switchabl 2020/09/12 15:16
@Marco
Yes, thank you! I should have done the math. Maybe I was also fooled a bit by the zoomed in plot. So the measurement from ok1vaw is maybe not ideal, but not as awful as I thought (and the port 2 return loss of his device is clearly quite good for a NanoVNA).
@ok1vaw
Yes, in principle with SOL calibration you have full correction of directivity, source match and tracking errors for 1-port measurements. In practice, there are factors that cannot be corrected, so residual errors remain:
- Calibration standards: the calibration is only as good as the standards. Or rather the characterization of the standards, they do not need to be perfect, but you need to know their parameters.
- Non-static effects: calibration is done at a single point in time and of course it cannot correct for anything that changes after that, like
* drift in your couplers (through temperature drift and relaxation processes)
* flexing your test port cables (and resulting phase/loss change)
* connector repeatability (the connectors of your DUT will be different from those of your calkit and connecting them at a different angle may change the results)
* in particular: bad cables/connectors
- Non-linear effects: the error model assumes that the errors can be described by a linear two-port, but there may be non-linear errors
* non-linearity of the receiver (e.g. the IF amp of the NanoVNA V2 has different gain settings and if the gain calibration for the different settings is not perfect, the resulting errors can not be corrected by a linear network)
* non-linearity of the source (this is normally not an issue because the source output is measured by the reference receiver through a power splitter/coupler, but the NanoVNA doesn't have one)
- "Actual" errors (as in, mistakes):
* software bugs (the 2-port calibration in NanoVNA-saver looks strange to me and I think it may be wrong, but have to look at it again in more detail; 1-port looks fine)
* user error (maybe the most common one :-)
If we wanted to have real confidence in the accuracy of the NanoVNA, in theory all these would have to be characterized.
You can tell that the ripple in our measurements is very likely due to residual errors if you consider the time domain. To get 400MHz ripple, you need two reflections with a delay of 2.5ns (roughly 2x30cm...), and it doesn't seem reasonable that the port 2 circuit can generate a reflection with that much delay. And indeed, looking at the impulse response calculated from my measurement, there is a second (small) reflection at 2.5ns negative (!) delay. Some of the signal is reflected before it even reaches the reference plane. So clearly this is a residual reflection at port 1 that has not been completely corrected through calibration.