I have a question about the amplitude of my spectrum analyzer. I'm producing upconverted white noise over some bandwidth (say 150 MHz), in the 7 GHz range. In this sense everything is as expected; there is a 150 MHz white noise band, which drops off to the noise floor on both sides. But the amplitude of the noise, this is behaving rather unexpectedly when varying the RBW. Note that the spectrum analyzer used in all that follows is the E4407B Spectrum Analyzer.
I understand the situation as follows: the spectrum analyzer shows the power in dBm/RBW, so to convert it to dBm/Hz one has to incorporate the RBW. I do so by converting the power to Watts (10^(dBm/10)*10^-3) and then dividing by the RBW; this should be the proper Watts/Hz power spectral density, should it not?
Now what puzzles me is that these values are not consistent for different RBW. Taking values between 100 kHz and 10 MHz, each time I end up with a vastly different Watts/Hz power spectral density, differing by substantial factors. How is this possible? I thought that the main factor in choosing the RBW was just that smaller RBWs mean your trace takes longer and is more accurate, but the scaling seems absurd. How do I determine which one is the most faithful representation of the true PSD, if any?
As for the calibration of the spectrum analyzer itself, I know that in principle the amplitudes can be trusted; using a signal generator generating a 20 MHz signal at some known power, I get a single peak with this exact power on the spectrum analyzer.