All else being equal, a larger capacitance has a slower MINIMUM "speed" than a small capacitance. Not a slower MAXIMUM "speed".
ESL and ESR can be unequal, but paralleling small caps with large ones solves this,
Smaller caps do respond quicker.
Unfortunately, they respond to high power demands by collapsing the supply voltage. OTOH, they recover faster too, so the full supply voltage is restored very quickly when you don't need it.
Well, obviously. That's because discrete designers need to know something about electronics, otherwise they wouldn't be able to design discrete circuitry.
In contrast, a lot of chipamp "designers" chose to go the chipamp route in the first place precicely because they don't know much (if anything) about electronics. They just copy the simplest circuit they can find in the app-note (or on the internet, for those who wouldn't recognize an app-note if it slapped them in the face). Then they put all their "design" effort into making the box look pretty (and choosing what color capacitors to use, of course).
Fortunately for those with commercial aspirations, this plays right into the hands of certain magazine reviewers who judge audio products strictly by sound quality, and base their evaluation of sound quality strictly on look&feel and price (Don't be a dumbass and try to sell an LP demagnetiser for $35.00 - Everybody knows a decent demagnetiser costs at least a couple thousand bucks).
Unfortunately, gullibility tends to be proportional to ignorance, so most of them will have swallowed the "capacitance is bad" kool-aid even before they built their first chipamp, and this gets reflected in their "designs".
I was following JoeryTech his build because his use of materials (which looks good).
I also have a LM4780 (dual mono) (Audiosector) which sounds fine!!!!, but after a while I found out that my setup had a "lack" of bass. Reading JT his experience it seemed that he does not seem to have a "lack of " bass. To my surprise the thread went off topic due to the bass question. And JT started this thread. There are many replies but none of them gives “THE” answer. I do understand that “THE” answer is hard to give.
But what to do?
Well first of all I think that Peter Daniel brings us a perfect kit so I first checked the rest of my setup.
At the source-end of my setup there was progress to be made. My Musical Fidelity V-Dac II and my Musical Fidelity V-LPS II were in need of a better power supply. So I build a new PS for these devices. The new PS (still on a bit of wood) made a HUGE!! difference and I wished that I had done this upgrade a bit earlier.
But still my other diy amp (Arjen Helder TA2020 MKIII) seem to give a better bass with this new.. source, but in the highs and mids the LM4780 won at this point.
But what about the bass and capacitance?
Reading many threads these two seem to be linked. There seem to be pro’s and con’s about adding caps. My skills on this DIY are a bit poor so some off the replies are a bit difficult to understand and I don’t have a pile of parts laying around to test different setups. But than it came to me , I have a snubberized PS on my LM3886 from Chipamp.com with 2x 10.000uf , which is a bit offline at the moment.
I got my soldering iron out and put these snubberized PS units in my LM4780.
The result?
It is hard to tell if the “high’s” and the “mid’s” had any benefit of this change but what I can say that the bass in “this” setup increased dramatically. Is 10.000uf (one cap, with snubber) at each rail the answer? Well I am pleased so far, so I am going to build a new PS for the LM4780 and add the possibility to use large caps or many small caps.
I realize that this is not “THE” answer but yet another opinion or/and possibility which I am happy with.
I also have a LM4780 (dual mono) (Audiosector) which sounds fine!!!!, but after a while I found out that my setup had a "lack" of bass. Reading JT his experience it seemed that he does not seem to have a "lack of " bass. To my surprise the thread went off topic due to the bass question. And JT started this thread. There are many replies but none of them gives “THE” answer. I do understand that “THE” answer is hard to give.
But what to do?
Well first of all I think that Peter Daniel brings us a perfect kit so I first checked the rest of my setup.
At the source-end of my setup there was progress to be made. My Musical Fidelity V-Dac II and my Musical Fidelity V-LPS II were in need of a better power supply. So I build a new PS for these devices. The new PS (still on a bit of wood) made a HUGE!! difference and I wished that I had done this upgrade a bit earlier.
But still my other diy amp (Arjen Helder TA2020 MKIII) seem to give a better bass with this new.. source, but in the highs and mids the LM4780 won at this point.
But what about the bass and capacitance?
Reading many threads these two seem to be linked. There seem to be pro’s and con’s about adding caps. My skills on this DIY are a bit poor so some off the replies are a bit difficult to understand and I don’t have a pile of parts laying around to test different setups. But than it came to me , I have a snubberized PS on my LM3886 from Chipamp.com with 2x 10.000uf , which is a bit offline at the moment.
I got my soldering iron out and put these snubberized PS units in my LM4780.
The result?
It is hard to tell if the “high’s” and the “mid’s” had any benefit of this change but what I can say that the bass in “this” setup increased dramatically. Is 10.000uf (one cap, with snubber) at each rail the answer? Well I am pleased so far, so I am going to build a new PS for the LM4780 and add the possibility to use large caps or many small caps.
