Spectrum analyzer DANL: measuring low-level spurs without fooling yourself

Displayed average noise level looks like a single number on a datasheet, but on the bench it moves with almost every setting you touch. When I need to measure a low-level spur that sits close to the noise floor, I work through a fixed checklist before trusting any reading.
First, set the resolution bandwidth deliberately. Narrower RBW lowers the displayed noise, but it also slows the sweep and can hide a spur that drifts. I start wide to find the spur, then narrow the RBW in steps and watch whether the spur level stays stable while the noise floor drops. If the spur moves with RBW, it is probably not a real continuous signal.
Second, check the attenuator and preamp combination. Too much input attenuation buries the spur; too little risks compression and creates false spurs of its own. I step the attenuation and confirm the spur level changes by the expected amount. A spur that does not track attenuation is usually generated inside the analyzer.
Third, use averaging correctly. Trace averaging reduces displayed noise variance, but it does not improve DANL the way a narrower RBW does, and it can smooth away an intermittent spur. For drifting or bursty signals, I compare a max-hold trace against an averaged trace rather than trusting either one alone.
Fourth, verify with a different span and sweep time. A spur that only appears at one span or one sweep setting deserves suspicion: check for image responses, LO feedthrough, and residual responses by shifting the center frequency slightly. A real spur stays put; an analyzer artifact moves or vanishes.
Finally, record the full setup: RBW, VBW, attenuation, preamp state, and detector type. Two engineers measuring the same spur with different settings will get different numbers, and without the setup the numbers are not comparable.
What is your go-to sanity check when a spur sits within a few dB of the noise floor? Do you trust max-hold, averaging, or a narrower RBW first?

Thank you, RF_Bench_Notes — the “DANL is not a single number” framing, plus the RBW / attenuator-step / max-hold vs average / residual-response checklist, is exactly what we will keep linking under every “spur that sits on the noise floor” report.

What we will keep quoting:

1. RBW vs spur stability. Start wide to find the candidate, then narrow RBW in steps and watch whether the spur level stays put while the floor drops. A “spur” that moves or collapses with RBW is usually not a continuous tone worth quoting.

2. Attenuator-step false-spur test. Step input attenuation and require the spur to track by the expected amount. A spur that does not track is generated inside the analyser (mixer / residual / compression), not at the DUT.

3. Max-hold vs average for intermittent energy. Trace averaging lowers displayed variance but does not buy DANL the way a narrower RBW does, and it can erase a bursty spur. Compare max-hold against an averaged trace before calling the spur absent or stable.

4. Residual / image checks. Shift the centre frequency slightly (and change span / sweep time). A real spur stays at the same absolute frequency; an image, LO feedthrough, or residual response moves or vanishes. Record RBW, VBW, attenuation, preamp state, and detector type next to every number.

Open question for the room: when a candidate spur sits within ~3 dB of the displayed floor, which single check do you run first on a mid-range SA — attenuator step, centre-frequency shift for image/residual, or max-hold vs average — and at what margin above the floor do you stop trusting a single-trace reading without that checklist?

— rfforum 编辑部 / Editorial Team