Designing without the caps
I think Jan's comments are right on the mark, and the design route I'm taking in all new designs.
Not only do caps distort the signal path, but polarised types degrade with time.
Removing the caps eliminates their effects and brings long-term consistent performance.
Good PSU design can reduce the necessity for local decoupling, but these components are less problematic than the signal path.
It makes the design harder, but designing for good DC stability also tends to equal consistent operating area and performance.
My current project is to replace my existing DC coupled class A headphone amo with a totally discrete design. It's a much more challenging design process than using an op-amp / follower, whether it's worth it remains to be seen, if nothing else I'm learning masses about discrete design!
Andy.
I think Jan's comments are right on the mark, and the design route I'm taking in all new designs.
Not only do caps distort the signal path, but polarised types degrade with time.
Removing the caps eliminates their effects and brings long-term consistent performance.
Good PSU design can reduce the necessity for local decoupling, but these components are less problematic than the signal path.
It makes the design harder, but designing for good DC stability also tends to equal consistent operating area and performance.
My current project is to replace my existing DC coupled class A headphone amo with a totally discrete design. It's a much more challenging design process than using an op-amp / follower, whether it's worth it remains to be seen, if nothing else I'm learning masses about discrete design!
Andy.
Re: don't forget compensation capacitors.
ALSO feedback capacitors;very important. Change that ceramic one and you will be surprised.
-------------------------------------------------------------mirlo said:
ALSO feedback capacitors;very important. Change that ceramic one and you will be surprised.
"Good PSU design can reduce the necessity for local decoupling, but these components are less problematic than the signal path."
Andy,
Don't forget that, except for complete symetric H structure in pure class A (which drains strictly constant current), last capacitors in PSU are in signal path... (Load current flow across them, in part or totality)
Regards, P.Lacombe
Andy,
Don't forget that, except for complete symetric H structure in pure class A (which drains strictly constant current), last capacitors in PSU are in signal path... (Load current flow across them, in part or totality)
Regards, P.Lacombe
Agreed
"Don't forget that, except for complete symetric H structure in pure class A (which drains strictly constant current), last capacitors in PSU are in signal path... (Load current flow across them, in part or totality)
"
Agreed, but their effect is attenuated by the circuit PSRR, which varies from good to non-existent - admittedly in the same design, sometimes 😉
Designing for high PSRR will be the subject of a future thread, since I'm deep into some discrete design at the moment, and it's one of the factors I'm interested in. The obious benefits of current sources are evident, yet I'm not sure they are universally applicable or even desirable.
Andy.
"Don't forget that, except for complete symetric H structure in pure class A (which drains strictly constant current), last capacitors in PSU are in signal path... (Load current flow across them, in part or totality)
"
Agreed, but their effect is attenuated by the circuit PSRR, which varies from good to non-existent - admittedly in the same design, sometimes 😉
Designing for high PSRR will be the subject of a future thread, since I'm deep into some discrete design at the moment, and it's one of the factors I'm interested in. The obious benefits of current sources are evident, yet I'm not sure they are universally applicable or even desirable.
Andy.
Andy,
Even with high PSRR ratio, the problem remains : the alternative current which flows across the load must return to the amplifier power supply rails, half to the positive, half to the negative. So, it must flow across electrochemicals capacitors, and the "common ground" at this point is not as stable as desirable.
On the other hand, high PSRR are usually obtained by high FB factors, which can cause very unpleasant IMD.
This is why very high quality capacitors are needed. Use of regulated supplies displace the problem, but not solve it : signal flows across the regulation amplifiers, which become parts of the whole amplifier, adding their distortions...
Regards, P.Lacombe
Even with high PSRR ratio, the problem remains : the alternative current which flows across the load must return to the amplifier power supply rails, half to the positive, half to the negative. So, it must flow across electrochemicals capacitors, and the "common ground" at this point is not as stable as desirable.
On the other hand, high PSRR are usually obtained by high FB factors, which can cause very unpleasant IMD.
