Poul-Henning,

On 01/05/2016 10:37 PM, Poul-Henning Kamp wrote:

You still raise the noise-level which is in the pass-band of those
filters. This will be true both for white noise and flicker noise.
The white noise will be particularly annoying as it then converts to
jitter through the slew-rate limitation as you go into the
trigger-circuit. To reduce this effect, we amplify up the signal in
steps, with higher and higher bandwidth to balance noise contribution
with slew-rate incrementation. Using noisy mixers rather than quieter
mixers makes this more worthwhile. The diode mixers needed does not have
to be very rare, just look at the 2N2222A based mixer out of NIST,
actually being a Harris chip with four transistors and a pair of off the
shelf transformers.

Yes, I've played with this, ran into the issues.
Tried to build a DTMD, but didn't manage to handle some problems before
it got side-tracked.

As always, choosing the trigg-point to have the highest slew-rate have
always been key to reducing timing jitter.k Turns out that most counters
isn't optimized for this property.

Cheers,
Magnus

Poul-Henning wrote:

You could have mentioned any of dozens of popular analog multipliers,
and the answer would have been, "because they are way too
noisy."  The AD835 is also substantially noisier than diode mixers,
but it at least begins to bridge the gap.  The folks at CERN have
been improving phase detector S/N by averaging the output of several
AD835s for the TPMON project, with promising results.  There is a
preliminary report in "EUROTeV Report 2006-005-1."

See also:

RF-based electron beam timing measurement with sub-10fs resolution,
A. Andersson and J. P. H. Sladen, CERN (EUROTeV Report 2008-015)
[phase detector with 8x AD835 analog multipliers].

ANDERSSON, A. and SLADEN, J. P. H.: "First tests of a precision beam
phase measurement system in CTF3" (Proc. PAC07).

"PRECISION BEAM TIMING MEASUREMENT SYSTEM FOR CLIC SYNCHRONIZATION,"
A. Andersson, J. P. H. Sladen, CERN (Proceedings of EPAC 2006).

Best regards,

Charles

You mean DMTD =  dual mixer time differencenotDDMTD = Digital dual mixer timer difference.The latter uses a pair of synchronisers / shift registers instead of a pair of mixers.
Bruce

On Wednesday, 6 January 2016 12:03 PM, Charles Steinmetz <csteinmetz@yandex.com> wrote:

Poul-Henning wrote:

You could have mentioned any of dozens of popular analog multipliers,
and the answer would have been, "because they are way too
noisy."  The AD835 is also substantially noisier than diode mixers,
but it at least begins to bridge the gap.  The folks at CERN have
been improving phase detector S/N by averaging the output of several
AD835s for the TPMON project, with promising results.  There is a
preliminary report in "EUROTeV Report 2006-005-1."

See also:

RF-based electron beam timing measurement with sub-10fs resolution,
A. Andersson and J. P. H. Sladen, CERN (EUROTeV Report 2008-015)
[phase detector with 8x AD835 analog multipliers].

ANDERSSON, A. and SLADEN, J. P. H.: "First tests of a precision beam
phase measurement system in CTF3" (Proc. PAC07).

"PRECISION BEAM TIMING MEASUREMENT SYSTEM FOR CLIC SYNCHRONIZATION,"
A. Andersson, J. P. H. Sladen, CERN (Proceedings of EPAC 2006).

Best regards,

Charles

time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

In message 568C46B9.4020303@rubidium.dyndns.org, Magnus Danielson writes:

Digitize the LPF output and do a curve-fit to find the zero-crossings ?

--
Poul-Henning Kamp      | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG        | TCP/IP since RFC 956
FreeBSD committer      | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

Hi Poul-Henning,

On 01/06/2016 12:28 AM, Poul-Henning Kamp wrote:

That would work. You could least-square fit it with very cheap
processing. The LPF would mainly need to reject the sum frequencies to
act as anti-aliasing filter, and the noise would be filtered out by the
least-square processing.

Estimating the phase and slew-rate, and then use those to calculate the
actual through-zero phase would not be too hard. As a consequence you
get a slew-rate monitor, which act as an observation of signal level.

Cheers,
Magnus

Hi

Ok, so what needs to be done with the output of the mixer (no matter how you do it)?

Assume you start from 10 MHz and head down to 10 Hz.

Assume you are mad at your 5370 and want significantly better performance.

Where does that get you?

