AC mains filter design

[Cicero]

Hi guys,

I haven't been here a while, but just thought of something that always bothers me, and perhaps some of you deal with this problem as well.

I'm talking about issues with conducted emissions on the AC lines. Now I use a pretty standard input mains filter I put on everything, and then tweak where necessary during testing. But I find this quite stressful at the time, because testing is firstly very expensive, and time is limited as well. But its worse because I dont have a complete grasp of AC filter design, so its almost a trial and error response to any issues seen.

I just place a couple of Y caps, a couple of X caps, and common mode choke on every product.

How do you guys deal with this kind of problem? Do you have any set designs you use? Ways of simulating the filter response, additions etc?

 

[ronsimpson]

The PCB design under the filter parts is very important.

I have a setup to measure power line noise but it is very old and not calibrated.
You are right I just guess at the filter design.
Then I get testes and see where the filter/power supply is very quite and where it is noisy.
Back home, I know that I need to reduce the noise by 20db at 3mhz and I probably have too much filtering at 100-500khz. I can see the 3mhz noise and I can see 20db.
Part if what I need is to know if the noise is common mode or differential.

I use spice to model the response of the filter. No one can model the entire electronics. I can get good simulation of the filter.
If you use spice like this you must know the ESR and ESL of each capacitor. The inductor is also a complicated LRC+LRC.
The model does not need to be perfect. You do need to know how each component effects the filter.
example: Changing a capacitor by 2X will move the frequency here but the self resonance frequency of the cap is now in a bad place.

Chasing noise is not something you can do with math and SPICE only. You must relay on experiance, guessing, and knowledge. (some luck) I look at other peoples designs.

This is not easy. If it was easy you would not have the job.

 

[4pyros]

I have used off the shelf line filters.
https://www.mouser.com/Power/Power-Condition/Power-Line-Filters/_/N-axgnp/

 

[Cicero]

I use spice to model the response of the filter. No one can model the entire electronics. I can get good simulation of the filter.
If you use spice like this you must know the ESR and ESL of each capacitor. The inductor is also a complicated LRC+LRC.
The model does not need to be perfect. You do need to know how each component effects the filter.
example: Changing a capacitor by 2X will move the frequency here but the self resonance frequency of the cap is now in a bad place.

Chasing noise is not something you can do with math and SPICE only. You must relay on experiance, guessing, and knowledge. (some luck) I look at other peoples designs.

This is not easy. If it was easy you would not have the job.

True, glad to know that there is a good bit of guesswork for others as well though. I have a home made LISN, which is probably a similar setup like yours. But its currently out of action at the moment.

I definitely need to do more spice simulations though. What spice program do you use? Generally I only use LTspice. Do you simulate the equivalent circuits for the differential filter, and common mode filter?

I have used off the shelf line filters.
https://www.mouser.com/Power/Power-Condition/Power-Line-Filters/_/N-axgnp/

I would love to, but they're often too expensive to justify for the product.

 

[ronsimpson]

I use LT Spice.
Usually I have two files. Common and Differential

 

[4pyros]

I would love to, but they're often too expensive to justify for the product.

We did it mostly for UL approves. Its cheaper then having a home brew filter tested.

 

[Cicero]

We did it mostly for UL approves. Its cheaper then having a home brew filter tested.

Cheaper as in you dont have to submit for testing as its already UL rated? Surely your whole system still needs to be submitted though?

 

[cct314]

Just curious about your application: Are you more interested in containing the RFI produced by the product within, reduce susceptibility of the product to conducted RFI/EMI on the AC mains input, as a "passive" means of providing protection against/suppression of power line transients (e.g. spikes), or some combination of these three?

There are minor tweaks to the standard design that can tremendously alter the efficacy of such a filter, but the degree to which they may be applied (or not!) can significantly affect the costing of the parts involved.

The last item in the above list (e.g. protection against/suppression of power line transients) is a potential application of these devices that is often overlooked as a properly-designed filter of this type can do an excellent job in this respect, doing better both effectively (and safely!) in many respects than a voltage-triggered clamping device!

 

[Cicero]

Just curious about your application: Are you more interested in containing the RFI produced by the product within, reduce susceptibility of the product to conducted RFI/EMI on the AC mains input, as a "passive" means of providing protection against/suppression of power line transients (e.g. spikes), or some combination of these three?

