Re: [time-nuts] HP 5071A Electron Multiplier of Cesium Beam Tube
That's an interesting answer. Can you explain what you mean by
"faster digital noise analysis capabilities"?
The 3048A is relatively cumbersome to use, compared to a modern phase-noise
test set with high dynamic range ADCs. Conceptually, a software radio with
multiple ADC channels could be used to measure phase noise directly as well
as to perform other timing-related measurements. The devil's in the
details, though, because the state of the art in digital PN measurement is
down below -170 dBc/Hz, and the front-end requirements (noise, jitter,
channel isolation...) are accordingly strict. To compete with the better
commercial gear you need to employ cross-correlation and various other
error-cancellation techniques. It starts to look like real work before
long.
It would be relatively trivial to build a mediocre digital PN test set, but
such an instrument probably wouldn't be useful for characterizing
high-quality crystal oscillators by itself. It would be more challenging to
build one that could routinely compete with the 3048A's analog front end in
the general case.
The reason this interest me is I'd like to get the low phase noise
of a Wenzel 100MHz ULN, but I understand the price is $1,500 which
is a bit too high.
Wait by the river, and one will eventually come floating by. Or...
Some guys at NIST got very good noise performance with a DRO at
10GHz. This is interesting, since MiniCircuits sells inexpensive
low-noise microwave amplifier ic's and mixers. So it might be
possible to get a low noise cavity DRO at 8GHz and use regenerative
dividers to get down to 1GHz (8 / 2^3), then use injection locking
to get down to 10MHz. This could be an inexpensive solution to a
difficult problem. And you have shown you can put 10GHz on FR4, so a
Rogers pcb may not be needed:
http://www.thegleam.com/ke5fx/hpll.htm
Possibly true, but don't kid yourself: such a divider chain would cost you
way more than $1500 worth of your time. And don't forget that you'll have
to build two to test it!
One of the biggest problems would be the effect of the DRO control loop. I
haven't seen the NIST papers you're mentioning but the best X-band DRO I've
played with has a loop bandwidth of 300-400 kHz. Within that bandwidth, it
will just scale up the noise of whatever you're using as a reference, so any
attempt to get low VHF phase noise with a DRO and divider chain will just
end up giving you back the noise of your reference, plus any residual
effects.
A better approach IMHO is to work on pushing the limits of what can be done
with homebrew crystal oscillators. The excellent broadband floor of Wenzel
and similar oscillators is not due to their use of exotic crystals, but to
their use of good oscillator circuit topologies (and no buffering to speak
of). The crystal's job is stability, not noise, and unlike low noise, good
stability is relatively cheap and trivial nowadays thanks to cheap GPS
clocks, rubidiums, and good-quality OCXOs.
So the question is what kind of tweaking is needed to get the best
performance in a regenerative divider, and what kind of equipment is
needed to do it? Then, is perfection really needed in order to beat
the Wenzel ULN? Maybe put up with lower performance in the
beginning, then upgrade later.
In practice many applications for ULN-class oscillators put the broadband
floor at risk in other ways. Very few buffer amplifiers have a noise floor
below -170 dBc/Hz, for instance. Fortunately, apart from timing metrology,
ULNs often end up driving high-end ADCs, where the application is likely to
be a good test bed in itself.
One trick I have found that really helps isolate circuit blocks is
to put them on their own small island pcb, which is then soldered to
the main ground plane to hold it in place. Then find the location of
ground connections that give the lowest crosstalk. A brief
description is here.
Yep, totally, and the islands become reusable components in their own right.
That's a valid thing to do, although I find that when I'm that concerned
with isolation, I probably want a full shield anyway (hence the use of lots
of discrete Hammond boxes). Sometimes even this approach is self-defeating,
as when I find that my tightly-sealed Hammond enclosures make good cavity
oscillators. :-P
-- john, KE5FX
"John Miles" jmiles@pop.net wrote:
That's an interesting answer. Can you explain what you mean by
"faster digital noise analysis capabilities"?
The 3048A is relatively cumbersome to use, compared to a modern
phase-noise test set with high dynamic range ADCs. Conceptually, a
software radio with multiple ADC channels could be used to measure
phase noise directly as well as to perform other timing-related
measurements. The devil's in the details, though, because the
state of the art in digital PN measurement is down below -170
dBc/Hz, and the front-end requirements (noise, jitter, channel
isolation...) are accordingly strict. To compete with the better
commercial gear you need to employ cross-correlation and various
other error-cancellation techniques. It starts to look like real
work before long.
