The 'signal path' is IMHO makes very little sense as typically the representation of the 'signal', i.e. a proportional copy of the input voltage, often varies from current to voltage and back as we progress through the stages of the amp. And since currents make loops and no loop has zero impedance, some (partial) representation of the input voltage in the form of a proportional current or voltage tends to appear in most and often all components of an amplifier. In a strict sense the 'signal path' ends right at the input of the amp.
In the same manner we tend to think of active components as having inputs and outputs, with the only influence of relevance being that of the input to the the output. In reality, every terminal of an active component is an input and output at the same time, with varying influences between them, as well as influences to themselves.
If you look at things this way, then if you can find a representation of some portion of he input signal to the amp in form of voltage or current (although strictly speaking other electrical and physical characteristics may be considered as well) at some point in the amp, you can logically assume that injecting a 'signal' in some form to that point will also influence other points. The trick is knowing that the influences may not be symetrical, i.e. if something close to a representation of the input signal comes out somewhere, it does not mean putting an external signal of the same kind on that spot will influence the internals of the amp in the same way. In this manner we can disregard many of the influences and come to an 'idealized' representation of the amplifier. HOWEVER when we do this in audio, we MUST look at this problem on a logarithmic scale, because this is how our target of interest, that being our ear-brain complex, perceives things. Once you look at things that way, it is often found that much less can be disregarded.
Some of the changes attributed to different caps may actually be the consequence of charging pulses getting where they shouldn't, via induction or common grounds. Or possibly poor PSRR at higher frequencies. Adding a resistor in series with the rectifier will help, as it tames the pulse. The real solution may be to fix the grounding, which is often wrong even when people think they have taken great care over it (because some of the advice you will see is wrong!).
I can see we all look at this cap issue in a different way.
If the amp and its perceived sound quality is "sensitive" in some way to the PSU caps used or the value of the caps etc then that indicates a problem inherent in the amp design itself, in that in some way, ripple/noise/modulation of the rails by the load current etc makes its presence felt... that isn't the same as saying the PSU caps have no influence, or saying that the load curent passes through the caps as such.
If an amp "improves" in sound quality for the few seconds it runs due to "pulling the plug" as was mentioned above, and that its suspected the reason is "charging currents/pulses" are in some way affecting sound quality then again, its an issue of poor design and implementation.
If the amp and its perceived sound quality is "sensitive" in some way to the PSU caps used or the value of the caps etc then that indicates a problem inherent in the amp design itself, in that in some way, ripple/noise/modulation of the rails by the load current etc makes its presence felt... that isn't the same as saying the PSU caps have no influence, or saying that the load curent passes through the caps as such.
If an amp "improves" in sound quality for the few seconds it runs due to "pulling the plug" as was mentioned above, and that its suspected the reason is "charging currents/pulses" are in some way affecting sound quality then again, its an issue of poor design and implementation.
I agree.
The smoothing caps pass audio signal.
Similarly at very high frequencies the decoupling caps pass audio signal.
Look back at Mooly's diagram. The two smoothing caps and the 741 have two loops that form a figure of 8. The +ve and -ve half cycles flow around that 8.
the 8 can be smoothing caps for LF signals or can be decoupling caps for HF signals.
That is why I think it is important that the decoupling must be integrated into the PCB so the the route length around the decoupling 8 is kept VERY VERY short.
The rail capacitance plays a very active role in passing audio signals.
BTW,
playing for tens of seconds after pulling the mains plug only applies when the ratio of bias current to smoothing capacitance is high.
When that ratio is low, tens becomes hundreds of seconds.
I have a few amplifiers that will "play normally" for 2 to 3 minutes after cutting mains power. +-20mF +-50Vdc, 8ohm speaker, ClassAB, quasi output stage, bias ~7mA.
Last edited:
Anyone care to back up their assertions with measurements of:
AC signal current through PSU caps.
AC signal voltage across each and both?
Mooly has already demonstrated an error of imagining PSU caps to be
simply in series with the signal with his 'scope posts referenced earlier in the thread.
OR.....
Why not just measure how much AC current flows through battery supplies
to a dual rail amp? C'mon, what's in the PC already, sim guys?
AC signal current through PSU caps.
AC signal voltage across each and both?
