AnsweredAssumed Answered

4 port phase deskew, any ideas?

Question asked by jprairie on Aug 7, 2006

About 3 weeks ago we had a conference call with several of the Agilent VNA gurus regarding calibration of a 4 port VNA using 90 degree thrus. Allow me to lay some of the foundation again.

This is a wafer probe measurement with 4 seperate GSG probes in a N,S,E, and W configuration. We have a 4 port hybrid structure  that we need to determine if the 2 output ports are 90 degrees apart from one another within a +/- 1 degree window at 36 GHz. Since this is a relative measurement I have some confidence that it can be done even if the spec seems very tight at first glance.

At this point we have tried several different calibration methods mostly focusing on the 4 and 6 thru calibrations using PLTS software (with thrus defined as simple electrical lengths as well as measuring the 90 degree thrus with 2 port unknown thru calibration and using that data to make a databased thru standard in PLTS).

One point of contention that comes up is that to accommodate certain thru paths during the calibration we must use a 90 degree angle thru structure. While not perfect, this structure, when measured after an unknown thru calibration, using 90 degree thru as the calibration unknown thru structure, shows a return loss of about 20 db at 36 GHz. I'm by no means a guru, but my intuition tells me that is a fairly good thru and isn't radiating too much (maybe the idea is how much is too much).

In any event, lately, we have been getting reasonable performance from calibrations using an SOLT type cal. Reasonable meaning that we feel the verification of the calibrations is within specified residual error that is listed in the manual.

With all of that said here is what we are dealing with at this point. Lets assume a good calibration has been performed on the 4 port analyzer. I want to get an idea of the residual phase skew (and amplitude for that matter) of the ports. I take a transmission line on the ISS and measure an open reflect of that line (only one port is down, other ports/probes are held in the air, cant measure thrus because of the probe arrangement). I do this for each of the 4 ports and look at the phase differences for what should be the same physical structure except for any probe placement error (which through experimentation we have seen on the order of 0.25 to 0.50 degrees at 36 GHz). These differences come out to be within the specified 3.5 degrees or so.

But now keep in mind I want to make this a relative measurement, in other words i want to measure S31 and S41 and subtract the phase of those 2 measurements from one another. Is it wrong to assume that I can take reflect measurements with the same structure on each of the ports and in software skew the remaining phase and amplitude error of ports 2, 3 and 4 to overlay what port 1 measured (essentially deskewing everything to port 1)? The thought is that this method is essentially creating an error vector consisting of the phase and amplitude difference at each frequency point with respect to some reference port, port 1 for example. I don't so much care that the absolute accuracy of the measurement of S31 and S41 is extremely precise, but rather that the delta phase and magnitude between them be accurately known.

The only potential problem I see (assuming repeatable probe contacts) is that this would only work for one calibration and need to be recomputed for each cal, and it also assumes that the residual error that is left after calibration is constant and repeatable for a given port and frequency point. In other words I expect if I remeasured port 1 reflect on this line it should overlay very closely with the previous measurement, which will not necessarily be the same as the measurement obtained from the other 3 ports.

Does this seem totally unreasonable? or is there something here that makes sense with some maybe some tweaking or re-thinking?