Simple Way To Reduce Distortion From Electrolytic Capacitors

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

Samuel Groner

Well-known member
Joined
Aug 19, 2004
Messages
2,940
Location
Zürich, Switzerland
Hi

It occured to me several years back that an anti-parallel series combination of two electrolytic capacitors (i.e. two caps in parallel but with opposite orientations regarding polarity) might reduce their low-frequency distortion compared to either a standard single capacitor with twice the value or a standard parallel combination (i.e. polarities orientated the same way).

Today I finally found time/interest to verify this. I used a 1k series resistor and two 22 uF caps to ground, fed from a +20 dBu source. The caps were used either in standard parallel combination or the new anti-parallel arrangement. Here's the result:

elco_distortion.png


The anti-parallel combination cancels even-order distortion products; hence there is appreciably lower distortion, particularly towards higher frequencies. The rise above 150 Hz/50 Hz is due to noise, not actual distortion--the noise increase occurs because the low-pass filter configuration reduces test signal level at higher frequencies. Likely the cancellation shows some sensitivity with respect to capacitor matching; I've not yet investigated this, at least the two caps used for the measurements were just randomly selected devices.

To my best knowledge I'm the first one who writes about this, despite the striking simplicity of the arrangement. The use of series combinations has been suggested before (i.e. by Bateman). While this is probably even more effective in reducing distortion (essentially it lowers the AC swing seen by the capacitor) it also reduces the effectively available capacitance, which is inconvenient as usually the reason to use an electrolytic capacitor instead of a film dielectric is size and cost.

So I might suggest that such anti-parallel combinations should be used for cases where sonic transparency is the main goal. The cost increase is modest, and likely this will help much more than the let's-add-a-100 nF-film-cap for which I've not yet found any objective evidence for an improvement within the audio frequency range.

Samuel
 
Interesting.

The problem with two electrolytic capacitors in antiparallel would appear to be that the combination doesn't tolerate much or any DC, as in this configuration at least one of the caps is 'the wrong way around'.

Not looking to generate more work for you, but have you considered testing the influence of DC bias?

JDB.
[funnily enough I was just looking into polar caps in parallel/series for a transistor test jig which I'd like to make as polarity-agnostic as possible]
 
:-[

I gotta learn to read through the text better!

I was confused of the mention of parallelling:
The use of series combinations has been suggested before (i.e. by Bateman). While this is probably even more effective in reducing distortion (essentially it lowers the AC swing seen by the capacitor) it also reduces the effectively available capacitance,

Yes, now it adds up: Parallel = double capacitance, Series = half (if the caps got the same value)
No difference if anti parallel/series.

Well, somone learnt someting today!  ;)
 
JDB beat me to the observation that one has to be biased wrong by definition, but your measurements seem to suggest they are happier with one back biased?

I looked into a somewhat similar topology (22uf to ground with 360 ohm R in series). This was in a gain stage for a phono preamp (did I say this was a while ago  8)  ) . Looking into the large aluminum electrolytic as a HPF I found the ESL and ESR were compelling sources of error.  I forget the actual numbers but phase shift at 20 kHz due to the non-ideal characteristics of electrolytic of the day caused tens of degrees of phase shift in my actual circuit.  Note: since I was in a preamp front end, my cap was only seeing tens of mVolts of AC swing.

While some will argue that phase shift isn't audible (or THD at -100 for that matter) but screw them.. this is science...

Regarding the old popular wisdom of paralleling electrolytic with small film caps, I found on my bench, at least for phase shift, you are wasting your time unless you parallel with no less than 1/10 of the larger cap's capacitance. So paralleling a 22 uf electrolytic with a 2 uf film is still ridiculously large.  While I did find tantalum to deliver superior HF impedance at 20 kHz compared to aluminum of the day, despite their bad rap for dielectric absorption et al.

If you are really serious about opening up this can of worms, there is a lot to look at.

I wonder how the modern low ESR caps do for voltage coefficient and DA... I expect they will be dramatically different than my aluminum bench samples from 30 years ago.  

But at the end of the day, the only good cap is no cap... I hear you can make HPF and LPF in the digital domain without having to kill any capacitor trees. ;D

JR



 
IIRC John Curl wrote about anti parallel caps in TAA or audio many years ago.  I will look for my issues
look for Curl in this link
http://waltjung.org/PDFs/Picking_Capacitors_1.pdf
fig 1b,B

 
Now I see the light: Gainclones Galore on the rampage! 8)

(perhaps with two cheap Si diodes across the nfb caps for better sleep...)

------

A question: did you by any chance test a couple of tantalums as well? Do they "cancel" as well?
 
