Which Capacitors for Audio?

[Newmarket]

Me too, but when they have a tool missing from their toolbox it's not so great.

Ouch

 

[thor.zmt]

Probably, but they are nothing like as bad as people suggest.

Generic polarised ones are terrible. Especially very small size ones. Without DC bias and with.

And they are commonly seen in coupling positions in audio gear without being placed inside the feedback loop..

Very often they are far better in tweeter sections than some brands of film capacitors.

NOT POLARISED ONES.

You mention "rolling" capacitors... I will happily roll in Alcaps over cheap film.

I will always replace them with Wima MKP or equivalent and adjust the L-Pad for the ridiculously high ESR of these.

Worse, Alcap's (or other bipolar electrochemical capacitors) have poor stability, over time capacitance reduces and ESR rises. Not what you want in a crossover.

There's often much more detail, and more body to the sound - which is the perfect combination. In another thread (either here or on the Naim forum) someone replied to me on the subject mentioning that he'd had his Kan Mk1s redone with Solen caps (which I would have expected to work) and it killed the music for him.

Not surprised.

For those on the wrong of the Pond, the Linn Kan is a tiny Speaker, kinda like the BBCLS3/5 in size and using the same drivers as the BBC LS3/5, but a rather incompetently designed crossover, that causes a distinct and not neutral tonal balance (overly forward and upper mid treble emphasised) and due to using electrochemical capacitors in the high pass to save a few bob tended to sound rather different the old and wet english winter and the warm summers.

They are part of the original Linn/Naim cult, which did a lot to give a bad Naim to "Audiophiles" in the UK and Europe.

Anyway on top of that generic "Solen" brand metalised film capacitors have a long history of being considered "bright" sounding. Which is why I always kept good relation with the Ansar Factory who made much superior Film Capacitors and not because of the "supersound" branding.

Anyway, take an already overly bright and forward speaker, replace an ALCAP high esr electrochemical capacitor on the tweeter which tends to have a somewhat "dark" and "opaque" sound, with a very low ESR and "bright" sounding capacitor the result must be a sonic disaster.

Note, the BBC LS-3/5 uses all film capacitors, in part to ensure long term stability instead of a constantly drifting and temperature dependent frequency response.



Compare Linn Kan:

Somewhere I have a jpg of some impedance measurements from a decade or so ago of an Alcap vs MKP10 vs a Rubicon ZA and, IIRC, the Alcap is the one that looks most like a theoretical capacitor.

I doubt that. Cyril Bateman shows Alcap much worse than MKP but better than polarised.

I agree. I would go so far as to say they are better in some power positions too.

Maybe. I more tend towards polarised types there.

If you have an RC on your rail, I would use a bipolar there too.

If you place snubbers, usually the series R is so large compared to the polarised cap's ESR and there is always a large boas voltage, I would find the use of bipolar capacitors questionable.

I think Meridian are hip to this too. I saw a part photo of one of their internal amplifiers and it had axial caps in the familiar Alcap blue, right next to the output transistors. There wasn't enough of the circuit board shown to be completely sure of what they were doing but I suspect they were between the rails and ground

Maybe something else?

Not untrue. I seem to remember this is in Ott and meant to find it for you but was behind in replying here already. I certainly remember my reaction at the time was "Oh, for God's sake! Really?". But of course there are currents in the casework.

If there are, and not at 30MHz-GHz (which I would not worry about in our context) you have a HUGE Problem in your design. I'd in fact call it highly questionable.

For Audio, there should only be one connection between earth and chassis (if any) and only one connection between signal ground and chassis (if any).

In RF systems we do tend to use the case for ground for connectors, but ideally this too is not done. Better to have a sub enclosure that handles this and openings in the otherwise isolated outer enclosure that is in turn earthed and a single connection between the two.

It's a very good reason to keep the transformers outside the amplifier part.

Or fitted with their own enclosures that also provide shielding of the magnetic fields.

Like so, mild steel:



I'd worry a LOT LESS about currents circulating in the chassis induced by the magnetic field of the mains transformer in turn inducing noise into large size capacitors, than I would worry about about noise DIRECTLY induced by those fields.

