K7ITM wrote:
On Sep 12, 9:22 am, "David" nospam@nospam wrote:

I have a RF switching unit that multiplexes 4 input RF ports to 12 output
ports. The unit carries RF of a frequency range from DC to 1 GHz. It carries
GPS signals and other RF from an antenna.

GPS is 1GHz.. 1.2 to 1.5 or thereabouts


Can you advise me of the design of a test jig to accurately test this unit?
The test jig will contain a Network Analyser.
FWIW:
Rhode+Schwartz, Agilent, Anritsu and others would be happy to tell you
how to do it (when you spend $100K on the core of the measurement
system, the field application engineers are more than willing to spend a
few hours or days figuring out how to make your measurements)

Does your test jig have to make all the measurments "hands off"? or is
reconfiguration of the test port cables between measurements ok?

If the former, you need a 16 port measurement system with a suitable
relay box.


Where I have a split path or differential measurement, do I always have to
use phase matched and batch matched cables?

In theory, you'll be able to calibrate out all the various port
configurations. If you're using one of the newer PNAs, this is pretty
straightforward with an automated script. You'll connect your cal
standards to the ports you need, configure your test jig relays, run the
cal, store the cal configuration, go on to the next.

Then, when making the measurements on the UUT, you just do something
along the lines of:
For each measurement
Configure test jig
Load Cal set corresponding to configuration
Make Measurement
Save data
repeat






Does my test jig switch calibration loads between ports? The cal load being
an open, a short and a load.


It can. This is sort of like what an "E-Cal" box does. You'll do a
one-time calibration of your cal stuff. What kind of performance do you
need?


Do I get the paths the same by using phase matched & batch matched cables,
and then do a vector gen to save a 'footprint' of the test jig when it is in
a known good
state?

Sort of.. there's a bit more to it, but essentially that's what's done.
For a full bidirectional two port measurement, there's potentially 16
calibration terms for each frequency.

How do I minimise errors introduced by imperfections in the test jig? I am
looking for the test jig to measure the absolute SWR and insertion loss of
the RF switching unit, using S parameters.

To what precision do you need to know these things. 20dB return loss is
a heck of a lot easier than 50dB. 0.1dB insertion loss is harder than
1dB. What sort of isolation do you have between ports of the UUT and do
you need to measure it. (i.e. is the isolation good enough that you just
need to check it, and after that, you can ignore the ports that aren't
connected, or do you have to do a full N-port characterization)


We do testing of a somewhat similar nature here, using RF relays to
switch the test configuration for different tests. It's important to
calibrate such a system, and you need to establish how often the
calibration should happen. Because of the number of ports involved,
your calibration may be a bit tedious, so it could be an advantage to
insure that you use especially stable cables and other components so
the calibration doesn't have to be done too often. It's also
important to do a careful error analysis so you understand the source
of the errors and their expected magnitudes. At first look, I'm not
seeing that you need matched cables in your system, but you do need to
calibrate the net effect of the cables and whatever switching you use,
and account for the variability as cables are flexed and as relays
close with slightly different contact resistances each time. It's
possible that your accounting will tell you that those variabilities
are so far below your allowed tolerances as to be unimportant, but you
should think about them in any event.


The uncertainty analysis is going to drive your system design.