[time-nuts] Sub Pico Second Phase logger

[Joseph M Gwinn]

Bruce,


time-nuts-bounces at febo.com wrote on 12/11/2008 07:44:01 PM:

> Joe
> 
> Isolation from mixer RF to LO port may be too low when the mixer input
> frequencies are different.
> Injection locking can then occur all too easily (just ask Ulrich about
> this) when the mixer RF ports are driven by 2 separate OCXOs.
> 
> Detailed in line post follows:
> 
> Bruce
> 
> Joseph M Gwinn wrote:
> > Bruce,
> >
> >
> > time-nuts-bounces at febo.com wrote on 12/10/2008 08:38:13 PM:
> >
> > 
> >> Joe
> >> Joseph M Gwinn wrote:
> >> 
> >>> Bruce,
> >>>
> >>>
> >>> 
> >>>> Reflecting the sum frequency back into the mixer is actually 
necessary
> >>>> to reduce the noise at the IF port.
> >>>> I believe that one of Agilent's simulation application notes 
mentions
> >>>> this effect but I don't recall the actual application note number.
> >>>> This will affect the mixer RF and IF port impedance so adding a 
> >>>> series resistor may be required to improve the SWR.
> >>>>
> >>>> 
> >>> How big an effect is this?  Is the absolute noise decreased, or does 

> >>> it remain the same while the signal increase?
> >>>
> >>>
> >>> 
> >> With the same difference frequency IF port termination impedance, 
noise
> >> is actually decreased along with the mixer conversion loss.
> >> 
> >
> > OK.  Complicated beasts, those mixers.  Do you know of a paper (or 
book) 
> > on the subject?
> >
> > 
> 
> 
> Not offhand, but this crops up in lots of places usually when one least
> expects it..

I've noticed.  Someone has to have poured his soul into a monograph.


> >> However if the sound card input noise dominates, reducing the mixer
> >> effective output noise won't help.
> >> 
> >
> > Yes.  In the plots you posted in a different email, there was a big 
rise 
> > below 1 KHz (scan stopped at 1 KHz, so don't know the shape).  Why is 
> > this?
> >
> I'll expand the frequency scale and take another snapshot for the region
> below 1kHz.
> This rise may be due to ADC and/or input differential amplifier flicker
> noise.

Saw it.  Thanks.  Does look like flicker noise.  Although it wasn't large 
enough to be a real problem it seems.

 

> > 
> >> If one is time stamping the zero crossings an increased zero-crossing 

> >> slope is an advantage. 
> >> For relative phase measurements a trapezoidal beat frequencywaveform 
> >> may be less useful.
> >
> > Fitting to the approximate waveshape, sine or trapezoidal, 
> > should yield a very robust estimate, due to the large data support, 
and zero-crossing 
> > slope won't much matter.  Hmm.  Actually, if the slopes of the 
trapezoid 
> > are too steep, we may not have all that many slope samples.
> >
> >
> > 
> 
> If one believes the NIST papers the trapezoid zero crossing slope only
> increases by a factor of 3.
> If one uses a cascaded filter limiter the slope gain can be adjusted for
> optimum results.

The implication would be to not sharpen things up too well in before 
digitizing.


> > [snip]
> > 
> >> Of course with a capacitive IF port termination, matching the RF and 
LO
> >> ports becomes more critical as does the reverse isolation of the 
various
> >> amplifiers driving the RF and LO ports.
> >> It may be simpler in fact to use a level 17 mixer with high LO to RF 
and
> >> LO to IF isolation with the RF port unsaturated as it relaxes the
> >> reverse isolation specs for the isolation amplifiers.
> >> 
> >
> > Another tradeoff.  I'll have to think about it.
> >
> > I'm thinking of 6 db and 10 db attenuators on the LO and RF ports 
> > respectively, but no isolation amplifier.
> 
> You may get away with that if you use mixers with very high RF to LO
> port isolation.
> Minicircuits have at least 3 level 17 mixer models that typically have
> 80dB LO to RF isolation at 10MHz.

I've used the ZRPD-1, which claims about 75 dB isolation at 10 MHz.

 
> Using a passive splitter for the LO drives will gain at least another
> 30dB in isolation between the 2 RF inputs if you use an appropriate
> splitter.

True. 

I may have lost the thread here.  If we have one oscillator driving 
everything, one cannot have injection locking even if isolation isn't 
perfect.  What isolation does gain us is a reduction in undesired phase 
shifta.


