1
\$\begingroup\$

Issue: My lab power supply lines (2-3 feet) pick up noise that propagates into my prototype electronics.

Noise desc.: Multiple bursts (10-20 periods in each burst) of 43 MHz sinus-like waves in a certain pattern repeated every 32.5 us (i.e. the sequence of bursts repeats itself every 32.5 us).

Most likely source: I believe the noise is generated by the IKEA light strips I have installed over my workbench. When I turn the light off the noise goes away and it comes right back the second I turn the light back on (100% repeatable). When I dim the light, the 32.4 us repetition period gets longer (it clearly follows how much I dim the light - 100% repeatable).

I speculate (more investigative work needs to be done here) that the AC/DC power supply for the LED light strips is a switch mode PS and that is the source of the noise. I can imagine how the light strip converts the conducted emission (from the switch mode PS) to radiated emission. The long light strips are almost perfect antennas. But as I said this last paragraph is pure speculation.

Setup: I have my lab PS and prototype board separated by ~2 feet and the power lines are not much longer. Since these power cables are not twisted (creates a big loop) it sort makes sense that the setup is susceptible to radiated electromagnetic waves.

Question: But my oscilloscope (350 MHz BW) picks up the same amount of noise when I connect my probe to a 2 feet long cat 5 UTP cable (the 2 leads are connected at the other end, i.e. it is a closed loop system). I'm trying to figure out what kind of coupling that takes place here? What kind of coupling is not suppressed by the twisting? There is no difference between a loop created from a twisted pair and a loop created from an not twisted pair (2 feet out, 2 feet home and 6" apart) in terms of the amount of noise (40 mV pp) I see on my oscilloscope (into a 10 Mohm impedance)... obviously, I am not an RF engineer (I apologize for my ignorance).

Repeated sequence of bursts of noise: enter image description here

A single burst of noise: enter image description here


(The following is added on Jan 14, 2024 based on the comments below)

I inserted a common mode choke in the AC power line to my scope but unfortunately it didn't yield a meaningful difference (see pics below). Maybe the burst dies out quicker but the peak-to-peak is almost bigger now.

Without common mode choke: enter image description here

With common mode choke: enter image description here

Common mode choke: enter image description here

\$\endgroup\$
9
  • \$\begingroup\$ Two questions are verboten here. Don't be surprised if this gets removed. \$\endgroup\$ Jan 13 at 0:13
  • \$\begingroup\$ Thank you… deleted the 2nd question. \$\endgroup\$
    – t34hansen
    Jan 13 at 0:24
  • 1
    \$\begingroup\$ What probe are you using? Do you see the same noise if you connect probe tip to ground clip to cable? \$\endgroup\$ Jan 13 at 2:50
  • \$\begingroup\$ The standard single ended probe that came with the scope. The signal is clean when I connect the ground clip to the probe tip. I also tried to supply the power to the LED light strip from a different power panel (I have two main panels in my home) to make sure the noise didn’t come into the scope via the power line (but that didn’t change the outcome). \$\endgroup\$
    – t34hansen
    Jan 13 at 3:44
  • \$\begingroup\$ Try putting a common mode choke (clamp-on ferrite type) on the oscilloscope power cord. It could be you're seeing AC line conduction. Try reversing the AC plug, if that's possible. Grounding the tip doesn't get rid of induction pickup, but does reduce capacitive pickup, and one two-prong polarity can have more of that than the other. \$\endgroup\$
    – Whit3rd
    Jan 13 at 4:04

1 Answer 1

1
\$\begingroup\$

I have spent a considerable amount of time trying to find the answer to the question I raised above - both by debugging in the lab as well (and mostly) by building a Spice model of my setup (more on that further down). First of all, thanks to Whit3rd and Tim Williams for guiding me down the 'common mode noise' path (w/o that investigation & what I learned from it, I would never have reached the conclusion I did). I am going to leave out detailed descriptions of failed ideas (to shorten this write up) but will mention them and the outcomes of those investigations.

Common Mode Noise Injected into the AC Power Supply Lines

Based on the comments above, I decided it was worth to look into the possibility of common mode noise somehow getting injected into the AC power supply lines to my oscilloscope (and thus into the vertical amplifier and eventually to be displayed on the screen of the scope).

