impedance measurement with VNA using series, shunt/series through methods, graph

[joeqsmith]

I guess he is using a ground spring ? but I haven't seen any.

The signal was probably pre-recorded anyway and he is "lip synching" the manip 

Lip synch,      I would expect him to use the spring or wire wrapped around the tip with that 6" removed.  He may have very well used the springs but I couldn't see it.  If there's no spring, I wonder what the reference was.  I would have thought the little board he shows was isolated. 

To Joe:
Why not ask VirtualParticles to open it up and see how it's made?

There are a few reasons.  I'm not so brash that I would ask anyone to pull their equipment apart.   He stated he does work with them and I think helping reverse engineer their product would be a pretty unprofessional move.  Brian collected the S-parameters for us which is really all I needed to get an idea what it was.   Last, I thought that it may be potted.  I was actually thinking I could borrow one to X-ray if it came down to it. 

[coppercone2]

making those spring bushings is really hard to get right. at least I have an assortment of spring wire now

[PartialDischarge]

I was thinking if it would be a good idea in order to measure the ESR of a capacitor in shunt mode, to intentionally lower the resonance frequency by adding in series a low ESR inductor, ie a power inductor for example.
The main reason is that ceramic capacitors do have a high resonance freq and maybe the real use mode is going to be at much lower switching frequencies, where the real ESR is going to be lower too.

So for example a 100pF capacitor in series with a 1mOhm 4.7uH would have resonance at only 7.3MHz instead of at hundreds of MHz, a region in which it would be easy to make measurements even with an oscilloscope.

[joeqsmith]

ESR of a capacitor in shunt mode

Starts about 12 minutes in. 

[PartialDischarge]

ESR of a capacitor in shunt mode

Starts about 12 minutes in. 

I know the method, maybe I didn't explain very well. I'm referring to lowering the resonance point on purpose with an L.

[virtualparticles]

Interesting idea. This would probably work well as long as the inductor resonance was high enough to not get in the way.

I can think of an alternative though. Let's imagine you accurately extract the S-parameters of the shunt capacitor alone without the traces or connectors. You would need to use automatic fixture removal to do this properly or adjust gating very carefully. (See https://coppermountaintech.com/automatic-fixture-removal-plug-in/) Once you have the S-Parameters of the Capacitor alone you could examine the real (ESR) and imaginary parts (X) of the impedance using shunt Z conversion.

[JohnG]

The method you are proposing is the operating principle behind the "Q-meter" of old. A good practical reference on this would be the manual for the HP 4342A Q-meter: https://manualzz.com/doc/6376599/hp-4342a-operating-and-service

John

[virtualparticles]

It is! Now there's a blast from the past. 

I remember creating extremely high Q coils for certain G-jobs using copper tubing and polishing the tubes to eek out a bit more Q. Good times.

[nctnico]

https://www.mwrf.com/technologies/test-measurement/article/21849791/copper-mountain-technologies-make-accurate-impedance-measurements-using-a-vna

So I have finally put together a full decent set of equipment for my 300MHz VNA, including a resistive splitter and a directional bridge. I see the different ways of setting the system up offer a good amount of measurement range, but I am a little short on details.
The main point of the thread is to show the error graph, I thought it was ALOT worse for some reason. Maybe this will increase peoples interest, because I thought it was a seriously dodgy solution, but it looks practical, like I thought it was something like 20x the error. The author does a measurement at 90Mhz

I have a few reservations about the graph and the article. I think there is an error in the graph as well. It doesn't make sense that the accuracy for shunt and series measurement are different.

Secondly, AFAIK an error margin is a tolerance. Every piece of test equipment should perform within that tolerance so the measurement result has a defined uncertaintity level. I don't follow why the author of the article converts the tolerance into a 3 sigma value. IMHO that just doesn't apply to measurement uncertainty as you don't know what the error is of the unit that sits in front of you. You have to assume the worst.

Anyway... I went ahead and put an Excel sheet together to make my own graph:



I used an error of 0.2dB (just like the article does) and you can see both shunt and series errors level out towards 2.33% which is the linear error that 0.2dB represents. I have attached my Excel as well. For simplicity and to verify the article, I went for a numeric approach to calculate the numbers instead of deriving a function. The Excel sheet calculates S21, applies the error, calculates Z from that and last but not least calculates the error. For the shunt part the error seems to be limited to 100% but that is due to to formula used for the percentage being unsuitable for that range of the impedance but since that area is not interesting anyway, its not relevant to calculate correctly anyway (so meh).

