3
\$\begingroup\$

I am testing an LC power filter and my real world measurements of high frequency (~1.5Mhz) switching spikes show that the attenuation is not remotely close to what my design model suggested.

Situation / Requirements

I have a 6A 7.5V Mean Well SMPS. GST60A07-P1J

It powers ~250 WS2812 addressable RGB LEDs

It's noisy, but an otherwise excellent supply. Even with only a small resistive load of 22 Ohm at ~300mA, (without the WS2812 attached, which run PWM internally so that would be worse) the output looks like this - AC coupled, so zero centered:

SMPS output

The main control loop of the SMPS is rippling at ~ 1Khz. This part is fine because we can clean that up with an LDO. What is not fine, is those switching spikes. They are about 400mV+ peak-to-peak and have very high frequency content which an LDO will not filter. Here is one of those spikes:

Spike zoomed in

Just a visual estimate says that fundamental oscillation of that spike is about 1-2Mhz. My rudimentary FFT capability says there is content above the noise floor up to 5Mhz, which seems about right.

The challenge is that this same power supply needs to also supply the micro controller which runs the WS2812s and also includes some relatively simple analogue circuits and some rudimentary single channel audio. So I am trying to clean this mess up before feeding it to the uC and the analogues. I only need about 200mA of clean supply for the "brains". During testing I am using a 5V LDO, but the final design will be 3.3V.

Attempted solution

The only solution I know how to filter such high frequency content in a power line is a passive LC filter. Semiconductors can't keep up. LDOs PSRR drops quickly. "Capacitor multiplier" type circuits suffer from the same issue - the BJT can't keep up.

I am using the highest HF PSRR LDO I know of downstream of my LC filter: The slightly dated LM2931. This deals with the 1Khz SMPS loop ripple just fine, but it barely touches those spikes.

So I attempted to design and test such an LC filter. I loosely followed this process.

I selected a corner frequency (fc = 1/(sqrt(L*C) * 2pi) of ~1kHz by chosing this RLB0712-101KL 100uH inductor this 100uF Tantalum Capacitor. Note these are through hole components, the final design will be SMC.

I modelled the filter in ngspice with this model circuit (parasitic model components calculated from datasheets, according to that article):

Spice model

The AC sweep simulation shows this:

Simulation results

Which is as expected from design and shows Vload / Vinput = -52db down (~ 66 - 14) @ 1Mhz and better up to 5Mhz. That would be a great result indeed, as the 400mV spikes would be squashed by factor 400x ( 10^(52/20) ) to ~ 1mV.

Reality

I built a simple test rig on vero board. I kept all leads super short and the whole circuit is just 15x15mm including LC filter and LDO stage with 2 more caps:

enter image description here

So what's the result? Quite disappointing really. I get this:

enter image description here

The yellow channel is the input from the SMPS, and the blue channel (note the scale is 10x less) is the "FilterOutput" node on above schematic - equivalent to Vload in simulation. The peak-to-peak averages in that screenshot show at best a 10:1 attenuation. Not 400:1. (The LDO does fine in getting rid of the 1Khz ripple, that's not the focus here, but it barely further improves the spikes).

Questions

  • Is my approach "sane"? Did I miss a trick?
  • Why am I getting 40x less attenuation than the design simulation suggests? Component model? Through hole? Vero board construction ? 1-5Mhz is not very HF, really?
  • What can I do to squash the spikes? Use a second stage LC filter? Recommended approach to design equations? 4th order Butterworth? Link to article?