This is why very high quality capacitors are needed. Use of regulated supplies displace the problem, but not solve it : signal flows across the regulation amplifiers, which become parts of the whole amplifier, adding their distortions...
Regards, P.Lacombe
Voodoo
"Even with high PSRR ratio, the problem remains : the alternative current which flows across the load must return to the amplifier power supply rails, half to the positive, half to the negative. So, it must flow across electrochemicals capacitors, and the "common ground" at this point is not as stable as desirable."
Whilst the current path is not an area of disagreement, surely any distortions introduced by the capacitors must be attenuated by the circuit PSRR. Unless we support the idea of magic and voodoo so often attributed to audio. Any anomalies introduced by the power supply components must manifest themselves as an error somewhere within the amplifier, ultimately producing an error at the amplifier output. Whether this is measurable by conventional means, or by simply listening tests is irrelevant, it must be an error of some sort to be audible / measurable.
Your comments w.r.t. regulation are indeed valid, but since no capacitor can ever be as effective (at least at LF) as active regulation, this is a necessity IMO. I would support the argument that the regulation circuitry's dynamic performance must be at least as good as the circuit being powered - preferably better, in order to reduce the effect of it's sonic signature. The better the active regulation the closer it approximates the 'perfect' supply, that of a DC rail with no AC content.
The sonic signature results from degradations in the PSU dynamic performance that gives rise to AC components (again errors) at the PS o/p.
With regard to the common ground stability, is it's stability with reference to an absolute ground, or relative to the circuitry the relevant factor?
If the former it's always going to be limited, but if the latter very careful consideration of current paths can eliminate the bulk of the effects. Reducing current flows in the critical signal reference makes a massive difference to sound quality. It's something I discovered long ago in a simple headphone amp - decoupling both +ve and -ve rails to 0V ruined the sound completely, despite total star wiring. Decoupling +ve to -ve, using the same capacitor type, produced a stupendous improvement, for less cost.
Andy.
"Even with high PSRR ratio, the problem remains : the alternative current which flows across the load must return to the amplifier power supply rails, half to the positive, half to the negative. So, it must flow across electrochemicals capacitors, and the "common ground" at this point is not as stable as desirable."
Whilst the current path is not an area of disagreement, surely any distortions introduced by the capacitors must be attenuated by the circuit PSRR. Unless we support the idea of magic and voodoo so often attributed to audio. Any anomalies introduced by the power supply components must manifest themselves as an error somewhere within the amplifier, ultimately producing an error at the amplifier output. Whether this is measurable by conventional means, or by simply listening tests is irrelevant, it must be an error of some sort to be audible / measurable.
Your comments w.r.t. regulation are indeed valid, but since no capacitor can ever be as effective (at least at LF) as active regulation, this is a necessity IMO. I would support the argument that the regulation circuitry's dynamic performance must be at least as good as the circuit being powered - preferably better, in order to reduce the effect of it's sonic signature. The better the active regulation the closer it approximates the 'perfect' supply, that of a DC rail with no AC content.
The sonic signature results from degradations in the PSU dynamic performance that gives rise to AC components (again errors) at the PS o/p.
With regard to the common ground stability, is it's stability with reference to an absolute ground, or relative to the circuitry the relevant factor?
If the former it's always going to be limited, but if the latter very careful consideration of current paths can eliminate the bulk of the effects. Reducing current flows in the critical signal reference makes a massive difference to sound quality. It's something I discovered long ago in a simple headphone amp - decoupling both +ve and -ve rails to 0V ruined the sound completely, despite total star wiring. Decoupling +ve to -ve, using the same capacitor type, produced a stupendous improvement, for less cost.
Andy.
Cap distortion etc
Andy,
I'm there with you completely. I agree with your comments re: regulater psu. It is sometimes difficult to reason these things unbiased by so called "common knowledge".