I realize that this is not “THE” answer but yet another opinion or/and possibility which I am happy with.
I agree with you. THAT is the purpose of an smoothing capacitor. To store energy for a brief period of time while the sinuoidal signal from the mains passes through the X axis, thus providing 0 volts and 0 current. So to avoid the supply voltage to drop to 0, you put capacitance there. But capacitors are finite in their capabilities, so their energy gets "exahusted" and they reflect that by suplying a lower voltage as time goes by. Then when the sinuoidal cycle start going up from 0 volts, as soon as that voltage is above the one from the capacitors, they get charged.
So in an unregulated PSU you get what is called ripple, nothing more than the variation of that supply voltage. It is supposed to be constant, but it isn't. The more constant it is, the less the performance of the amplifier varies with time. Take the LM3886 datasheet and see all the parameters that are affected by the supply voltage. Now think they will be varying like 100 or 120 times per second. Not a good thing.
That's why the more capacitance, the better. With some 'but's.
Well some still listens to mono recordings
Anyway, 4700uF minimum. I like it 10mF or much more. If it is even more then soft start is needed.
What's your opinion on regulated PSU's for Gainclones? I used some LM338 on one of them. Have you heard one of them?
How do you measure how "fast" is a capacitor? By its ESR?
And yes, as you said, those caps are small enough to provide some quality music. But I don't think diyers strive for "enough". It isn't enough, at least for me.
You talk about "mythical" bypass capacitors, so you are implying that bypassing isn't an extended industrial and professional practice and its only merit is being used by some audio diyers?
There isn't a measurement for speed. It's done independantly by people that either make faster capacitors, or need them in their equipment. For use lamers we can just use ears and principles. I know, it sucks...
I use multiple small capacitors. While one could let the voltage sag greatly, creating large ripple current, one can also use enough capacitance that it's very negligable. PD figured out very good balances, which eliminate other problems discussed that can come up with a need for snubber. He said snubbers don't sound right.
Search for bypass capacitors. There's some big threads on how they're basically total BS in most situations.
I use multiple small capacitors. While one could let the voltage sag greatly, creating large ripple current, one can also use enough capacitance that it's very negligable. PD figured out very good balances, which eliminate other problems discussed that can come up with a need for snubber. He said snubbers don't sound right.
Search for bypass capacitors. There's some big threads on how they're basically total BS in most situations.
I have read the following article:
http://www.tnt-audio.com/clinica/ssps2_e.html
And I quote from that:
[FONT=Arial, Helvetica]"Capacitors
As noted before, electrolytic capacitors serve two functions - to filter out the rectified supposedly DC voltage and to act as energy storage for those peaks which may require very large currents. Hence, they also stabilize the voltage by acting as energy reservoirs which offload the transformer, but this function is not of prime importance (however, it is far from being unimportant). I have set the priority list here: 1) filtering and 2) energy reserves.
In practice, manufacturers tend to use them as a panacea or cure-all for all other in-built power supply shortcomings. There is no doubt that filter capacitors have crucial influence on the sound obtained - that much is agreed upon by all, no matter what school they belong to. Therefore, they require special consideration."
It sounds like this guy have a point.
[/FONT]
No.
The ESL tells you how it reacts to changes in current demand.
Then you MUST add on the inductance of the connections and cables and traces from the supply to the device that changes the current.
The connection length dominates the total inductance.
That's where small capacitors can show an improvement. They can be placed close to, or in, the circuit that changes it's current demand.
Big caps must by size be located outside the circuit.
It is for those reasons that properly designed decoupling is in at least two stages.
Stage 1 is the HF decoupling located right at the point of changing current demand. Tiny X7R are the only size that gets right inside the circuit.
Stage 2 is the MF decoupling and is usually smallish electrolytic located a bit farther away, maybe 20mm to 40mm, or so.
Generally smoothing capacitance is at the transformer. This can never operate as decoupling. It is simply too far away.
Omitting the smoothing capacitance at the transformer forces the designer to bring the charging pulses to the capacitors at the amplifier.
No one will ever convince me that bringing those charging pulses to the amplifier is good for sound quality. Think about it!
One downside first come to my mind is that it isn't nearly the same to carry high amperage AC along the internal cabling compared to carrying DC, regarding EMI on near placed low signal cabling.
Low Capacitance
I am going to make a few bold statements here, all based on theory and simulation why low capacitance is a (very) bad thing to have. (Bold = could be wrong)
1. The rails are connected to the input via the output and feedback resistor.
2. The rail ripple voltage is added to the input signal via the output and feedback resistor.
3. Because the rail ripple voltage has a sawtooth wave shape, it will add HF to the input signal and could this make the chip oscillate?