Mooly has already demonstrated an error of imagining PSU caps to be
simply in series with the signal with his 'scope posts referenced earlier in the thread.
OR.....
Why not just measure how much AC current flows through battery supplies
to a dual rail amp? C'mon, what's in the PC already, sim guys?
Last edited:
@DF96, Mooly, AndrewT
Agreed on most accounts. Ripple current injection into ground as well as induction from stray fields is a major factor, exactly what I meant by saying that putting the same caps literally in series with the audio signal (assuming some DC bias on them) tends to show much less insidious problems with said caps. Same problem with current loops being longer than they need to be (bad decoupling) and often enclosing a large area (wiring techniques!!!) end up putting various voltages and currents, either by injection or induction, where they should not be in the first place. In particular various resonant circuits arising from parasitics of capacitors and various other inductances having their resonant frequencies where the amp has little to no PSRR available is asking for trouble. I'm not even going to start on marginal transformer designs (ripple currents should never drive the core into deep saturation!!!) that result in induction of various voltages in the amp circuits completely bypassing the amps ability to treat t hem as foreign signals. All of these problems are not THAT difficult to solve to a sufficient degree.
I have to mention that working with designs that have poor inherent PSRR, or have PSRR mechanisms that do not rely on NFB, exposes some mechanisms which may be quite well hidden otherwise, yet have their subtle influence - in said amps the influence is not at all subtle. The idea is similar to basic good amp design - design an amplifier that is largely immune to problems before you apply methods to fix the remaining vulnerabilities. Tube amps are a good example but some things need to be scaled to take into account that energy storage in caps increases with voltage squared.
@AndrewT, amps can indeed play a long time from storage caps, but for a fair comparison care has to be taken in their design so that operating conditions do not change significantly with power supply voltage.
Another experiment which i tried just recently is centered on the idea of trying to separate the charging ripple current loop and the amp power supply current loop from each other. The theory behind it is that if the PSU capacitors have any influence, it must also come from these two currents sharing a common path through the capacitor, presenting a source of intermodulation. If the capacitor was ideal, it's impedance would approach zero so it would be far smaller than either the impedance of the rectifier + transformer, or the amp as viewed from the power supply terminals (in fact most typical semiconductor amps look similar to an input modulated current source from the PSU terminals, so very high impedance). To this end, a high bias follower was used as a proto-amp, driving peak currents up to 3.5A into a 4 ohm load. The power supply was made out of the same toroidal transformer, two bridge rectifiers (one per PSU side to prevent any asymetry of a center tap wiring injecting ripple currents into the ground). The bridges were classic high current ones, no fast diodes or anything like it. The total caapcitance per side was 20000uF. 4 different setups of caps were then investigated:
two 20000uF caps, one per side. Cornell-Doublier high ripple current computer grade.
Four 10000uF caps, two in parallel making the same 20000uF per side. Pins were soldered directly to one another for the shortest possible connection.
Same four 10000uF, two per side separated by 0.47 ohm resistors.
Finally, same four 10000uF, two per side, separated by a simple current source setup using BJTs, set to about 4A (higher than the maximum peak current). The idea here is to have a much larger impedance betwen the two caps in a pair.
The two PSU voltages were treated as separate and joined into a ground at the negative output terminal of the follower, from which a signal ground was then returned to the follower input and bias. Local decoupling is also present with the same center point, the loops were kept very short, but the wires to the actual follower were not reduced to a minimum, in fact about 25cm of wire was braided to that point to gve enough resistance so that any wire inductance formed resonant circuit was somewhat damped.
