An experiment with NanoVNA and series through impedance measurement concluded with:
The NanoVNA-H4 v4.3 and Dislord firmware v1.2.30 using 5 term calibration does not appear to give good accuracy on series through measurement of impedance.
The method
The focus is on the corrected s21 measurement as that is what drives the conversion to series through impedance.
The test fixture was designed to be very close to ideal at 1MHz, the test Zx is a 1% SM 200Ω resistor, taken as being 200+j0Ω.
The method of error correction based on a one path calibration does not correct all errors. In particular it does not correct load mismatch error, but an ‘enhanced response’ calibration of s21 (done properly) should correct source mismatch errors. Of course there remain other errors that cannot be corrected but in a good implementation and effective application, they should be relatively insignificant.
Above is an extract of the published schematic of the NanoVNA-H4 v4.3, the circuitry around Port 1 and Port 2 jacks is of particular interest.
Some notable observations:
- raw (ie uncorrected) s21 measurement reported by the combination of hardware and firmware tested has a gross error, phase of s21 at low frequencies is approximately 180° where it should be approximately 0°;
- cursory circuit analysis of the Return Loss Bridge (RLB) associated with Port 1 would suggest that the ‘detector’ impedance (the circuitry connected between both sides of the RLB to measure the imbalance or ReturnLoss) would appear to be substantially higher than 50+j0Ω which compromises the bridge; and
- cursory circuit analysis of Port 2 circuity suggests that the designer did attempt to obtain a Port 2 load impedance close to 50+j0Ω, measurement indicates quite good ReturnLoss.
A superficial response to observation #1 is “never mind, correction will fix it”, but it does question the rigor of the firmware design and testing.
Same thing for observation #2, no need to design hardware to have an equivalent source impedance of 50+j0Ω when correction fixes it. But in this case, the corrected s21 reported does not fix it, the firmware is probably defective (or lacks Enhanced Response correction).
Observation 3 is very important as it is uncorrected in the simple one path calibration. This area has been redesigned with successive hardware versions, and is fairly good in this version. ReturnLoss could be improved with an external precision attenuator… at cost of dynamic range. Note, “precision attenuator”.
Let’s run a series of tests, same fixture and parts used for all tests.
Using the NanoVNA measurement with internal correction
From the dataset used in the previous article, internally corrected measurement of the series through 200Ω DUT yielded s21=0.305207264 -0.000040260.
This is the scenario discussed at An experiment with NanoVNA and series through impedance measurement which concluded:
The NanoVNA-H4 v4.3 and Dislord firmware v1.2.30 using 5 term calibration does not appear to give good accuracy on series through measurement of impedance.
Using the NanoVNA for raw measurement with external correction
A set of raw measurement were made for SOLT calibration and saved to .sxp files. A raw s21 measurement of the series through 200Ω DUT.
In both the calibration and measurement files with relevant s21 data, the s21 value needed inversion (see observation #1 above).
Scikit-RF was used to correct the raw measurement and obtain s21=(0.33152261949111933-0.0045085939553834485j).
Now this is significantly different, it is actually quite close to expectation (based on DC measurement of the DUT and s11 reflection measurement).
Using the VNWA-3E measurement with internal correction
The same test was conducted using a one path calibration of the VNWA-3E.
Above is a screenshot of the sweep 1-30MHz. Note how flat the lines are, the fixture and DUT do not seem to have significant imperfections.
Above is calculation of Zx using the measured internally corrected s21. It appears very close to expectation (based on DC measurement of the DUT and s11 reflection measurement).
Conclusions
The NanoVNA-H4 v4.3 NanoVNA-D v1.2.30 using 5 term calibration does not appear to give good accuracy on series through measurement of impedance.
Further investigation identifies several possible contributions, the main one being the s21 correction algorithm and implementation.