Parallel connecting of big capacitors with smaller capacitors for optimising the sound??

[Brian Roth]

Are there any 105 degree bi-polars?

After a bit of digging tonight, all I can find are the Nichicon UEP series and the Rubycon NXA series. I have no idea if they are "hi fidelity" as Nichicon brags about the MUSE caps (which are 85C rated).

I recall using Panasonic bipolars with 105C ratings several decades ago when doing a partial recap of a MCI JH-636 desk, but can't seem to find anything in year 2024.

Bri

/Edit....all I searched for were radial through-hole caps. There were surface mount options.

 

[rock soderstrom]

Phew, I've just read through the whole thread again. Firstly, thank you very much for the rich and informative input.

If I had to summarise, I would say that the situation is complex and opinions vary greatly.

Some don't see it as an issue at all, as long as the capacitor has the right electrical values for the job all is well. There is no need to think about this stuff.

The other extreme is the perspective that down to the dedicated material properties and other unmentioned parameters all play their part and need to be considered.

Of course, there are also all the shades of grey in between, which in my view probably come closest to the truth.

What does this mean in detail, what do I take away for my practical work?

I try to avoid capacitors that I can avoid, sounds strange but is meant seriously. My namesake probably takes a similar view, it would be interesting to see how this could be applied in practice using the example given.

Alternatives would be to ground the cathode and go for "contact bias" or restructure the circuit so that the cathode moves into the negative supply line or use fixed bias. At 1.5V bias, you can get tiny coin cells to connect in series with R5, negative towards gate.

Yes, sounds logical and I think it would be exciting to try biasing a tube with batteries. I've really wanted to try this out for some time, I'm curious to see how it works.

Yes and no. I would argue a lot of this sound is easy to measure, you just need to know WHAT to measure.

In a world where the great unwashed from the "objectivist" group will come with Pitchforks and Torches to the door of the Heretic who states that not everything sound the same and that not all differences are imaginary and on the other side the "subjectivists" will insist that a holographic sticker treated with "electro cream" will improve the sound, there seems to be no middle ground from which to investigate the subject with sufficient science.

It's true, it's always easier to find something if you know what to look for. But overall I have a similar view, our world tends to polarise, and not just in the electronics sector.

Well, did you actually "test the test"?

Well, in this case it was quite simple. The test apparatus masked the process and my ears were the sensors. I am very aware of the limitations of our sense of hearing, but on the other hand this is ultimately decisive for me, as I perceive the result only through my senses. The differences were really very small to non-existent despite my very biased test perspective, because I wanted to hear differences. Overall very disillusioning but just a specific application that cannot necessarily be transferred to all other scenarios.

Thanks for the additional info, Thor. I guess I'll throw away my 1% rule of thumb when I wire capacitors in parallel, see also JR's experience.

ChatGTP is worse than useless for electronics.

I clearly disagree with you on this point. Even in its current state, ChatGPT already offers considerable advantages for those interested. Yes, there are still plenty of problems with the quality of the answers, but I think that can be solved in the near future.

You also have to bear in mind that the level of knowledge of potential users varies and therefore the usefulness can also be different. I would also like to draw attention to the simplification of knowledge transfer for people with disabilities. Think about it.

Beyond that, AI is a disruptive technological revolution that will change everything. The use of AI-based assistance systems in professional life will be commonplace and necessary, because otherwise we will no longer be competitive. You don't have to like it, but it will happen.

1

I agree with moamps's grading of the ChatGPT response for this topic. Yes, you get the impression that the "student" didn't really understand the topic (which is true!) and it also sounds like a lot of hearsay, but let's be honest. What would be your summarised answer to these questions? I would just like to point out once again the very different perspectives of interested parties and experts in this thread. With this in mind I think the answers from ChatGPT were pretty good!

 

[abbey road d enfer]

With this in mind I think the answers from ChatGPT were pretty good!

IMO, they are not, just because it doesn't result in any conclusions. It just says that some do it, some don't and it's up to you to find out if it results in something beneficial to you.
In short, it could have answered "I don"t know, find for yourself".
Hammering it again, ChatGPT is a political animal.
Did you really learn something from it?

 

[Tubetec]

My conclusion about chat GPT is , it makes up believable ******** , and without fact checking can easily misslead people .
One researcher from Ireland used chat GPT to check his own credentials , the bot apparently made up several false links to their work in well known online publications ,that seemed plausible ,but was utterly false .
Dont trust a word this thing spits out without double checking the references they purport to supply .

 

[Newmarket]

OPA1678 which is dirt cheap is 4.5nV|/Hz, drives 600 ohm, has < 0.0001% (-120dB) THD & N into 2k, 9 V/uS slew rate, 2mV max offset and is FET input, folded cascade internal structure with rail to rail output.

Supply range is +/-2.25V to +/-18V, the main limitation compared to NE553X +/-22V not that you see +/-22V used that often.

Agree that +/-22V is minority interest. For thermal reasons apart from anything else.

However, if the rails are lowered to +/-18V and OPA167X are used, the loss of headroom is rather modest, as the OPA167X output swings to 0.2V from the rail into 2kOhm loads.