I did a MC input Phono Stage with internal Transformers. Shielded in mild steel.

Copper/Alu Chassis so no magnetically induced circulating currents.

In the end I had to use Mu-Metal shields for the MC level part to get noise as low as I wated, as a combination of the C-Core transformers and relative mechanical arrangements still coupled sufficient noise into ground & signal lines and the mechanical arrangements could be not altered sufficiently to reduce coupling.

This is interesting. Maybe those orange drop caps really are the reason only Nakamichi managed to make decent cassette decks. Such an overlooked medium!

I used a Akai 3-Head Unit that was next level. It totally killed any NAK's.

I think you've misconstrued my position on things audiophile. I would defend almost everything that's usually dismissed as "audiophoolery". In fact I am militantly against anyone who believes that SINAD encapsulates the performance of any piece of equipment (while also not being pro injecting euphony).

Ditto. Appy Polly Logie.

I'd say PP is pretty well damped by comparison to PE, which is why it was used after bextrene in loudspeakers. Sure, it is usually loaded with carbon black in loudspeakers, which helps too, but that was mainly for UV blocking (which it has done amazingly well if you look at a B110 SP1228, 30 odd years on). It's also used in composites for toughness and damping. It is 1/3 as stiff as PET and you can test its properties yourself on any milk container. Or see here: Plastic Rigidity & Material Stiffness, Units, Formula & Table

Not a lot difference between any of them.

FKPs and MKPs are my go-to capacitors. I usually stick to MKP4s as MKP10s can get a little expensive. I thought MKPs were wound. Not sure about FKP but I love them.

Modern production is much easier with stacked film. Most of the cut views in datasheets show stacked film.

I'm not actually pro DC coupling these days but I would be very interested in how you think a servo ought to be done.

I showed it before. It went into a huge kerfuffle.

Phono, MC Pre-Pre & MM. Optional high pass on output with elliptical filter.

Naturally direct coupled inputs (makes no sense otherwise).

I was very taken with how Bruno Putzeys did the input of a preamp by subtracting a low pass from the signal.

I do something rather different, but there are many ways.

I prefer to direct-couple inputs where I cannot enclose capacitors in feedback loop's.

The classic servo is a lowpass that inverts the passband signal which is the summed (subtracted because of inversion) from the signal. How is what Brüno does different? He has a way promoting the most obvious things as amazing new inventions.

I was thinking more a 70mm long capacitor with a 50mm diameter, rated at 600V, which is the kind of thing that sometimes appears at the input of "exotic" MC phono preamps.

An MC Pre-pre is a good place for a direct coupled input, see above.

Thor

 

[thor.zmt]

I don't think anybody denies the fact that loudspeaker cables interact with the amp and speaker. However, in most cases, the differences are way outside the domain of perception.

Interconnect cables are a potential source of distortion. For a long time I used chin ese starquad cable for my AP2 connections. after a few month the softeners in the PVC insulation had degraded the conductors sufficiently to cause observable increases in measurable distortion.

I switched to RG316 (silverplated copper with TFE), twineed for balanced and the problems went away.

In most interconnects with mains powered gear circulating leakage currents make the ground impedance in interconnects critical, more-so with RCA connections, hence locking RCA's and "exotic" cables can make oversized measured differences.

Even the DCR of the cable is not such a significant factor,

Maybe not in loudspeakers. But even there I'd be careful in making such pronouncements.

Ground resistance in interconnects?

Allowed current leakage in Class II it is <= 0.25mA, now 0.25mA * 1 Ohm = 0.25mV which with a 2V line signal is 78dB SNR).

Now lets reduce to 0.01Ohm and we get 118dB SNR instead.

Thor

 

[thor.zmt]

I used to write a magazine column called "audio mythology" back in the 80s (40 years ago).

Some old myths refuse to die.

Probably because they often are not Myth?

Thor

 

[gyraf]

Or because there's good money to be made from the confused and half-informed masses..?

This is the tradition we try to avoid around here

 

[gyraf]

very, very long posts that contains very little real content other than subjective claims if you read closely.

At first I was (im)pressed into tl:dr-mode - now I wonder if this is a technique for obfuscation?

