paralleling film caps with electrolytic caps
 

I've seen various website modding their gear where they parallel 0.1uF metallized polypro with the electrolytic cap.

I understand that this is called bypassing.

But when I searched the net, it seems bypassing is commonly done on the supply pins of the chip (opamps for example).

Is there any value to paralleling film caps to electrolytic caps? If yes, is film caps the best type to use?

Thank you very much

ps. I'm not sure if this fits with power supply design but in most instances, the power supply caps are the ones that are bypassed.

Those high speed bypasses are normally to be used at the point of load, that is at the chip or output transistors. If you bypass the 'lytics in the PSU the way to the point of load still has wiring inductance which degrades the effect of the bypasses. That inductance is also the reason why bypassing both the supplies and the point-of-load is a no-go, this forms a C-L-C circuit that can easily ring at RF frequencies. If bypassing is used at the PSU to filter incoming RF noise, there should be some resistance on the way to the load (ferrite bead, e.g.) is needed when the load is also bypassed. Any good film capacitor will do, ceramic capacitors will also be ok. The return point (GND connection) is critical, when the GND gets disturbed by high/fast current pulses from the bypasses not much is to be gained. One will have to measure things (a lot of things) to make sure that bypassing actually impoves perfomance, there a quite a few gotchas...

Regards, Klaus

the issue is that there is significant ESR and ESL in the larger caps, they make poor caps at higher frequencies. So some designs call for the addition of film caps, which act like caps to a higher frequency.

in industry, ceramic caps are used for the smaller size. because the ceramic caps are not in the signal path, there is less concern about possible issues.

I have read that bypass caps should/must have higher voltage that the electrolytic. true?

no,
each cap must be rated to exceed the highest operational voltage.

This is a myth. Larger caps exhibit lower ESR and overal lower impedance. Also, electrolytic capacitors are much less inductive that things such as the PCB traces connecting them (unless power planes are employed).

Furthermore, parallel capacitors usually resonate with parasitistic inductances leading to increased impedance.
There are only a few situations in which paralleling capacitors is actually advantageous.

The concept is simple, but the reality is anything but. Yes, electrolytics are far better than their reputation, but at some tens of kilohertz, they can become more inductive than capacitive. I have some nice manual capacitance and inductance bridges, but at work we have a fancy HP LCR meter that covers 20hz to 2Mhz, a difficult range for most of the equipment a hobbiest can get his or her hands on. It's also fast, so I ran measurements on a bunch of different caps and put the results in an Excel spreadsheet. If I did the math right, you'll see ESR in ohms vs frequency for each cap. Note the log scale. If you were going to parallel caps, you'd need to convert the series model to a parallel model, but it's still interesting to see the differences in ESR between say, a big tubular polypropylene, a dipped tantalum, and a low value mil spec that's either polypropylene or teflon- not sure which.

Excel Capacitor Comparison (.xls file)

thank you all for the help

If you look at the 47Lab Gaincard, there are no bypass caps. At $3000, you would think if it really would matter, they would have added a bypass cap....

I agree, most time, bypassing large caps with smaller caps at 1/100 the bigger cap is a myth.

If you can hear the difference with the bypass and it is better, then go for it. Otherwise leave 'em out...

At $3000, one would also expect to get a decent amplifier... the Gaincard doesn't seem to be one (I'd say it rather is a money making machine)...
http://stereophile.com/solidpoweramp...47/index4.html
I guess that bypassing would not have improved it's performance too much, as it is limited by other design factors.

Regards, Klaus