It's actually a resymbolised OPA1652/54, which won a blind Op-Amp shootout we did, 5532 came in the very bottom.

Interesting. I was looking at getting some OPA1654 to repurpose a poorly designed DI.
Looking at the datasheets I can see that the OPA1652 typical figures are better than those for OPA1678 for a few parameters eg bandwidth and distortion. Not that I'm going to worry about 16MHz vs 18MHz

 

[soapfoot]

After a bit of digging tonight, all I can find are the Nichicon UEP series and the Rubycon NXA series. I have no idea if they are "hi fidelity" as Nichicon brags about the MUSE caps (which are 85C rated)

Does anybody know what “hi fidelity” is supposed to mean in a capacitor? Like… what is the claim being made by the manufacturer?

How is it measured?

 

[JohnRoberts]

It is a marketing appeal.

JR

 

[ruffrecords]

What have you compared them to? I was surprised at how much I liked the 5532/5534. Wonder if it's a familiarity thing? If you grow up listening to so many records mixed on SSLs it becomes somewhat of an ingrained preference

One thing that sets them apart os the position of the dominant pole. It is a lot higher that in most 'modern' op amps. The benefit is higher open loop at high frequencies and a lot less of the rising distortion with frequency at high gains.

Cheers

Ian

 

[Script]

Does anybody know what “hi fidelity” is supposed to mean in a capacitor?

Fidelity comes from Old French and means 'faithful', but interestingly it also means 'devotion'.

And 'hi' is an abbreviation referring to the effect of substance abuse, coming from Old English heh (Anglian), heah (West Saxon) "of great height, tall, conspicuously elevated; lofty, exalted".

So, it means 'exalted devotion' -- and it's not 'in' a capacitor but 'to' a capacitor or whatever.

 

[living sounds]

It's actually a resymbolised OPA1652/54, which won a blind Op-Amp shootout we did, 5532 came in the very bottom.

Well, which 5532 (manufacturer, date of production)? And how was it decoupled? They do vary greatly between manufacturers and production dates, and can sound quite different depending on the decoupling scheme used. It's a big can of worms...

 

[JohnRoberts]

One thing that sets them apart os the position of the dominant pole. It is a lot higher that in most 'modern' op amps. The benefit is higher open loop at high frequencies and a lot less of the rising distortion with frequency at high gains.

Cheers

Ian

[veer alert] This is getting a little off into the weeds..... but I never loved the open loop transfer function of the 5532, but admittedly this was a cosmetic concern (I didn't like the way it looks, phase weirdness in the middle of the audio mid range). When properly applied using copious loop gain margin my imagined concern never expressed. Over the decades I have used truckloads of them with no regrets.

The TL07x their obvious competition back in the day (70s) had less GBW, 3 MHz vs 10 MHz. That extrapolates out to roughly 10 dB more loop gain margin at 20kHz all else equal. I also used truckloads of TL07x with no regrets. I preferred the look of the TL07x internal compensation being a simple pole beginning low in the audio band.

[esoteric alert] I used this open loop transfer function characteristic in one of my too many phono preamp designs. By configuring the RIAA EQ into the NF network across an inverting topology op amp. The open loop pole roughly lined up with the RIAA 3,180 uSec time constant pole. This coordination of open loop and closed loop poles makes the error voltage present across the op amp inputs in phase with the input signal. This was subtle and evidence that I over thought this stuff. [/esoteric]

To repeat when properly applied (using adequate loop gain margin), there should not be audible differences between popular op amps. Of course better is always better.

JR

 

[thor.zmt]

Well, which 5532 (manufacturer, date of production)? And how was it decoupled? They do vary greatly between manufacturers and production dates, and can sound quite different depending on the decoupling scheme used. It's a big can of worms...

TI, current production, JRC current production. They were rated identical in tests.

Decoupling for all was identical, 10nF C0G across power pins, 1uF X7R from power pin to ground, below the IC. 100uF Os-Con on each rail within 10mm of the IC's, power lines decoupled via ferrite beads.

All IC's SMD, soldered. Configuration differential amplifier as balanced DAC post I/U conversion summing amplifier from dual DAC per channel.

Multiple otherwise identical units off the production line were modified, secondary testing was with a headphone amplifier, again as above.

Testing was blind, different devices identified by geometric symbols. Testers were unaware what difference, if any, existed between units.

Tests were preference / marks out of ten with decimals across multiple categories as well as a numbers of psychological proxies.

Testers were allowed to interact freely and according to their desire with the units, except opening up to see what's inside.

Testers could select music to their liking and were compensated for their time on a time basis.

Thor

 

[thor.zmt]

By configuring the RIAA EQ into the NF network across an inverting topology op amp. The open loop pole roughly lined up with the RIAA 3,180 uSec time constant pole.

The same works for non-inverting. I use it in many of my own phono designs using intentionally the similar first order low pass slope of RIAA EQ to keep loop gain near constant with frequency.

To repeat when properly applied (using adequate loop gain margin), there should not be audible differences between popular op amps. Of course better is always better.

Maybe there should not be, but alas to my chagrin I found there are.

