practical drift characterization

[Michael Smith]

I'm characterizing drift on one of our PNA's using the prescribed Ecal drift characterization method from our tools training and Euramet. The problem I'm encountering is the excessive amount of resources needed to compute the error terms. I'm measuring from 10MHz to 67GHz with 1601 points. With 7 impedance states plus the switch terms, there are over 1500 files to compute. I feel like I'm doing something wrong but neither of the guidelines go into detail about practical steps, point size, etc for accomplishing this. Aside from using a server with 1.5TB of memory, there's no way I can process the data. I'm currently using 100's of GB to store the data before processing with with the drift analyzer matlab script.

Does anyone have a PRACTICAL set of steps/resolution, etc that's good enough for analysis without requiring unrealistic levels of computing resources to measure?

[Michael Wollensack METAS]

1. If you perform drift characterization by performing periodic ECal calibrations, e.q.: 96 per day then you need to compute 96 times the calibration error terms (*.calb). There is no need to perform the error correction of all your raw data and all your 96 calibrations. So you should end up with 96 calb files which have a size about 96 x 100 MiB. Then you can load those calb-files from MATLAB and do statistics on each error term.

2. If you would just perform one calibration and then measure a thru connection for 24 hours then you can observe the drift on the error corrected data. Here you can check the Without uncertainties checkbox in the error correction configuration to reduce the file size, because in that case you're just interested in the values and not in the uncertainties.

Hope this helps.

Regards

Michael

[Michael Smith]

Thank you Michael. I wonder if the problem is that I'm using the measurement series, "ECU Drift" and it's performing calibrations every 15 minutes (96 times) but it's measuring all 7 impedance states (A and B, plus AB1 and AB2). That totals 1536 vdatb files. Would it be better then to do a manual cal and use the measurement series "Drift" instead of using the ECU drift method?

[Michael Wollensack METAS]

> Would it be better then to do a manual cal and use the measurement series "Drift" instead of using the ECU drift method?

You can do both. 

With the ECU drift method you get the drift of each error term.

With the drift method you get the drift of the error corrected data.

The assumptions here are:

- The transmission drift (S21, S12) of a thru connection is caused by the tracking terms.

- The reflection drift (S11, S22) of a thru connection is caused by the directivity drift

We normally do the second method in our lab.