As mentioned above, I built a Spice model of my setup (partly included further down). I assumed common mode noise from the LED PS unit made it via the AC power lines into the oscilloscope. At this point it doesn't really matter how that noise made it into the AC power lines, conducted or some kind of coupling - in Spice I used a step function from GND to both the hot an neutral to simulate the injection of common mode noise as result of the change in load that happens when the LED PS switches on/off. I assumed stray capacitance coupled the common mode noise from the primary coil of the PS transformer in the scope to the secondary coil and thus into the DC power line to the vertical amplifier (the rest is shown in the schematic included below). Irrespective of what I did (and I spent days trying different schemes out), I couldn't replicate (qualitatively) what I saw in the lab. Sure - the Spice simulation did show a lot of noise (i.e. differential voltage that would be shown on the screen) but nothing that qualitatively came close to what I saw on the screen in the lab. Then I added the common mode choke to my Spice simulation (I also did this in the lab a few days earlier - the original question was updated on Jan 14, 2024 to reflect this). In my Spice simulation the choke did wonders - absolutely wonders! All the differential noise that would be displayed on the screen was gone! In real life not so much if any change at all. I know the pictures posted under the question indicate minute improvements but I think this is probably a coincidence. My guess is I snapped the picture when the scope triggered on a smaller burst - I regret posting that picture instead of going back to my lab to retake it. Finally, if the issue is common mode noise injected into the scope via the AC power lines it should show up with different loads (DUTs) connected to the probe. I tried different resistors (incl. the characteristic impedance of the UTP cable). Absolutely nothing similar to what I saw when the UTP cable was connected to the probe was displayed on the oscilloscope screen. Basically, I just saw the same low level noise as when the GND clip was connected directly to the probe tip.

In summary, I have to reject the idea of common mode noise injected into the oscilloscope via the AC power lines because:

  1. I cannot (qualitatively) replicate the phenomenon in Spice
  2. The common mode choke does wonders in Spice but has no meaningful impact in real life
  3. The different loads (between probe tip and ground clip) do not yield the outcome expected

Further Investigation

Back to the drawing board (I went back to the lab and tried a lot of different things - mostly probing under different situation). And here I have to correct myself. I wrote in my original post that when the GND clip is connected directly to the probe tip then I don't see the noise on the scope. This is mostly correct but at a specific location this configuration does pick up a little bit of noise (much smaller amplitude, slightly different frequency) that qualitatively looks like the noise I see when the probe is connected to the UTP cable. It takes a lot of effort to set up the trigger to capture this - see picture below.

enter image description here

The location that yields this result is -not surprisingly- right over the LED PS (mounted underneath the table). Now this (combined with what I learned from just rejected theory - Common Mode Noise Injected into the AC Power Supply Lines) led to a new theory - common mode noise injected somehow into the DUT where the DUT is the GND clip connected to the probe tip, the UTP cable loop etc.

Common Mode Noise Injected into the DUT

So I went back to my office and moved the expected root cause (the step function generated by the change of load when the LED PS switches on/off) to the DUT (i.e. UTP cable instead of the AC power lines). The diagram used for the Spice simulation is shown in the following picture.

enter image description here

The incomplete circuitry to the right is the leftover stuff from when I tried to explain the issue as "Common Mode Noise Injected into the AC Power Lines". It is all inactive and can be ignore at this point (don't look at anything to the right of the label 'osc_inp'). Now let's look at the schematic from left to right...

I modeled the UTP cable as a series of LRC stages plus stray capacitance from both conductors to the noise source (labeled Earth in the schematic above - the name is a leftover from the previous investigation). Since a UTP cable is symmetrical I split the L and R into two components - one per conductor.

enter image description here

I got the LRC parameters from the cable data sheet (derived the L value from the characteristic impedance and the capacitance) - Bicc General CAT 5 (leftover cable from the time wired 100Mbps Ethernet was in). I reused the value for the stray capacitance I had computed for the previous investigation (estimated by plugging the numbers into the tool provided at https://www.emisoftware.com/calculator/wire-over-ground-plane-capacitance/). The distance from the UTP cable to the suspected source is less than from the UTP cable to the ground but the suspected source is not exactly a huge ground plane - so a little bit give and take. This is not accurate but the order of magnitude should be in the right ballpark (assumed to be an OK starting point).