 

[abbey road d enfer]

Interconnect cables are a potential source of distortion. For a long time I used chin ese starquad cable for my AP2 connections. after a few month the softeners in the PVC insulation had degraded the conductors sufficiently to cause observable increases in measurable distortion.

What kind of "degradation"? Something that alters the chemical composition of copper in a way that makes it non-linear?

Ground resistance in interconnects?

Allowed current leakage in Class II it is <= 0.25mA, now 0.25mA * 1 Ohm = 0.25mV which with a 2V line signal is 78dB SNR).

That supposes zero CMRR. Playing with unbalanced connections is a receipe for headaches.
BTW, generations of vinyl listeners have lived with much less S/N.

 

[thor.zmt]

Subjective listening tests? Not double blind?

What, to you is "double blind"?

ABX as promoted by the "ABX company in Troy, Michigan", a commercial entity that promoted and sold the commercially failed ABX Comparator?

I have no issues with well designed double blind testing, but setting something that cannot be criticised the way classic ABX is routinely, to a point that means the results have in essence no confidence whatsoever and are equivalent to doing the same thing as ****** PMC in RuSSia, is non trivial.

For example, if the test subject is actually aware of the change being tested and has a strong opinion on the subject, this bias provides a strong tendency towards null results regardless of the actual situation.

I demonstrated that back in the 90's at the London HiFi club, pretending to test mains cable using the ABX protocol, while actually reversing the speaker cable polarity for one channel instead. We had a "strong cable disbeliever" (to put his opinion charitably), one "strong believer" (who incidentally believed my dual coax based main cable design was garbage) and several "true sceptics" who were doubtful but open minded. All "sceptics" scored 10/10 while believer and disbeliever both has less than chance results (3/10 & 4/10).

Another example, if you have a subtle difference and a single trial according to ABX, you have arguably a very low risk of a false positive, but a near certainty of a false negative. So what is the use?

Thor

 

[Walter66]

The main reason was loudspeakers were designed for efficiency. If bass was needed, a large speaker was used.
The use of an "underhung" voice-coil, where all the voice-coil is in the magnet's magnetic field results in permanent motional control of the voice-coil movement, but also limited excursion.
Modern speakers use "overhung" voice-coils, which results in only a fraction of the voice-coil being under motional control. This provides better linearity at large excursions, at the cost of impairing low-level efficiency.
The actual control the amp has on the speaker, as suggested by damping factor, is almost irrelevant since the output impedance of amps have usually been much smaller than the DC resistance of the voice-coil. Some designers have conducted tests with varying degrees of damping, from no damping (current-drive) to negative-impedance, as JR reminded.
In particular, current drive has its followers, but implies the use of specifically designed transducers; it works well for midrange, not so well for bass.

Interesting fact with those underhung coils you mentioned.
I read rules about those damping factors to control a speaker best. On the other side, we strive for greatest power transmission from amp to speaker, thats why tube amps should have the same output impedance as speakers. A tube amp will run into problems with too low impedance in speakers, a transistor amp can handle this better I think. Tube amps shouldn't be run with open outputs, transistor amps not with shorted outlets.
I have an 8 Ohm speaker system and use the 8 Ohm outlets of my tube amp, a perfect match. Sounds very controlled and damped, I think it has something to do with the output circuit, mine is including negative feedback. It makes the speaker sounding more controlled and damped.

 

[thor.zmt]

Doesn’t that depend on the actual test? Difference is just one goal of listening tests. I could imagine tests (e.g. using a semantic differential) where the goal is to find out what sounds best. For statements that can be generalized to some degree (in contrast to subjective listening), the effort is quite high (as already mentioned).

I personally, outside of demonstrating the uselessness of the Audio ABX test as promoted by ABX company of Troy Michigan never did this kind of test.

I have organised and performed tens maybe even into the 100's of blind listening tests for preference. All invariably were blind insofar that subjects were unaware of:

1) What actually was being tested

2) The actual identity of Circle or Square or Triangle (I used neutral symbols for identification as A/B or 1/2 tended to add bias

Subjects filled in complex questionnaires designed including psychology input to tease out actual unbiased preference. Natuarry such things as level matching and excluding known audible factors were attended to.

