After working with a quadrature PLL/LNA, I feel very strongly that anyone
who's serious about PN measurement should go this route.  I wasted a lot of
time myself, sitting around wishing I could afford a quieter spectrum
analyzer.  It's simply the wrong question to ask.

It's true that the HP 8590s are among the noisiest spectrum analyzers out
there, but the difference between the phase-noise floors of an 8596E and an
8560E is only about 20-25 dB.  The difference in cost is several thousand
dollars.  If you invest in a nice 8662A instead -- or even an 8640B! -- you
can use your existing 8596E to make measurements 30-40 dB below what even
the 8560E series can do.

This is a seriously-nice thing to be able to do at home.  You can use a
quadrature PLL to do things like compare different crystal-oscillator
circuits and look at DDS phase-noise floors... and lots of other things you
would never be able to do with a spectrum analyzer alone.

I'll see if the list will let me attach a small .GIF showing the difference.
The 11729C's noise floor is about 6-10 dB below the green trace, which is a
comparison of a couple of decent-quality 100 MHz OCXOs.  Note how much
higher the direct measurement floors of your current spectrum analyzer and
an 8560E are.  For $25 in parts, you could make the measurement in the green
trace with your 8596E!  Spend the extra money on signal generators, and/or
collecting VCXO references at various frequencies.

'A' models go down to 10 Hz RBW, while 'E' models have 1 Hz RBW.  The 'E'
models also have better phase noise than the 'A' models from what I've seen.
8560E-series analyzers are the quietest I've tested at most offsets.  I
don't know offhand what the 'B' model designation implies.

Still, you should keep what you already have, and add a quadrature PLL and
LNA to it.

-- john, KE5FX

As a more-concrete answer to your question, since you mentioned a need for
coverage into the 6-GHz region, an 11729B/C and 8662A would actually be a
good choice.  Together they'll still be much cheaper than the 8561E I'd
recommend otherwise.  Figure $2500 at most for the 8662A and $1200 at most
for the 11729B/C.

-- john, KE5FX

On Jan 21, 2008 11:44 AM, John Miles jmiles@pop.net wrote:

John,

I'm a little confused as to what you are suggesting.  An 8662A is
about $1500, and the 11729C is about $3k.  What would I get for $25?
I don't know exactly what is involved with the 11729 and how it makes
measurements.

If I just connected a quadrature PLL and LNA, I would still need a
very clean VCO at the same frequency, right?

Thanks,
Matt

On Jan 21, 2008 12:09 PM, John Miles jmiles@pop.net wrote:

That's where you lose me -- the 8662A only goes to 1.28 GHz, right?
Do you use multipliers?

Matt

The parts needed to implement Wenzel's app note:
http://www.wenzel.com/documents/measuringphasenoise.htm

Take an hour and look through this HP app note (large file, but only about
50 pages):
http://www.thegleam.com/ke5fx/gpib/5952-8286E.pdf

It is not all that specific to the 11729B/C despite making frequent
references to it.

Basically, the 11729B/C is a decent implementation of the technique in
Wenzel's app note, plus some microwave plumbing that will, if you have a
clean 640 MHz source available, downconvert 1-18 GHz signals to less than 1
GHz, where they can be further mixed down to baseband with any convenient
DC-1 GHz source.  An 8662A is ordinarily used with the 11729 becaus it fills
both the 'any convenient DC-1 GHz source' and the 'clean 640 MHz source'
roles.

If you decide to homebrew a PLL/LNA box, the notes on baseband LNA design
that Bruce Griffiths has posted are better than Wenzel's, assuming you are
aiming for ultimate performance.  Modern low-noise opamps would be fine as
well, at least for a first attempt at getting a PN rig working.

Yes.  The reference VCO is what determines the PN floor in most cases,
unless you start spending thousands of dollars on low-noise crystal
oscillators.  8662A-class signal generators are good but not "great," which
is why it's also nice to have a bunch of varactor-tunable crystal
oscillators around.  Often you can structure your measurements to use them
instead of a noisier general-purpose reference generator.