Sims are of a LM3886 with single cap per rail. The load is 4 ohm.
Sim.1 is 22 Hz sin wave with 1000 uF cap
Sim.2 is 22 Hz sin wave with 10000 uF cap
Sim.3 is 22 Hz sin wave with 1000 uF cap showing charging pulses.
V(OUT) = Voltage Pin 3 - GND
V(n001) = Voltage Pin 1 - GND
V(N001,OUT) = Voltage Pin 1 - Pin 3
I(C1) = Capacitor Current
I am going to make a few bold statements here, all based on theory and simulation why low capacitance is a (very) bad thing to have. (Bold = could be wrong)
1. The rails are connected to the input via the output and feedback resistor.
2. The rail ripple voltage is added to the input signal via the output and feedback resistor.
3. Because the rail ripple voltage has a sawtooth wave shape, it will add HF to the input signal and could this make the chip oscillate?
Sims are of a LM3886 with single cap per rail. The load is 4 ohm.
Sim.1 is 22 Hz sin wave with 1000 uF cap
Sim.2 is 22 Hz sin wave with 10000 uF cap
Sim.3 is 22 Hz sin wave with 1000 uF cap showing charging pulses.
V(OUT) = Voltage Pin 3 - GND
V(n001) = Voltage Pin 1 - GND
V(N001,OUT) = Voltage Pin 1 - Pin 3
I(C1) = Capacitor Current
Are you saying that
omitting the transformer located smoothing capacitors has a downside
or
keeping the transformer located smoothing capacitors has a downside.
Are you claiming:
that AC currents in wiring emit EMI
or
that DC currents in wiring emit EMI.
or
that changing currents in wiring emit EMI
Last edited:
I have underlined it, I am sure of the first one, but now you gave me something to think about regarding the second question. My bet goes to changing currents, but this is not a gambling game, so I would be glad if you could answer to that in a pair of lines
I studied the psu and decoupling capacitance value problems for quite a while, along with the closely-related parasitic inductance problem. The best way to think about it is to realize that the psu current makes the audio signal that we hear. The psu voltage is much less important than the current. The voltage only has to stay high-enough to prevent clipping, and keeping it more constant can improve linearity. But the current is what needs to be agile and accurate.
The minimum required bulk capacitance value can be calculated, assuming we don't want the amp to run out of current in between charging pulses. And the decoupling cap values and their maximum distances from the power pins (inductance) can be calculated based on the worst-case transient response or highest frequency that could be needed.
For the minimum bulk capacitance, find my spreadsheet but make sure you get the latest version. For the decoupling, just make sure that you place them right AT the chip's power pins, if possible, and there it can be better to use multiple parallel caps, but use a total of at least 470 uF for each power pin, for an LM3886, per the datasheet. (It can be better and cheaper to use multiple parallel caps for the bulk capacitance, too, but to get the most benefit they would need to be laid out properly. Sorry, no time to explain, right now. Do some searches.)
Also use a physically very small (lead spacing) cap for high-frequency bypass, directly from each power pin to ground, for hf stability.
Cheers,
Tom Gootee
The minimum required bulk capacitance value can be calculated, assuming we don't want the amp to run out of current in between charging pulses. And the decoupling cap values and their maximum distances from the power pins (inductance) can be calculated based on the worst-case transient response or highest frequency that could be needed.
For the minimum bulk capacitance, find my spreadsheet but make sure you get the latest version. For the decoupling, just make sure that you place them right AT the chip's power pins, if possible, and there it can be better to use multiple parallel caps, but use a total of at least 470 uF for each power pin, for an LM3886, per the datasheet. (It can be better and cheaper to use multiple parallel caps for the bulk capacitance, too, but to get the most benefit they would need to be laid out properly. Sorry, no time to explain, right now. Do some searches.)
Also use a physically very small (lead spacing) cap for high-frequency bypass, directly from each power pin to ground, for hf stability.
Cheers,
Tom Gootee
Last edited:
Mark,
Yes, there is a hidden positive feedback loop through the rails, for high frequencies, in almost every transistor-based amplifier. The small bypass caps should prevent any problems.
Also, ripple voltage is only a sawtooth when the load is a constant DC current (i.e. usually only in trivial textbook examples). Try sims with ltspice, with WAV files of real music as the input signal, for example (see the link in my signature, which shows a single drum strike in the intro to "Highway to Hell" by AC-DC). The ripple current is the music signal and the ripple voltage is the integral of that, combined with the sawtooth effect as the caps drain between charging pulses, which depends on the total power output moving average, or something like that. But the music signal can easily dominate the ripple voltage.
Yes, there is a hidden positive feedback loop through the rails, for high frequencies, in almost every transistor-based amplifier. The small bypass caps should prevent any problems.
