Maybe that is what you were talking about but I'm trying to stay more
general. The formula and explanation I provided can be applied to RF
circuits in general not just mixers. Mixers comprise just one to three
stages in a radio generally. The rest of the circuits are filters and
amplifiers and detectors. With the right generalized approach you can
analyze any linear circuit or the whole radio.

When a receiver is being characterized for IP3, the 1st mixer is what is
being characterized in every case.


Well, its easy if you know all the specifications of the amplifiers or
have made the measurements on the test setup by itself. This is
something I can't do and would be making an assumption about.

This is something that is generally done in the industry today. It is more
or less the testing standard.

I don't understand why the amplifiers are need in the first place. Maybe
their generators have weak outputs.

Most RF generators today have an output of at least +15 to +20dBm, so
generator output is not an issue. The main reason for selecting these
amplifiers is twofold. First of all, they have very high reverse isolation,
and second, they have very low harmonic distortion and very good multitone
response.

The test setup is overly complex and has unneeded equipment. When I see
something like this it generally
means that someone is having problems making the measurement and not
understanding the problems they are having.

From the above statement, I have to deduce that you haven't really made too
many mixer characterizations. If you had, you would have discovered that a
"complex" setup is necessary when characterizing low distortion mixers.
If you are measuring the IP3 of the newer RF switches and digital step
attenuators, this "complex" setup just barely makes the grade.

It could be they are having
reflections from the DUT screwing up the measurements and the amplifiers
and attenuators is their way of dealing with it. I could keep making
pointless speculations about this but you must see the point by now.

No, I don't see your point.

The need of attenuators and power divider on the generator outputs is
well understood. The generator outputs must have some degree of
isolation from each other so the test setup itself does not generate the
intermodulation products to be measured. I've made these measurements on
unity gain amplifiers with just two generators, two attenuators, a power
combiner and a spectrum analyzer. That's all you need.

Perhaps this setup is adequate for unity gain amplifiers, but it doesn't
even come close if you have to characterize high performance mixers. You
might thing the setup is working well, but you would discover very quickly
that one of the newer mixers isn't meeting its specifications.
understand their problem with that.
above.

If you are using low-pass or bandpass filters at the output of each
generator and make sure that the tones are at the required level,
this is a non-issue.

Well OK for you to say that but I'm not so sure other people that make
measurements on SW radios and publish the results have paid proper
attention to this.

I am not sure about that.......if they are not following these guidelines
and if different reviewers are using different topologies with their test
setups, there is no standard to compare by.
A similar problem occurs with microscope specifications. As an example, with
planachromatic objectives one manufacturer might specify a plan objective as
having a flat field with good color correction over at least 90% of the
viewing field while another manufacturer would use 85% as a specification.

Here is another problem when making IP3 measurements, or any distortion
measurement...........when you are looking at distortion components that are
at -100dBc or lower, even connectors can contribute to the measured result.
When selecting attenuators, the ones with the transverse heatsink fins do
have lower IMD properties, so these are the ones to use.

Sometimes, you might only have a filter that has a corner frequency
very close to your highest frequency of interest. Part of calibrating
the test setup is making sure that you have the correct power level
at every frequency that you are making the test at. What I would do
is measure the power level of the RF generators and LO generator at
every frequency of interest, and either use a correction factor for
setting the generator output manually, or I would enter the
correction factor into the Labview program when applicable. Since we
are making sure that the power levels are correct at all frequencies,
condition #1 is being met.

Very conscientious of you to follow proper procedure in calibrating out
the affects of support circuitry. It is the right approach but again for
the sake of discussion I can't know what it is you or Mini-circuits may
or may not be doing unless it is explicitly stated.

That is why I provided the PDF link from Mini-Circuits..............they are
pretty succint in the appnote.

