Paralleling small value caps with large value PS caps

[xyrium]

I've seen some instances where manufacturers have enormous amounts of capacitance in their power supply design, and place small caps (0.01mfd) in line with the larger lytic caps (10000mfd or more). Is there a method or formula used to employ this technique for cleaning up PS noise? See the tiny red Wimas ni line with the large electrolytics below:

[Netlist]

This theory with calculator might help you.

/Hugo

[xyrium]

Thanks Hugo. ALthough I don't think that can help me in my situation (the large caps already exist) , the whitepaper was certainly interesting. Thanks!

Originally posted by Netlist
This theory with calculator might help you.

/Hugo

[sam9]

The reason is simple. Large electrolytic caps have a mimited bandwidth due to a property called self-resonance. If they were ideal they would totally block DC and as frequency above 0Hz increases the they would present a continuously decreasing resistance to AC. In fact they opperate this way quite well up to a point - certainly enough to remove nearly all of the ripple. However, at some frequency which may even be in the higher audio band depending on design specifics, the resistance (impedance, to be more correct) stops declecreasing and starts increasing and reduces the big caps ability to direct HF signals away from the DC power rails.

The small ceramic or film caps you see have a much higher bandwidth, sometimes way into the Ghz range. This provides a low impedance path for HF signals and noise away from the power rails. They add nothing significant to the total capacitance of the PS but extend the bandwidth of the PS-filter network. If the big electros were not subject to the limitation of self resonance and had an ideal bandwidth there would be no need for the little caps.

[xyrium]

Ok, this makes a tremendous amount of sense. However, how does one determine the value of the small caps if they wanted to add them to an existing design?

Thanks!
Paul

Originally posted by sam9
The reason is simple. Large electrolytic caps have a mimited bandwidth due to a property called self-resonance. If they were ideal they would totally block DC and as frequency above 0Hz increases the they would present a continuously decreasing resistance to AC. In fact they opperate this way quite well up to a point - certainly enough to remove nearly all of the ripple. However, at some frequency which may even be in the higher audio band depending on design specifics, the resistance (impedance, to be more correct) stops declecreasing and starts increasing and reduces the big caps ability to direct HF signals away from the DC power rails.

The small ceramic or film caps you see have a much higher bandwidth, sometimes way into the Ghz range. This provides a low impedance path for HF signals and noise away from the power rails. They add nothing significant to the total capacitance of the PS but extend the bandwidth of the PS-filter network. If the big electros were not subject to the limitation of self resonance and had an ideal bandwidth there would be no need for the little caps.

[sam9]

It's non-critical.

0.1uF is pretty standard because it covers nearly all circumstances. If the large caps were REALLY big individually, 0.22uF or o.33uF.

Some people use smaller vales, 0.01uF but I don't see a reason to do that.

[Eva]

The resulting system from paralelling very dissimilar capacitors is going to resonate like crazy, usually at several hundred Khz, except when a resistor of high enough value is inserted to isolate one capacitor from another. But inserting such a resistor increases the output impedance at high frequencies. On the other hand, big electrolytic capacitors actually doesn't tend to be so inductive at high frequencies, they just turn into more resistive, which is not so bad at all since it's not likely to cause any resonance trouble. When output stages resonate, they do because of PCB track inductance, not due to big electrolytic capacitor inductance.

That's how things actually work, the rest are just tales from people that has never tested the actual setup with dissimilar paralelled capacitors and a pulse generator to excite them at HF.

You can actually build a nice resonator just by paralelling two ceramic capacitors of equal value with enough PCB track inudctance inbetween. And it will look absolutely harmless at first sight...

[sam9]

Eva,

Since we are taking power supply capacitor "banks" it seems to me you can protect yourself from HF insertion by using a good metal enclusure and with a "Y" (or "X") cap across the mains. Both of these are pretty standard.

Also when using big screw terminal caps a lot of people connect them with copper plates. The last time I dealt with that type of cap I made two each assemblies of a bleed resistor parralleled with a .22uF film cap with soldered together along with a ring terminal. Then these were jusy slipped over the screw terminals. I thought I was being original until I saw a catalog with some sand cast power resistors made built in brackets just for this purpose. If they were to add an appropriate film cap that would be real cool. Would any of these means be make the circumstances worse, or better?

[DSP_Geek]

Parking the small caps across the big electrolytics is the Wrong Thing, anyway. You're better off snubbing the rectifiers and putting the smaller caps right at the output devices. McCormack, as a matter of fact, didn't even have large main supply caps: they used smallish electrolytics right up against each output device.

[Eva]

I agree: This is obviously the worst place to connect those small capacitors, since wiring inductance will cause the amplifier modules to see a high and inductive power supply impedance at high frequencies anyway...

Furthermore, following my SMPS experience related to managing high frequencies, and EMI suppresion/inmunity principles, I would put the output stage and the storage capacitors all together in the same PCB (with no small capacitors paralelled at all), and I would place the small-signal gain stages in a daugher PCB in a corner, away from output stage, in a different spatial plane from the main PCB, and with a tight SMD layout.