You can also do residual tests by building two of whatever it is you're
testing, and beat them against each other in quadrature.

Finally, you can sometimes use a coaxial delay line to test a single source
effectively.  This is a more error-prone process; I haven't played with it
much and don't consider it of primary interest, compared to 2-source
phase-detector measurements.

-- john, KE5FX

Yes; that's the rest of the stuff in the 11729 box.

There are two rather-large caveats with the 11729.  First, make sure the one
you buy has all the microwave filters.  All of the labels in the buttons
should be populated.  When it was new, customers could save a lot of cash by
specifying "options" that deleted some of the filters, or all but one of
them.

Second, you need to add a ~2 MHz low-pass LC filter to the LNA input.  By
default the LNA path has a 20 MHz LPF to allow flat response for offsets up
to 10 MHz.  That effectively keeps you from being able to measure sources at
10 MHz or less, because the other sideband coming out of the quadrature
mixer is at 2x the carrier frequency, and if that sideband leaks into the
LNA it degrades the accuracy at baseband.

There are 8662As available in the $1500 range, but most of them are kind of
ratty.  IMHO it's best to buy two of them anyway, so you have a parts mule
for troubleshooting.  They can be temperamental.  Plan on spending $2000 for
two beaters, or $2500-$3000 for a really-nice one.  Look for option 003, for
use with an 11729B/C.  You can add this option yourself but it's just
something else to deal with.

There are 11729B/C boxes available in the $3000 range, but that's overpriced
in my opinion.  $800-$1500 is more like it.  They are not all that rare.
Something like eBay item #120168988485 would be fine.

-- john, KE5FX

On Jan 21, 2008 12:41 PM, John Miles jmiles@pop.net wrote:

Makes sense now.  One problem -- the 8596E only goes down to 9kHz, at
least according to the specs.

Matt

Yeah, Grant's got a point there.  It will probably tune lower than that, but
will be limited by its narrowest RBW filter selection.

There are a couple of ways to tackle that, ranging from the exotic (run the
600-ohm output from the 11729B/C into a sound card and FFT it) to the easy
(buy an 8568A/B, which will work great down to 100 Hz.)

I actually have an 11848A on the way, and a 3561A FFT analyzer waiting for
the time to support it in PN.EXE, but that's getting kind of pathological
even for the time-nuts list. :)

-- john, KE5FX

John Miles wrote:

Hi John,

How does a HP-8660C compare to the HP-8662 and HP-8640B in your opinion ?

Bill....WB6BNQ

On Mon, Jan 21, 2008 at 01:18:15PM -0800, Matt Ettus wrote:

Spectrum analyzer front ends often either have a blocking

capacitor (to protect the mixer from DC) or don't.    The kind that
don't usually start to roll off pretty significantly below 10 KHz, and
are typically spec'd only to 9 KHz.  I suppose if one wants to live
REALLY dangerously the cap  might be removable in a few cases (NOT
RECOMMENDED)....

But would I be too simple minded to suggest that maybe some form

of A/D PC/workstation input device with high dynamic range and decent
sample rate (certainly available in high end audio stuff to 192 KHz)
would be the logical vehicle for close in measurement in a quadrature
locked PLL type phase noise setup ?  Otherwise why would you care about
performance below 9 KHz ?

Why exactly does one need a wideband SA that goes down to 100 Hz

(common spec) or 30 Hz (nicer and newer...) for this ?  Maybe I am missing
something here.... (probably am, I often do...)

And should one want an actual SA for these measurements instead

of a soundcardish thing and FFT software, I know that LF/HF boxes with
better specs than any of the affordable general coverage SA families on
Ebay show up regularly for not too much money.

--
Dave Emery N1PRE/AE, die@dieconsulting.com  DIE Consulting, Weston, Mass 02493
"An empty zombie mind with a forlorn barely readable weatherbeaten
'For Rent' sign still vainly flapping outside on the weed encrusted pole - in
celebration of what could have been, but wasn't and is not to be now either."