I like the black box concept: The power supply can be seen as a black box with 2 terminals. The object is to make the box in such a way the there is pure DC at the terminals with zero impedance. If you put a regulated supply in the box and you get closer to the object, then that is better. Talking about "electrochemicals" inside the box, or "the regulator becomes part of the amplifier" trouble the issue IMO.
You have a black box, and if the impedance is not zero, the load current will cause an AC component on the box terminals, and through the amp PSRR some of this component reaches the output, either pure or a spart of some complex internal IMD.
And yes, that must be measurable/audible. If not, it isn't there.
Jan Didden
Andy,
I'm there with you completely. I agree with your comments re: regulater psu. It is sometimes difficult to reason these things unbiased by so called "common knowledge".
I like the black box concept: The power supply can be seen as a black box with 2 terminals. The object is to make the box in such a way the there is pure DC at the terminals with zero impedance. If you put a regulated supply in the box and you get closer to the object, then that is better. Talking about "electrochemicals" inside the box, or "the regulator becomes part of the amplifier" trouble the issue IMO.
You have a black box, and if the impedance is not zero, the load current will cause an AC component on the box terminals, and through the amp PSRR some of this component reaches the output, either pure or a spart of some complex internal IMD.
And yes, that must be measurable/audible. If not, it isn't there.
Jan Didden
Andy, Jan,
I agree with you, except : I do have nothing to do with Voodoo or another magic consideration of audio. All we are talking about must be explained by the common laws of physics.
But, can we establish that all that is audible, is measurable ? Which procedure is used to insure that all kinds of sound degradation are taken in account ? How can we be sure that unknown phenomenon cannot occurs under some circonstances ? This has been the case for TID, some years ago... Einstein said that knowledge is necessarly limited, but imagination isn't.
This is the reason why I suggest to study the behaviour of an audio amplifier on both analysis and synthesis. And I cannot agree with the idea to consider a black box, without taking care about that is exactly inside the box : Make perfect ketchup with rotten tomatoes is impossible, any chemical analysis cann't overcome this.
Regards, P.Lacombe.
I agree with you, except : I do have nothing to do with Voodoo or another magic consideration of audio. All we are talking about must be explained by the common laws of physics.
But, can we establish that all that is audible, is measurable ? Which procedure is used to insure that all kinds of sound degradation are taken in account ? How can we be sure that unknown phenomenon cannot occurs under some circonstances ? This has been the case for TID, some years ago... Einstein said that knowledge is necessarly limited, but imagination isn't.
This is the reason why I suggest to study the behaviour of an audio amplifier on both analysis and synthesis. And I cannot agree with the idea to consider a black box, without taking care about that is exactly inside the box : Make perfect ketchup with rotten tomatoes is impossible, any chemical analysis cann't overcome this.
Regards, P.Lacombe.
P.Lacombe said:
Mr Lacombe (Pierre?),
I absolutely agree that anything that is audible is measureable.
If we cannnot measure it, it is because a) we only thought we heard something, or b) we are not measuring right (yet).
I'm not sure what you mean by "analysis and[B/] synthesis". But to come back to the black box: my point is, if you have a kettle with perfect ketchup, why do you want to know whether the tomatoes inside are rotten, good quality or maybe just potatoes? I'm not sure we are talking on the same wavelength here, but what I mean is that all the talk about electrolyse and "common knowledge" takes the focus away from the issue at hand.
Am I making sense?
Jan Didden
Seconded
Jan has summarised my views - I was not arguing 'measurment vs audibility'. I consider my ears part of my measurement apparatus, indeed they are the only thing that matters to the end result.
The problem is my brain gets in the way, the ears are an exceptionally sensitive transducer, the brain is a totally rubbish analyser, since my moods, health, weather etc. all affect it's function 😉
The fundamental point though is that I feel the perfect PSU (a theoretical entity) would have no sound if it does it's job - i.e. provides DC with no AC content. It matters not one jot what capacitor type is used in my opinion if that ideal is achieved. I suspect we have the same views, from different angles. I of course carefully consider the capacitor choice in my own circuits and PSU's, but solely with a view to acheiving the desired and result. What I do not do, as some have done (and I'm not accusing you of this) is assign magic properties to certain constructions or material types - the 'it'll cure your asthma too' to paraphrase a Frank Zappa lyric.