That is a very interesting answer. No wonder Stein pushes ease of
use so much for the 5120/5125. But they are $40k to $50k in Canada,
so obviously it's time for a new approach.
-
Where would you find ADCs with enough speed and resolution to
capture the noise signal from the phase detector? -
What do current systems use for a reference oscillator to reach
-170dBc? I'm not talking about the 5120/5125, or the Rohde
It would be relatively trivial to build a mediocre digital PN test
set, but such an instrument probably wouldn't be useful for
characterizing high-quality crystal oscillators by itself. It
would be more challenging to build one that could routinely
compete with the 3048A's analog front end in the general case.
I tried to identify the U1 and U2 ics on the A12 LNA board in the
11848A. The best I could come up with was the part number -
1826-2081. But there was no cross-reference in any of the HP lists
on the the HP Museum.
Anyway, technology has far surpassed what was available back in the
80's when the 3048 was designed. Wenzel and Rubiola both published
front ends for PN that probably match anything currently in use:
http://www.wenzel.com/pdffiles1/pdfs/lowamp.pdf
http://www.femto-st.fr/~rubiola/pdf-articles/archives/2005-arxiv-0503012v1-ampli.pdf
In "The Measurement of AM noise of Oscillators", Rubiola states "The
measurement systems described exhibit the world-record lowest
background noise."
Since AM noise is generally less than PM noise, the amplifiers he
describes should be pretty close to state of the art. Table 6 on
page 18 shows the noise parameters of some selected amplifiers:
http://arxiv.org/PS_cache/physics/pdf/0512/0512082v1.pdf
So the amplifier front end doesn't appear to be the gating item. I
think the biggest problem is to find low noise oscillators that can
be used as a reference. One approach might be to use 8 Wenzel 100MHz
ULN's in a cross-correlation analyzer. That gets expensive.
The reason this interest me is I'd like to get the low phase
noise of a Wenzel 100MHz ULN, but I understand the price is
$1,500 which is a bit too high.
Wait by the river, and one will eventually come floating by. Or...
As above, I'm looking for more than one:)
Some guys at NIST got very good noise performance with a DRO at
10GHz. This is interesting, since MiniCircuits sells inexpensive
low-noise microwave amplifier ic's and mixers. So it might be
possible to get a low noise cavity DRO at 8GHz and use
regenerative dividers to get down to 1GHz (8 / 2^3), then use
injection locking to get down to 10MHz. This could be an
inexpensive solution to a difficult problem. And you have shown
you can put 10GHz on FR4, so a Rogers pcb may not be needed:
Possibly true, but don't kid yourself: such a divider chain would
cost you way more than $1500 worth of your time. And don't forget
that you'll have to build two to test it!
I still don't see why it should take so much time to tweak. There is
not that much to adjust, and a good network analyzer should be able
to show the response of each section. So once you have one working,
it whould be easy to duplicate. And if they were that touchy, it
would be difficult to sell them commercially. The slightest bump
would knockthem out of spec.
But as described below, I have scrapped the whole idea. It turns out
the performance may not be much better than a Wenzel.
One of the biggest problems would be the effect of the DRO control
loop. I haven't seen the NIST papers you're mentioning but the
best X-band DRO I've played with has a loop bandwidth of 300-400
kHz. Within that bandwidth, it will just scale up the noise of
whatever you're using as a reference, so any attempt to get low
VHF phase noise with a DRO and divider chain will just end up
giving you back the noise of your reference, plus any residual
effects.
The idea was to use the 10GHz oscillator as a low phase noise
source, then divide down to use at lower frequencies. So it is the
reference. One application would be to lock it to the oscillator
under test to make PN measurements, so the loop would be pretty
slow. But it turns out the whole concept probably won't give better
phase noise, so I scrap the idea.
Here's a bunch of links - you don't have to download them since the
last one demolishes the concept. But here they are as a reference.