Cases 1 and 2 showed intermodulation products around the basic sinewave at 100Hz intervals, with the actual spread being dependant on the cap and the output current (which of course influences ripple current). The interesting thing to note is that the intermodulation products do not fall off proportionally to the output current (which is proportional to the input signal), suggesting that a number of mechanisms are at play - such as RFI, capacitive injection, induction, etc. Forgot to mention that the follower was driven from a 75 ohm source (now immagine how much this influence would increase for a typical very high output impedance VAS!), and the transformer was dislocated somewhat in a shielded enclosure. I did also measure stray field from the transformer using a small feritte core, and I'm not even going to start on the horrors one sees - this is a standard power supply toroid wound for 1.6T flux density. In other words, local decoupling often behaves as RF filtering as well ant this is not to be dismissed easily. This sort of thing obscures low level detail at low listening levels. Another interesting thing are spurs at relatively high frequencies, ~200kHz and up. Several were present in case number 2, ssuggesting multiple resonances occuring between the C and ESL of two parallel caps, and then there were the various intermodulation components. In my case there was nothing lower than about 30kHz so not easily audible. (* two further experiments were tried on this setup, one was heating up the caps using a bow drier. As temperature goes up, ESR seems to increase because the resonances are damped. Then fast diodes were used to replace the ordinary bridges and this produced a marked INCREASE of the resonances as well as appearing of further intermodulation components, including inside the audio band).
I'll take case 4 next, which basically cleared up everything to the point where my wave analyser could not detect anything non-harmonic with any certainly. It should be noted that the power supply waveform looked quite 'funky' at maximum output current but you couldn't really see any trace of the mains x 2 ripple, and it was all nice and smooth, no kinks and sharp corners. For all intents and purposes, the PSU was completely isolated.
Now for case 3, which is the most interesting. Essentially, the results were the same as with case 4, showing clearly that even using half an ohm to make up a PI filter out of two pairs of caps very effectively separates the PSU and amp current loops. The spectrum was different, and even though some remenants of cap resonances were still visible, they were so much lower taht one could not reliably tell them apart from harmonics. Another interesting experiment then followed - three different types of 10000uF cap were tried on either side of the 0.47 ohm resistor. While there were subtle differences on the PSU side (the best spec cap showed the least hash), there was practically NO difference at all between them on the amp side. Since I am using a wave analyzer to sweep from ~0 to 32MHz basically by hand, i have no option for trace averaging, which would probably show things more clearly, but IMHO even these simple experiments are indicative.
I have to say that some of the results were not at all unexpected. I have learned about capacitor resonances from SMPS designs, as well as the caveats of paralleling caps with little impedance inbetween. It is surprising how much better a bank of caps will filter when there is actually a snake-shaped trace connecting the caps rather than a copper plane, with next to no extra loss - where someone else would consider using very high spec and very high price caps, often with exactly the opposite end effect.
Agreed on most accounts. Ripple current injection into ground as well as induction from stray fields is a major factor, exactly what I meant by saying that putting the same caps literally in series with the audio signal (assuming some DC bias on them) tends to show much less insidious problems with said caps. Same problem with current loops being longer than they need to be (bad decoupling) and often enclosing a large area (wiring techniques!!!) end up putting various voltages and currents, either by injection or induction, where they should not be in the first place. In particular various resonant circuits arising from parasitics of capacitors and various other inductances having their resonant frequencies where the amp has little to no PSRR available is asking for trouble. I'm not even going to start on marginal transformer designs (ripple currents should never drive the core into deep saturation!!!) that result in induction of various voltages in the amp circuits completely bypassing the amps ability to treat t hem as foreign signals. All of these problems are not THAT difficult to solve to a sufficient degree.
I have to mention that working with designs that have poor inherent PSRR, or have PSRR mechanisms that do not rely on NFB, exposes some mechanisms which may be quite well hidden otherwise, yet have their subtle influence - in said amps the influence is not at all subtle. The idea is similar to basic good amp design - design an amplifier that is largely immune to problems before you apply methods to fix the remaining vulnerabilities. Tube amps are a good example but some things need to be scaled to take into account that energy storage in caps increases with voltage squared.
@AndrewT, amps can indeed play a long time from storage caps, but for a fair comparison care has to be taken in their design so that operating conditions do not change significantly with power supply voltage.
Another experiment which i tried just recently is centered on the idea of trying to separate the charging ripple current loop and the amp power supply current loop from each other. The theory behind it is that if the PSU capacitors have any influence, it must also come from these two currents sharing a common path through the capacitor, presenting a source of intermodulation. If the capacitor was ideal, it's impedance would approach zero so it would be far smaller than either the impedance of the rectifier + transformer, or the amp as viewed from the power supply terminals (in fact most typical semiconductor amps look similar to an input modulated current source from the PSU terminals, so very high impedance). To this end, a high bias follower was used as a proto-amp, driving peak currents up to 3.5A into a 4 ohm load. The power supply was made out of the same toroidal transformer, two bridge rectifiers (one per PSU side to prevent any asymetry of a center tap wiring injecting ripple currents into the ground). The bridges were classic high current ones, no fast diodes or anything like it. The total caapcitance per side was 20000uF. 4 different setups of caps were then investigated:
two 20000uF caps, one per side. Cornell-Doublier high ripple current computer grade.
