paralleling film caps with electrolytic caps

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
metre is internationally recognised.
Circuit is a common term to both languages.
Nothing to do with Amercanisation of the english language.

Actually Americans do spell it meter, they dont care what the rest of the world does.
Being psuedo Brits but living so close to the States we Canadians see this stuff all the
time. Litre liter, colour color, pasta pawsta, okay that last ones a joke.
 
Gootee,
read what I said, last sentence...
You cannot design a power supply by ears alone, that is what my point was, I am quite aware of the problems with power supplies...and as I've stated before, the power supply isw the MOST important part of any electronics design, get it right and the rest will go OK, get it wrong and nothing else is gonna work.
BUT when designing a PSU you need to look at the output with a scope, is the rf present, whhat ripple is there etc etc
You cannot hear DC, that is a fact, you can hear noise intereference etc iterposed on the DC but DC is silent, clean and wonderfully unobtainable, even using a battery cannot guarantee noise free DC, so PSU's are designed to WORK in the equipement they are designed for, if they dont as I said before its a BAD design.
My pet peeve is people designing power supplies with their ears...
Again with Andrew, signal has a return path

marce,

I believe that we almost fully agree on every aspect that you mentioned. And also, I did not intend to sound like I was "railing" at you. I am sorry.

You are correct that we cannot hear DC. But again, "DC" refers only to the VOLTAGE rails of the PSU. The CURRENT output from the PSU is _NOT_ DC, when there is music power being supplied through a power amplifier by the power supply.

This thread started (I thought, anyway) in order to discuss paralleling capacitors in a power supply, presumably a power supply that would be meant to supply power for the production of music through loudspeakers.

Since the power supply must supply the dynamic current that IS, usually extremely directly, the music signal which drives the loudspeakers, it seems to me that we are right on target in discussing parallel capacitors in the power supply, along with any parallel capacitors that are used to bypass/decouple the high-power current-control valves that are controlled by the input signal chain, such as the final power transistors or the output stages of chipamps or whatever is used in the output section of the "power amplifier". I often like to think of the bypass/decoupling caps as a sort-of local "point of load" power supply, anyway.

The possibilities for high-frequency resonances being formed between smaller capacitors and the parasitic inductance of larger co-located capacitors, or with the stray or parasitic inductance of attached wires or PCB traces, have been fairly-adequately discussed, already, in this thread.

I think that at least AndrewT and I are interested in understanding more about which capacitors supply which time and frequency components of the music signal current, and how to try to ensure that their implementation has been optimized for the most-faithful reproduction of the music.

My primary goal for an audio system is the most-faithful reproduction of the input, so that the output from the speakers is as close as possible to what was intended by the makers of the source material.

My two main questions related to this thread's topic seem to be:

In terms of the PSU and all of the capacitors upstream from the power pins of the output stage power devices (at least), what and how significant can their effects on the faithfulness of the music reproduction be, and, how would one best attempt to ensure that the cause of any flaw in the faithfulness of the music reproduction would not originate there?

Cheers,

Tom
 
Last edited:
NO the SIGNAL is actualy voltage. An amp is a voltage device, x volts in gives y volts out regardless of the frequency or load. And y volts out no matter what the current draw.

Try feeding your speakers with a 100ma sine wave and sweep the freq, you will not get a constant SPL output like you should from a 100mv sine wave.

Oh come on! We're saying exactly the same thing. The rest is just semantical (more or less). Voltages just appear between two places. But current is the juice that actually goes through the load and does the work (yes, because of the voltage). Think about it some more.

Either way works, to think about it, but it's actually always both ways at the same time, as I'm sure you know. But in the context of this thread's discussion, it doesn't seem as useful to think about the VOLTAGE on the PSU rails, since that's almost static. In the PSU, the CURRENT is where the action is, especially as it relates to the functioning of the capacitors in supplying the music current that is pushed and pulled through the load (yes, by the voltage; or is it the other way around? <grin>).
 