The results have usually been educational, interesting and often humiliating.

Thor

 

[thor.zmt]

There are cases where it's not possible. For example the parametric EQ section in a mixer. High-value pots are not desirable for noise reasons.

Well, the values of capacitors needed are commonly available in Film. At least for ALL AND ANY parametric EQ topologies you would see me using.

Using servos, not only at the ins and outs, but at the potentiometer nodes is unconceivable.

Well, again, I posted elsewhere my "AC only" Op-Amp transposition of a standard OPA. Noise levels are low. They would perfectly.

The only practical solution is to use electrolytic caps.

I STRONGLY disagree. That's a wimp out for persons lacking imagination and competence, or the will to do something other than "0815".

Until only recently FET input opamps were too noisy for this application.

Define recently?

A NE5534 and 2SK389 were available at least in the late 70's.

That makes a very low noise FET input OP-Amp and they used to be quite cheap.

So, sorry, on evidence I must reject all your arguments as not valid except in a "we try do this the cheapest way that will kinda work" way. In which case performance compromises are routine and not an issue.

It is easy to do cheap no-fi. What place that has in mixing consoles for live performance or recording is another question.

Thor

 

[Newmarket]

On the other side, we strive for greatest power transmission from amp to speaker, thats why tube amps should have the same output impedance as speakers.

Do we strive for that ? Aren't we more interested in efficiency (not burning off energy in our amplifier) ?

 

[Walter66]

Do we strive for that ? Aren't we more interested in efficiency (not burning off energy in our amplifier) ?

Sorry, my english is bad. Max. power output can be transmitted if the load equals the source impedance. It's one of the basic theory rules.
We strive for max. power when designing a working point for output tubes of power amps, too. If the speaker is highly efficient, we need lower power output from the amp, that's favorable of course. To design a monster amp for a low efficiency speaker is what's been done for quite a while. Surely not the best solution but everybody does to achieve good "speaker control".

 

[JohnRoberts]

Sorry, my english is bad. Max. power output can be transmitted if the load equals the source impedance. It's one of the basic theory rules.

Equal source/input impedance is an archaic (old) strategy for audio interfaces. Instead of maximum power transfer, modern interfaces use "bridging" interfaces to maximize voltage transfer. Bridging input impedance is typically 10x source impedance so inputs don't excessively load down signals. Lots of older legacy gear standardized on 600 ohm source and input impedance for maximum power transfer. Modern gear using bridging interfaces use much lower source impedance and much higher input impedance.

We strive for max. power when designing a working point for output tubes of power amps, too. If the speaker is highly efficient, we need lower power output from the amp, that's favorable of course. To design a monster amp for a low efficiency speaker is what's been done for quite a while. Surely not the best solution but everybody does to achieve good "speaker control".

Vacuum tubes have high source impedance compared to typical loudspeaker loads. Output transformers step down the output voltage to better match speaker impedances. Transformer impedance changes with the voltage ratio squared.

Modern solid state amplifiers have output impedances much less than 1 ohm, and are in fact bridging terminations to typical loudspeaker loads.

JR

 

[Newmarket]

Sorry, my english is bad. Max. power output can be transmitted if the load equals the source impedance. It's one of the basic theory rules.
We strive for max. power when designing a working point for output tubes of power amps, too. If the speaker is highly efficient, we need lower power output from the amp, that's favorable of course. To design a monster amp for a low efficiency speaker is what's been done for quite a while. Surely not the best solution but everybody does to achieve good "speaker control".

English seems fine. I know the maths there thanks. But isn't impedance matching what output transformers are for ? I'm not a "valve person" but am aware of a trend to use lowish power valve amps to drive high efficiency speakers with UK hi fi circles at least

 

[thor.zmt]

In particular, current drive has its followers, but implies the use of specifically designed transducers; it works well for midrange, not so well for bass.

Current drive works great for bass. It is just that the highly nonlinear damping of the Amplifier attempting to short-circuit the back-emf (aka dumping fuctor) needs to be replaced by mechanical (and usually highly linear) means.