Also, ripple voltage is only a sawtooth when the load is a constant DC current (i.e. usually only in trivial textbook examples). Try sims with ltspice, with WAV files of real music as the input signal, for example (see the link in my signature, which shows a single drum strike in the intro to "Highway to Hell" by AC-DC). The ripple current is the music signal and the ripple voltage is the integral of that, combined with the sawtooth effect as the caps drain between charging pulses, which depends on the total power output moving average, or something like that. But the music signal can easily dominate the ripple voltage.
Last edited:
changing current along a conductor leads to transmitted interference.
That changing current can be AC, or non constant DC current.
It is the CHANGE in current that matters and it is the rate of change that determines how bad the interference gets.
an example of non constant DC current is the supply line running to an audio amplifier.
The current flow is always in ONE direction.
The current varies a lot and depends on the demands made by the load.
ClassB and ClassAB and overloaded** ClassA are all bad for this type of interference.
In each of these amplifier types the current can change from near zero instantaneously to some higher value. it is that instantaneous change that is bad.
Cordel and others, have described how to reduce the transmission of this interference.
Couple the +ve and -ve supply together so that the combined current flow becomes more like the AC supplied to the load AFTER the amplifier output.
Then couple the return wire to the two flow wires to create a low loop area triplet.
This becomes very important when low level stages are nearby.
overloaded** is not intended as a damage inducing overload. I am referring to an output current transient that exceeds the maximum ClassA capability of the amplifier.
That changing current can be AC, or non constant DC current.
It is the CHANGE in current that matters and it is the rate of change that determines how bad the interference gets.
an example of non constant DC current is the supply line running to an audio amplifier.
The current flow is always in ONE direction.
The current varies a lot and depends on the demands made by the load.
ClassB and ClassAB and overloaded** ClassA are all bad for this type of interference.
In each of these amplifier types the current can change from near zero instantaneously to some higher value. it is that instantaneous change that is bad.
Cordel and others, have described how to reduce the transmission of this interference.
Couple the +ve and -ve supply together so that the combined current flow becomes more like the AC supplied to the load AFTER the amplifier output.
Then couple the return wire to the two flow wires to create a low loop area triplet.
This becomes very important when low level stages are nearby.
overloaded** is not intended as a damage inducing overload. I am referring to an output current transient that exceeds the maximum ClassA capability of the amplifier.
Last edited:
OK, I got all of it right. Thank you for that.
So it comes to say that having local decoupling capacitors near the IC helps in reducing that changing current along the whole wire, and keeping those drastic changes in current flow in the piece of wire between the local decoupling caps and the IC.
So it comes to say that having local decoupling capacitors near the IC helps in reducing that changing current along the whole wire, and keeping those drastic changes in current flow in the piece of wire between the local decoupling caps and the IC.
Follow on from #32
We are talking about a LM3886 with only one Cap per rail (1000uF).
If the chip had no feedback, the Load voltage would change with the rail voltage, because the output voltage is set by the feedback, the load voltage is upheld. this is accomplished by increasing the conductivity of the power transistors of the chip.
Demo:
Input = 0.5V DC
Gain = 20
Output = 10 V (Load)
Rails = 30 V
When the cap is fully charged, the Load voltage is 10V with a rail voltage of 30V the voltage over the chip (pin1 to pin3) = 20V
as the Caps drain, the Rail voltage reduces to 20V, the Load voltage will stay at 10V, so the voltage over the chip (pin1 to pin3) = 10V
Switching abruptly between these two conditions at a rate of 100 time per second for both rails will mean that the output transistors will be working very hard even with a DC input.
With an audio input , the rail ripple is combined with the audio input. If you plot the signals between pin 1 and 3 and between pin 4 and 3 you can see what is happening (Red line in the diagrams at #32).
We are talking about a LM3886 with only one Cap per rail (1000uF).
If the chip had no feedback, the Load voltage would change with the rail voltage, because the output voltage is set by the feedback, the load voltage is upheld. this is accomplished by increasing the conductivity of the power transistors of the chip.
Demo:
Input = 0.5V DC
Gain = 20
Output = 10 V (Load)
Rails = 30 V
When the cap is fully charged, the Load voltage is 10V with a rail voltage of 30V the voltage over the chip (pin1 to pin3) = 20V
as the Caps drain, the Rail voltage reduces to 20V, the Load voltage will stay at 10V, so the voltage over the chip (pin1 to pin3) = 10V
Switching abruptly between these two conditions at a rate of 100 time per second for both rails will mean that the output transistors will be working very hard even with a DC input.
With an audio input , the rail ripple is combined with the audio input. If you plot the signals between pin 1 and 3 and between pin 4 and 3 you can see what is happening (Red line in the diagrams at #32).
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.