So yes, every component in the power supply is in the main sound-signal path, and could possibly affect the sound quality even more-directly than the components in the input "signal path".
Except for the huge isolation between PS rails and output provided by the OP semiconductors. Your view of the PS rejection is wrong, the output current is proportional to output voltage and thus the error in OP current with rail voltage change is determined by the PSRR.Otherwise that 5 volt ripple on the supply voltage would be deafening.

AGAIN, you are thinking ONLY in terms of VOLTAGE. There is the opposite of isolation of the CURRENT, by the semiconductors, between the PSU rails and the load. The semiconductors are opening and closing their channels precisely in order to modulate the flow of current through them, on its way to the load. They FLOOD the load with always-changing current!

(The following rant is NOT directed at you.)

Why is it seemingly so difficult for so many people to think in terms of PSU output current? Why this fixation on a voltage-centric view of everything? Yes, keeping the power rails at a fixed DC voltage can be very important. But (in my mind, at least) that's not what this discussion is about!

Where do you people think that the current comes from, for the speakers? Does it come in through the signal input connectors? Does amplifying the input voltage magically make current appear from somewhere else? Or do the speakers only need VOLTAGE, alone? No!! Whether your final power amplification components are BJTs and current-controlled or FETs and voltage-controlled, their OTHER two terminals are small-signal-controlled higher-power-capable CURRENT VALVES that enable more or less CURRENT to flow through them, depending on the voltage across them and the control input level. THAT'S just what they DO. They are basically just like small-signal-controlled RESISTORS.

Or maybe I should have just said, "Can we talk about the current, for a little while, and not so much about the voltage?".
 
Why is it seemingly so difficult for so many people to think in terms of PSU output current? Why this fixation on a voltage-centric view of everything? Yes, keeping the power rails at a fixed DC voltage can be very important. But (in my mind, at least) that's not what this discussion is about!

Where do you people think that the current comes from, for the speakers? Does it come in through the signal input connectors? Does amplifying the input voltage magically make current appear from somewhere else? Or do the speakers only need VOLTAGE, alone? No!! Whether your final power amplification components are BJTs and current-controlled or FETs and voltage-controlled, their OTHER two terminals are small-signal-controlled higher-power-capable CURRENT VALVES that enable more or less CURRENT to flow through them, depending on the voltage across them and the control input level. THAT'S just what they DO. They are basically just like small-signal-controlled RESISTORS.

Or maybe I should have just said, "Can we talk about the current, for a little while, and not so much about the voltage?".

Hi gootee,

Finally, somebody who says it clearly. In fact, power supply buffer capacitors are (merely) to be considered current sources. Otherwise, when thinking in voltage, how who it be possible the feed 2 (or more) channels from a single power supply without crosstalk?

The big 'thinking' problem is that merely all text books / reference documents draw the power supply and amplifier circuits separately (without speakers), noting rail voltage as the unit to focus on, while the reality drawing should be a closed circuit schematic of power supply caps + OPS BJT/FET + speaker. Even better: draw up to schematic like this with both channels. This would demonstrate very well your current view story.

And, it would make an explanation and calculation of damping factor much more 'visible' and understandable as well (but that's off-topic).
 
just what is the relevance of this test to audio equipment? The pics may look nice, but relevant?
Superficially, that looks like plain common sense.
But in fact, it is a very naive and limited view of the problems involved.
Nowadays, audio gear consists of high speed DACs and ADCs, dsp cores, SMPS's, class D amplifiers, etc, with simple analog baseband applications being the minority.
Even simple analog sources like FM tuners handle base frequencies in the tens of MHz range.
Even when dealing with 20KHz- signals is the behavior at higher frequencies still relevant.
Try to feed an audio amplifier through an LCL lowpass filter of 20KHz, you will see the problem.
High speed, high performance amplifiers require a low supply impedance in the MHz range, or more.
Anyway, this thread is about the paralleling of small caps with big caps, and the supposed improvement it is supposed to bring, and in this case, nothing much happens below 1MHz.
Also, large power caps behave much worse below 20 kHz than the small caps tested in the above test, like I measured in this thread. There is much more to caps then only impedance, such as knowing the capacitor's own resonance frequency (where its ESR and impedance equals), before adding a bypass (or not, that I leave in the middle).
In absolute terms, a 1nF polypropylene cap is always better than a 1,000µF e-lytic, and that is valid at any frequency: the 90° angle between voltage and current will always be better respected with the PP.
The ESL too will be smaller.
Does this make the 1nF a better decoupling cap at 20Hz?
Of course not. The important thing, for the circuit to be bypassed is the module of the impedance, and this includes all the parameters of the complex impedance at that frequency.

