differential return loss measurment

[biff44]

If you had an IC with a true differential input, and you were trying to measure the differential return loss using a typical network analyzer (single ended port 1 and port 2 available for S11 and S22 measurement), can you make a differential return loss measurment?

I tryied calibrating S11 with a broadband balun in place, and did not get a very good return loss indicated when measuring two 50 ohm loads.

I also tryed calibrating S11 with a DC-6 GHz balun with 3 dB pads at the two ouputs, and once again, got very poor VSWR when measuring two 50 ohm loads, (even worse than the hybrid).

Any tricks, other than renting a 4 port network analyzer?

 

[E Kafeman]

I do a bit different in each situation. Last time was a CC2420, 2400-2480 MHz. As it was relative narrow band, did I make a lamda/2 coaxial balun and did open/short/load calibration. As load was a 200 Ohm SMD resistor used, soldered directly between the two center conductor tips. Verified result by replacing resistor with a known capacitor and it showed expected result. Distance between tips fitted perfect to then do further probing directly on PCB.
I did set system impedance in my measurement software at 200 Ohm. Calculating LC balun and matching network worked out perfect, verified by finally measuring output power at 50 Ohm.
Did actually also first try with a broadband balun but resulting curves did look too irregular, so I tried not to do a calibration.

 

[biff44]

yes, I am trying something similar using either coaxial connector baluns or surface mount ones, and trying to calibrate the network analyzer with differential load/short/open. I am making some progress, but there are some odd results so far too.

 

[FvM]

In principle, differential and common mode impedance (or respective S11) can be calculated from a single ended two port measurement. Assuming a symmetrical differential port, the two values S11 and S21 contain all information.

 

[E Kafeman]

I have the coax balun still laying here but no camera so directly from my photo-scanner:

It is no deep in the photo sharpness but it gives an idea.
First did I cut main cable and adjusted electrical time delay at VNA, S11 Smith chart, so I got a correct measurement point.
Next did I solder its center pin with first part of balun cable. This balun cable is cut a bit too long.
Then did I shorten balun cable in very small pieces each time, until VNA showed that I almost was back in same spot.
After removing a bit of outer braid, was also the third cable braid soldered together with the other two.
S11 delay result was verified once more, and then was the impedance calibration performed.

 

[biff44]

I have a marki balun and a krytar 180 deg hybrid available. But I was thinking that a toko 617db-1023 smt balun might work best--no connectors in the way or lines to match up.

 

[E Kafeman]

In my particular case did I need something for probing directly on an existing PCB. Due to inductive losses in PCB traces did I want to measure what values to use for optimized balun and impedance matching for each component position, adding that component and then measure next position. Distance between the tips did then fit well for 0402 solder pads when I did probing on free hand. Did first try with something like the toko balun but resulting calibration was less good when verified with known caps and it was mechanical more complicated to build the probe. Guess I had some kind of coupling to ground that caused measurement problem as I also was holding in the probe. The probe showed above, I did clamp coaxial wire with ferrite tubes, but it had just minor effect so it was really not needed.

 

[Mazz]

biff44

a way I used many times is the following one.

Once you get calibrated to the input pins of your differential device, apply the equation:

**broken link removed**

and get the result.

There is a paper explaining the math behind this, I'm not able to find it now.

Easy, but not a direct measurement.

I hope it can help.

Mazz

 

[biff44]

thanks guys. the equations are helpful.