NanoVAN Test Fixture - Capacitor results

Here I present two results: measuring a supposedly high Q 110pF capacitor with and without the 50 ohm chip resistor in series.

110pF capacitor in series with 50 ohm chip resistor

Immediately some questions are posed, and some pleasing results. VNA suggest 110pF capacitor is actually 112pF. This is true over most of the HF range but by 54MHz the effective capacitance has already fallen to 109pF. It gets progressively worse after that. Worth knowing in the design of filters and matching networks.

Overall the fixture with a 50 ohm series resistor appears useful.

110 pF capacitor with no series resistor

Keen readers will notice that the 110pF capacitor, tested in this fixture, measured 110pF, not 112pF. That could be because I lost the first capacitor in the shed floor. Sorry, but the time taken to present these results didn't warrant re-measuring the capacitor in the first fixture. ( Seconds to measure and view on the nanoVNA screen, many minutes to collate and present. )

What is interesting is that while the nanoVNA display showed a trace around the edge of the Smith Chart, the software did not. Is this evidence that at the extremes there are errors as data is transferred and manipulated to re-display the trace, or underlying accuracy issues with the NanoVNA? Loss of one too many decimal places? The S2P file I exported with "nanoVNA partner" went to 9 decimal places, however that does not mean the software worked at that level of precision. Displaying answers to 4 decimal places is meaningless without knowing the precision of the underlying calculations.



Viewed on the nanoVNA it suggested by the software the capacitor is useful up to 2m before it departs from being a 110pF capacitor.

At this point both test fixtures have, within reason, results and findings consistent with each other.

I then examined in turn capacitors of 100nF, 10nF, 1.5nF and 150pF. The nanoVNA screen clearly showed where these capacitors departed from being the stated value and confirmed my understanding of why the value and type of decoupling capacitor requires thought. Unfortunately, with no series resistor once the readings are transferred to the PC the data translation/transfer/rounding issues prevent a meaningful display of the results when the S parameter file is later opened with "nanoVNA partner".

However, all is not lost.  The exported S2P file, viewed in RFSIM99, is a close representation of what the nanoVNA screen was showing.

100nF capacitor, no series resistor

The 100nF 1206 capacitor measured 93nF at 50kHz. At 9MHz it was already a 1nH inductor. Not what I expected from a 1206 part. A trip to the workshop to check. Indeed, this seems correct. While the 50 ohm resistor fixture showed the transition was at 18MHz, the difference between the two is within the accuracy I wanted.

For decoupling purposes the series impedance at 36MHz is 0.05R+j0.65R. Above that the decoupling gets worse.

How does 10nF look? A 10nF 1206 "recovered" capacitor was tested. The series impedance at 72MHz is 0.09R+j0.63R, getting worse as frequency rose beyond that.

The 1.5nF capacitor demonstrated why two values of decoupling capacitor are sometimes shown. It would be poor at decoupling until the frequency rose to 72MHz. Above 300MHz the response was erratic so this needs checking.

The 150pF capacitor was unaffected by parasitics until the frequency rose to 234MHz.

I overlaid the results of 3 39pF capacitors of different construction, mica clad, 1206 ceramic and a SMD mica? film capacitor. There was no real difference until the frequencies rose above 300MHz. After this all displayed a similar erratic pattern though it was shifted.

3 39pF capacitors of different construction.

Advantages of 50 ohm test fixture

There is some suggestion in published papers that this fixture should have greater accuracy. I see nothing in the results to date to support this. Perhaps my test fixture for SMD parts was better than hte authors test fixture.

In any event, since you have to build one of these to calibrate the other fixture ( OSL ) then for now I put aside the "accuracy" debate. 

Advantages of the test fixture without 50 ohms

A/B comparisons of decoupling applications at spot frequencies gets easy since for the same frequency A is worse than B if it is further away from the 0+j0 point of the Smith Chart.

Conclusion to date

Either fixture appears useful at characterizing capacitors down to very small values. The lowest I tested was 1.5pF and the results gave me confidence in the value and presence of parasitics.
Accuracy appears degraded above 300MHz but for my purposes will still be adequate.

For A/B comparisons I would use the test fixture without the series resistor since the deviation is easier to spot on the small display.