"Ultra-Low-Noise Cavity-Stabilized Microwave Reference Oscillator
Using An Air-Dielectric Resonator"
http://tycho.usno.navy.mil/ptti/ptti2004/paper16.pdf
Siemens App Note 002 shows the pcb layout for a 10GHz DRO:
http://www.taconic-add.com/pdf/technicalarticles--resonator-oscillator.pdf
The next paper shows the phase noise of a 10.24 GHz x-band sapphire
oscillator divided down to 640 MHz using regenerative dividers. The
plot in Figure 10 on page 5 shows the result is barely 15 dB better
than a Wenzel at 1 KHz, and it looks like the Wenzel pretty much
matches the performance past 10 KHz. On the other end, it looks like
a Wenzel 10 MHz crystal would match the sapphire performance below
100Hz.
"Low Phase Noise Division From X-Band To 640mhz"
http://www.psi.com.au/Pages/LibraryPublished/fcs_2002_lnrd_paper.pdf
Since a cavity stabilized DRO oscillator at 10 GHz wouldn't come
close to the performance of a sapphire, it means the best practical
source is a Wenzel. So I scrap the idea and start looking at better
crystal oscillators as you discuss next.
A better approach IMHO is to work on pushing the limits of what
can be done with homebrew crystal oscillators. The excellent
broadband floor of Wenzel and similar oscillators is not due to
their use of exotic crystals, but to their use of good oscillator
circuit topologies (and no buffering to speak of).
This is very interesting news. I thought it took excellent high
quality quartz and very good low noise circuitry.
Can you tell more about how it is done? Do you happen to know of any
schematics? What kind of crystal would be suitable? I would be very
interested in any additional info.
The crystal's job is stability, not noise, and unlike low noise,
good stability is relatively cheap and trivial nowadays thanks to
cheap GPS clocks, rubidiums, and good-quality OCXOs.
Yes, I very much agree. GPS solves a lot of problems.
So the question is what kind of tweaking is needed to get the
best performance in a regenerative divider, and what kind of
equipment is needed to do it? Then, is perfection really needed
in order to beat the Wenzel ULN? Maybe put up with lower
performance in the beginning, then upgrade later.
In practice many applications for ULN-class oscillators put the
broadband floor at risk in other ways. Very few buffer amplifiers
have a noise floor below -170 dBc/Hz, for instance. Fortunately,
apart from timing metrology, ULNs often end up driving high-end
ADCs, where the application is likely to be a good test bed in
itself.
I thought the noise in a 50 ohm resistor set the lower limit to
-174dBc. Modern amplifiers are better than that. For example, a 50
ohm resistor has 0.894nV/sqrt(Hz) noise, but you can get wideband
amplifiers with 0.7nV/sqrt(Hz) noise, which is equal to the noise in
a 30.6 ohm resistor. (Of course, flicker noise is not included)
High speed adcs have very low jitter requirements to maintain ENOB,
so anything that can improve the noise is helpful.
One trick I have found that really helps isolate circuit blocks
is to put them on their own small island pcb, which is then
soldered to the main ground plane to hold it in place. Then find
the location of ground connections that give the lowest
crosstalk. A brief description is here.
Yep, totally, and the islands become reusable components in their
own right.
That's a valid thing to do, although I find that when I'm that
concerned with isolation, I probably want a full shield anyway
(hence the use of lots of discrete Hammond boxes). Sometimes even
this approach is self-defeating, as when I find that my
tightly-sealed Hammond enclosures make good cavity oscillators.
I'm probably preaching to the choir, but do you find the waveguide
cutoff frequency for the box? It's pretty easy - you can do it in
your head. For example, the cutoff frequency is
fc = c / 2w, where
fc = cutoff in GHz
c = speed of light, 30 cm/ns
w = width in cm
So a box 4 inches wide would be
fc = 30 / (2 * 10)
= 30 / 20
= 1.5 GHz
Here's a calculator that gives the attenuation at any desired
frequency below cutoff:
http://www.k5rmg.org/calc/waveguide.html
Another problem is the pcb will resonate at some frequency, just
like a patch antenna.
For example, a 100mm x 50 mm (4 inch x 2 inch) pcb will resonate at
700MHz. But drop the size to a 1 inch square, and the resonance
moves up to 2.768 GHz. This is a bit more difficult to do in your
head, so here's a calculator to help:
http://www.emtalk.com/mpacalc.php
So the trick is to use smaller parts and put them in smaller boxes.
Then fill them with Eccosorb:)
john, KE5FX
Mike