Four 10000uF caps, two in parallel making the same 20000uF per side. Pins were soldered directly to one another for the shortest possible connection.
Same four 10000uF, two per side separated by 0.47 ohm resistors.
Finally, same four 10000uF, two per side, separated by a simple current source setup using BJTs, set to about 4A (higher than the maximum peak current). The idea here is to have a much larger impedance betwen the two caps in a pair.
The two PSU voltages were treated as separate and joined into a ground at the negative output terminal of the follower, from which a signal ground was then returned to the follower input and bias. Local decoupling is also present with the same center point, the loops were kept very short, but the wires to the actual follower were not reduced to a minimum, in fact about 25cm of wire was braided to that point to gve enough resistance so that any wire inductance formed resonant circuit was somewhat damped.
Cases 1 and 2 showed intermodulation products around the basic sinewave at 100Hz intervals, with the actual spread being dependant on the cap and the output current (which of course influences ripple current). The interesting thing to note is that the intermodulation products do not fall off proportionally to the output current (which is proportional to the input signal), suggesting that a number of mechanisms are at play - such as RFI, capacitive injection, induction, etc. Forgot to mention that the follower was driven from a 75 ohm source (now immagine how much this influence would increase for a typical very high output impedance VAS!), and the transformer was dislocated somewhat in a shielded enclosure. I did also measure stray field from the transformer using a small feritte core, and I'm not even going to start on the horrors one sees - this is a standard power supply toroid wound for 1.6T flux density. In other words, local decoupling often behaves as RF filtering as well ant this is not to be dismissed easily. This sort of thing obscures low level detail at low listening levels. Another interesting thing are spurs at relatively high frequencies, ~200kHz and up. Several were present in case number 2, ssuggesting multiple resonances occuring between the C and ESL of two parallel caps, and then there were the various intermodulation components. In my case there was nothing lower than about 30kHz so not easily audible. (* two further experiments were tried on this setup, one was heating up the caps using a bow drier. As temperature goes up, ESR seems to increase because the resonances are damped. Then fast diodes were used to replace the ordinary bridges and this produced a marked INCREASE of the resonances as well as appearing of further intermodulation components, including inside the audio band).
I'll take case 4 next, which basically cleared up everything to the point where my wave analyser could not detect anything non-harmonic with any certainly. It should be noted that the power supply waveform looked quite 'funky' at maximum output current but you couldn't really see any trace of the mains x 2 ripple, and it was all nice and smooth, no kinks and sharp corners. For all intents and purposes, the PSU was completely isolated.
Now for case 3, which is the most interesting. Essentially, the results were the same as with case 4, showing clearly that even using half an ohm to make up a PI filter out of two pairs of caps very effectively separates the PSU and amp current loops. The spectrum was different, and even though some remenants of cap resonances were still visible, they were so much lower taht one could not reliably tell them apart from harmonics. Another interesting experiment then followed - three different types of 10000uF cap were tried on either side of the 0.47 ohm resistor. While there were subtle differences on the PSU side (the best spec cap showed the least hash), there was practically NO difference at all between them on the amp side. Since I am using a wave analyzer to sweep from ~0 to 32MHz basically by hand, i have no option for trace averaging, which would probably show things more clearly, but IMHO even these simple experiments are indicative.
I have to say that some of the results were not at all unexpected. I have learned about capacitor resonances from SMPS designs, as well as the caveats of paralleling caps with little impedance inbetween. It is surprising how much better a bank of caps will filter when there is actually a snake-shaped trace connecting the caps rather than a copper plane, with next to no extra loss - where someone else would consider using very high spec and very high price caps, often with exactly the opposite end effect.
Anyone care to back up their assertions with measurements of:
AC signal current through PSU caps.
AC signal voltage across each and both?
Mooly has already demonstrated an error of imagining PSU caps to be
simply in series with the signal with his 'scope posts referenced earlier in the thread.
