Capacitor distortion

[SIY]

Properly sized coupling caps are the key. Do that and it doesn't matter what the cap type is. I have piles of measurements demonstrating this over at my website.

 

[BR52]

Properly sized coupling caps are the key. Do that and it doesn't matter what the cap type is. I have piles of measurements demonstrating this over at my website.

Hi, I can't agree fully because the size isn't everything. If we say designed in properly, I can agree fully. Means capacity is the common parameter for all capacitors, but every type has its own behavior (size, bias voltage, micro phonics, .....).
(my English is not the best, so hopefully I'm not driven by a misunderstanding)

 

[solderdude]

Probably SIY does not only hint at physical size (though that is certainly a factor with MLCC) but also the value.

22uF driven by 50 ohm and loaded by 50 ohm has a -3dB point at 72Hz which would not be a good coupling cap for audio but does highlight the difference between film and bipolar in technical performance which is kind of the idea of this measurement.
When the -3dB would have been at 7.2Hz it would already be questionable (audible even) which would be the case with a 500ohm load.
5Hz -3dB would have required a 320uF cap so in practice 2x680uF in anti-series.
That would probably have shown 2 (nearly ?) identical traces or at least much much lower distortion.

 

[SIY]

Hi, I can't agree fully because the size isn't everything. If we say designed in properly, I can agree fully. Means capacity is the common parameter for all capacitors, but every type has its own behavior (size, bias voltage, micro phonics, .....).
(my English is not the best, so hopefully I'm not driven by a misunderstanding)

Don't worry about the English. I hope I conveyed that by "size," I mean "amount of capacitance." There is some minor effect of physical size on noise pickup, but emphasis on "minor."

For coupling, microphonics are a very minor effect since there will be almost no signal voltage across a properly sized cap. Ditto biasing. My measurements bear this out.

 

[BR52]

Don't worry about the English. I hope I conveyed that by "size," I mean "amount of capacitance." There is some minor effect of physical size on noise pickup, but emphasis on "minor."

For coupling, microphonics are a very minor effect since there will be almost no signal voltage across a properly sized cap. Ditto biasing. My measurements bear this out.

OK is now clear PS: nice website

 

[pma]

@solderdude , we are not talking about coupling capacitors only. We are talking about capacitor distortion in general. Applications cover filters, crossovers, equalizers as well. We are able to tell which types of capacitors are prone to distortion and which are not. We can prove it and show data, not only debates and anecdotal stories.

 

[syn08]

No need to state the obvious; every engineer worth the salt knows what cap type to use where. Those who don't (like using high K caps for coupling, undersizing coupling electrolytics, ignore the bias conditions, etc...), are paying the price of ignorance. BTW, bipolar electrolytics are widely used in crossovers and only audio fear mongers are freaking out.

@SIY went through the practical use cases on his web site; you pushed the the experiment to absurd corner cases. Have you ever seen an audio appliance with 50ohm input impedance?

You still didn't answer the question as of why

I would never use and electrolyte as a coupling cap.

IMO, that's FUD it it's purest form.

BTW, your measurements above smell like you put the conclusion first, then search for facts that would support it (blind to constraints, practical use cases, without error source analysis, etc...). Based on what you provided, nobody would be able to reproduce your quantitative results.

 

[solderdude]

@solderdude , we are not talking about coupling capacitors only. We are talking about capacitor distortion in general. Applications cover filters, crossovers, equalizers as well. We are able to tell which types of capacitors are prone to distortion and which are not. We can prove it and show data, not only debates and anecdotal stories.

SIY and I were and given the tested values and impedances I assumed it was about coupling caps.

Passive XO is another matter entirely though. Higher voltages, higher currents, low impedances and varying ones. No tantalums, MLCC and no polar electrolitics either.

Post filters in DACs all use relatively small values so could be NP0/C0G and don't think we see much of these problems in DACs. Most do not have output caps either.
Of course in active crossovers and analog equalizers the capacitor choice sure is an issue.

