Capacitor selection, an engineering choice

[Christopher]

Hello all,

Many beleive that listening is the final judge on what does or does not make a difference, and which difference is better. Others however, beleive that reproducing sound is a purely engineering based exercise and most, if not all the principles involved are well known and can be allowed for and objectively tested. It is the second group that I am appealing to, to enter into several discussions involving such topics as component selection, bypassing et al. The intent here is to discuss these topics from a purely physical point of view, so with all due respect, please do not post views or beliefs based on personal listening tests or what magazine reviewers have said about the subject.

I would like to leave cables and interconnects alne for now, as it has been hotly debated on the DBLT thread (Blind listening tests) and from an engineering point of view, is really quite elementary.

The first topic I would like to address is capacitor selection. There are many sound reasons(pun intended) for choosing one type of capacitor over another, and many of us would not know what these are. Electronics World magazine recently published a 6 part article on the topic, which I will reference when I get a chance. I only discovered this recently and have only had the opportunity to read article #5. It seems the author looked at many of the various types of caps by applying signal and measuring distortion. I beleive many of the audiophile 'truths' were supported in his research, but many were struck down.

So my first question would be, given that capacitors will distort the signal, apparently more so when a DC bias is applied to them, would spending some extra time to engineer direct coupling in a device be time well spent? I know that it's a "that depends" kind of question, that's why I'm asking it.

The lines are open, first caller please?

Chris

[Sch3mat1c]

Hmm...

First of all, where are caps used? Power supply filtering, signal coupling and bypass, signal filtering and (rarely in audio) resonating with inductance to provide peaks or troughs in frequency response.
The distortion is produced from a change in voltage or current (current I guess, but they go hand-in-hand), so we can overlook most applications such as coupling and bypass (which only experience a significant voltage change at LF). For PSUs, currents can get very high, but capacitances are generally quite high as well, swamping the AC voltage (as is its intended purpose).
Now, filters, especially at speaker level (crossovers) will experience much more voltage and current, so this is a possible problem area (especially with the standard bipolar 'lytics.. yow!).
Lastly, LC resonance is often used to tune as in a radio, so capacitive distortion is N/A here. (I would imagine it manifesting itself as dielectric loss or ESR, simply reducing the Q of the circuit.)

Now... let's come up with some typical distortion figures. I heard .005% to .001% THD before <IIRC> so we'll be pessimistic and go with the upper range. First of all, that's -43dB below the fundamental. Pretty quiet, but it's certainly there, compared to the audible ~90dB range.
But...compare that to the typical >1% THD of _any_ amplifier, and the -43dB becomes .0043dB on top of -20dB of distortion. And I say _any_ amp because even a triode at the signal levels we're talking (to squeeze the .005% out of the cap) will be producing that much distortion. (What about op-amps? Take away the NFB.)

Tim

[Christopher]

Sam, I'm a skeptic at heart, and while I may have chosen better wording in my intro, my intention in starting this thread was to elicit responses like Tim's. I'm hoping to keep this discussion within the bounds of engineering practice and not get off on a head-butting tangent.

SY, I had no intention of saying that we know everything, or that listening during the design process has no place, simply that there are some things that are painfully obvious when you understand the engineering, and incredibly murky when you do not. I am a true believer in the idea that good design is good sound, component choice is second to that. There are many innovative topologies out there waiting to be discovered by clever people.

Getting on with the capacitors Tim, I can only speak on what little I've read, but my understanding was that at 1khz and one volt, (the article is not with me at present) the author was seeing second harmonic distortion down I think 100db or so. Now this was a bipolar electrolytic, and he said distortion was 50% worse than his reference 'best' capacitor which was a PET. I would think that in a coupling role, that second harmonic would be made worse would it not? I promise I'll dig up the numbers when I get a chance, as the debate is somewhat moot without them.

