Why are electrolytic caps bypassed with film caps?
Jan 10, 2007 at 11:52 PM Thread Starter Post #1 of 21


500+ Head-Fier
Nov 21, 2005
Newbie question (did try the search function):

1. Why are electrolytic caps bypassed with film caps?

2. The best answer I found with searching was that:

Bypassing will remove the shortcomings of electrolytic caps, but also removes the "sound signature" of higher quality electrolytic caps.

Can anyone give a more technical explanation of why this occurs, or point me towards a previous thread?

Jan 11, 2007 at 12:03 AM Post #3 of 21
electrolytic caps are terrible quality compared to other caps. They only get used because they offer very high capacitance (their only redeeming feature). They are bypassed so you can have the high capacitance, but with some of the qualities of the film cap. Idealy getting the best of both worlds.
Jan 11, 2007 at 12:07 AM Post #4 of 21
Electros sound like doggy doo. Bypassing them with a plastic cap (or better yet an oil) reduces, but not elliminates, their nastiness.

Personally, I'd just take out another morgage and buy a high value oil in the first place, but maybe that's just me.
Jan 11, 2007 at 1:25 AM Post #5 of 21
a somewhat relivant thread.

FWIW, i requested that the thread i linked to be locked, so please dont ban my sorry ass for linking to it.

i initally asked a similar Q as you, and have since given up on bypassing electrolytics with film. i will still stoop as low as paralleling similar caps when necessary. i think it would take a lot of work (or the cap would have to be WAYY upstream in the psu, and well before a choke and other caps) to get me to bypass an electrolytic.
Jan 11, 2007 at 3:01 AM Post #8 of 21

Originally Posted by dizzyorange /img/forum/go_quote.gif
Nikongod, have you done a comparison of with and w/o film caps in a millet? If so I'd be interested to hear what you found.

based on that thread and my cheap tube amp, i built my millet with no film across the output. i liked it, and then removed the film across the plate and cathode caps of the tube. the cap for the buffer still has its cap, AND i added a small film cap to the input cap.

yea, i liked the film so much i removed it....
caps are:
plate: black gate 220uf/100V nf i think (the premimum grade)
cathode: elna cerafine 470uf/25V
output: elna tonerex 470uf/100V

most of the caps on my millet are singificantly "overrated" for voltage because higher voltage caps have lower esr. the cathode caps (elna cerafine) required some creative mounting, but since im not using the film, i mounted them where the film WAS. a perfect fit.

if you will be at the nyc meet in a few months, my millet will be too.
Jan 11, 2007 at 3:10 AM Post #9 of 21
I think the general idea with bypassing is that electrolytics do not have very good high frequency characteristics (at least they didn't used to... modern electrolytics are far better than they used to be 10-20 years ago). By bypassing with a film cap which has (hopefully) much better HF response, it was thought that you could compensate for the electro.

One problem with this is by choosing the wrong types/values, you could actually create an oscillator. Also, the phase responses will differ between the two caps, so a mismatch in that sense can also be created when bypassing.

Many people feel that the best electros, such as Black Gates, do approach good film caps today. In a thread over on diyhifi (long and technical), jocko and some others have done some plots (Smith charts, which are a complicated complex impedance chart that I used to know how to plot and read) of many current electros, and also one of the Black Gate "L" cancelling pair. His conclusion to this point is that the best Black Gates (NH series) are quite close to good film caps (and that the L cancelling pair doesn't work as advertised... there's still L there
Jan 11, 2007 at 3:13 AM Post #10 of 21

Originally Posted by nikongod /img/forum/go_quote.gif
a somewhat relivant thread.

Ah, yes, good times - the final death knell for DIY on Head-fi.
Jan 11, 2007 at 3:47 AM Post #11 of 21

For BYPASSING ELECTROLYTIC CAPACITORS, attaching film capacitor on the side will reduce the impedance of ELECTOLYTIC capacitor at high frequencies.

Huh? Right? Well, electrolytic capacitors tend to be slower. This makes it have higher impedance. This higher impedance thingie is important for power supply. Electrocity is like people; go for easy paths. And, if the PSU impedance is high, stray electrocity may go elsewhere. This is a major doo-doo, but not because it makes the sound doo-doo. This will make the amp UNSTABLE and you might hear HISS.

On the other hand, film capacitors are relatively fast. This means its impedance will stay relatively low at high frequency. So theoretically, less likely to hiss.

If you recall the basic electronics, by PARALLELLING electrolytics with film caps, you will THEORETICALLY have lower impedance AS THE WHOLE even at high frequencies.

This is the reason we do this film cap parallelling jibba-jabba. Note however the improvement is highly dependent on the quality of the electronics. Some electronics are extremely fast. In such case, you do better by NOT having film cap; because you can save yourself from extra inductance. (which is also a disruptor of stability)

Note also, using film caps in this particular application may not be appropriate; as in cost effective. New electrolytics are pretty fast even the generic ones, so you only need impedance-CAPPING at pretty high frequency. Then you might as well use good ceramic caps which cost considerably less. (And ceramics perform well even beyond 10-100 MHz. That is not so with film caps.)

I think we haven't discussed this question much for like 2 yrs. Thank you, dizzyorange. I was starting to forget all these.

Jan 12, 2007 at 2:41 AM Post #12 of 21
Added: "Big" "Slow" compared to "Small" "Fast" is wrong. Capacitors are charged at different speeds because different time constants are used. T=R*C. Decrease R and the charge will be rapid.