On a final note, I haven't done any multitone testing of
amplifiers..........my tests were limited to harmonic distortion,
noise figure, S-Parameters, and 1dB compression point. As I have
mentioned in the past, it sounds like you have been in the industry,
and I appreciate your input. One thing I didn't mention was a piece
of test equipment that makes these tests a little bit easier. Instead
of using a swept spectrum analyzer, a Vector Signal Analyzer (VSA) is
used. This instrument has a very wide dynamic range, with a noise
floor of -140dBm, even in a very wide passband. Since this is really
an FFT analyzer vs a swept analyzer, you aren't limited by very long
sweep times of the swept analyzer. Another new tool that has become
available from Agilent is the PSA. This is a spectrum analyzer with
added functions, but the best thing about this analyzer is the very
low sideband noise from its internal LO. This makes it possible to
look at the phase noise sidebands from an 8657 for instance, even at
1MHz away from the carrier. There were several different generators
from Agilent, Rohde and Schwarz, and Fluke, but the quietest units
they had around were still the 8642B. When I characterized one of
those unit at a 100kHz offset, I measured the noise down at -154dBc.
The new R&S stuff isn't bad, but the 8642 generators are still "king
of the hill".

I have some experience making these measurements. One thing about being
in electronics is change. I'm sure you know the drill. I have
experience in many areas out of necessity. Digital, analog, RF, mixed
signal, electromechanical, military, computers and so on.

I have no doubt about your ability.......it appears that you have quite a
bit of experience in this field.

Agilent make some good equipment. Always always have since they were
HP. The company where I work now has an Agilent PSA. The thing is still
slow on the noise measurements and I can't see how that type of
measurement could be speeded up while maintaining accuracy. I checked
the formula I posted against the IP2 and IP3 measurements it calculated
on some wide band amplifiers and they matched so I have some confidence
in them AND I ran the generators at different output levels and got the
same IP2 and IP3 numbers so it can be done with consistency at
different levels so there you have it Mini-circuits personnel.

I have noted the same thing, as long as I was working within the linear
range of the DUT. As far as noise testing, I think you are talking about
phase noise. If that is the case, Agilent has a new phase noise/VCO test
system that can measure down to below -160dBc. We were using it with some of
the new VCOs (in-house design). Still, when measuring down at that level,
external factors can affect the measurement. When were were testing the VCOs
at -40C, there were some spikes in the phase noise spectrum.
The firing of the SCRs in the environmental chamber were causing
perturbation of the tuning voltage from radiated emissions. These were very
low frequency harmonics of 60hZ...............the 3rd and 4th harmonic were
causing the problems.

The Agilent PSA is an OK piece of equipment that automates some
measurements.

Anritsu bought Wiltron a US manufacture of RF signal generators with
very good phase noise. The R&S claim to fame I saw as important at the
the time I was buying them was phase locking ability. Most RF generators
have frequency lock but the R&S generators have phase lock. Most people
(very smart people by the way) didn't understand the difference and
would fail BER measurements due to the multiple generators in the test
setup being frequency locked that occasionally slipped a cycle on a
telecommunications link. What is a few bits lost every few minutes
between friends anyway? Oh well, details, details...

I remember when I was working at Rockwell-Collins. They had buffered 10MHz
signals running through all of the labs that we used to connect all of the
generators and spectrum analyzers to. This was also important when
characterizing the frequency hop synthesizers.
Same thing at Motorola.......we used to sync all of the equipment when we
were measuring BER of some of the newer radios.
Those R&S generators were pretty good. When we were looking at new
generators years back, we had samples of the Agilent E series and of the R&S
SMY-01s. Those SMY-01s were pretty nice, but not as clean as the 8642s.
We used them as the reference oscillator with new synthesizer designs. At
wide frequency offsets, the phase noise of the synthesizers were about the
same with both generators, but at a 7kHz offset there was a 25dB disparity
between the two, with the 8642 coming out better.
Another thing about the SMY-01s...............many of them had failures in
the phase lock loop section.
I haven't seen this problem with the newer R&S units..........they have been
very reliable. Wouldn't be bad for a home unit!
I remember the 8640s used frequency locking, and if you connected the
generator to a high resolution frequency counter you would see the output
frequency warp up and down by a few Hz over a period of time.
That is the reason that I still have my old Boonton 103D. This generator has
phase noise in the -143dBc range and it has a dedicated OCXO for the phase
locking function.

Oh well...........time to get ready for work again.............nice talking
(bantering?) with you. I don't mind intelligent debates!

Pete