To make it clear, the info I above is more for the casual reader, I am quite sure pma certainly is well aware.

 

[egellings]

Yep, and the DC bias also has to respect the + and -. Especially with tantalum. These little buggers can't stand reverse voltages.

Indeed if one has to use electrolytics as coupling caps and there is no voltage difference between either side and you need a high value capacitor and small size then you can still use electrolytics but must use 2 of them (so you need caps with double the intended value) in anti-series.
So like this:

Does not matter if the + are tied together or the - of both caps.
I would not recommend this with tantalums.

When there is very AC little voltage and you need a high capacitance in a small space you can also but 2 capacitors (with a high as possible voltage rating) in anti-parallel.
In such case the total capacitance doubles. Could be used for line-level audio signals while using 25V to 63V caps.

Would also be fun for @pma to measure this (along side with anti-series) to see how good this works. Should have lower distortion.

Placing high value bleeder resistors across each cap will swamp out the leakage current of the caps and cause equal DC bias voltage division across them.

 

[solderdude]

The bleeders would be a good idea if the caps were in the same polarity and the goal is to distribute the voltage equally over 2 capacitors.
In this case that trick is not an option nor desirable. The caps are supposed to DC block regardless of voltage/current direction when only polar capacitors are available.
Adding resistors there that evenly distribute the voltage is not desirable as it won't block DC anymore.
Also the whole trick is to have the smallest reverse voltage on the 'reversed' cap and most of it in the cap in the proper 'direction'.

 

[egellings]

Typically, the leakage caused by the high value resistors across the series caps will be swamped out by the much lower impedances on both ends of the caps. Granted, it's not the ideal audiophile-grade solution, but usually the added leakage won't upset the circuit's operation. If surrounding circuit impedances are excessively high, then the equalizing resistors can cause problems. Most of the time, though, they don't. It's a YMMV situation.

 

[syn08]

Placing high value bleeder resistors across each cap will swamp out the leakage current of the caps and cause equal DC bias voltage division across them.

Yes, but for coupling applications it is not needed (let’s put it bluntly: it is useless), if both caps support the peak voltage. One cap will limit the current through during the half cycle it is biased normally to it's leakage, which is a non dangerous values. Bipolar caps are build exactly so, two caps in anti-series, rated at nominal voltage, without any bleeder resistors inside.

Bleeder resistors are required when, for example, you need to filter a B+ of 800V and have only 450V caps. Put two in series with bleeder resistors (with correct values, to accommodate the caps worst case leakage) so to keep the worst case voltage across both caps to under 450V.

 

[syn08]

Typically, the leakage caused by the high value resistors across the series caps will be swamped out by the much lower impedances on both ends of the caps.

No. This is a DC (or quasi static, if you prefer) problem, not AC.

 

[egellings]

True, if both caps can individually stand off the DC bias voltage, then the resistors are a moot point. The only possible problem I could see is that if one cap has an excessive share of the DC bias voltage across it, metal migration could possibly come into play for that cap. I am grasping at straws with that, though.

 

[syn08]

Metal electromigration happens only at high current densities (voltage is largely irrelevant), 8-10 orders of magnitude higher than currents in bipolar caps.

 

[syn08]

Properly sized coupling caps are the key. Do that and it doesn't matter what the cap type is. I have piles of measurements demonstrating this over at my website.

Here's an independent confirmation:

1. The setup schematic

- DUT is allegedly the worse electrolytic I was able to find in my junkbox. Got them brand new about 20 years ago at a liquidation, 25pcs. for $1 and never used any of them. They are polar 100uF/100V certainly of Eastern origin, no branding other than a big letter A in a thin frame.
- My setup is differential, so I inserted the cap on one of the branches.
- The ADC has the inputs isolated from the input stage common mode voltage (2.25V) by two bipolar Nichicon MUSE ES 47uF/16V electrolytics.
- The ADC differential mode load is 2.8K, which is at the low end of any audio appliance.
- DAC differential source impedance is 210ohm.
- Loop high pass cutoff frequency is about 2.4Hz. With the DUT in place is about 3.5Hz (calculated). In the 20Hz-20KHz frequency range, the setup FR drop is well under 0.1dB (see above).