By adding a small, insignificant film capacitor impedance doesn't matter enough to even spend the cash on the capacitor. i.e.

C2=Large electrolytic
C1=small film capacitor

Zc1+Zc2 = 1/((jw)^2*C1C2) / s(C1+C2)/((jw)^2*C1C2)=1/jw(C2+C1)

It helps a wee bit, but drowns in component tolerances, and wont change a thing. The model does not show any issues. The world is perfect and everything is acting ideal. But in these situations discussing capacitor types it will actually now be revealed that the impedance is still remarkably high at higher frequencies in real life, so the added capacitance won't help at all


Dizzyorange, What you read there to begin with is one of the dangers of the internet. A lot of people who believes in something, but can't prove anything with theory, but just excuses it all with less good sound signature and some jibber-jabber. I will be writing all this once - yes i'm bored - and people can use the search function later on, because this question pops up from time to time. :)

The trick with electrolytic capacitors is that in reality, they are a capacitor C1 in series with a resistance, the ESR, we call it R1. Film caps does not act like this, because of the physical differences (i.e. stack versus a roll, and depend on electrolytic as well)

The R1 of the electrolytics is also in series with a small inductance, L1, and the capacitance in parallel with another resistance R2.

Z(C1)=R2||(1/sC1) + sL1 + R1
Z(C1)=(R2 + (sL + R1)(sC1R2 + 1)) / (sC1R2 + 1)
Z(C1)=(s^2L1C1R2 + s(L1+C1R1R2) + R1 + R2) / (sC1R2 + 1)

This is the impedance of the electrolytic capacitor with capacitance C1, measured at the frequency s=jw (w = omega, rad/sec)

According to http://www.johansondielectrics.com/technicalnotes/dcd/ , typical values for L1~2nH, R1=15mOhm .... R2 is considered to be extremely high, because the current leakage typically is ~10 uA (!!). Thus assume R2=20 MOhm.


Now, let's put a film capacitor, C2 in parallel with the electrolytic C1. Film capacitors have a much lower ESR and generally lower leakage current, thus I'll keep it at Z(C2)=1/sC2. This will still demonstrate the point.

= (S^2*L1C1R2 + s(L1 + C1R2) + R1 + R2) / (sC2(sC1R2 + 1)) / (sC2(s^2*L1C1R2 + s(L1+C1R1R2) + R1 + R2) + sC1R2 + 1) / (sC2(sC1R2 + 1))

= (S^2*L1C1R2 + s(L1 + C1R2) + R1 + R2) / (s^2*L1C1R2 + s(L1+C1R1R2) + R1 + R2) + sC1R2 + 1)

= (S^2*L1C1R2 + s(L1 + C1R2) + R1 + R2) / (s^3*L1C1C2R2+S^2*C2(L1+C1R1R2) + sC2(R1+ (C1/C2)*R2) + 1)

Divide by R2, and i think it will become clearer:

= (S^2*L1C1 + s(L1/R2 + C1) + R1/R2 + 1) / (s^3*L1C1C2+S^2*C2(L1/R2+C1R1) + sC2((R1/R2) + (C1/C2)) + 1/R2)

Recall that C1 was your large capacitor. and C2 was the parallel small film cap, i.e. C2<<C1

L1 was in the 10^-12 henry area, so we might as well neglect it totally. Thus we achieve

Ztotal= 1/sC1 ... I.e. the same as the capacitance of the electrolytic when frequency->large. The small film cap will be significant because of the variable s=jw increasing. There's something else we don't like: The third order poles in the denominator of the function. One is introduced with L1.

Inserting typical values:
C1: 470uF
C2: 0.1nF
R1: 15 mOhm
R2: 10 MOhm
L1: 2 nH

We get:

Ztotal=(s^2*9.4e-13 + s0.00000705 +1)/(s^3*9.4e-20 + s0.00047 + 0.5e-7)

Plotting the impedance function in matlab (as below) or similar, we get that the minimum real part of the impedance is achieved at w=1e6 rad/sec, where |Ztotal|=0.015 Ohm. However, introducing the real model for the electrolytic reveals a nasty peak in the bodeplot at w=7e7 rad/sec. Here the impedance is |Ztotal|=1.995 MOhm.

This is revealed from the bodeplot below:

As we see, the impedance curve for the "corrected" circuit follows the ideal world for a number of frequencies, but it differs greatly, and becomes unacceptable at points.
Jan 12, 2007 at 9:01 AM Post #14 of 21

Thanks, Daroid. I went through your derivation myself and came to the same conclusion. I always saw these things near the chips so I assumed the false assumption.

After I did your derivation, I took the last equation and graphed it. I got 15mohm straight up beyond 100kHz. Very cool. To make it even more cooler, I got the same spike at 7*10^7 rad/s.

Does this mean it is unstable at about 11MHz?

Thanks again for the derivation. I really enjoyed it.

Jan 12, 2007 at 1:59 PM Post #15 of 21
I'll look into the nyquist plot... never used nyquist diagrams before

The function was analytic for all frequencies, so it is stable everywhere. No reports on that by matlab, although division by 0 was close
Not sure about the Q factor, since the model was modified....

It's not bad, it just shows that one should avoid electrolytics in high-speed digital circuits where the impedance is of importance.

Something not accounted for was the fact that ESR varies with frequency, so the "improved model" is not even good enough.
Also, the capacitance of the film caps turns out to be affected by ESL and ESR, so that needs to be improved too ... some day...

Users who are viewing this thread