I suppose the main issue with this arrangement is that in the real world you often use a cap to get rid of a DC offset. And that will of course upset one of the two caps. So even if the reverse bias is not large enough to make it explode, that cap won't be happy. Thus the results may not look as good as in this test anymore, as both caps work under unequal conditions. No?
 
My personal view is that this arrangement could be used (for an "I want the max. quality" DIY-er) in the following ways:

1) as I/O caps in f.e. line-level apps. with opamps (where there is a negligible DC offset)
2) "NFB" caps in opamp-based gain stages (and micpres) where there is a negligible DC offset.
3) Gainclones (a whole universe on it's own)


Perhaps the most funny application would be to use the anti-pair in micpres: lowers ESR _plus_ lowers distortion plus looks good...

==========================
Samuel, what about TANTALUMS?
==========================

Can you test some tantalums as well?
 
Samuel Groner said:
To my best knowledge I'm the first one who writes about this, despite the striking simplicity of the arrangement.

Samuel

I remember a similar test in Elektor. I'll try to dig out but God help me. Hundreds of issues.

However, you might certainly be right but I think the principal has always been there. Although on a completely different matter, I parallelled two linear feedback potentiometer in opposite direction to cancel out each other's deficiencies. The linearity improved with a better positional accuracy on either side in relation to the centre position of the potentiometer. However, the potentiometers were from the same batch.
 
Samuel:

Just for grins, try a pair of caps in anti-series, + ends connected to one another, and a 1M resistor connected from the junction between them to V+. I'd be interested to know how that compared with anti-parallel.

Peace,
Paul
 
Thanks for your interest.

The problem with two electrolytic capacitors in antiparallel would appear to be that the combination doesn't tolerate much or any DC.

So even if the reverse bias is not large enough to make it explode, that cap won't be happy.

True, 100 mV might be about the maximum I'd suggest. However when designing with opamps the effective offset across the cap is usually less than this; and it has been standard practice to use single caps in such applications without further considering the polarity in production, so this seems to work well.

John Curl wrote about anti parallel caps in TAA or audio many years ago.

True, thanks for pointing this out. In this case it's even more surprising that this arrangement is not better known--31 years later I can't remember having seen any schematic except my own which would show it.

Perhaps the most funny application would be to use the anti-pair in micpres.

In fact I've been mostly thinking about caps in the feedback network of transformerless micpreamps; here the impedance levels are really low (approaching 1 Ohm), so series combinations ain't a practical solution. Also the addition of a servo is not as straight forward as it is for other applications. However DC offset at this place is usually less than 10 mV, so the anti-parallel arrangement should be a decent solution.

Not looking to generate more work for you, but have you considered testing the influence of DC bias?

Did you by any chance test a couple of tantalums as well?

Try a pair of caps in anti-series, + ends connected to one another, and a 1M resistor connected from the junction between them to V+.

At the moment I've hardly time to investiagate this much further; I'm rather confident though that tantalum caps will show the very same improvement.

Your measurements seem to suggest they are happier with one back biased?

At these frequencies there is no true reverse bias happening, i.e. the voltage across the cap is pretty low. I presume the capacitance has some voltage coefficient which of course cancels in the anti-parallel arrangement.

Presumably there is a reason why you don't just use NP caps?

These are to a good extent equivalent to a series back-to-back arrangement with two polar caps. So similar space and cost restrictions apply.

Samuel
 
Samuel Groner said:
In fact I've been mostly thinking about caps in the feedback network of transformerless micpreamps; here the impedance levels are really low (approaching 1 Ohm), so series combinations ain't a practical solution. Also the addition of a servo is not as straight forward as it is for other applications. However DC offset at this place is usually less than 10 mV, so the anti-parallel arrangement should be a decent solution.

Samuel

When used in the gain leg of a mic preamp, that could range down to low single digit ohms. I would look for ESL and ESR effects that could be significant in the context of low ohms impedances. While I expect modern low impedance caps to be much better these days than what I had to play with.

My personal preference is to DC couple that leg and deal with DC elsewhere, but customers like to change mic pre gain noiselessly, so cap coupled or an effective servo are indicated for that circuit node in commercial products.

JR
 
Samuel Groner said:
True, thanks for pointing this out. In this case it's even more surprising that this arrangement is not better known--31 years later I can't remember having seen any schematic except my own which would show it.

The AMEK 9098(i) does it. It's hard to see on the micpre schematic that's floating around, but the full manual for the 9098i console clearly shows two 4700uF elcaps with reversed polarity.

Olaf
 
>>> The AMEK 9098(i) does it.
Sneaky basstards. And they kept it all for themselves.
 
Back
Top