The common resistive Return Loss Bridge discussed the role, characteristics and behavior of an ideal Return Loss Bridge.
Let’s look at the Return Loss Bridge embedded in the NanoVNA-h4 v4.3.
Above is an extract from the schematic for a NanoVNA-h4 v4.3. It includes the source, Return Loss Bridge circuit and detectors. Other NanoVNA versions may have similar implementations, but small differences can have large impact on behavior.
Let’s create a calibrated LTSPICE model of the Return Loss Bridge, source and detector input.
Above is a DC LTSPICE model.
The model assumes that the clock output pin of the si5351A is well approximated by a Thevenin or Norton equivalent circuit, and the output impedance is taken from the datasheet Rev. 1.1 9/18. The datasheet hints that the output impedance is drive level dependent.
Model calcs / measurement
NanoVNA firmware calculates raw S11 as approximately (Va-Vb)/Vs. So let’s tabulate some results from the model for various Zu.
Above are three tables, all are from the model.
Analysis
Note that the the calculated voltage (Va-Vb)/Vs for sc should be -oc, it is not nearly, demonstrating significant departure from symmetry, from ideal. That said, the calculated voltage (Va-Vb)/Vs s11 for the 50Ω load case is quite good.
These values from the model reconcile well with saved sweeps of raw s11 for sc and oc terminations, s11m.
Equivalent source impedance Rth calculated from the model is very close to 50Ω. A load pull test on Port 1 came in very close to 50+j0Ω.
The right hand table compares:
- the model voltage (Va-Vb)/Vs (which is used by the firmware to calculate raw s11 with (Va-Vb);
- the reference voltage Vs; and
- the model bridge unbalance voltage.
Note that:
- Vs, the reference voltage is load dependent.
- (Va-Vb) for oc is very nearly -(Va-Vb) for sc, so this is only a small departure from ideal.
- (Va-Vb) for the 50Ω load is very small, corresponding to raw ReturnLoss=60dB, substantially better than the Vs adjusted case above.
- (Va-Vb)/Vs for oc is quite different to -(Va-Vb)/Vs for sc, a large departure from ideal. So, raw s11 is calculated based on a load dependent reference voltage Vs… building significant error into the raw measurements.
Realist that this is a DC model, but will be a good predictor of AC behavior at say 1MHz, things change as frequency increases and there are step changes where each of the harmonic modes are engaged.
The common resistive Return Loss Bridge set out:
Essential requirements of a good resistive RLB:
-
The source has a Thevenin equivalent source impedance of almost exactly Zref;
-
three of the bridge resistors are almost exactly Zref; and
-
the detector is a two terminal load almost perfectly isolated from ‘ground’ and has an impedance of almost exactly Zref.
The NanoVNA-h4 v4.3 fails two of them:
- the Thevenin equivalent source impedance seen by the bridge itself is not close to Zref (~25Ω); and
- the detector impedance is a long way from Zref and is not isolated, the grounding of R20 and R21 in the original schematic provides a current path to ground which disturbs the bridge. It is hard to understand why this point is grounded as the bridge output voltage is not balanced, nor is one side grounded, so the load needs to be isolated from ground.
Notwithstanding those problems, it appears the source impedance looking into Port 1 in the simulation happens to be close to 50+j0Ω, and a load pull test supports that, but measurement experience is that there is significant Port 1 mismatch error that needs correction.
These all contribute to less than ideal response report in the model results and saved raw s11 measurement.
The response to this criticism might well be a nonchalant “never mind, VNA error correction will paper over all problems.”
Perhaps… though tests on s21 correction indicate that it does not properly deal with error: An experiment with NanoVNA and series through impedance measurement… more.
That article has inspired review of s21 correction, and Enhanced Response Correction has been added to NanoVNA-D v1.2.32 by Dislord, and it has significantly improved s21 measurement accuracy.