EMI filter model gone wrong? Or, why do I have a gaussian at 25khz?

[mclute0]

So I was trying to build a noise free bench power supply for doing bench testing using a 117VAC@96VA to 24vac transformer. I wanted to include some filtering on the mains side and I have a couple EMI filters sitting around.

 I tried to model this filter to see if I wanted to add any components on the mains side besides a fuse and switch, but using the manufactures data that shows the schematic and the component values gave me an AC sweep graph that I did not expect.

Is my approach wrong or is this multisim use of coupled inductors causing a problem, or is this normal and I am just confused?

[Benta]

You don't have a "gaussian", you have peaking, which is normal in second-order filters. You'll need to look a your damping factors in the transfer function.

[Conrad Hoffman]

Haven't done a sim on this but remember that you're measuring voltage to ground, not to the other side of the line after the common mode choke. The filter does seem to give a nice sharp cutoff. The choke coupling might be a bit high, but not sure, and real ones are designed to be lossy at higher frequencies (more damping?).

[jonpaul]

Linear PSU with mains trsf need no EMI filters.

your simulation is not documented to show CM choke windings polarity, nor the simulation source/ load configuration.

Unclear if the plot is a CM or DM transfer function.

Suggest that you read about EMI filter design and common mode vs normal  or differential mode noise and filters.

Finally an EMI filter test or plot needs  a LISN termination.

Simple solution...nix the filter!

Jon

[T3sl4co1l]

You have no source or load impedance.  Load, well, that's probably capacitive anyway (isolation) so not much changes there, but the source is generally assumed to have 50 ohm source impedance at radio frequencies (>50kHz say).  This is implemented with a LISN.

Example mockup for an SMPS:



Note that the offending element in an SMPS is typically the switching node in the transformer, the noise source being that switch node, and its coupling being the capacitance in the transformer (C3).

To test immunity, move the source over left of R1/R3.  Or, in general, consider that any source and load is a source -- you're just setting most of them to zero, and looking at the current into the "load" one(s).  These are linear systems where superposition applies!

As for flatness -- EMI filters tend not to be very well behaved, they just need to do well enough in the bands of interest.  Having a prominent peak, or odd notches and stopband peaks, isn't necessarily a show stopper.  If the damping of the LISN doesn't do the job, you can always add losses elsewhere -- usually an R+C somewhere, or at higher frequencies, ferrite beads -- to dampen such things.

Tim

[mclute0]

The reason I am looking at an emi filter is the noisy environment I have here within ~1 meter of my bench. 2x (109cm 4k monitor, keyboard, mouse, 600watt computer SMPS), a fan, three wall warts, an RF amplifier, a laptop, a portable AC unit, soldering iron, several LED task lights, HEPA filtration system, and what ever other thingy I am working/playing on.

 Having a linear PS in a shielded metal case will, hopefully, stop me  from having to investigate emi on whatever I am working on and I will be able to get rid of some wall warts. Or maybe I am just paranoid...

on the plus side, Benta was correct I believe, as adding another stage to the packaged filter cleared up the problem in the simulation that was being caused by the coupled inductors. I just didn't expect that to be a problem with a packaged filter and my sim does not resemble their insertion loss graph.

 As the manufacturer says in their datasheet, "the best way to select and qualify a filter is for your engineering to test the unit in your equipment". Still not sure I trust this sim, but it looks ok now.

Thanks everyone for the responses.
Not looking at perfection, just for it to be close.

Mclute

(I did change the graph to DBu as it makes it easier to read, but the spike is gone.

[mclute0]

@Tim, thanks. You are two steps ahead of me. I was trying to make sure I had a reasonable sim of the filter and have deleted/not added components diagnosing my problem. I will get there eventually.

[T3sl4co1l]

The k of 0.99 by the way, looks like a guess, but it's not a bad guess, that's fairly typical of CMCs like that.

Also note that the filter functions in CM and DM, and it helps to use equal or complementary sources to model that (rather than measuring one port at a time).

CM is usually the higher priority while DM kinda comes along for the ride, but DM is also dealt with by just using really large capacitors (~uF instead of ~nF) for which the mismatch is kind of a helper (for leakage inductance ~uH and C ~uF, Zo(DM) ~ ohms).

As for your situation, none of those should be especially noisy, at least if they're reputable products (and for the RF amp, set up with a well balanced antenna, avoiding feedline currents).  You're more likely to find noise from a cheap (unfiltered!) phone charger, or your neighbors' even.

Tim