Also note that DF values <0 have no physical sense.
 
Superficially, that looks like plain common sense.
But in fact, it is a very naive and limited view of the problems involved.
Nowadays, audio gear consists of high speed DACs and ADCs, dsp cores, SMPS's, class D amplifiers, etc, with simple analog baseband applications being the minority.

Elvee,

As you point out very well, I forgot to mention relevant 'to regular analogue amplifier', as to put my question in the proper context. Indeed, it is very relevant for class D and digital audio gear, albeit I would not exclude anything below 300 kHz either as done in that test (and imposed by his meter).

You are propably very knowledgable and much more advanced than many people out her, but nevertheless no need to call other forum members 'very naive and limited view' who just try to understand and learn.

Thanks.
 
marce,

I believe that we almost fully agree on every aspect that you mentioned. And also, I did not intend to sound like I was "railing" at you. I am sorry.

You are correct that we cannot hear DC. But again, "DC" refers only to the VOLTAGE rails of the PSU. The CURRENT output from the PSU is _NOT_ DC, when there is music power being supplied through a power amplifier by the power supply.

A minor correction or clarification:

Of course each PSU rail's curent is also "DC", since there is only a single polarity on each rail. What I meant was that it's not a fixed value of DC. When many people say "DC", they are also assuming a fixed value of DC, and I fell into the same trap.

We can't hear FIXED-value DC but we can certainly hear dynamically-changing DC, as we do every time we hear a "push-pull" amplifier.

This thread started (I thought, anyway) in order to discuss paralleling capacitors in a power supply, presumably a power supply that would be meant to supply power for the production of music through loudspeakers.

Since the power supply must supply the dynamic current that IS, usually extremely directly, the music signal which drives the loudspeakers, it seems to me that we are right on target in discussing parallel capacitors in the power supply, along with any parallel capacitors that are used to bypass/decouple the high-power current-control valves that are controlled by the input signal chain, such as the final power transistors or the output stages of chipamps or whatever is used in the output section of the "power amplifier". I often like to think of the bypass/decoupling caps as a sort-of local "point of load" power supply, anyway.

The possibilities for high-frequency resonances being formed between smaller capacitors and the parasitic inductance of larger co-located capacitors, or with the stray or parasitic inductance of attached wires or PCB traces, have been fairly-adequately discussed, already, in this thread.

I think that at least AndrewT and I are interested in understanding more about which capacitors supply which time and frequency components of the music signal current, and how to try to ensure that their implementation has been optimized for the most-faithful reproduction of the music.

My primary goal for an audio system is the most-faithful reproduction of the input, so that the output from the speakers is as close as possible to what was intended by the makers of the source material.

My two main questions related to this thread's topic seem to be:

In terms of the PSU and all of the capacitors upstream from the power pins of the output stage power devices (at least), what and how significant can their effects on the faithfulness of the music reproduction be, and, how would one best attempt to ensure that the cause of any flaw in the faithfulness of the music reproduction would not originate there?

Cheers,

Tom
 
OK, I just ran some simulations with LTSpice, using a schematic/model that I had built up five or six years ago, with a linear regulated power supply and an OP541E chipamp model (inside the feedback loop of an AD845j opamp, in this case), including a lot of the parasitics and strays of both the components and the conductors, and a lumped-component model of the speaker cables, plus lumped-component models for crossovers and speakers. I had a square wave as input (with controlled slopes for the rising and falling edges).

It's easy to see the current from the power rail not quite be able to keep up with the current drawn by the chipamp power pins, and watch as the bypass caps send out exponential-shaped current pulses to make up the difference that's needed, mostly for the transient responses. A 2.5 microsecond lag from the power rail when needing to go from 0 to over 5 Amps in 6 or 7 us means that the bypass caps have to very quickly supply several amps, for a short time, in order to keep the square wave edge linear and prevent undershoot and overshoot (keeping the "corners" sharp).