enter image description here

I operated the probe in 10x mode. The probe diagram is guesstimated from the data sheet for the Siglent SP2035 probe (I doubt Siglent will share the probe design details with me and didn't feel like dissecting an almost brand new fully functional probe - if someone has a non functional SP2035 probe they'll like to surrender I'd be happy to take it off your hands for further investigation). I couldn't quite get the -3 dB point to be at 350 MHz (current values yield a -3 dB point at 275 MHz but with a virtually flat frequency response, i.e. bode plot). Thus I am aware of the discrepancy here. The oscilloscope input RC values are straight from the manual. The RC lowpass filter following the oscilloscope input impedance circuitry was supposed to imitate the limitation of the vertical amplifier (a moot point since the scope probe's -3 dB point is at 275 MHz). Now that we got all this in place I ran the transient analysis. The result is shown in the following couple of pictures.

Zoomed out... enter image description here

Zoomed in... enter image description here

I'd say qualitatively there is a lot of resemblance with the pictures I took of the oscilloscope screen and the Spice simulation result. To provide additional support for this theory (i.e. the entrance point/mechanism is common mode noise coupled to the DUT) I tried to cover the DUT in aluminum foil (the probe shield was also connected to the aluminum foil). The following two pictures show the difference when the DUT (the UTP cable) is wrapped in aluminum foil and not wrapped in foil.

DUT not wrapped in foil... enter image description here

...versus...

DUT wrapped in foil... enter image description here

The only difference in the setup between these two cases besides the shielding is the trigger point. In both cases I decrease/increase the trigger point so the scope (in normal mode) only got triggered every 2-3 second (I used this scheme to filter out low level noise). I say there is a visible difference between the shielded scenario and the unshielded scenario confirming that the entry point is the DUT (and not the AC power line). Obviously the solutions isn't to wrap the DUT in the aluminum foil.

Conclusion

Thus my assertion is this problem is mostly caused by capacitive coupling between the suspected noise source (which may include the power code to the suspected noise source - more work needed here) and the DUT.

Take Aways

  • I was surprised to learn how huge stray capacitance is (learned this from estimating the values needed for the Spice model). The values for stray capacitance can easily run into several pico farads (same order of magnitude as some of the values associated with the probe circuitry)... this is why my conclusion is 'capacitive' coupling (anything else I did in Spice didn't generate anywhere near the same differential voltages at the scope input)
  • All these 'green' switched-mode power supplies may save energy but they surely generate a lot of noise. The IKEA LED strip light was not the only source I identified. Our water heater (fortunately, it doesn't go on that often) knocks things out of the park... and so on... all these switched-mode PS powered devices are a nightmare for folks who have an electronic lab in their garage
  • The oscillation is the result of the resonance frequency of your system (in my case UTP cable, probe and oscilloscope input) and not what is injected into your system (what is injected into your system excites your circuitry)
\$\endgroup\$
3
  • \$\begingroup\$ There are some factors missing from your model, but it seems it's still a reasonable introduction. 1. I 'm not familiar with the X1 or X2 models shown (and I guess they're .SUBCKTs anyway); a potential confusion with SPICE TLs is, they are also ideal ports, i.e. no common mode impedance between ends. You therefore need to patch in CM Z separately. I don't know if this is included in your model, other than obviously nothing would come through if X2 had no CM Z. 2. ProbeGndClip should be 100nH or so, which strongly contributes to the CM-DM mode conversion of a probe in most circumstances \$\endgroup\$ Jan 21 at 1:12
  • \$\begingroup\$ (e.g. probing a PCB with CM noise). 3. Lhouse_gndWireToWorkbench is probably much more (e.g. ~1µH/m), and of course has couplings to other wiring, within the scope (who knows how much of the PSU should be modeled; assuming it's a short circuit probably isn't unreasonable, as the Y-type caps likely dominate at these frequencies) and outside. But modeling those probably isn't important for illustration purposes. 4. Notice the system is symmetical, i.e. we can model from the LED PSU's switch node and isolation capacitances, through mains wiring, to the scope; or backwards from the scope just as \$\endgroup\$ Jan 21 at 1:12
  • \$\begingroup\$ well (which maybe isn't a great description of how this is drawn, but to say it's more scope-centric). EMC systems largely obey reciprocity, so it doesn't matter whether you consider aggressor or victim, at least until you consider the circuits within them (which are decidedly non-reciprocal). Putting EMI networks (e.g. LISN) into this model may also be valuable illustration, though not required for a simple demonstration like this. \$\endgroup\$ Jan 21 at 1:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.