There are minor tweaks to the standard design that can tremendously alter the efficacy of such a filter, but the degree to which they may be applied (or not!) can significantly affect the costing of the parts involved.

The last item in the above list (e.g. protection against/suppression of power line transients) is a potential application of these devices that is often overlooked as a properly-designed filter of this type can do an excellent job in this respect, doing better both effectively (and safely!) in many respects than a voltage-triggered clamping device!

All of the above preferably, haha! But I'm more concerned about conducted emissions, so containing any emissions within the device and stopping them coming out the power lines. However, it does also help with conducted immunity as well and for surges and eft's. I also generally I put MOVs on the lines for overvoltage protection, and several tranzorbs around DC regulators.

 

[Cicero]

The PCB design under the filter parts is very important.

I have a setup to measure power line noise but it is very old and not calibrated.
You are right I just guess at the filter design.
Then I get testes and see where the filter/power supply is very quite and where it is noisy.
Back home, I know that I need to reduce the noise by 20db at 3mhz and I probably have too much filtering at 100-500khz. I can see the 3mhz noise and I can see 20db.
Part if what I need is to know if the noise is common mode or differential.

I use spice to model the response of the filter. No one can model the entire electronics. I can get good simulation of the filter.
If you use spice like this you must know the ESR and ESL of each capacitor. The inductor is also a complicated LRC+LRC.
The model does not need to be perfect. You do need to know how each component effects the filter.
example: Changing a capacitor by 2X will move the frequency here but the self resonance frequency of the cap is now in a bad place.

Chasing noise is not something you can do with math and SPICE only. You must relay on experiance, guessing, and knowledge. (some luck) I look at other peoples designs.

This is not easy. If it was easy you would not have the job.

Ron, excuse me dredging this up again. But I've had some time to look into this again and just have a few questions for you, if you dont mind.

Here is a sample filter design, what do you think of the equivalent circuits for simulating? Are they similar to what you use/simulate? Do you add your ESR and ESL as discreet additions, or change the specs of the caps in LTspice? Also how do you even find out your ESR/ESL for these mains caps though, majority of them dont seem to spec that in their datasheets? And one last question, could you explain the inductor LRC+LRC, I'm not quite following that?

 

[ronsimpson]

I do not have much time today.
Here is a graph in a capacitor data sheet.
I have PDF graph paper like this page i could send to you. It will help.
Look at the 10uF curve. 5 ohms at 1khz and drops at higher frequencies lake a capacitor should.
BUT
At high frequencies the curve is going up like a inductor. With a little math you can find what inductor has that slope.
Where the two lines (C and L) cross is the self resonant frequency.
Note that the curve stops going down at 0.01 ohms.
SO
You should be able to get spice to give you this curve with (10uF + ?uH + 0.01 ohms) in series.
Note: The 10uF cap will not be as good at 10mhz. Using a 10uF & 1uF & 0.1uF in parallel makes a good capacitor.

You also need to find data in your inductors. They are not a simple L.
The resistance can be found from a ohm meter.
There is a capacitor across your inductor. You can find that by finding the resonant frequency of the inductor.
Using the same idea you should be able to find the C.

You can find information with a network analyzer.
With out a network analyzer you can find the resonant frequency of an inductor by wrapping a small number of turns of wire around the coil and driving it with a RF signal generator. (you are looking for some thing small like 1pF so you can't put a 10pF scope probe on the coil, but you can place the probe very close to the coil and try to see radiation. Sweep the signal and look for a frequency where the coil resonates.

In many SPICE programs there is a way to add R and L to a capacitor. (set properties)

 

[Cicero]

Understandable, thanks for the time, appreciate it. Really clears some things up that I was wondering about.

I'll try post some extra info from an actual product where I had issues pre filter, and then simulate with the filter I used to fix the issues. Will tie up this thread nicely if it actually makes sense.

 

[ronsimpson]

Filter design (with out spice) can be done with a piece of graph paper.
Like in post #11 you can make graphs.
Draw a line for 10uF. \
Draw a line for 0.1uH (what ever the inductance is ) /
Draw a line for 0.1 ohms -------
Then you will see the graph in #11.

This also works for the RC around the error amplifier.
It is good for the frequency response of audio amplifiers and filters. (will take some practice)

RonS