If the ideal can be acheived by active circuitry, with the capacitors as a support role (stability etc) I'm not convinced that changing a capacitor for a different type (with otherwise identical AC characteristics) will make any difference.
The black box analagy is a useful one to enable understanding of the fundamental problem - view an amplifier as a device with one input and one output and things go wrong, but view it as a device with several inputs, of which one or two are the PSU and one starts to understand the sources of performance degradation.
Surely a theoretical amplifier with infinite PSRR must be immune to the effects of any of the components within it's PSU?
It's these extremes that can often bring focus and perspective to the desired goals.
And of course, 'better' capacitors have a role in that, in many circumstances.
A.
Jan has summarised my views - I was not arguing 'measurment vs audibility'. I consider my ears part of my measurement apparatus, indeed they are the only thing that matters to the end result.
The problem is my brain gets in the way, the ears are an exceptionally sensitive transducer, the brain is a totally rubbish analyser, since my moods, health, weather etc. all affect it's function 😉
The fundamental point though is that I feel the perfect PSU (a theoretical entity) would have no sound if it does it's job - i.e. provides DC with no AC content. It matters not one jot what capacitor type is used in my opinion if that ideal is achieved. I suspect we have the same views, from different angles. I of course carefully consider the capacitor choice in my own circuits and PSU's, but solely with a view to acheiving the desired and result. What I do not do, as some have done (and I'm not accusing you of this) is assign magic properties to certain constructions or material types - the 'it'll cure your asthma too' to paraphrase a Frank Zappa lyric.
If the ideal can be acheived by active circuitry, with the capacitors as a support role (stability etc) I'm not convinced that changing a capacitor for a different type (with otherwise identical AC characteristics) will make any difference.
The black box analagy is a useful one to enable understanding of the fundamental problem - view an amplifier as a device with one input and one output and things go wrong, but view it as a device with several inputs, of which one or two are the PSU and one starts to understand the sources of performance degradation.
Surely a theoretical amplifier with infinite PSRR must be immune to the effects of any of the components within it's PSU?
It's these extremes that can often bring focus and perspective to the desired goals.
And of course, 'better' capacitors have a role in that, in many circumstances.
A.
The issue I have always had with measuring CAP distortion (or amplifier distortion for that matter), is the way it is done. It is not an instantaneous measurement, it is an "averaged" measurement. As such, transient problems do not show up. The other issue with single sine wave measurements of capacitors, is that a primary "sound problem" of caps does not completely show up. Beyond the basic model of a capacitor of pure capacitance, resistance and inductance, dielectric absorption contributes almost a "delay" function to the transfer function of the capacitor. If you take a signal of 0 distortion and add a slightly delayed version of that signal with 0 distortion, you still get a signal of 0 harmonic distortion. That may sound just fine, but start talking complex musical signals and suddenly things do not sound so good.
Alvaius
Alvaius
"dielectric absorption contributes almost a "delay" function to the transfer function of the capacitor."
"but start talking complex musical signals and suddenly things do not sound so good."
Alvaius, I agree with you here as DA being a subtle but strong factor in the audibility of capacitors.
The nature of the way (spectral response characteristics) that this DA stores then returns energy is dependent on the type of dielectric, and I agree that this energy return can give a micro low level delay/reverb with it's own additional soundprint and this imo is part of sounds affects/effects attributed to capacitor types in sonics testing.
Eric.
"but start talking complex musical signals and suddenly things do not sound so good."
Alvaius, I agree with you here as DA being a subtle but strong factor in the audibility of capacitors.
The nature of the way (spectral response characteristics) that this DA stores then returns energy is dependent on the type of dielectric, and I agree that this energy return can give a micro low level delay/reverb with it's own additional soundprint and this imo is part of sounds affects/effects attributed to capacitor types in sonics testing.
