Very High Frequency noise on POE PD Flyback converter

[FotatoPotato]

Hello everyone!

I just finished building my POE power converter. Its purpose is to take in 48V + Data and output 12v @1A + Data. The design is heavily based on the STEVAL-TSP003V2 Board and my schematics and theirs are essentially the same. The problem I'm having is that there is HUGE AC ripple on the output of the supply. To add insult to injury the ripple is jumping between 20Mhz and 70Mhz but it tends to stay in the mid to high 40's. Does anyone know what could be causing this ripple? No matter what I try it wont go away.... and trust me I have tried everything I can think of.

Reference Schematic: https://www.st.com/resource/en/data_brief/steval-tsp003v2.pdf

Scope Images:

img 1: http://www.mediafire.com/file/14kd5gcg2hc27xb/No_Load_Noise.JPG/file
img 2: http://www.mediafire.com/file/ne3u3jakcxuiu2e/1A_Load_Noise.JPG/file

Thanks!

[T3sl4co1l]

RF bursts, coincident with switching edges = lack of damping/snubbing on the transformer windings (primary at turn-off and secondary at turn-on).

I don't see any common mode chokes (aside from what's in the Ethernet connector itself), and it's not obvious if the "Y cap" is even doing its job (or it looks like they intend it as a placement option with alternatives, and I don't know which is present).  That's probably the rest of the explanation.

Well, that, and probably the transformer isn't wound nicely.  The winding order has a huge impact on EMI, but you'd have to take it apart to tell.  You rarely if ever see such details in the datasheet.

But anyway, didn't you know?  It's a power supply, noise above 20MHz doesn't exist, we never measure it that way!...

 

Also, you'd hope that eval kits would illustrate good EMI practices at least, if not a proper full realization including conducted and radiated emissions tests.

Alas, eval kits and appnotes are always hopelessly clueless about these matters; often, FAEs aren't much more help, either.

Eh... all the more business for me I guess.  (Which, speaking of... )

Tim

[FotatoPotato]

Hey Tim,

Thanks for the reply!

In the App Note for the Eval board they show that their board has sub 100mV peak to peak ripple and when I measured the eval board I confirmed this myself. Also with regards to the transformer being wound badly, I doubt that's the problem, its a Coil Craft part and from my past experience I have never had issued but with that being said, there's always that one time when even a trusted manufacturer sends out a bad part.. who knows...

Also, you mentioned "Y Cap" in your explanation, could you clarify a bit for me please?

And lastly, I forgot to mention that I did try adding more capacitance to the output, adding a 470uH inductor, and a capacitance multiplier to the output with little to no effect. So I'm not quite sure what to do other than just call it a day and ignore the ripple (Which I dont want to do).

Thanks so much for the input, I really appreciate it!

[T3sl4co1l]

"Y Cap", C24, normally a Y1 rated cap for noise suppression around mains connections (mains-to-ground safety rated).  That exact rating might not be required here, but people often use that to refer to the purpose as well as the rating.

Right, additional (diff mode) filtering has no effect.  The noise is from ground to ground.  It gets into your probed signal at the ground clip, which is a good 50nH or so.  In other words, F-all of a ground at this frequency.  So the probe measures that signal plus the differential signal.

Quality has little to do with how the transformer is wound.  It can be a quality part with the most expensive tape and everything, that just means it's consistent and meets specifications -- it's what's specified in the drawing, the order of windings, that matters.

Suppose the primary winding is a single layer, left to right; if the secondary is 1:1, wound on top (also single layer), left to right, then the left (start) terminal is used for +V (primary) and GND (secondary), and the right (end) terminal is used for switch (primary) and diode (secondary).  Two things:
1. Note that no AC (signal) voltage appears across the isolation barrier.  +V is at GND (for RF purposes), and the switch and diode nodes are in phase.  You will never have to filter the (voltage on switch node) * (capacitance between windings) because that voltage isn't applied across that capacitance.
2. For the entire length of the winding, the primary and secondary wires are on top of each other, forming a twin-lead transmission line.  This is very good for predictable behavior -- characteristic impedance will be around 100 ohms, and it's not folding back on itself or anything.

Some voltage will actually be applied across the transmission line: the voltage due to leakage inductance.  Which is exactly the stray inductance of the transmission line length, or about 0.3uH per meter of wire length in this case.  This still needs to be filtered, but it is fairly small magnitude, and all high frequency content.

Now suppose the secondary was wound right (start) to left (end).  It has to be connected backwards relative to primary, for correct flyback phasing.  But now, the entire switch node voltage is applied across the entire length of the winding.  It's still a transmission line (it's still a pair, from primary start and secondary end, to primary end and secondary start), so we have one electrical length of transmission line connected across the isolation barrier.  Which is equivalent to its entire inductance (the leakage inductance) and capacitance.  The full switch node signal goes across the capacitance, making this much more demanding on common mode filtering.

A single layer pair, giving ~100 ohms transmission line impedance, is kind of high for power applications (at 48V and probably ~2A peak, a 48/2 = 24 ohm impedance would be preferable), so more layers (interleaved, alternating primary and secondary) should be used to make that up (probably P-S-P-S would be fine).

If it was wound some other way, it will probably be worse.  If a single layer doesn't have enough inductance, then two or more layers, in series, are needed (per winding).  That raises the impedance considerably (the layers act like transmission lines on top of each other, roughly doubling the impedance between windings, while also dropping the cutoff frequency).

Or you can alternate single layers, connecting layers in series as needed.  That gets a transmission line structure that's... kinda split and interleaved.  The response will certainly be more complex, and you should expect to see multiple high frequency peaks as a result.

Or if the turns ratio isn't 1:1, you can still use the same general build, but vary the number of turns while changing the number of strands in parallel.  Example, a 2:1 transformer might use one strand for primary, and two strands (bifilar) in parallel for secondary.  This gives a weirder transmission line structure (two primary turns span one secondary turn), but unless the switch is very fast (risetime comparable to the electrical length of a few turns), this will be averaged over, and it will still look like a transmission line sort of structure.  Obviously, the switch and diode nodes don't have the same voltage on each other, which introduces some EMI -- but the common ends are still both at (RF) ground, so the EMI is half the worst-case condition (phased backwards).

The advantage to winding your own transformers is you can play with all of this directly.  Tedious to iterate on (it should take you about half an hour to design and wind a single transformer, once you know what to do), but you aren't locked to off-the-shelf components.  Downside: if you want more than one, you have to make them all over again (or pay a shop custom pricing for low quantities -- not actually as horrible as it sounds, as winding shops are usually fairly priced when you have a drawing in hand).

Tim