One could, for example, make a single piece, non-slotted aluminum voice coil carrier of the correct thickness, to give our driver a Qm that is equal to desired Qt, with current drive We will approach infinity, so Qt ~ Qm.

Seas had a few drivers with a low Qm (~ 1.3 IIRC) which when used with current drive on open baffle of the right size produced spectacular results.

Applying motional feedback is another option.

Thor

 

[thor.zmt]

What kind of "degradation"? Something that alters the chemical composition of copper in a way that makes it non-linear?

The tin plating turned black, the copper surface underneath was black too. The cable wax in principle of good quality, OEM factory for some big brands.

And the measured degradation was an increase in general noise floor on the AP2 FFT as well as increased levels of individual harmonics doing loopback.vs selecting genmon as input.

That supposes zero CMRR. Playing with unbalanced connections is a receipe for headaches.

Unbalanced connections are the normal case for bomestic audio, regardless of price.

BTW, generations of vinyl listeners have lived with much less S/N.

And?

Thor

 

[abbey road d enfer]

Interesting fact with those underhung coils you mentioned.
I read rules about those damping factors to control a speaker best. On the other side, we strive for greatest power transmission from amp to speaker, thats why tube amps should have the same output impedance as speakers.

This is seldom the case, particularly with pentodes.
Let's take the case of a push-pull of 6L6. Rp is about 30 kohms. For a nominal 8 ohms output, each sees a half primary of about 1600 ohms, which reflects at the secondary as about 20 ohms. Only NFB reduces the output impedance to a workable value.
Triodes are usually closer to a "matching" value, because Rp in triode mode is way smaller than in pentode mode.
It may be a reason why triodes are preferred for non-NFB amps.

I think it has something to do with the output circuit, mine is including negative feedback. It makes the speaker sounding more controlled and damped.

That's the idea.

 

[abbey road d enfer]

Well, the values of capacitors needed are commonly available in Film.

I'm not talking about teh capacitors used in filters, I'm talking about the caps that are necessary to prevent DC in the pots. People don't want to hear ugly scratches when they select frequency or change gain.

Well, again, I posted elsewhere my "AC only" Op-Amp transposition of a standard OPA. Noise levels are low. They would perfectly.

Not a reasonable choice when you want 4 parametric sections in a mixer channel. Each section having 4 or 5 opamps.

That makes a very low noise FET input OP-Amp and they used to be quite cheap.

Yet SSL, Neve et al didn't do it; not even Studer or Focusrite.

So, sorry, on evidence I must reject all your arguments as not valid except in a "we try do this the cheapest way that will kinda work" way. In which case performance compromises are routine and not an issue.

You're totally neglecting the industrial aspects, where cost optimization is a vital factor.
I'd be curious to see the BOM for an analog music recording mixer designed after your principles. Not mentioning real-estate and the nuclear plant to power it.

It is easy to do cheap no-fi. What place that has in mixing consoles for live performance or recording is another question.

It's often an argument between pro audio and HiFi. If a recording mixer was made using the same methods and principles that pertain to HiFi, it would cost several $M.
When they listen to their million $ stereo, they should realize the signal probably went through dozens of electrolytics caps, TL07x and Blackmer VCA's.

 

[abbey road d enfer]

Current drive works great for bass.

At moderate levels, yes. When the voice-coil starts to reach the point where induction decreases, distortion increases, and there's no counter-effect from the back emf.

It is just that the highly nonlinear damping of the Amplifier attempting to short-circuit the back-emf (aka dumping fuctor) needs to be replaced by mechanical (and usually highly linear) means.

Resulting in loss of efficiency.

One could, for example, make a single piece, non-slotted aluminum voice coil carrier of the correct thickness, to give our driver a Qm that is equal to desired Qt, with current drive We will approach infinity, so Qt ~ Qm.

I don't see how a piece of Al can result in strong damping. Maybe I lack vision...

Seas had a few drivers with a low Qm (~ 1.3 IIRC) which when used with current drive on open baffle of the right size produced spectacular results.

There is a commercial example of a loudspeaker using small speakers with extreme mechanical damping. It takes 1.7kW on the bass section to provide decent spl.

Applying motional feedback is another option.

Motional FB is just another way of applying electrical damping.