It's also easy to sum the currents of the PSU's flter caps and compare the total to the regulator input current and see that almost all of the current comes from the caps (the difference was only about 10-15 mA out of 5-6 A).

It was also easy to see severe high-frequency ringing when I added some very small film caps in parallel with the 2200 uF electrolytic bypass caps at the chip's power pins.

I'm interested in how different sizes and types of bypass caps can handle the demands for transient currents. At first I thought that fast transients would need small-value caps. But why would that be true? If it is true, would it be only because their ESR and ESL are usually able to be lower? I know that smaller caps can be fully-recharged faster. So maybe size would matter, then, for fast repetition rates. But considering a one-shot event, how would the capacitance value, itself, affect the initial rate (amps per second, say) of a capacitor's transient current response? Oh well, I guess I should also consider the external series resistance. But that would be very (relatively) low, roughly just the speaker impedance, when a large fast transient is demanded. I'll have to review the basic equations and circuits that used to be so familiar (and try it out in simulations). It's been too many years. But I'm out of time for now.
 
Last edited:
I guess maybe part of the question is about how many amps per second a real capacitor can give, initially, and then how long it can sustain that current-flow trajectory.

Just to try to put some ballpark numbers to it:

The magnitude of the maximum rate of change of a voltage sine wave is:

slew rate max of sine (in volts per microsecond) =

[(2 x Pi) x (freq in Hz) x (amplitude in volts)] / 1,000,000

So, for example, for a 60 V p-p sine at 20 kHz, the maximum slew rate would be about 7.54 Volts per microsecond, which occurs at the zero-crossing.

If the voltage is across a resistive load, the current amplitude would be the voltage amplitude divided by the resistance. So for the case above the current would also be a sine, of 7.5 A p-p. The maximum slew rate for a sine of that amplitude at 20 kHz would be about 0.94 Amps/usec.

So 0.94 Amps/usec, or maybe double that, should be about as fast as an 80-Watt amplifier should have to be able to change its output current, with an 8 Ohm purely-resistive speaker. So let's say 2 Amps per microsecond in order to be able to reproduce up to about 40 kHz, at that amplitude and with an 8 Ohm purely-resistive speaker load.

I guess that, at that rate, a lot of types of electrolytics might exceed their ripple-current ratings after 3 or 4 us.

-------------------

So what makes a smoothing or bypass capacitor release current at the precise time that it's needed? What makes a charged capacitor release current, in general, in the presence of an external voltrage across it? The external voltage must drop. I noticed in my simulations that small downward changes in the voltage across a large capacitor would be accompanied by relatively large flows of current.

In theory, a capacitor's current is the capacitance multiplied by the time rate-of-change of the voltage across the capacitor. But in my simulations (which I am quite sure do solve the differential equations correctly), the non-ideal capacitors generally didn't have enough time to react in order to conform to the ideal capacitor differential equation. What I "measured" from the output plots was that whenever the voltage had a transient downward blip of magnitude "delta V", then the magnitudes of the current pulses from the large electrolytic caps were about (delta V) / ESR.

And that makes sense, because that would be the maximum discharge current given by the solution to the capacitor equation, at time = 0 (assuming the ESR is the only resistance in the discharge path), and the times involved here were only long-enough for the cap to start trying to discharge and ramp quickly up to its maximum discharge current after 2 or 3 us, but then the voltage differential across the cap would go back to zero and so would the current. For example, in my simulations a temporary drop of the voltage by 70 mV would produce a capacitor output current pulse with a magnitude of about 2.3 Amps, from a capacitor with an ESR of 0.03 Ohms.

So I answered one of my own questions: The initial discharge current of an ideal capacitor does not depend on the capacitance. It is merely (roughly) the sudden change in the voltage across the capacitor divided by any nearby series resistance in the circuit.

I still don't quite know exactly what determines how much time it would take for a non-ideal capacitor to change its current from zero to the initial discharge rate that is called for. I'm guessing it involves the parasitic inductance.
 