Many thanks

\$\endgroup\$
6
  • \$\begingroup\$ For one thing, your parasitic model is off. The SRF of your inductor is given as 4.0 MHz, which would imply that the equivalent shunt capacitance is more like 16 pF, not the 0.4 pF that you show. Also, the link for the capacitor doesn't work for me, but I'm wondering why you would choose tantalum over ceramic in an application like this. \$\endgroup\$ – Dave Tweed Jul 29 '20 at 11:19
  • \$\begingroup\$ @DaveTweed Yup, thanks, that's a good spot. The 0.4pF is a mistake. I had 16pF in my calcs, but failed to transfer to the model correctly. That changes the HF knee in the Xfer function to 4Mhz. Still down a huge amount compared to real world measurement (although see doubt about that belowr). I am not experienced in picking caps for this kind of application. I appreciate that ceramics have lower ESR, but I only had 1uF in ceramic, and wanted a lower fc. Is that thinking wrong? In fact the datasheet of the LDO calls for a suggested 0.1uF ceramic on the input. that's in the test circuit. \$\endgroup\$ – Oliver Schönrock Jul 29 '20 at 11:29
  • \$\begingroup\$ Start by putting one of your 1 uF ceramics directly across the output of the Meanwell. Try putting a second one in parallel with your tantalum. Consider putting ferrite beads (or a common-mode choke) in series with both leads coming from the Meanwell. \$\endgroup\$ – Dave Tweed Jul 29 '20 at 11:38
  • \$\begingroup\$ @DaveTweed Thanks. So you are suggesting (once I get my measurement reality sorted - below), perhaps I don't even need an inductor? The meanwell already has one of those EMC ferrite bead lumps on the low voltage lead about 10in from my test circuit. Is that the same thing? \$\endgroup\$ – Oliver Schönrock Jul 29 '20 at 11:46
  • 1
    \$\begingroup\$ @DaveTweed I have now solved my crazy grounding/probing issues (see below). So I may try to make a version with a 10uH inductor or 1uF ceramic cap as you suggested. Probably not both unless its a 2nd stage (overkill?), as modelling suggests attenuation won't be enough. Many thanks for your help! \$\endgroup\$ – Oliver Schönrock Jul 29 '20 at 19:24
3
\$\begingroup\$

I suspect that your scope probing is not ideal. Did you use this type of probing measurement to reduce to "loop" pick-up formed by a conventional scope's probe: -

enter image description here

enter image description here

You can do a simple test to prove it is your probe technique - connect probe top to where the earth clip connects and see what is picked-up - ideally there should be no signal picked up because probe tip = grounding point. I'll guess that you pick-up quite a bit of noise and spikes if you try this.

SMPS's are notorious for screwing with a badly connected probe and you probably need to use a ground spring: -

enter image description here

See also this article on the same subject entitled Power Tip #6: Accurately Measuring Power Supply Ripple.

\$\endgroup\$
13
  • \$\begingroup\$ Yes, thanks. That did occur to me, and no I can't rule it out yet. I fiddled around with a less than ideal spring ground clip and got the same result. But I am not convinced I have 100% percent eliminated this possibility. Is my approach/design otherwise sane? \$\endgroup\$ – Oliver Schönrock Jul 29 '20 at 11:07
  • \$\begingroup\$ I would't use the inductor you chose or I might put one extra inductor in series that deals with stuff greater that the 4 MHz SRF of the one you chose. There is a possibility that the core is saturating - I didn't really see what DC current it was taking during the test but, do the simple measurement I mentioned of putting the probe tip to the grounding point and seeing how much noise you pick up. \$\endgroup\$ – Andy aka Jul 29 '20 at 11:10
  • \$\begingroup\$ After some more fiddling around, I would say scope probing is probably part of the issue. I get a similar size spike on the ground pin when probing with croc ground lead/antenna, so that's bad. BUT I still the same size spike on the ground pin when probing with a spring clip - one ground pin to another ground pin. Earth difference/loop between the SMPS and the scope? I have a new scope arriving in an hour, so that should provide a "second opinion". \$\endgroup\$ – Oliver Schönrock Jul 29 '20 at 11:42
  • 1
    \$\begingroup\$ Yeah, you're right. Many thanks. I think I am making progress by progressively removing more wires. ;-) Will report back when (!!) I solve it. \$\endgroup\$ – Oliver Schönrock Jul 29 '20 at 17:23
  • 1
    \$\begingroup\$ Yeah, it's just grounding. That's the first time I have been forced to really dig into that . But if I ground my circuit by pressing one of the ground pin headers up against the scope's front ground terminal (!!), then the spikes all but disappear into the noise floor. 5mV p-p which is 10x better. And that's without using the ground spring on the probe, because I ran out of hands! Any length of wire for the grounding of my circuit and I have 50mV spikes. So does that mean my circuit works?? Guess so. Anyway thanks. I am going to accept your answer, it was close and got me in right direction. \$\endgroup\$ – Oliver Schönrock Jul 29 '20 at 17:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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