Hi everyone,I'm curious about the SA option in PNA-X.It can do specrtum measurements without hardware preselector.In Keysight's help files it says it has software-based preselector function which can eliminate image signals.And in the SA setup there is a image reject menu with six options.I'm not sure how does it works.Does anyone know about the algorithm?And another question is the confidence level.Does this function eliminate all of erroueous signals?Can I trust all of the signals displayed on the screen?Thank you.

Well yes. JP Teyssier and I developed the algorithm. It is based on taking overlapped FFT's over a band of frequencies (default is over about 68 MHz band). Each FFT has a different LO frequency, shifted by something like 68/n+-~1 MHz. N is the image factor (it's slightly more complicated than this, but this is a reasonable approx.) The overlapped FFTs make the RF spectrum appear in different portions of the IF, and we keep careful track of where each signal should be. We then compare the N acquisitions on a FFT bin-by-bin basis to see if in each RF bin, the signal is the same level (meaning a real signal) or if the level changes. If the level changes, then there's a more complicated algorithm to decide if it is a real signal or an image, and if it is an image (because it shows up in only one of the n acquistions, for example) then the other acquisitions are used (usually taking the average or the min of what's left). The exact algorithm is proprietary.

If the signal you are measuring is stationary (doesn't change with time) then it is essentially always correct. If the signal is time varying, but in a non-random way, then if you set the acquisition time to be longer than the variation, it will also be always correct. If the signal is ephemeral (transitory) then it depends how long it stays around (swept SA has the same issue). If it disappears before the N-1 acquistions occur, then it will likely be missed. You can improve the odds of catching it by decreasing the VBW (which actually increases the FFT averaging before image rejection and helps a lot to capture transient signals). If the transient signal comes and goes, you need the FFT acquisition time to be long enough to see it (e.g. a pulsed signal that is on 10 usec and off 1000 usec, with a 2 usec FFT will mostly be missed because the FFT is most likely to occur on the off time; setting the VBW narrow increases the FFT acquisition and ensures the FFT is acquiring when the transient signal is present). Ok. Already too long. Shoot me an email if you have more detailed questions or a specific use case.