Best cap TYPE for bypasing a 20A 12VDC Power Supply

[GonzoTheGreat]

What capacitor type would you recommend for bypassing the output of a 20A 12VDC Power Supply ?

It is a switching power supply that has 400mVp-p ripple on the output and the spectrum analyzer shows that most of this ripple energy is between 30kHz-90kHz.
I have a 50mH choke and ferrite beads before this cap already.

Most EEs tell me that the electrolytic caps have too high ESR for 20A output current and that they have a too high impedance at 90kHz anyway.  Some recommend polypropylene film caps, some recommend tantalum caps, some recommend ceramic caps ... but there are so many types, I thought I'd ask here.

[Cliff Matthews]

http://www.designers-guide.org/design/bypassing.pdf

[GonzoTheGreat]

http://www.designers-guide.org/design/bypassing.pdf

I read it but I still don't know which cap types can handle these frequencies and currents well.

[T3sl4co1l]

Uhhh, what are you doing exactly?  (Specs, circuit, layout.)

50mH is about 10^3 larger than typical here, so I really wonder if you're having the problem you think you're having...

About caps in general, electrolytic are perfectly fine, with 10s mOhm ESR available.  If those still aren't enough, polymer caps are available.

Tim

[Sceadwian]

If you want low ripple, probably a couple different types. There's almost never 'one best type' for filtering. This is a great thing to use a simulator for, punch in parameter values from cap data sheets and watch the results, try it in the real world and compare your spectrum results.

[T3sl4co1l]

Well... given that most capacitors don't have SPICE models.

Don't think I've ever seen one for electrolytics.  Ceramics usually do (KEMET SPICE for one), but keep in mind these are models approximated around a center frequency and need not be accurate over the whole range (even the range of data provided in the utility).

I normally use a series RLC model, optionally with leakage, absorption or dielectric loss in parallel with the capacitor, or dampening in parallel with the inductor.  If values are not given, I made educated guesses (like, a small electrolytic might be ballpark 5nH ESL).

Tim

[Sceadwian]

Just find a datasheet from a manufacturer that makes caps like what you have that do specify the basics, for a 'good enough' simulation RLC is probably fine. Unless you're really trying to control the ripple to a high degree in which case you should probably be using specified caps anyways.

Comparing the output of the simulation to the output of the real world circuit after you build it should give you an idea of what part of the parasitics you're missing as well. It's an adventure in educated guessing.

[GonzoTheGreat]

Uhhh, what are you doing exactly?

I am powering a load (digital controller) with a commercial switching power supply rated 25A @ 12.5VDC.
The power supply is 40cm away from the load and connected to the load by a four 2mm diameter copper wires (two for ground and two for 12VDC).
The power supply does not have voltage sense terminals.

(Specs, circuit, layout.)

Circuit and layout of the powers supply have not been divulged by its manufacturer.  (the same goes for the digital controller)
PS' specs are 230VAC 1.5A in and 12.5VDC 25A out.
The Digital Controller (the load) is rated at 11-14VDC 20A.

50mH is about 10^3 larger than typical here, so I really wonder if you're having the problem you think you're having...

I think I have this problem. because loading these power supplies with automotive incandescent light bulbs have indicated that there is a 400mVp-p switching noise present at the output when 20A of current flows and most of the energy of these ripples are in the 30kHz to 90kHz range.
This PS noise causes the digital controller (the load) to lock-up.    Powering the same load through a large choke, connected in series with the output of the power supply, decreases the frequency of these lock-ups significantly.  I did not measure the self-capacitance of this choke.  The same controller powered by a 12V sealed lead acid battery does not lock up at all.

I want to put a decoupling capacitor after the choke and I am asking what type of capacitor is best for it when the load draws 20A.

[T3sl4co1l]

A 50mH choke at 20A stores 10J.  That's a big fucking choke, umm, I think on the order of a microwave oven transformer, a large one.  Are you sure that's what you have there?

So, off the shelf supply, full of ripple, common mode unknown?  Can you measure common mode noise as well?

Does the controller not have a supply separate from the 20A load?  It would be much more reasonable to do, say, a ~10kHz filter for the controller at less than an ampere (which would be, let's see, maybe 100uF and 10uH 1A), than the whole freaking thing (maybe 10uH 20A + 10,000uF).

Tim

[GonzoTheGreat]

A 50mH choke at 20A stores 10J.  That's a big fucking choke, umm, I think on the order of a microwave oven transformer, a large one.  Are you sure that's what you have there?

Yes, it is has a large inductance but the size is not so big because it is wound on a Nanoperm core.  However it must saturate or have a high self-capacitance to still let some of that noise through.

So, off the shelf supply, full of ripple, common mode unknown?  Can you measure common mode noise as well?

Common mode is only 42mVp-p.

Does the controller not have a supply separate from the 20A load? 

No, the controller itself is the load.  It has 8 CPUs that draw 20A when fully tasked.

It would be much more reasonable to do, say, a ~10kHz filter for the controller at less than an ampere (which would be, let's see, maybe 100uF and 10uH 1A), than the whole freaking thing (maybe 10uH 20A + 10,000uF).

I guess I could wind my own choke that has less inductance with the turns far apart so it also has a small self-capacitance ...but my LCR meter shows that these large 10mF electrolytic caps have above an Ohm of ESR at 100kHz so are you sure they will do  me any good ? 
Would Tantalums be better?  I heard good stuff about the ESR of polypropylene film caps @ 100kHz, too...  Is there any truth to it?

[T3sl4co1l]

Yes, it is has a large inductance but the size is not so big because it is wound on a Nanoperm core.  However it must saturate or have a high self-capacitance to still let some of that noise through.

Oh.. that's a common mode choke.  Only useful when you run a twisted pair through it (or, it doesn't matter that the wires be twisted together, they can be separate -- as long as they are wound on the core the same direction).

Those cores saturate with a few mA so you're getting bupkis if it's not being used as a CMC.

Common mode is only 42mVp-p.

Not horrific, but still well outside of FCC limits.

No, the controller itself is the load.  It has 8 CPUs that draw 20A when fully tasked.

Oh.. well, that's a lot of control.  That's not what comes to mind when I hear "controller".  You can control a small corner of the internet with that kind of CPU power...

I guess I could wind my own choke that has less inductance with the turns far apart so it also has a small self-capacitance ...but my LCR meter shows that these large 10mF electrolytic caps have above an Ohm of ESR at 100kHz so are you sure they will do  me any good ? 
Would Tantalums be better?  I heard good stuff about the ESR of polypropylene film caps @ 100kHz, too...  Is there any truth to it?

You need a lot of film caps to do the job (20uF+), but that works.  Tantalum, no, sooner or later it will catch fire, with that kind of available current.

Either your caps are toast, or your ESR meter stinks -- whatever the case, a proper, say, 25V 2200uF low-ESR electrolytic should be in the 10s of milliohms, just fine for this.

Same kind of thing should be found inside the power supply too, if you care to snoop around inside it.  Although with that much ripple, I'd be worried how long it's going to last, and if it's made with poor caps in the first place.

For the inductor, this is a perfect application for those powdered iron toroids you find in ATX supplies (yellow-white or green-blue core), if you've parted out any.

Tim