Thor

 

[thor.zmt]

One thing that sets them apart os the position of the dominant pole. It is a lot higher that in most 'modern' op amps. The benefit is higher open loop at high frequencies and a lot less of the rising distortion with frequency at high gains.

Not true. The 5534 is externally compensated, it should, if not operating unity gain compensated, compared to decompensated Op-Amp's.

The 5532 has 46-49dB OLG @ 100kHz.

The OPA1678 has a similar OLG @ 100kHz (> 46dB per Datash!t).

The reason is obvious, unity gain stability. Move the pole higher and it becomes unstable.

Thor

 

[moamps]

I agree with moamps's grading of the ChatGPT response for this topic. Yes, you get the impression that the "student" didn't really understand the topic (which is true!) and it also sounds like a lot of hearsay, but let's be honest. What would be your summarised answer to these questions? I would just like to point out once again the very different perspectives of interested parties and experts in this thread. With this in mind I think the answers from ChatGPT were pretty good!

Thank you. The questions were general, and so were the answers. The machine just summarized the current state of available information on the subject, and that in itself is great information.

On the other hand, does the questioner know the answers to those questions? Do any of us know the exact detailed answers to these questions and can anyone else confirm this? If we don't know, how can we judge the quality of the answers when we don't know the correct answers?........ I don't know what you expected.

Next time, ask him to find 100 designs where a parallel capacitor is used, find an explanation for why it was done that way, and do a statistical and electrical evaluation of each solution.

 

[thor.zmt]

Looking at the datasheets I can see that the OPA1652 typical figures are better than those for OPA1678 for a few parameters eg bandwidth and distortion. Not that I'm going to worry about 16MHz vs 18MHz

Way I heard it from the deep throat at TI, it's the same die in both (and in an industrial targeted version under still another symbol, I forgot which).

Difference is the amount of tests on the line. Any die can pass OPA167X tests, if it functions, in other words, performance guaranteed by design ant testing is minimal, lowering cost.

It was actually his big idea, to get sales in the TL07X/08X market by offering a superior product at similar cost (and I guess use up rejects).

My understanding is that the OPA1652/54 go through much more extensive testing and have more parameters guaranteed. Same for the industrial re-symbolisation, but testing for different parameters.

So you get same die, tested for the parameters guaranteed for either industrial, premium audio or "common garden" versions.

For most audio the differences if any are immaterial.

This is also not the only such part. TI (and I suspect others) have the same die under different part numbers (symbols) with different guaranteed parameters.

I'm aware of dozens of parts, due to recent chipageddon shortages, where audio op amps were 108 Weeks lead time for prepay-preorder, zero stock. While the industrial version had several kU stock.

Thor

 

[thor.zmt]

does the questioner know the answers to those questions?

More critical, does he know the actual question and how to formulate a question so basic machine learning with SQL based queries against databases can answer in a meaningful way?

Next time, ask him to find 100 designs where a parallel capacitor is used, find an explanation for why it was done that way, and do a statistical and electrical evaluation of each solution.

It cannot actually do that.

Chat GPT is a mix of lawyer and marketing futzie. It can tell you what you already know, with precedent, qualifications and flowery language that sounds important.

Kind of like the pointy haired boss in Dilbert.

Thor

 

[ccaudle]

It cannot actually do that.

I guess that's why they call it a large language model and not a large physics model.

 

[Newmarket]

Way I heard it from the deep throat at TI, it's the same die in both (and in an industrial targeted version under still another symbol, I forgot which).

Difference is the amount of tests on the line. Any die can pass OPA167X tests, if it functions, in other words, performance guaranteed by design ant testing is minimal, lowering cost.

It was actually his big idea, to get sales in the TL07X/08X market by offering a superior product at similar cost (and I guess use up rejects).

My understanding is that the OPA1652/54 go through much more extensive testing and have more parameters guaranteed. Same for the industrial re-symbolisation, but testing for different parameters.

So you get same die, tested for the parameters guaranteed for either industrial, premium audio or "common garden" versions.

For most audio the differences if any are immaterial.

This is also not the only such part. TI (and I suspect others) have the same die under different part numbers (symbols) with different guaranteed parameters.

I'm aware of dozens of parts, due to recent chipageddon shortages, where audio op amps were 108 Weeks lead time for prepay-preorder, zero stock. While the industrial version had several kU stock.

Thor

Sounds about right. And 081/071 differentiation was similar back in the day based on production results. Along those lines I had some of the first OPAx134 opamps back from the BB rep when I was working professionally in pro-audio - talking so new that the marking was in handwritten pencil - and IIRC they were same as OPAx132 without the DC precision.
That was looking at alternatives to OPA627 for ADC front end. The 627 did measure marginally better in the circuit (taken from Crystal Semiconductor App Note) but not enough for our MD to swallow the cost

 

[Cqwet Dbdfte]

The OPA1677 Fig.6-9 THD%+N curves looks like it could be a fixed ~1uV input noise until it gets suppressed by other non-linearities at around 3V rms out. This may be the low frequency random noise from Fig. 6-2. If it was some monotonic noise it would be bad.
Simplified schematic Section 7.2 shows a true input differential topology with output stage having gain with a common source connection.
Nice part.