 
> > [snip]
> > 
> >>> 
> >> The only configuration for which it makes any sense is an inverting
> >> input amplifier with a finite input voltage offset.
> >> 
> >
> > Why would non-inverting not work?  Both inputs source or sink bias 
> > currents, and non-inverting presents a very high impedance.
> 
> Non inverting amplifiers usually have lower noise and generally work 
very well.
> I was only trying to come up with a preamp circuit for which the
> comments in the Minicircuits application note on the effect of amplifier
> input offset voltage made any sense.

Ah.  It may be hopeless.

My reading was that they were worried about bias currents from the amp 
flowing into the mixer and causing offsets, not amplifier offset voltages 
per se.  The amplifier offset voltage does not cause a mixer offset, and 
may be reduced by use of a chopper amp or very good balance.


> The only risk with a noninverting amplifier, is that under fault
> conditions (missing supply) a very large current can flow back (with
> some low noise opamps as Enrico has experienced) into the mixers and
> destroy them.

Yes, I recall the discussion on your website.

 
> For this particular application the mixer preamp gain need only be
> sufficient to boost the mixer phase detector output (1V pk?, 350mV pk??
> depends on mixer and its operating conditions) to the sound card input
> (FSR ~ 5.6V pk for an AP192). The resultant preamp gain is relatively
> low ( 5 - 15X depending on the mixer etc) and the sound card noise will
> dominate (~ 100nV/rtHz midband for an AP192) thus using an ultra low
> noise mixer preamp isn't necessary.

Yes.



> > We are converging on a soundcard wishlist:
> >
> > 1.  True balanced inputs on XLR connectors.  And good ground design, 
so we 
> > aren't bedeviled by ground loops.
> >
> > 2.  24-bit ADCs, and similar DACs.
> >
> > 3.  Very good isolation all around. 
> >
> > 4.  Digital access via firewire (or USB3 I suppose), with the 
soundcard in 
> > its own box.
> >
> > 5.  High-level input direct to the ADCs.
> >
> >
> > While use of AKM ICs may be a very good idea, it is not a requirement 
per 
> > se.
> >
> > [snip]
> > 
> 
> 
> Optical isolation of the ADC from the noisy digital interface to the PC
> would also be nice.

Good point.  Part of ground design I suppose.   Although with noise floor 
of -150 dB down there cannot be so much leakage.

 
> If we design our own PCB then the AD7760 series ADCs are another
> possible option.
> These have a built in differential input differential output amplifier.

Yes.  But aren't we trying to use commonly available soundcards?

 
> 
> >>>> Can alleviate [oddities at end of phase range} to some extent by 
> >>>> driving a pair of such phase detectors so that their outputs are in 

> >>>> quadrature.
> > 
> >>>> One just selects the phase detector output that is in the linear 
> >>>> range.
> > 
> >>>> The quadrature outputs also allow unambiguous assignment of the 
sign 
> >>>> of any phase change.
> >>>>
> >>>> 
> >>> The Symmetricom 5120A does something very clever to alleviate this 
> >>> problem.  Explained in US patent 7,227,346 and "Direct-Digital 
> >>> Phase-Noise Measurement"; J. Grove, J. Hein, J. Retta, P. Schweiger, 
W.Solbrig, 
> >>> and S.R. Stein; 2004 IEEE International Ultrasonics, Ferroelectrics, 
and 
> >>> Frequency Control Joint 50th Anniversary Conference, pages 287-291.
> >>>
> >>> Joe
> >>>
> >>>
> >>> 
> >> I've read the patent.
> >> 
> >
> > The paper is also worthwhile, and available on the web somewhere 
(don't 
> > recall where, but google found the pdf).  I had to read the patent 
> > multiple times to figure out what's going on.  The correlation 
approach is 
> > old as the hills, and only the digital phase detector was patentable.
> >
> 
> It may be feasible to achieve the same effect by purely digital means at
> least for low sample rates where FIR filters with tens of thousands of
> taps are feasible.

It *is* feasible, and Sam Stein is doing it.  I've perhaps lost the thread 
here.


> Of course 64 bit or higher precision arithmetic is then mandatory to
> avoid excessive calculation roundoff noise.

It may be 64 bit *integer*, actually.

My understanding is that the 5120A is built upon a DSP or more likely FPGA 
(of unspecified make and model).  The 5125A will have a top frequency of 
400 MHz, so the DSP and/or FPGA better be damn fast.  Little analog stuff 
remains.

Joe