Last edited:
And that makes sense, because that would be the maximum discharge current given by the solution to the capacitor equation, at time = 0 (assuming the ESR is the only resistance in the discharge path), and the times involved here were only long-enough for the cap to start trying to discharge and ramp quickly up to its maximum discharge current after 2 or 3 us, but then the voltage differential across the cap would go back to zero and so would the current. For example, in my simulations a temporary drop of the voltage by 70 mV would produce a capacitor output current pulse with a magnitude of about 2.3 Amps, from a capacitor with an ESR of 0.03 Ohms.

So I answered one of my own questions: The initial discharge current of an ideal capacitor does not depend on the capacitance. It is merely (roughly) the sudden change in the voltage across the capacitor divided by any nearby series resistance in the circuit.

I still don't quite know exactly what determines how much time it would take for a non-ideal capacitor to change its current from zero to the initial discharge rate that is called for. I'm guessing it involves the parasitic inductance.
I am following you intently.
You are one of a few that majors on current capability.
Cordell talks of current slew rate and current clipping as well as the conventional voltage slew rate and voltage clipping.

I agree that the "size" value of the current spike supplied by a capacitor is not controlled by the capacitance value.

The duration of that current spike is determined by the value of capacitance.
That's precisely why HF decoupling must be right next to the main current consumers in the very shortest tightest loop possible.
The MF decoupling can tolerate a bit more in the way of resistances/impedances and so they are located next on a slightly longer loop.
The smoothing in the PSU is the LF decoupling (that's a new term from my fingers) can be located much further away. Some even put them in a remote chassis.
Look at Peter Daniel's implementation of his chipamp. HF decoupling at the chip. MF decoupling on board a little further away. LF decoupling dispensed with completely. The result is his claim that this gives superb mid & treble performance. It is obvious that he follows the instantaneous current demand/supply philosophy.
 
I am following you intently.
You are one of a few that majors on current capability.
Cordell talks of current slew rate and current clipping as well as the conventional voltage slew rate and voltage clipping.

I agree that the "size" value of the current spike supplied by a capacitor is not controlled by the capacitance value.

The duration of that current spike is determined by the value of capacitance.
That's precisely why HF decoupling must be right next to the main current consumers in the very shortest tightest loop possible.
The MF decoupling can tolerate a bit more in the way of resistances/impedances and so they are located next on a slightly longer loop.
The smoothing in the PSU is the LF decoupling (that's a new term from my fingers) can be located much further away. Some even put them in a remote chassis.
Look at Peter Daniel's implementation of his chipamp. HF decoupling at the chip. MF decoupling on board a little further away. LF decoupling dispensed with completely. The result is his claim that this gives superb mid & treble performance. It is obvious that he follows the instantaneous current demand/supply philosophy.

AndrewT,

Thanks. I was beginning to wonder if this thing was on or not, or if I was alone, here. But, in this, compared to Cordell and Daniel (and many others here), I am like a child... or only an egg.

I worry that maybe this angle is not even well worth pursuing and was hoping that someone who has already investigated it would jump in with some enlightenment. (Or maybe this is all just the most recent skirt passing by in the crowd that I randomly happened to notice.)

It "seems" like it would be very helpful to be able to actually quantify the capacitance requirements and characteristics, and quantify the effects of decoupling cap parameter changes (in terms of the capabilities and limitations of the output's reproduction fidelity). [Edit: (Engineer may be starting to wake up a little: ) We can break this down and first define the ideal current behaviors and requirements without thinking about capacitors. Then we could express how those might be able to be produced or satisfied, ideally and then in reality.]

I worry about transient reproduction accuracy, and "believe" that it's very important to be able to reproduce all of the Fourier components with the correct amplitude and phase. Otherwise, edges could become blurred, or exaggerated, or otherwise distorted, et al. This might all be merely "chasing the last 0.1% improvement" but that seems to be one reasonable definition of hi-fi.

Maybe it will make things clearer to me if I start from the other end of the spectrum and try to see what the absolute bare minimum capacitance requirements would be, under varying conditions (power levels, signal types, etc), and why.

I am pretty sure that the equations for the ideal case will be quite simple. Thirty years ago I probably could have had a complete closed-form solution within a few minutes. I deeply regret having allowed myself to gradually spiral down from deft to inept, in this area.

