I was wondering, what integration time does the noise receiver of the 029 option in the PNA-X use for noise measurements?

Based on the typical noise figure trace noise of 0.07 dB, one might assume that 1 ms per frequency point is used. Is this value fixed for all bandwidths?

Furthermore, I found this in the help file:

"Increase the number of averages to reduce jitter. This also increases measurement speed."

(http://na.support.keysight.com/pna/help/latest/Applications/Noise_Figure.htm)

If only this could be true.

Based on the typical noise figure trace noise of 0.07 dB, one might assume that 1 ms per frequency point is used. Is this value fixed for all bandwidths?

Furthermore, I found this in the help file:

"Increase the number of averages to reduce jitter. This also increases measurement speed."

(http://na.support.keysight.com/pna/help/latest/Applications/Noise_Figure.htm)

If only this could be true.

For a sufficiently large BW-time product, the errors can be estimated by

sigma(T)/T = 1 / sqrt(Navg * BW * t).

This is how I derived the 1ms integration time.

Now if the integration time is also dependent on the bandwidth, it is difficult to use this equation.

Do you have any more quantitative information?

Or is the integration time inversely proportional to the bandwidth? This way increasing bandwidth does not improve the error.

I am sorry, I currently have no access to our PNA-X, so I am unable to perform tests.

In 029, a different receiver and digitizer is used, and the digitizer first converts to power (at a high sample rate) and then does (I believe) 10,000 averages of the samels for each "noise acquisition". The BW of this receiver is set by an analog filter (of a few values, like 4 MHz, 8 MHz, 24 MHz and some lower ones). I think you are close on the acquision time,the digitizer has a real rate of 100 nsec noise measurement with the 10,000 integration to give your time of 1 msec. Each noise average adds another 10,000 acquisitons. This doesn't change with IF BW.