4P Ecal error corrected file sizes: ENORMOUS!

[Michael Smith]

I'm in the middle of performing an unknown thru error correction on a 4-port measurement, calibrated with an Ecal.

I kept the cal config simple, using the minimum number of terms: only 3 reflect standards (A1-3/B1-3) out of 7 impedance states for all 4 ports, plus the switch terms and the thru's but only using AB1 for the thru.

The error corrected calibration standards range from 400MB to 500MB to 1GB, EACH.

The 4port DUT has just crossed over 4.2GB and the file is still writing.

Understandably a 10MHz-67GHz range with 10MHz step sizes is going to be large, but at the rate VNA Tools is writing such large files, our server will run out of space - and the processing time is very large too.

What are the realistic options to reduce file size and processing time? If I add more impedance states (A4-A7, B4-B7) will this consume even more space on the error corrected DUT? With a typical measurement session ranging from 5-20+ Sparameter files that would need error correction, the practical limitations are clear. Any ideas?

[Juerg Ruefenacht]

Hi Michael,

as you are well aware, the unique uncertainty ID's managed by VNA Tools allows to keep track of the whole traceability chain including uncertainties, correlations and dependencies. If you do not take on board the correlation between the standards the resulting uncertainties can easily be under- or over-estimated depending on the used calibration schemes.
The drawback is that the *.sdtab files, describing the calibration standards, can easily grow rather large (even more than 100 MiB per standard).

There are mainly four steps to tackle this problem:
1) Check under VNA Tools, Options, Measurement Model: that the Small Unc Limit is set to its default value of 1e-07 and NOT set to zero!
2) Reduce the Ecal impedance states files to S11 data only (might be 2-port measurements from the used characterization process).
3) Cut away the dependencies by using the sdatcv or scolcv file format.
4) Reduce the frequency resolution.

More detailed information about the above mentioned points:
1) The Small Unc Limit does cut away all uncertainty dependencies which are smaller than the defined value.
This limit should only be set to zero for didactical reasons as you have done it during the VNA Tools training course. Otherwise the uncertainty matrixes will just explode and the file sizes will get very large using the settings you have described (like the fine frequency resolution of 10 MHz).

2) I assume that you have measured all Ecal states with a 2-port setup. Therefore also your single impedance states measurements will consist of all four S-parameters (S11, S21, S12, S22).
Reducing these files with the Change Port Assignment will allow you to reduce the single impedance states data to smaller S11 data files only.
This is automatically done if you import a 2-port file to the database using the 'Database, Create Database Standard functionality'. There you can define the Ports you want to use for the 1-port impedance state of the characterized Ecal unit.

3) If you still run in computational resources issues then you can cut away the dependencies by transferring the sdtab files in sdatcv files.
This will generate much smaller files but keeps the important correlation information between the Ecal states.
The scolcv file format was specifically developed to save the definitions of a whole cal kit (collection). The benefit of scolcv is that it can store multiple calibration standards in a single file and it keeps the correlation between all standards in that collection. That would be the ideal file format to store the definitions of a mechanical calibration kit or electronic calibration units if file size matters but access to the dependencies is not needed.

4) Normally we recommend to characterize your transfer cal kits with the same frequency resolution as you perform your production measurements. This will avoid any additional interpolation steps.
The calibration standards typically show a rather steady frequency response. Therefore a very fine frequency resolution is not needed.

At METAS we use the following standard frequency list for the characterization of the customer calibration kit standards:
9 kHz to 9.5 kHz in 0.5 kHz steps,
10 kHz to 95 kHz in 5 kHz steps,
100 kHz to 950 kHz 50 kHz steps,
1 MHz to 9.5 MHz in 0.5 MHz steps,
10 MHz to 95 MHz in 5 MHz steps,
100 MHz to 950 MHz in 50 MHz steps,
1 GHz to xx GHz in 100 MHz steps.    (xx: upper frequency limit)

Some general remarks:
a) We would strongly recommend to use all impedance states for an Ecal based calibration and not select only a few states.

b) The initial characterization of the Ecal unit, based on a traceable mechanical cal kit, may result in rather large *.sdatb files. However, once you have created the smaller sdatcv files of your Ecal states, all further calibrations will produce much smaller file sizes.
At METAS we are using for the VNA measurements the following PC system: Win10, 64-bit, Intel Core i7-8700K CPU, 3.7 GHz, 32 GB of RAM. With this we have no problems to manage larger sdtab file sizes for our customer based calibrations - even without cutting away the dependencies.
Only for our primary traceability calibrations we are using a special designated system with more RAM and CPU power J

Hope this helps and gives you an idea how you can tackle the large file size issues,
Juerg

[Michael Smith]

Thank you, this is helpful. I did have the Small Unc Limit set to 0 from the training. I'm using a lab system now that has it set to 1e-07.