make an oven for the mixer, as 
> the level of 
> > control needed is far less stringent than for a crystal.
> >
> >
> > 
> >> ADEV noise level of around 2E-14/Tau (1s < tau <100s).
> >> Haven't yet seen [or] have data for longer tau.
> >> 
> >
> > Yes.  Need at least 10^4 seconds.
> >
> >
> > 
> >> With identical beat frequency outputs, crosstalk between 
> >> channels within the sound card shouldn't be a great problem.
> >> 
> >
> > I'm not sure I believe this, as there is likely ground coupling within 
the 
> > soundcard and the ear is famously insensitive to phase.  Channel 
isolation 
> > of 60 dB isn't enough to prevent phase shifts.
> >
> > 
> It will be present but its effect in some cases (when the phase shift
> between channels is such that the crosstalk phase is at 90 degrees to
> the signal of interest) will be negligible, in other cases it is easily
> measured and compensated for.

Isn't 90 degrees (quadrature) the worst case for causing phase shifts? 

To get a one picosecond change at 10 MHz by injection of an attenuated 
quadrature copy of the main signal requires a relative voltage ratio of 
Tan[(10^-12)(10^7)(360)] = Tan[0.0036] = 0.0000628 of the main signal, or 
20 Log[0.0000628]= -84 dBc.  This is well exceeds the interchannel 
isolation of many sound cards.

Cancellation by mathematical means could be possible, but will require a 
dynamic range well exceeding 84 dB.  This ought to be easy to arrange.


> >> One concern particularly for low beat frequencies is the phase shift 
in
> >> the sound card input coupling capacitors (usually electrolytics).
> >>
> >> It should be easy to test the sound card phase shift stability for 
this
> >> application by driving both inputs from the same signal source.

Or terminate one channel input and drive the other, and measure the 
amplitude and phase of whatever comes out of the terminated channel, 
compared to the driven channel.  Then swap channels and repeat.  The phase 
and amplitude will depend on frequency, so a sweep will be required, and 
some frequencies may need to be avoided.


> > I assume that the beatnote must be ~100 Hz for the soundcard to handle 

> > with low phase shift.  One might get to 10 Hz, but 1 Hz is likely 
> > hopeless.
> >
> > One thing that will be very useful is a list of sound cards by make 
and 
> > model, annotated with their advantages and disadvantages for time-nut 
use.
> >
> > "High-end" may not be a sufficient description.
> >
> >
> > By the way, I looked at the operating and service manual for the HP 
> > K34-59991A Broadband Linear Phase Comparator.  Very interesting little 

> > gadget, but little performance data was given.  Does anyone know how 
well 
> > phase change can be measured?  It would be easy to duplicate this with 

> > modern ICs.  Also, about when was this unit made?  The manual has no 
date.
> >
> >
> > Joe
> >
> > 
> Joe
> 
> I suspect that slow phase changes much less than 1ns or so are hard to
> distinguish from gain drift given the gain tempco of the ECL 
> phase detector.
> 
> A beat note near 1kHz appears to be even better if one is using
> something like an enhanced Costas receiver or even using WKS
> interpolation to locate and time stamp zero crossings.

But it limits the phase slope gain.  I suppose there is an optimum 
somewhere.

 
> So far only the M-Audio AP192 has been used.
> Tests with an embedded motherboard 16 bit sound system show
> significantly increased noise.
> I've found that the noise level of motherboard sound systems varies
> enormously from one motherboard model (sample of 2) to another.
> 
> Any 24 bit sound card with a performance close to or better than that of
> the AP192 should suffice.

In general, firewire connected sound cards should be better, because the 
soundcard maker has complete control of what's inside the box.  Unlike 
inside a PC. 


> Other cards using AKM 24 bit ADCs should also be suitable.

Who is AKM?

20 Log[ 2^24 ] = 144 dB, so something else will be the limit.


> Ideally an external sound card with balanced  XLR inputs would be best.
> 
> HP produced a number of different phase comparators each with a
> different type of phase detector.
> The K34-5991A design can't be older than the early 1970's because the
> MECLIII devices used weren't available until then.

OK.  I recall MECL.  RIP.  But we have PECL now.


> Warren built a similar phase detector (differential XOR or XOR + XNOR)
> using CMOS ICs and for a common 10MHz input with a phase difference near
> zero found short term output noise of of around 10uV or so (10V phase
> detector FSR) using a passive low pass filter.

(10uV)/(10v)= 1 ppm.  (100 nS)(10^-6)= 0.1 pS.


> In principle an ADC like the LTC2484 could be used with a 2.5V CMOS
> OR/XNOR phase detector and passive low pass filters.
> The ratiometric conversion capability will significantly reduce the
> sensitivity to the XOR gate supply if the XOR gate supply is also used
> for the ADC reference voltage.
> If one used 5V logic a resistive output attenuator would be needed which
> reduce the gain stability somewhat.
> 
> All such phase detectors suffer from substantial nonlinearity near the
> ends of the range due to gate output slew rate limitations.

If one is tracking through multiple phase cycles (as did the HP unit), 
this would matter.

As for slew rate, one can buy faster logic, but probably at the expense of 
jitter (as bandwidth must be larger).

The HP unit used balanced logic all the way up to the flipflops.  This 
doubles the relative crossing rate for a given slew rate.  But doubles the 
noise, but it must have been worth it, or HP wouldn't have done that. 

In beatnote zero crossing detectors, differential signals help keep ground 
loops and bounce down.


> However if operated near the centre of the range subpicosecond
> sensitivity/short term stability may be possible.

Yes, 0.1 pS.


> Since the circuit volume is small and the tempco relatively low (few
> ps/C at most) regulating the circuit temperature should be relatively
> easy to do.

We ought to be able to achieve less than 0.1 K change.



Joe