Eric.
DA probably isn't even linear
All modelling I have seen of DA is linear (i.e. added capacitor and resistor), and all the measurements I know of use a constant charging voltage and constant discharge and measurement delay times.
I would almost bet that neither the voltage nor the time dependence is what one would expect from the linear model.
What do we call this? Harmonic delay????
All modelling I have seen of DA is linear (i.e. added capacitor and resistor), and all the measurements I know of use a constant charging voltage and constant discharge and measurement delay times.
I would almost bet that neither the voltage nor the time dependence is what one would expect from the linear model.
What do we call this? Harmonic delay????
?? Maybe I am just bad at math, but how isalvaius said:
f(sin(t))=sin(t)+sin(t+k)*c
a linear function on sin(t)?
I believe this effect will be measurable as distortion.
It isn't a linear function of sin(t) but a linear function of amplitude. Harmonic distortion will cover only new frequencies (FFT as it is usually done is phase-blind), and the only way to get new frequencies is to have a nonlinear transfer function.
It is a distortion, no doubt, but in time. Serves to show that the defintions of THD and IMD are not fit to describe all the distortions that can occur.
Eric
It is a distortion, no doubt, but in time. Serves to show that the defintions of THD and IMD are not fit to describe all the distortions that can occur.
Eric
I should have said "amplitude forefactor" rather than the shorthand amplitude. I guess its clear enough to a physicist...
Input voltage: A0*sin(omega*t)
Output voltage of an ideal system:
B*A0*sin(omega*t-phi) phi: constant delay
output of system with time delay but still linear in amplitude:
B*A0*sin(omega*t-phi) + C*A0*sin(omega*t-phi-theta)
Will not show up in THD analysis as everything is at omega and the amplitude is always a linear function of the input amplitude but will exhibit some smearing.
Distorted system without DA delay:
B*A0*sin(omega*t-phi)+ D*A0*sin(2*omega*t-phi)
This component at 2*omega will show up as second harmonic distortion. The only way to get this is a nonlinear amplitude transfer function.
Clear enough?
Eric
Input voltage: A0*sin(omega*t)
Output voltage of an ideal system:
B*A0*sin(omega*t-phi) phi: constant delay
output of system with time delay but still linear in amplitude:
B*A0*sin(omega*t-phi) + C*A0*sin(omega*t-phi-theta)
Will not show up in THD analysis as everything is at omega and the amplitude is always a linear function of the input amplitude but will exhibit some smearing.
Distorted system without DA delay:
B*A0*sin(omega*t-phi)+ D*A0*sin(2*omega*t-phi)
This component at 2*omega will show up as second harmonic distortion. The only way to get this is a nonlinear amplitude transfer function.
Clear enough?
Eric
OK,
Before this becomes a battle of arrogance... this site has some practical measurements of different capacitor types so one can see the distortion introduced by capacitors.
http://members.aol.com/sbench102/caps.html
He is also an EE with all the knowledge of those neat formulae, etc. I think seeing a picture (why calculate what you can measure?) makes more of an impact than just calculating it or parroting off formula. The proof is in the pudding.
Gabe
Before this becomes a battle of arrogance... this site has some practical measurements of different capacitor types so one can see the distortion introduced by capacitors.
http://members.aol.com/sbench102/caps.html
He is also an EE with all the knowledge of those neat formulae, etc. I think seeing a picture (why calculate what you can measure?) makes more of an impact than just calculating it or parroting off formula. The proof is in the pudding.
Gabe
Capslock,
Hmm...I still don't get it. The problem was not the notational representation of the amplitude function (we basically wrote the same thing) but rather what is the meaning of "linear amplitude transfer function." If we agree (do we?) that B*A0*sin(omega*t-phi) is the amplitude function, then to me, this means if G(t) is "a linear function" of F(t) then exists {c,k} such that k*F(t)+c=G(t).