Since a lot of people seem to be willing to shell out big bucks for special coupling caps, (I was told that the best of the best were polystyrene) and distortion is only 005%, then if there is an audible difference, and I'm beginning to think that that's a bigger 'if' then I originally thought, what could be accounting for an audible difference?

Chris

[ThomasLMcLean]

SY:

We have met. I will admit to being one of those.

Christopher:

Try these threads;

Sound of Capacitors

Capacitor Distortion

Capacitors and double-blind listening tests

I consider this an engineering approach:

http://www.capacitors.com/pickcap/pickcap.htm

[sam9]

Ah, capacitors. I may be naive and uninformed about what folloes but here are a couple of observations.

Thake a typical amplifier schematic and trace the audio signal path. Usualy the onlt cap directly in the path is the DC blocker at the input.

There are the caps between rail to ground whose function (expressed crudely) is to shunt as much AC as possible to ground and leaving the rail as close to pure DC as possible. Doesn't work perfectly, but well enough most of the time. If in the process the AC suffers distortion -- so what? If you are paranoid, design you amp with separate power and signal/NFB return ground networks.

There is the cap that one finds across the bias transistor. Not long ago I think I read Nelson Pass opine that since it's not in the signal path, practically any cap of the appropriate value does the job.

Then there is the tricky one - the Miller compensation cap. This one is in many regards at the core of making the whole shebang actually work. Yet being down in the pF ranges, the chance that any audio signal passes through it. But if it is not behaving right other parts of the amplifier circuit don't act right either. I've been using silver mica on the grounds they are better than ceramic. I haven't found any of the pollies or metal-pollies that run low enough values to use. Still I suspect that I could put the cheapest ceramic discs of appropriate value and I could never detect an difference. If threre is one, it probably is beyond my meagre ability to measure or hear. Yet I wonder . . .

A lot of amps have a zobel at the output. Since this relates directly to the speakers, tell me anything you like and I'll believe it all even if it is self-contradictory. To me speakers are stange devices made from wood, paper, copper, iron and what-not that no sane person could expect to actually work let alone sound good. The fundamental principles go all the way back Alexander Grahm Bell and Edison. Since speakers are inanimate (sort of) objects they obviously don't care about this and go ahead and work anyway.

Conclusion regarding caps in amps: make sure you've got a metal-poly at the input, even if it is a bit large, and don't worry excessively about the rest.

[MarcelvdG]

Dear Christopher,

I also read Cyril Bateman's articles, the whole series. Polystyrene, polypropylene and class I ceramic capacitors came out best, class II ceramic capacitors were the very worst, when it comes to distortion.

However, the distortion values in his articles are extremely low for all but the very worst capacitors. Other authors have shown measurable differences in the small-signal step response between capacitors with and without serious dielectric absorption, however, these response differences are very small compared to the difference in step response between an ideal first-order high pass and no high pass at all.

Personally, this makes me belief that although there are real and measurable quality differences between capacitors, their influence on audio quality is overrated. I would definitely like to do a simple double-blind listening test with someone with a different opinion, though.


On the other hand, any coupling capacitor you use in your circuits, bad, good or even perfect, is going to give some loss of deep bass and is going to introduce low-frequency group delay errors (or, equivalently, phase errors). Alternatively, if you look at the step response of a bunch of cascaded first-order filters, you will probably be amazed to see how bad it looks. I think this can be a good reason for avoiding coupling capacitors when this is possible without too much effort, and for making the time constants much greater than 8ms when coupling capacitors are really required.

On the other hand, if the microphone picks up subsonic rubbish (or if it is generated by a warped record or so) and this is amplified and applied at a high level to the loudspeaker, the resulting intermodulation distortion in the loudspeaker will make the sound much worse than the sound you would have got with a subsonic filter somewhere in the signal path. This can be a good reason for including a subsonic filter with an on-off switch on your preamplifier, or for including subsonic filters for those signal sources which are likely to have a high subsonic content (like record players).

