Hello VNA team,

for quite a while I am trying to understand how the transmission uncertainty is to be calculated for our PNA-X VNA.

I saw in this forum that you are usually referring to document "08720-90393" for questions about VNA measurement

analysis. I have tried to use the datasheet in combination with the formula provided in your reference document.

However the results do not seem to be useful. I specially realized that when I was trying to compute the phase uncertainty

which fails due to an argument being larger than "1". Basically the error part computed is always smaller (about one

order of magnitude) as compared to the "real" S21. Hence Etm/S21 leads to some value >1.

I have used linear values for the effective system data, however, I think that my problem is the conversion of the

trace noise as well as the noise floor values to linear quantities.

I have used the following calculation PnoiseflorLIN = 10^(NoisefloordB/10) * 0,001 and TracenoiseLin = 10^(TracenoisedBrms/20) - 1

Is this correct or am I doing a mistake here. The order of magnitude for the systematic error seems to be ok, so probably

it is only an error in the noise calculation. (BTW, I calculate the crosstalk contribution as follows XTalkLIn = 10^(XTalkdB/20)).

If this is ok so far, do you have an idea what can lead to the calculation error for the S21phase part? Am I probably missing any

required normalization?

Thanks very much for your help.

MFinner

for quite a while I am trying to understand how the transmission uncertainty is to be calculated for our PNA-X VNA.

I saw in this forum that you are usually referring to document "08720-90393" for questions about VNA measurement

analysis. I have tried to use the datasheet in combination with the formula provided in your reference document.

However the results do not seem to be useful. I specially realized that when I was trying to compute the phase uncertainty

which fails due to an argument being larger than "1". Basically the error part computed is always smaller (about one

order of magnitude) as compared to the "real" S21. Hence Etm/S21 leads to some value >1.

I have used linear values for the effective system data, however, I think that my problem is the conversion of the

trace noise as well as the noise floor values to linear quantities.

I have used the following calculation PnoiseflorLIN = 10^(NoisefloordB/10) * 0,001 and TracenoiseLin = 10^(TracenoisedBrms/20) - 1

Is this correct or am I doing a mistake here. The order of magnitude for the systematic error seems to be ok, so probably

it is only an error in the noise calculation. (BTW, I calculate the crosstalk contribution as follows XTalkLIn = 10^(XTalkdB/20)).

If this is ok so far, do you have an idea what can lead to the calculation error for the S21phase part? Am I probably missing any

required normalization?

Thanks very much for your help.

MFinner

Could you, however, comment on the calculation of the linear values from tracenoise and noisefloor.

I am just curious to learn from the experts on how to compute these linear representations - or is it already ok to do it the way I

suggested the calculation.

Best regards

MFinner

I am really interested in getting a deeper insight in these calculations and it already took me quite some time and effort to get me to

the actual status of insight I have to the topic of measurement uncertainty calculation. I understand that you are quite busy in answering

various questions in the field of network analysis. However, I would like to stress that I would be really helpful if you could give me

an answer concerning my questions about transe noise and noise floor calculations. Could you please give me a hint on what the correct

conversion from dBRMS of the tracenoise to a linear value as well as the conversion from the noisefloor to its linear representation has to

look like? Your help is highly appreciated.

Best regards

MFinner

But, to you answer, you must also know the power level the signal to unserstand the effect of noise floor. Trace noise is a bit simpler, and I think you understand it already;

In CW, take the rms value of trace noise, in dB; convert to linear using 10^(x/20), then subtract from 1, take the absolute value, then take 20 log10 of that. That is the trace noise noise floor, it is the same regardless of power level, until it falls below the receiver power level (the trace noise comes mostly from phase noise).

For noise floor, take the source power times S21 to get the output power, compare it to the noise floor, to get a dBc value. If it is below the noise floor from trace noise, ignore it (or add it in as a linear sense to be fully rigorous).

Take the value on noise floor (dBc noise floor now), convert to linear, add/subtract to the linear S21, convert each to log to upper and lower errors.

I need to think this over and maybe come back to you at a later stage.

Thanks very much for the moment.

Best regards

MFinner