The 5370 already is in the ~ 20 ps range. A lot depends on your definitions and
how good your sample is running. Let’s call that 2x10^-11 at tau = 1 second. You
could indeed call it a couple of other things as well.

Simply moving up a decade with a whole bunch of gear and it’s limitations seems like
a waste. To me you want to go for 1 to 2x10^-13 as your target. It is an achievable target
and there are a number of papers that validate it as a reasonable DMTD target.

You get a 1x10^6 “amplification due to your down mix from 10 MHz to 10 Hz. You then
need another 1x10^7 to get you to your target. All errors from everything included, you need to
work out the location of the zero crossings to within 100 ns.

The practical examples of doing it include some fairly tight lowpass filtering as well as high
pass filtering ahead of the detection process. I have never seen it done without this filtering
as part of the setup. There is just to much noise at the detector otherwise. Most systems
have something like a 15 Hz lowpass and a 5 Hz high pass for a 10 Hz note.

With fairly good diode ring phase detectors and a less than perfect (not 25 stage Collins style)
analog limiter, you can indeed get to the target.

Doing it digitally assumes you have a pretty good clock and sampler. If you look at it as a
3V p-p triangle waveform at 10 Hz, you have a 60V / second slew rate. (a 1 V p-p sine wave
is pretty close to the same number). You need to filter that at 15 Hz and then resolve it to about
6 uV at the zero crossing. You can either keep a high sample rate and make your filter a
major nightmare or you can decimate ahead of the filter and turn the resolver into a headache.
Either way, there is some work to be done.

A couple of op-amp packages is about all it takes to do the limiter with the analog approach ….

Bob

On 1/5/2016 12:07 PM, Bruce Griffiths wrote:

Read Gilbert's paper or Gray and Meyers analog IC textbook and
you will see that the whole theory of operation of these
depends on keeping the signal levels in them very small,
especially if linearity (actually translinearity) is
important.  They always have current sources in the
emitters that contribute a lot of noise.  So you have
small signals and large noise.  The IC's that are
designed to be DC coupled have even more sources of
extra noise.

IMHO, they only make sense in low performance applications
where the lack of transformers is important or in DC
coupled applications.  The only time I have used an
analog multiplier IC was in Costas loop to demodulate
QPSK from weather satellite.  It needed to be DC coupled.

Rick Karlquist N6RK

On Wed, 6 Jan 2016 18:30:07 -0800
"Richard (Rick) Karlquist" richard@karlquist.com wrote:

How about using the Gilbert Cell as "digital" mixer,
ie driving the currents hard from one branch to the other
and replacing the current sources by resistors?

How much would that improve the noise? Would it be still much
worse than the diode mixer?

		Attila Kinali

--
It is upon moral qualities that a society is ultimately founded. All
the prosperity and technological sophistication in the world is of no
use without that foundation.
-- Miss Matheson, The Diamond Age, Neil Stephenson

On 2016-01-07 13:35, Attila Kinali wrote:

I think so.

I checked up the MC1496 (just a sample-point of a classic Gilbert cell
chip), it has 25 mV Peak, or -22 dBm as maximum input voltage before it
starts to compress. Looking at the VCWR curves it is clear that it
starts to misbehave there.

Comparing that to the SBL-1+ double-balanced mixer (another random
sample-point), which has an LO max of +7 dBm, you are looking at a
difference of about 30 dB (29 to be exact, but neither number is exact
to the 1 dB so).

The MC1495, which is a linearized variant of the MC1496 (only true to
some degree, it's more complex than that), allows 5 V signals easily,
but internally you then go down to about the same levels.

So, while you can drive things harder, you can do that on both sides. If
it where less of a difference I'd say it would not be such a big
difference, but it is relatively large difference here.

Anyway, just wanted to put a few numbers down to illustrate the
difference.

Cheers,
Magnus

On 1/7/2016 4:35 AM, Attila Kinali wrote:

You can drive the Gilbert cell as hard as you want, but
the active region is only about 100 mv so the extra
drive voltage doesn't help.  It is the same as if you
drove it with a 100 mV square wave.  Somewhat better
than a sine wave, but not a game changer.

You can of course try to replace the emitter current source
with a resistor, which works to the extent that you can
afford to throw away voltage across the resistor, but
you will never get a very high impedance.  No OTS IC's
are designed this way.  What would be better would be
to use an inductor.  A true noiseless current source.
Again no OTS IC's are designed this way.

You would have to homebrew the whole mixer from discretes.

Rick Karlquist N6RK