I also found an interesting article about making impedance measurements using a VNA or a 2 port as well: https://www.picotest.com/Rohde-Schwarz/latin-america-microwave-conference-paper.pdf

This describes a neat trick to shift the measurement range for 2 port shunt or series by adding series resistors for shunt measurements and shunt resistors for series measurements. In my Excel sheet it is possible to change Z0 (the reference impedance) to accomodate for these extra shunt / series resistors. What the resulting measurement range will become depends on the noise floor of the VNA / spectrum analyser that is being used. The Excel sheet also shows the S21 in dB.

Forum member oz2cpu was so kind to make a PCB design availabe that allows to fit these extra resistors as well:
https://www.eevblog.com/forum/projects/component-tester-board-for-sa-na-impedance-caps-inductors-filters/msg4026844/#msg4026844

[gf]

I don't follow why the author of the article converts the tolerance into a 3 sigma value. IMHO that just doesn't apply to measurement uncertainty as you don't know what the error is of the unit that sits in front of you. You have to assume the worst.

Whenever random errors are involved which are drawn from an unbounded probability distribution (like e.g. Gaussian, Chi-Square, etc.), then it does not really make sense to specify a maximum error, because the maximum error were always infinite in this case. It does not help you either to known that the worst error is infinite. A practical compromise is therefore to choose a confidence level (say 95%, or 99%, or 99.9%,...) and to specify the corresponding confidence interval instead of a maximum error.

[nctnico]

I don't follow why the author of the article converts the tolerance into a 3 sigma value. IMHO that just doesn't apply to measurement uncertainty as you don't know what the error is of the unit that sits in front of you. You have to assume the worst.

Whenever random errors are involved which are drawn from an unbounded probability distribution (like e.g. Gaussian, Chi-Square, etc.), then it does not really make sense to specify a maximum error, because the maximum error were always infinite in this case. It does not help you either to known that the worst error is infinite. A practical compromise is therefore to choose a confidence level (say 95%, or 99%, or 99.9%,...) and to specify the corresponding confidence interval instead of a maximum error.

I get that but I don't see how this would relate to test equipment. Typically test equipment has hard boundaries for the accuracy of the results. Outside those boundaries test equipment is considered 'broken' and needs to be repaired / adjusted. IOW: statistical outliers don't exist. For example: nobody wants to buy a multimeter of which 0.3% of the units may have an unknown tolerance.

[coppercone2]

well its bounded by being off scale (the error is so big it hits a rail), so if you have a 99% chance of being at ADC span, that is a bound.

[nctnico]

well its bounded by being off scale (the error is so big it hits a rail), so if you have a 99% chance of being at ADC span, that is a bound.

Ofcourse there is also a limit to the measurement range. However within the specified measurement range the results should be within the tolerance limit.

[virtualparticles]

It's good to question things. If you don't do so, you'll never really "own" the answers that you're looking for.

The accuracy for shunt and series measurements are different because the partial derivatives of the impedances with respect to the reflection coefficients are different. I used standard metrological methods for evaluating uncertainties. The reason why there are 3 sigma standard deviations is that metrology is a statistical science. There are no fixed, "defined" accuracies, only a confidence interval within which one can expect to find a particular measurement. This was explicit in the article. You can't arrive at the correct answers by banging away at it numerically in Excel. I used 3 sigma statistics in particular because almost all manufacturers of test equipment use this confidence interval when setting data sheet specifications. With a 3 sigma confidence interval, we can say that 99.7% of the time, your measurement should fall within a certain error tolerance. It goes without saying that 0.3% of the time, your measurement may be outside of the data sheet tolerance.

There is an error in the chart. As I stated in the article, the expected measurement error does not go to 0 for the shunt measurement. I was simply uninterested in the value at 50 ohms and only concerned about the crossover points with the two other curves.

Adding a series resistor for the Shunt-Thru measurement to extend measurement range is potentially useful, but there are limitations. Calibration must be performed with the resistors in place in order to compensate for them. This creates a problem for for the reflection measurements on each side. If the attenuation of the series resistors looking into the 50 ohm cal kit is greater than about 15 dB, the raw S11 and S22 measurements will suffer and the corrected S21 will have greater inaccuracy. If the attenuation is closer to 20 dB, then the S21 measurement cannot be trusted at all.

For reference, the most relevant doc for VNA uncertainties is "Guidelines on the Evaluation of Vector Network Analyzers (VNA), EURAMET calibration guide No. 12, Version 3.0
https://www.euramet.org/Media/news/I-CAL-GUI-012_Calibration_Guide_No._12.web.pdf

Best,

Brian

[cdev]

It is! Now there's a blast from the past. 

I remember creating extremely high Q coils for certain G-jobs using copper tubing and polishing the tubes to eek out a bit more Q. Good times.

Would you be up to tell us a bit more about this. What was the intended application?

