I would like to better understand what dynamic uncertainty is and how it should be used in uncertainty calculations.

For example, lets say that a measurement of the substitution loss of a step attenuator is required. Two VNA transmission measurements will be made: one with the step attenuator in the datum position and one in the step position. Then the SL could be calculated by taking the ratio of the results, etc.

Dynamic accuracy is, if I understand correctly, predominantly associated with linearity errors in this type of measurement. Is that correct?

Agilent presents DA as associated with a sytematic effect, which suggests that the errors do not change from one measurement to the next (unless the measurement conditions change). So, is there some potential for these errors to combine and cancel in compound measurements?

In a single transmission measurement (S21 say) two amplitudes are measured on different channels, and ratioed to get the transmission (reference and testport). There will be an error in each measured amplitude, but it is the net effect of these errors in different signal channels (reference and testport) that is presumably reported in specifications of the DA. Since the receivers are different, I suppose that it is fair to say that the errors are independent? In that case, the DA specification will be larger than the uncertainties associated with the two individual errors.

So, what should one think in the case of the SL measurement, which combines two transmission measurements?

One might expect that the (two) errors associated with the linearity of the (two) reference measurements will cancel (to some extent), because the reference level is unchanged and the receiver is the same. That would leave the (two) measurement errors at the test port, which have different amplitudes (one is attenuated). But how would the uncertainty of the combined error be related to the DA specifications? Since these errors now arise on the same receiver channel, I would (naively) expect there to be some cancellation (correlation) of the resultant error here too.

Does this makes any sense?

Comments would be appreciated.

For example, lets say that a measurement of the substitution loss of a step attenuator is required. Two VNA transmission measurements will be made: one with the step attenuator in the datum position and one in the step position. Then the SL could be calculated by taking the ratio of the results, etc.

Dynamic accuracy is, if I understand correctly, predominantly associated with linearity errors in this type of measurement. Is that correct?

Agilent presents DA as associated with a sytematic effect, which suggests that the errors do not change from one measurement to the next (unless the measurement conditions change). So, is there some potential for these errors to combine and cancel in compound measurements?

In a single transmission measurement (S21 say) two amplitudes are measured on different channels, and ratioed to get the transmission (reference and testport). There will be an error in each measured amplitude, but it is the net effect of these errors in different signal channels (reference and testport) that is presumably reported in specifications of the DA. Since the receivers are different, I suppose that it is fair to say that the errors are independent? In that case, the DA specification will be larger than the uncertainties associated with the two individual errors.

So, what should one think in the case of the SL measurement, which combines two transmission measurements?

One might expect that the (two) errors associated with the linearity of the (two) reference measurements will cancel (to some extent), because the reference level is unchanged and the receiver is the same. That would leave the (two) measurement errors at the test port, which have different amplitudes (one is attenuated). But how would the uncertainty of the combined error be related to the DA specifications? Since these errors now arise on the same receiver channel, I would (naively) expect there to be some cancellation (correlation) of the resultant error here too.

Does this makes any sense?

Comments would be appreciated.

However, the reference receivers are typically more linear than the test receviers, as they are usually padded down 5-10 dB to ensure they don't go into any compression even at highest test port powers. Further, in the newest instruments, the dynamic accuracy is becoming vanishingly small. I've posted a plot of the new specifications. Typical performance is less than 0.01 dB over an 80 dB range, and I have personally verified the performance down to -120 dBm

So, to account for dynamic accuracy, look on the curve at the power in the receiver for the reference measurement (i.e., where the calibration was peformed) and look for the change in power from that level for the effect of dynamic accuracy. In the case you mention, if the reference measuremnt is made at -20 dBm, and you measure a 50 dB attenuator, you must account 0.022 or so dB for the dynamic accuracy of the receiver.