So, if
G(t)=B*A0*sin(omega*t-phi) + C*A0*sin(omega*t-phi-theta)
is linear in amplitude to
F(t)=B*A0*sin(omega*t-phi)
then shouldn't a linear transformation from F(t) to G(t) exist?
I am unfamiliar with the term "amplitude forefactor".
BTW Gabevee I do enjoy discussing such matters...I think it is bothersome when people say something technical or mathematical) which may or may not be true, and it is simply accepted. I have a feeling I'm wrong about this issue, but I'd much rather discuss it and understand it than be ignorant.
Hmm...I still don't get it. The problem was not the notational representation of the amplitude function (we basically wrote the same thing) but rather what is the meaning of "linear amplitude transfer function." If we agree (do we?) that B*A0*sin(omega*t-phi) is the amplitude function, then to me, this means if G(t) is "a linear function" of F(t) then exists {c,k} such that k*F(t)+c=G(t).
So, if
G(t)=B*A0*sin(omega*t-phi) + C*A0*sin(omega*t-phi-theta)
is linear in amplitude to
F(t)=B*A0*sin(omega*t-phi)
then shouldn't a linear transformation from F(t) to G(t) exist?
I am unfamiliar with the term "amplitude forefactor".
BTW Gabevee I do enjoy discussing such matters...I think it is bothersome when people say something technical or mathematical) which may or may not be true, and it is simply accepted. I have a feeling I'm wrong about this issue, but I'd much rather discuss it and understand it than be ignorant.
Tiroth,
linear in your terms seems to be that the output function G can be expressed as a first order (i.e. constant offset and linear multiplier) function of the input F. We shouldn't worry about the offset because that would be DC (i.e. a battery rather than a cap) and here we are concerned with AC.
The reason that your approach does not work is that the output function will be delayed, i.e.
input F is a function of sin(omega * t)
and
output G is a function of sin(omega * t - phi).
Linear in the sense of distortion analysis means that the prefactor of the output function is a integer or non-integer multiple of the prefactor of the input function.
These prefactors A0 for the input or B*A0 for the output denote the maximum amplitude of the sine term.
Example
F(t) = 3V * sin(omega*t)
will be 0V when the argument of the sine is 0, pi or a multiple thereof.
When omega*t = pi/2, it is 3V, when 3/2*pi, it'll be -3V. I'd say that the prefactor A0=3V.
When you put this through a low pass filter, i.e. an ideal cap and a resistor, this filter is linear, i.e. it will not add new frequencies. It will, however, for any given frequency omega, attenuate the amplitude (=prefactor) by a constant factor and add a constant phase shift. For a different frequency, factor and delay would be different but again independent of the prefactor of the original input function. So for example, a 10 V sine would be attenuated to 5 V and a 3 V sine to 1.5 V.
Eric
linear in your terms seems to be that the output function G can be expressed as a first order (i.e. constant offset and linear multiplier) function of the input F. We shouldn't worry about the offset because that would be DC (i.e. a battery rather than a cap) and here we are concerned with AC.
The reason that your approach does not work is that the output function will be delayed, i.e.
input F is a function of sin(omega * t)
and
output G is a function of sin(omega * t - phi).
Linear in the sense of distortion analysis means that the prefactor of the output function is a integer or non-integer multiple of the prefactor of the input function.
These prefactors A0 for the input or B*A0 for the output denote the maximum amplitude of the sine term.
Example
F(t) = 3V * sin(omega*t)
will be 0V when the argument of the sine is 0, pi or a multiple thereof.
When omega*t = pi/2, it is 3V, when 3/2*pi, it'll be -3V. I'd say that the prefactor A0=3V.
When you put this through a low pass filter, i.e. an ideal cap and a resistor, this filter is linear, i.e. it will not add new frequencies. It will, however, for any given frequency omega, attenuate the amplitude (=prefactor) by a constant factor and add a constant phase shift. For a different frequency, factor and delay would be different but again independent of the prefactor of the original input function. So for example, a 10 V sine would be attenuated to 5 V and a 3 V sine to 1.5 V.
Eric