I want to build an Ethernet-connected HF filter board, maybe with some antenna switching too. Kind of like the one in the Hermes Lite 2.

The filters in it will be socket-ed and interchangeable for flexibility. They will be constructed in a DIP format, which should be fine for HF.

[gf]

I used 3 sigma statistics in particular because almost all manufacturers of test equipment use this confidence interval when setting data sheet specifications.

Can you cite some references that promiment manufacturers are really assuming 3 sigma, when they specify +/-U (without augmenting this specification with a coverage factor or percentage)?
Searching the web, I did not really find clear statements. Some search result rather indicated, that 2 sigma / 95% were quite common. So I'm still puzzled
EDIT: Or does any official standard exist, which requires 3 sigma?

[virtualparticles]

You probably won't find anyone stating that kind of information. Part of it has to do with expected production yields, proprietary information. It isn't practical to "cherry-pick" products which meet a tight specification and pitch the failures. Once a production process has been optimized and the product is stable, the final specifications can be determined statistically. These specifications must exceed the goals set forth in the product plan and allow for small process drifts. To be profitable, It is important that a very high percentage of products meet all specifications without re-work. A 5% fall-out rate would not be very exciting. Therefore, 3 sigma or greater compliance to specification is a common requirement.

This applies to the production of hardware. The noise figure of the receivers, the raw source and load match or the directivity of the bridges. But more importantly, each measurement is affected by random factors. Reflection measurements are limited by residual directivity. The leakage signal adds to actual reflections either in or out of phase depending on the distance to the reflection. Transmission measurements like S21 are affected by receiver noise and Raleigh statistics govern the interaction. With a knowledge of the noise figure it is possible to predict that S21 measurements will fall within a certain range. It is not that the VNA is part of a 0.3% that doesn't meet specifications. A series of measurements will fall within a Raleigh distribution. Our data sheet specifications are based on 3 sigma statistics and since equivalent hardware from the other guys have the same specifications, I can state that they are doing the same.

There are two videos which quantify these uncertainties.

https://coppermountaintech.com/video-vna-transmission-measurement-uncertainty/

https://coppermountaintech.com/reflection-vs-transmission-accuracy-in-vector-network-analyzer-measurement/

Hopefully this is helpful

[virtualparticles]

I needed a very high-Q inductor for an avionics filter. (land based).

[jmw]

Reviving since I'm building a setup for PDN measurement and built my own common mode transformer. The core I used was the Vac T60006-L2090-W518, with a 3D printed former to help with the split winding and maintaining minimums on bend radius for conformable RG-402 coax. Better performance than the J2102B in the HF range and UHF range, but worse in the VHF range. The ground loop error is only significant at lower frequencies though, correct? My measurements were done just like the picture on page 3 with pigtails with the shields shorted via a copper sheet. I only have a TG+SA so no vector measurement of S21.

[joeqsmith]

Yes, the cables have more loss as you start going beyond a few MHz. 

Xrunner had found a version of firmware for the H4 VNA that looks to be stable enough to use.   I think my H4 had better performance than my original NanoVNA at these lower frequencies.   I could attempt to remeasure the three standards I made up and see if it does any better. 

Since sorting out the common mode transformers, a few people here sorted out the keys for the Agilent PNA.  It would be easy enough to repeat these test now to get a better idea how mine compares with the data Brian collected from the Picotest transformer above 6GHz. 

My friend flipper is also very interested in this.  We were hoping after a member here had posted about getting the new LiteVNA that they would post some data.  Flipper was tired of waiting and we have new hardware on the way.   This would provide yet one more data point.   

[joeqsmith]

Attached images show the latest LiteVNA 3.1 hardware being used to measure our shunt standards (0.1, 0.025 and 0.001 ohms). 

Capture8 showing my original NanoVNA measuring these same standards.  It's still not great but for this measurement, it by far out performs any of the other low cost VNAs I have looked at. 

Capture10 showing my original NanoVNA with the 0.025, 0.001 and 0.0001 ohm standards.   

[joeqsmith]

Attempting to measure our shunt standards with the H4 with the firmware Xrunner provided (Version: 1.2.08 [p:401, IF:12k, ADC:384k, Lcd:480x320]).

H4_2k_10M_RStandards1:  2kHz to 10MHz with 0.1, 0.025 & 0.001 ohm standards attached.    The H4 was only rated to run to 50kHz and you can see the performance is pretty poor as we move below this.   

H4_2k_10M_RStandards3: 20kHz to 1MHz with 0.025, 0.001 & 0.0001 ohm standards attached.   It's much worse than my original NanoVNA.  When you consider that original NanoVNA was the lowest cost of these, there's a reason I recommended it.