Setup an S11 (reflection) or S21 (transmission measurement). you can either use a CW sweep or a frequency sweep. for the S21, make sure that you have an actual thru connected. for the S11, you can leave the port open or connect an open or a short calibration standard to the port. set the IF Bandwidth to the value that you want to measure trace noise levels at and then setup some averaging. if you are measuring a small IF Bandwidth, you don't need as many averages. my typical rule of thumb is to use 20 averages for 1 KHz IF Bandwidth. let the channel sweep until the averaging is complete. once the averaging is complete, while the channel is still sweeping, normalize (data -> memory and then data/memory) your S11 or S21 trace and then turn averaging off. if your analyzer supports it, turn on trace statistics over the full band of the measurement and the trace noise levels are the peak-to-peak values reported by the trace statistic function. if you don't have trace statistics, then just use 2 markers, one to max (tracking on) and one to min (tracking on) and subtract the two values.
Thanks a lot from detailed information. If the same test doing for 8510C VNA which has fixed IF bandwidth. Our default averaging factor is always 1024, so how this test method would use then? Would it do then, like taking first 20 sweeps (avg 1024) and calculating average from that data set (using external PC-software, which reading those each sweeps). After that (calculated averaged data set) / (one sweep result, avg factor 1024). Finding then max and min points from calculated value and subtract them from each other?
taking more averages before doing the normalization is not a problem. so if you want to take 1024 averages, that is fine; it will just take longer. but after you normalize, you'll have to take at least one more sweep without any averaging and divide it by the data in memory that is averaged. I never used 8510's, so I can't tell you the exact sequence of steps on that box, but the general concept should work on any VNA
Trace noise is defined in a CW state, so you need to set to a fixed frequency to charaterize trace noise.
In the 8510, the trace noise is defined (as I remember) with no averaging (and as you say, the 8510 has a fixed IF BW).
the 8510 did not have spec on trace noise, that I remember, but newer analyzers have a spec based on the RMS value of trace noise. (spec means warranted performance; it might have had a typical value).
To measure the trace noise take a single sweep (no need to normalize and average because you are at a fixed frequency). Read out the data trace and use a stats package to compute the standard deviation. This represents the rms trace noise.
As daras said, use a short for S11, use a thru for S21.
Thanks a lot from this adivce! It's rather time consuming to measure and calculate each CW frequency points up to 26.5GHz :-) Would it give rather good estimation of systems trace noise performance if takes reasonable set of frequecy points from each bands of 8510C?
yes, you can do freq sweep but it will give poorer results. How much poorer? it depends on the hardware performance, but I would guess 2x worse. We specify trace noise in CW so if you want to verify the spec, you really must meas in that way.
In the 8510, the trace noise is defined (as I remember) with no averaging (and as you say, the 8510 has a fixed IF BW).
the 8510 did not have spec on trace noise, that I remember, but newer analyzers have a spec based on the RMS value of trace noise. (spec means warranted performance; it might have had a typical value).
To measure the trace noise take a single sweep (no need to normalize and average because you are at a fixed frequency). Read out the data trace and use a stats package to compute the standard deviation. This represents the rms trace noise.
As daras said, use a short for S11, use a thru for S21.