David I. Emery wrote:

If one is measuring the phase noise of a good OCXO with a frequency of
less than 100MHz, there isnt usually much of interest above 10-100KHz.
However when measuring the phase noise of a VCO there may be significant
phase noise out to several MHz (limit depends on the VCO frequency phase
noise measurements at offsets above half the source frequency are
problematic.). The range of interest is even wider with microwave
sources, so the answer depends on what sources you are trying to measure.
For phase noise measurement of most OCXOs a soundcard system will
suffice and with accurate calibration a modern sound card will have
better performance over the [20Hz, 20kHz] range than most surplus
spectrum analysers. It is particularly important in the flicker phase
noise region that the noise bandwidth of the filter be significantly
less than the offset frequency. It is also important that a filter with
steep skirts be used for accurate measurement in this offset frequency
region where the phase noise spectrum is far from flat. These
requirements are easier to meet with an FFT filter than to implement
such a filter in hardware.

Thus for characterising the phase noise of OCXOs like an HP10811,
FTS1200 etc there is a good case for using a sound card based system.
Since sound cards have stereo inputs implementing a correlation phase
noise measurement system is also possible.

Bruce

Correct, that's why I was complaining the other day about the lack of any
good high-dynamic-range DAQ dongles in the medium-speed market segment.

You can buy an ADC outfit for audio work ($) or a complete GNU Radio USRP
with >20 MHz acquisition bandwidth ($$$$), but there is nothing economically
available in the 1-MHz neighborhood.  There are some superb 24-bit
sigma-delta chips in the 0.5 MSPS-2.5 MSPS range, but nobody has anything I
can plug into my USB port and talk to with a C compiler.

With quadrature conversion, the carrier is downconverted to 0 Hz, and the
analyzer sees the offset frequencies of interest directly.  If the analyzer
can't look down below 9 kHz due to limited RBW or a noisy LO, the
measurement won't be valid down there.

Phase noise at >= 10-12 kHz is important for adjacent-channel rejection in
narrowband communications work.  Below that, it's important for clean
recovery of the on-channel signal.  And when comparing crystal oscillators,
you'd like to be able to look down to 1 Hz, to bridge the gap between
phase-noise and Allan deviation measurements.

Conversely, you rarely need to look at offsets beyond 1 MHz.  Anything the
DUT is going to do, noisewise, it probably will do before that point.
(Spurs are a different story of course.)

I can only look down to 100 Hz with my 8566B, and I'm fine with that for the
most part.  Someone testing extremely stable sources, again, may want a
closer look at the immediate vicinity of the carrier.

The 3047A/3048A outfits are also quadrature downconverters, but they have
two parallel outputs, one for an HF spectrum analyzer, and the other for an
FFT analyzer.  The software merges plots from both analyzers into one
wideband plot (offsets from 1 Hz-10 MHz).

Yes, noting that the FFT analyzers generally are limited to DC-100 kHz.
They also work better with switchable lowpass/highpass filters that
eliminate signal components outside the decade currently being measured.
(That's one of the things you get with a 3047A or 3048A outfit that you
don't get with the 11729B/C by itself.)

Chances are you already have an RF/microwave SA on your bench already if
you're doing this kind of work, so it makes sense to use it, as long as its
minimum frequency limit isn't a problem with the measurements you want to
make.

-- john, KE5FX

I don't have an 8660C, but coincidentally, someone just sent me some PN
plots from one.  Here's a comparison with a typical 8662A and 8640B.

The 8660C is fairly noisy, as you'd expect given its age.  The noise
profile is reminiscent of the 8566 spectrum analyzer's first LO.  It would
not be a good choice as a reference for that reason (you might as well just
use the SA.)

Basically, it's about 25 dB noisier than an 8662A at all offsets between 100
Hz and 1 MHz.  The exact numbers would depend on a lot of things, including
the carrier frequency.

