[hpsdr] Sampling Phase Detectors

[John Miles]

> For simple synthesisers the PFD represented a huge step forward
> when it was
> first introduced, however, it came at a cost. The very close-in
> phase noise
> is degraded by uncertainties introduced by the detector, and
> shows as greater
> than expected spurious FM. That's why many IC implementations of
> the PFD add
> some form of dead-band reduction circuitry. Most make some
> improvement, but
> few, if any completely solve the problem. In the old days, we
> simply hung a
> resistor from the PFD output to ground in order to improve the
> spurious FM,
> and fought the reference frequency sidebands! The levels of spurious FM
> generated by the PFD wouldn't matter in a 'normal' synthesiser,
> but if we're
> trying to go for the best...

Yes, I remember having to do the 1M-resistor trick at the output of the
4046's PFD section, to keep it out of its "dead band."  PFDs have, however,
come a LONG way since the 4046 was designed.  Modern parts from Analog and
National are not subject to dead-band behavior in the least.

I'll confess that I have never tried an XOR-gate phase detector.  I've made
extensive measurements of sampling-loop PLLs in HP gear, though, including
the 8753 VNA and 8566 spectrum analyzer.  The 8753 is limited to about -100
dBc/Hz at 2.5 GHz due to noise added by its SRD multiplier driver, while the
8566 ends up at about the same level due to noise in the ECL M/N synthesizer
that drives the sampler.  However, as I recall, the sampler in the 8566 is
limited to around -112 dBc/Hz, presumably due to conversion loss.

Both of those figures are easy to beat with current ADI/NS/Hittite PFD
chips, even at much-higher N factors.

> In practice, the folding effects you posit don't seem to be that
> significant,
> presumably because in a harmonic locking system everything is
> synchronous.
> Also, with a decent VCO, the sampler should only be seeing a single
> sinusoidal signal...

Its input port is not noiseless, though.  Samplers are inherently wideband
devices, and there is a lot of kTB noise in a spectrum several GHz wide,
even with the VCO turned off.  My understanding is that the kTB noise ends
up folded down to baseband around each LO comb tooth.

> For the 1GHz case which we've been
> discussing over the
> last few days, I've looked at the use of packaged commercial DBMs as a
> sampler. Within their frequency range they work well, as might be
> expected,
> on odd-order harmonics, and seem to essentially be operating as samplers
> rather than harmonic mixers. I need to understand that better.

Yes, using DBMs as samplers is a really-interesting topic in itself.  There
are some rather-intriguing implications for homebrew VNA design if nothing
else.

> Don't forget that a sampler will have a sensitivity which (at least in
> theory!) is independent of harmonic number. Harmonic mixers are a
> different
> beast...
> A major problem is generating a suitable pulse.

I'm not sure I'm comfortable with the notion of sensitivity as being
relevant to the residual PN floor of a sampler-based PLL, but I'd have to
yield the point since I don't have a good argument either way.  All I know
is that sampler-based PLLs are relatively-difficult to design and optimize,
and that the effort doesn't seem to bring proportionate rewards in the PN
department.

> The use of sampling phase detectors for locking microwave
> oscillators has been
> standard practice for decades.

Largely because there hasn't been an economical alternative, I suppose.

> FWIW my 24GHz system relies on a
> commercial
> 'brick' which uses a sampler to lock a 12GHz VCO to a VHF oscillator with
> remarkably good results. I agree that careful design is needed,
> but if that
> is accepted, the sampling detector removes the noise
> contributions of the PFD
> and counter,

No; you pay the 20*log(N) penalty regardless of whether you use a sampler or
a counter/PFD.  A key difference, again returning to the example of the HP
8753A, is that the sampler LO is probably noisier due to the large amount of
gain needed to drive the multiplier element.

> and replaces them with potentially significantly smaller
> contributions of its own. The free lunch principle still applies!

Sad but true. :(

> In this
> case we have to trade some circuit complexity for performance.
>
> > Would be good to hear any pointers to the contrary...?
>
> I hope I've provided some indications in this very late night email.

Yes; certainly good food for thought.

> This
> isn't a straightforward problem, but I'm convinced that the
> harmonic sampler
> is the right way to go. Either that, or back to multiplier chains...

I'd definitely encourage you to spend some time with the newer Analog
Devices and Hittite parts before giving up on PFDs. They aren't perfect, but
they aren't 4046s either.

-- john, KE5FX