Marcel van de Gevel

[Christopher]

I'm listening to CBC radio 2 as I write this, my favorite classical music show. Today the host is doing a "classical one hit wonders" theme and I'm listening to Pachelbel's Canon, before that, Also Sprach Zarathustra. But I digress, I think the violin tuning analogy you make Tmac is rather a good one. I'm assuming you're talking in regards to the construction methods and not just turning the tuning pegs. I read a little on this several years back and while it is an art, I think that it is because the variables are so complex and difficult to define that taking a tried method and refining it may be the best route for sure. I think there are many in the Hi Fi world that would like to beleive audio engineering is quite the same. I don't think it is at all. Although speaker construction comes the closest. If we just stick to amplifier design for illustration, we know exactly what we would like to achieve with an amplifier design, and that can be approached or perhaps practically accomplished using sound engineering models. If you have done the due diligence in the design process, the chances that it needs to be "tweeked by ear" when you are done is nil. Flower Duet, nice...

Getting back to the capacitors for a moment, thanks for the links, I think I've been to the pickcap site but I'm not sure. Many sites of that nature are very informative, but a real expert on the subject may take issue with what is published, where the layman takes it verbatim, that's one of the reasons I started this thread, so informed individuals could discuss these things. It seems the consensus is that as far as signal path distortion goes, a good polyfilm is plenty, is there anyone who could see any other reason for a cap in a coupling role to alter the sound by it's composition? On the low freq attenuation front, it would seem to me that the largest value you could stick in for coupling would be the best, providing it won't mess up other things or cause instability. Which brings me to the next point, bypassing.

The theory goes that big electolytics are required to keep the dc constant under various loads, but that they don't pass high freq all that well, thus bypassing them with smaller values will lower noise on the DC rails by giving the highs a better cap to go through. Can anyone say to what extent this actually makes a difference? And further, if you do bypass, what possible difference could any boutique capacitor have. The only thing that my feeble little brain can concoct is that DA could play a role here by not allowing the voltage to be extracted from the cap fast enough to keep the DC constant. Which brings up another point, capacitor banks. I have been known to build them myself but from an economic advantage point of view. To get large values there is a point where a few parallel caps are actually cheaper than one big one, and you get better ripple current numbers, so you can choose a cheaper cap even still based on the ripple current handling you need. But am I gaining any performance in the process? I don't know.

Can you get one of those "reasonably informed designers" to accurately duplicate the three-dimensional sound field present at a live concert into a living room?

I think that you have hit it on the head SY, there is definitely work to be done in the reproduction field and things to be discovered, but so long as people spend all their time trying to convince the unsuspecting that a new set of interconnects or another amplifier for that matter, will make the difference, we're swimming against the current. What needs to be done is to spend less time trying to find the perfect capacitor, and more time trying to design the perfect transducer. So that it doesn't need to have a capacitor placed in front of it in the first place. Our nemesis is the loudspeaker for sure. I think we could get pretty close to the ideal by using binaural recording and headphone playback, but that somehow just doesn't appeal. I guess "feeling" the music is lost. But if one was to wear headphones and have the speakers going... Hmmm.

Chris

[Steve Eddy]

Originally posted by Christopher
So my first question would be, given that capacitors will distort the signal, apparently more so when a DC bias is applied to them...

What were the conditions for this? Was the bias greater than or less than the full signal swing? I'm wondering if the increased distortion here isn't akin to the increased distortion when biasing a complimentary output stage higher into class A but but not high enough that it stays in class A for the entire signal swing.

se

[MarcelvdG]

Cyril Bateman also measured increased distortion when the DC voltage exceeded the peak AC voltage.

Marcel

[Christer]

Originally posted by MarcelvdG
Cyril Bateman also measured increased distortion when the DC voltage exceeded the peak AC voltage.

Marcel

Does this apply to electrolytics also? It would seem reasonable
that it is better to bias an electrolytic above the peak AC value
so the polarity is always respected.