-- john, KE5FX

Sometimes, rightly or wrongly, phase noise in the systems I work with are
specified down to as close as 1 Hz from the carrier where the carrier may be
around 30 GHZ. Some of it is a leftover from older satellite systems with
real low data rates where the close in phase noise was considered
significant. Unfortunately, most of the time it is easier to copy an old
requirement, than really figure out what makes sense. Unfortunately, the
contractor is stuck with both meeting and demonstrating compliance.

The age old question is, and of course depends on the modulation scheme
used, is where does the carrier stop and where does the noise that creates
degradation to BER performance begin.

Now days we no longer deal with data rates in the few kHz rates but
typically in the 100KHz to 20 or so MHz rates. So, the close in stuff is not
as significant anymore. Some order wire data may still be at low data rates,
but even that has increased significantly.

With higher data rates higher modulation orders are now also in play. I am
ignoring different FEC rates and Turbo Codes. Anyway like 16 PSK modulation
the spacing between significant symbol locations become fairly close, and
even closer at higher rates. The integrated phase noise can indirectly give
one the RMS phase jitter, so, depending on the jitter the anticipated
location of the symbol may be, or then again may not, be close the
anticipated or intended phase. A good rule of thumb is to keep the
integrated phase jitter to be less than 10 % of the Euclidian distance
between symbol locations. With M-ary QAM, both the phase jitter and phase
noise is a concern.

Just look at DirecTV receivers with free running DROs inside. Not the best
as far as stability or noise, and yet due to the data rates work fine at Ku
band and now Ka.

Unfortunately the above tirade of mine did not address ham use, where
typically we do deal with real low data rates, and phase noise does become
much more of a concern, and hence our interest in being able to fairly able
to measure it and understand it. There are actually cases where uncorrelated
noise can be of a benefit. 73 - Mike

Mike B. Feher, N4FS
89 Arnold Blvd.
Howell, NJ, 07731
732-886-5960

-----Original Message-----
From: time-nuts-bounces@febo.com [mailto:time-nuts-bounces@febo.com] On
Behalf Of Bruce Griffiths
Sent: Monday, January 21, 2008 5:51 PM
To: Discussion of precise time and frequency measurement
Subject: Re: [time-nuts] Spec An for phase noise measurements

David I. Emery wrote:

missing

If one is measuring the phase noise of a good OCXO with a frequency of
less than 100MHz, there isnt usually much of interest above 10-100KHz.
However when measuring the phase noise of a VCO there may be significant
phase noise out to several MHz (limit depends on the VCO frequency phase
noise measurements at offsets above half the source frequency are
problematic.). The range of interest is even wider with microwave
sources, so the answer depends on what sources you are trying to measure.
For phase noise measurement of most OCXOs a soundcard system will
suffice and with accurate calibration a modern sound card will have
better performance over the [20Hz, 20kHz] range than most surplus
spectrum analysers. It is particularly important in the flicker phase
noise region that the noise bandwidth of the filter be significantly
less than the offset frequency. It is also important that a filter with
steep skirts be used for accurate measurement in this offset frequency
region where the phase noise spectrum is far from flat. These
requirements are easier to meet with an FFT filter than to implement
such a filter in hardware.

Thus for characterising the phase noise of OCXOs like an HP10811,
FTS1200 etc there is a good case for using a sound card based system.
Since sound cards have stereo inputs implementing a correlation phase
noise measurement system is also possible.

Bruce

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.

On Mon, Jan 21, 2008 at 05:31:14PM -0500, David I. Emery wrote:

I apologize to Matt - I was looking at SA specs this evening

and the 8596E has the capacitor out performance spec'd only down to 9
KHz (100 KHz with the capacitor in)... no real idea of what might
actually limit performance in the DC coupled case except the LO quality
of the analyzer itself...

It is quite surprising that the 859xE series is spec'd about

10-15 db worse on phase noise than the comparable 856xE instruments.  I
wonder if this difference exists in the performance or just marketing
domains and what the actual differences are in the implementation that
accounts for it.