I'll do some more research and thinking, and will hope to be able to report back with more than just whining.

Cheers,

Tom

P.S. I wonder why no one markets some type(s) of "multi-capacitor" packages that would fit neatly right across device power pins and contain a continuum of capacitance with an optimal geometry for decoupling/bypass, with a range of models for different applications or for frequency ranges of interest (or whatever the criteria should be). Or why not have easily-stackable packages so one could "roll their own" and still get the best-possible geometry? And why isn't there a single package containing the ubiquitous "10uF || 0.1 uF" with an optimal geometry?
 
Last edited:
Hi Gootee,
yes I agree we do agree:)
Regarding cappacitor supplying power, there is a good reference on this in one of Xylinx Virtes 4 or 5 data sheets. Though digital based it covers the same concept, providing power to supply instant current requirement for switching, while the main PSU reacts.
Any capacitor would (within reason) do the job if it was just capacitance that was to be considered, but the series ESL is the problem. That is why small physical lower value packages are placed next to the device pins, as the parasitic inductance is low they can supply the almost instantaneous requirement for current, then moving further away from the device power pins the larger reservoir caps, supply the next current requirements, then the power supply output cap(s), then finaly the voltage regulator, this being the slowest to react.
I will dig out the app notes etc I have on this when I'm back a work.
On local bypassing of large capacitors, I cannot find any designs that I have worked on where it has been done, on any power supply! it is not common practice (in the other world:))
 
Again,
there are multicap packages available, these are quite often BGA or other small form factor SMD package, or there is the proadlizer:
http://www.nec.co.jp/techrep/en/journal/g09/n01/090116.pdf

Again it is package size that dictates the effectiveness of a given cap value, so you can find 1n 10n combinations and 10n 100n combinations.
For digital designs the best method is closely coupled power and ground planes (sub 0.05mm dialectric) to give planar capacitance across the planes.
 
Hi Gootee,
yes I agree we do agree:)
Regarding cappacitor supplying power, there is a good reference on this in one of Xylinx Virtes 4 or 5 data sheets. Though digital based it covers the same concept, providing power to supply instant current requirement for switching, while the main PSU reacts.
Any capacitor would (within reason) do the job if it was just capacitance that was to be considered, but the series ESL is the problem. That is why small physical lower value packages are placed next to the device pins, as the parasitic inductance is low they can supply the almost instantaneous requirement for current, then moving further away from the device power pins the larger reservoir caps, supply the next current requirements, then the power supply output cap(s), then finaly the voltage regulator, this being the slowest to react.
Thinking in the time domain can be dangerous and misleading if you are unable to actually visualize the differential equations.
Maybe some people can do it, but I honestly cannot.
Compensating the connections inductance to the reservoir capacitors with a small, local capacitor looks like a good idea, but unfortunately, the solution of such a system is oscillatory, meaning in fact a resonance peak in the frequency domain.
Servoed electronic supplies displaying a delay in the reaction when a voltage drop appears at the output are in fact gyrators, ie synthetic inductance and have the same kind of problems.
Which is why high performance regulators have difficulties tolerating perfect capacitive loads at their outputs.
The problem is always the same, and is mainly incorrectly addressed, because people are unable to think vectorially in the time domain, and unwilling to think in the frequency domain.
In the frequency domain, only the module is important, and even if you are vectorially challenged, like myself, ie unable to think in multiple dimensions simultaneously, you can get away with it.

Compensating a reactive element with the opposite reactive element looks like a good idea until you realize this works at one, and only one frequency: the resonance.
 
Luckily there is cool software that does a lot of the work, and provides pretty liitle pictures, in my case this:
http://quadrasol.co.uk/useruploads/...werintegrityadvanced_draft_eng_2011_10_05.pdf
Trouble is the software isn't cheep (my complete PCB design system cost 10K a year in maintenance alone!!), but when you have a few 700,800+ pin devices on a board, plus memory plus analogue, plus audio out etc you need the kit.
As the guy who teaches people how to use the software says, you dont have to know the maths, thats what computers are for...
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.