2. Measurement setup
- ADC and DAC are both running @192KHz sampling rate. They are in sync mode (driven by the same clocks in slave mode).
- Differential voltage at the DAC output is about 2Vrms (-6dBFS)
- Frequency response of the loop: Note the Y scale is +/-0.1dB

- 10 harmonics are considered for THD. Since the bandwidth is limited to 96KHz, frequencies over 9KHz have less harmonics contributing to the THD, therefore the decaying THD towards 20KHz (see below).
- For both cases, the 3rd harmonic is dominant at low and mid frequencies. 2nd harmonic is at -135db at mid frequencies.

3. Measurement results:

- Red is the loop THD (with the DUT replaced by a SMA female-female through adapter) Orange is the THD with the DUT in place.
- The junk(?) Chinese cap adds 2-3dB of THD between 20 and 50Hz.
- I am quite suspicious about measurements below -120dB, since they are at the uncertainty limit of the setup. But assuming the LF distortion delta is real, I am not planning to lose my sleep over it.
- These chinese NOS caps could be successfully used in the most demanding audiophile appliances.
- I have no idea why the HF distortions of the cap are lower by about the same amount (compared to the loop).

That's it, folks. The myth of polar electrolytic distortions is once again debunked. Electrolytic distortions are a matter of proper design, and have almost nothing to do with the intrinsic properties of the devices. Pathological cases excluded, of course, I am sure High End Audio Priests could find electrolytics with euphonic distortions and sell them as a feature .

 

[restorer-john]

no branding other than a big letter A in a thin frame.

A is for Audiophile! Highly collectible. You should put them on eBay at $40 each.

 

[pma]

Sorry, “properly chosen” electrolytic coupling capacitor is a myth. It will either have LF distortion, or leakage current, or both.

 

[solderdude]

Anecdotal:
About 35 years ago I built 2 pre-amps based on the same schematic one with not biased electrolytic caps (very old pulls) in the input and output path and a TL071 and no power rail decoupling and ceramic disc caps on the input and feedback.
I also built it with an OP27 (in those days a great opamp) and MKP coupling caps and compound decoupling. It had silver-mica feedback and input caps.
In those days this was all the rage for 'affordable' high quality sound.
It had a switch on the back so you could choose the circuit. This was built to demonstrate how poor the TL071 sounded (with help from crappy electrolytics)
And I could clearly hear it when flipping the switch. No one will be surprised at that.
Sometimes (audio) friends came over to listen to my (own build) stats with active SW and then played with the switch always telling which was which.
Substantial differences would be confirmed.
Then one day I left the switch (which was on the back) in the 'poor' position.
From then on I listened for months to the 'crappy' amp fully convinced it was in the 'high performance' setting and intensely enjoyed music and did not detect any crappy sound at all.
Once I found out it flipped a switch in my thinking and from then on performed blind listening tests and sometime telling 'unsuspecting listeners' that I changed this or that but did not do that in reality. I even built a crappy 4 transistor SS amp into an old tube amp chassis with the tubes only connected to AC so they glowed.

That brought me to where I am today.

Sorry, “properly chosen” electrolytic coupling capacitor is a myth. It will either have LF distortion, or leakage current, or both.

That might well be so (perhaps repeat the 22uF cap measurements but using 100k in) but when it is below any audible levels it may not be of practical significance (until the caps dry out )

 

[HarmonicTHD]

Purify, Benchmark and others apply a significant amount of feedback to the signal path of their low distortion amps.
Question: Would this feedback loop “correct” the distortion introduced by caps within the signal path? (Let alone that the caps should be selected correctly in the first place as some previous poster pointed out). If yes, how relevant is the cap material for designing low distortion amps?

Thanks.