This was poorly phrased - I perfectly well know why one is

interested in phase noise on LOs and other signals closer to the carrier
than 9 KHz (no question there for me at least) but what I meant is why
one would want to use a general purpose wideband SA for making
measurements between 0-9 KHz or so... when FFT based technology is
available and cheap and very high dynamic range with very tight filters.
(Obviously this means high end soundcard type devices on a PC platform
these days),

I might also add that many of the cheap 24 bit 96/192 Kbs USB

sound input devices actually use codec chips which are DC coupled
internally all the way to the input pin though that is usually
referenced to half the analog supply voltage (3.3V/2 typically) rather
than 0.  So going all the way to DC with one of these chips is not
impossible with external non-audio signal processing feeding the analog
input pin, thus removing the 20 Hz high pass from the audio preamps
ahead of the codec chip.

--
Dave Emery N1PRE/AE, die@dieconsulting.com  DIE Consulting, Weston, Mass 02493
"An empty zombie mind with a forlorn barely readable weatherbeaten
'For Rent' sign still vainly flapping outside on the weed encrusted pole - in
celebration of what could have been, but wasn't and is not to be now either."

David I. Emery wrote:

Most of the phase noise originates in the local oscillator, the cost can
be reduced by using a lower performance oscillator.

As long as any internal digital high pass filter is bypassed and the
idle tone performance is satisfactory (more likely with a multibit
modulator) with a dc input.

Bruce

Yes, and I just documented it.  See my earlier post with the .GIF
attachment.

Usually phase noise in a synthesizer comes down to one question: "What's the
N factor?"

The older HP analyzers, dating back to the 8554/8555 plugins for the 141T,
stabilized their LOs by locking them with a sampler to a harmonic of a 1-MHz
VCXO.  You can imagine what N=2000-4000 or more does to the phase noise.  I
don't know if the 8590s used a similar synthesizer topology but it wouldn't
be surprising.  They may have been designed in a hurry; Tektronix came out
with reasonably-good portable analyzers before HP did, and took a lot of
business that historically belonged to HP.

The 8560s use a completely-different synthesizer architecture with a
much-lower N factor.  I think they call it a "roller oscillator," where the
synthesizer runs at something like 1/10 the frequency of the YTO.

It comes down to convenience, I think.  (1) Most people already have an RF
SA, many of which are perfectly usable down to 100 Hz or 1 kHz; (2)
Commercial phase-noise packages don't talk to PC sound cards; and (3) Most
people don't care about PN at offsets that their SAs won't cover directly.
Add to that (4): Newer SAs are FFT-based anyway, and don't need an auxiliary
FFT analyzer at all.

-- john, KE5FX

On Jan 21, 2008 11:44 AM, John Miles jmiles@pop.net wrote:

John,

I decided to get a new spectrum analyzer rather than an 11729 at this
time.  I bought an Advantest R3267 to replace the HP 8596E.  It seems
to have phase noise performance in the same neighborhood as the
8560-series, for about half the price.  In case I ever need anything
better, I can get an 11729 and use it with this analyzer, since it
goes down to 100 Hz, unlike the 8596 which only went to 9kHz.

I am seeing the following performance when measuring the built in 30
MHz reference:
100 Hz    -81dBc
1 kHz        -108 to -110
10 kHz      -117
100 kHz    -125
1 MHz        -131

All of those are 3-8dB better than the spec, except for at 1 MHz where
the spec is -135.  The measurements do jump around a bit.

I would really love to have PN.exe drive the analyzer so I can do real
phase noise plots.  I have an electronic copy of the manual, which has
all the GPIB info, but the last time I dove into the PN code (in an
aborted attempt to port it to Linux), I had a very hard time grokking
the code, plus I don't have a Windows development environment.  Is
there any way we can work together on getting this analyzer supported?
Several other Advantest machines are compatible with it, so any work
would cover multiple models.

Thanks,
Matt

If I change to "Digital mode for <100 Hz RBW", and turn on averaging,
I get much better results from the analyzer --

100 Hz    -100dBc
1 kHz      -110
10kHz    -120
100kHz    -129
1M            -135

This, of course, makes me question all the measurements....

Matt