One of the most stressful things in the lab is hearing (or saying):
“Is the analyzer broken? The noise floor is suddenly 10 dB higher today.”
Because this happens a lot.
Yesterday the bench looked clean. Same instrument, same frequency range, same cables (supposedly).
Today you power up and the entire floor is lifted. It feels like the whole system got worse overnight.
In reality, most of the time the analyzer didn’t “suddenly get bad,” and the DUT didn’t magically degrade either.
Much more often, a small variable changed—and what you’re calling “noise floor” is no longer defined the same way as yesterday.
This is the 10-minute sanity check workflow I use in R&D debugging.
The goal is not to find the “ultimate root cause” immediately. The goal is to quickly eliminate the most common traps so you don’t spend half a day guessing.
First: confirm you’re even comparing the same “noise floor”
The “noise floor” you see on a spectrum analyzer is not a fixed number. It depends heavily on measurement settings.
If the definition isn’t aligned, comparing “today vs yesterday” is a guaranteed trap.
At minimum, align these:
- RBW (Resolution Bandwidth): if RBW increases, the displayed noise floor rises—because you’re integrating more noise bandwidth.
- VBW (Video Bandwidth): VBW affects smoothing and the apparent stability of the trace.
- Detector / Trace mode: different detectors present noise differently.
- Averaging: averaging mode and method can change what you perceive as the “floor.”
I’ve seen “10 dB drift” cases where the entire mystery was:
RBW was 10 kHz yesterday, and 100 kHz today.
Nothing was broken—only the definition changed.
My 10-minute sanity check (the order matters)
You can copy this workflow directly. If you manage a lab team, it’s worth printing as a “noise floor anomaly checklist.”
Step 1: Lock the “four critical settings”
Before touching the DUT, confirm and record (screenshot is fine):
- Center / Span
- RBW / VBW
- Detector + Averaging
- Atten / Preamp / Ref Level
Simple reason:
Many noise floor “problems” are actually settings drift.
Step 2: Quickly rule out front-end overload/compression (the fake noise floor)
This is a blunt but effective move:
- Increase input attenuation (Atten) by one step (or adjust Ref Level by a step)
- Watch whether the floor behaves more “reasonably”
Near overload/compression, the analyzer can show a floor that is lifted—and sometimes even looks “stable.”
The dangerous part is: it doesn’t always look like obvious clipping.
So before blaming the DUT, push the front end back into a clearly linear operating region.
Step 3: Align Preamp / Atten states (a very common real cause)
People remember frequency and span—but often forget two switches that can completely change the floor:
- Preamp ON yesterday, OFF today (or vice versa)
- Atten 0 dB yesterday, 20 dB today (or vice versa)
If these aren’t aligned, 5–10 dB differences are not surprising.
And it’s easy to miss because the screen can still “look similar.”
Step 4: Turn the input into a known condition (separate variables)
If you’re connected to a DUT, you have too many variables: DUT state, cable routing, power ripple, clocks, transmit modes, etc.
So I usually do one practical move:
Temporarily switch the input to a known condition, such as:
- a 50 Ω termination (to confirm the port is clean)
- or a stable source / reference path (to confirm the system behaves predictably)
This isn’t about absolute accuracy. It answers the key question:
Is the lifted floor coming from the environment/measurement chain, or from the DUT?
If the floor is still high with a 50 Ω load, stop chasing DUT behavior and go back to settings/chain.
If the floor is clean on 50 Ω and rises only with the DUT connected, your direction is clear.
Step 5: Don’t ignore “today the environment is noisier”
Sometimes it’s not the analyzer or the DUT—it’s the lab environment.
Real examples I’ve seen:
- a new switching PSU / laptop / monitor nearby
- a shielding box lid not fully closed
- cable routing changed and moved closer to a noise source
- the DUT is in a different state (clock/Tx enabled) than you assumed
A quick environment test:
Shorten/re-route the cable, move the setup, add temporary shielding, or power down suspicious nearby devices—and see if the floor immediately drops.
If the floor tracks the environment, the problem is not “instrument accuracy.”
My probability order (most common → less common)
1) RBW/VBW/Detector/Averaging not aligned
2) Preamp/Atten/Ref Level not aligned
3) Overload/compression creating a fake floor
4) Cable/connector/termination issues (contact, loose termination, mechanical stress)
5) External interference / environment changes
6) DUT state changes (Tx/clock/power ripple/mode)
You’ll notice: “the analyzer is broken” is usually far down the list.
A mindset that saves time
“Noise floor up by 10 dB” feels like it demands one perfect root cause.
But in terms of engineering efficiency, the smarter move is:
Lock the definition, restore linear headroom, convert the input to a known condition, then isolate environment vs DUT.
Once you do that, many “mysteries” become simple.
If you’re dealing with this right now, send me a few lines:
- frequency range + span
- RBW/VBW + whether averaging is enabled
- Atten + Preamp state
- whether the input is connected to a DUT or a 50 Ω termination
I can help you prioritize what to check first so you don’t waste hours in the lab.
Website: https://maronlabs.com
Email: contact@maronlabs.com
Top comments (0)