OR.....
Why not just measure how much AC current flows through battery supplies
to a dual rail amp? C'mon, what's in the PC already, sim guys?
Good idea... I'd offer to do it
but the clue is in the signature.Seriously though, its really informative and interesting hearing all your thoughts on the subject.
Another example... if you loaded the rails or even a single rail in an amp with say a pulsed 1khz load that put a very heavy ripple component on the rail/s... would you hear it in the output of the amp ?
If the load such as a speaker and the PSU caps are in some way "in series" then it makes sense that just as you say the audio currents flow from speaker to cap, the same would be true in reverse and an outside influence causing modulation of the cap current (our 1khz pulsed load) would flow in the speaker... but it doesn't.
Of course amplifiers are not perfect and this is where PSU improvement can have a part to play.
Thanks Illimzn,
I would not want to fault anthing you have said and experimented with.
I agree wholeheartedly that a small amount of Pi filtering can have an enormous effect on what passes from mains to PSU output. This is why I changed completely from big thick PSU line cables to relatively thin twisted solid core. The tiny amount of added resistance of the thin cables made the Pi filter effect much more effective.
Once one realises the the last cap in the CLCRCRC, or whatever, passes most of the audio signal, it becomes obvious that this cap is where the money should be spent to improve audio sound quality. That is not the same as promoting exotic, silly money AudioFool capacitors.
I would not want to fault anthing you have said and experimented with.
I agree wholeheartedly that a small amount of Pi filtering can have an enormous effect on what passes from mains to PSU output. This is why I changed completely from big thick PSU line cables to relatively thin twisted solid core. The tiny amount of added resistance of the thin cables made the Pi filter effect much more effective.
Once one realises the the last cap in the CLCRCRC, or whatever, passes most of the audio signal, it becomes obvious that this cap is where the money should be spent to improve audio sound quality. That is not the same as promoting exotic, silly money AudioFool capacitors.
I did not add a few experiments with adding small caps in parallel to a large cap. I can safely say that putting many caps of the same size dead parallel can sometimes have rather bad effects. Not always, and I can't say they are unpredictable, but generally higher ESR caps will work better this way. Low ESR will have pronounced resonant peaks, which usually means exotics will also behave similairly.
It's not easy to measure using regular instruments because ESR is quite low at those frequencies but a wave analyzer is of great help. My old HP has a very large dynamic range and a narrow bandwidth filter, so even driving a cap directly from the generator's 75 ohm output gives very useful readings. ESR or, more properly, ESZ is on the order of tens of mOhms but that's 60-70dB attenuation and still easy to see on the Wave analyzer. A particulairly interesting example was a bank of 24 1500uF electrolytics connected by copper planes. There were so many combinations of resonance between the caps that it actually looked like very low Q 'wideband' resonance inside which the ESZ was almost constant - better than a single cap. So, this also lends credibility to some manufacturers swearing by huge filter banks made out of many tens of caps. On the other hand, two 'equal' caps can produce anything from 1 to 4 resonant peaks in ESZ, and if they are some sort of of special construction, even more. However, just make the caps sufficiently different, say 1:2 capacitance ratios, and you get much less pronounced peaks. When paralleling an electrolytic with a smaller non-inductive foil cap, you are in practice the safer the bigger the difference in capacitance. I had a 18000uF cap with a pronounced ESZ peak at ~84kHZ which became a small hump when a stacked foil 1uF SMD (leadless) cap was added in parallel directly at the terminals.
For many standard caps, ESZ rises after a point where increase due to ESL becomes larger than ESR while the portion of ESZ from the actual capacitance decreases below ESR. Since these are all in series, you get falling impedance to a certain turnover frequency, then it starts rising. This is often in the high 10s or 100-200kHz range for 'generic' caps.
This is the less of a problem the higher the internal impedance of the transformer secondary, and also (perhaps not intuitively) the faster the rectifier diodes are. Replacing a regular EI transformer with a toroidal one with far lower stray inductance and winding resistance does make a signifficant difference in such a case even though the frequencies in question seem far away from the audio band - granted, usually because of stray magnetic field induction and sloppy ground execution, but as I said, not excatly intuitive if you just 'change the transformer'.
Suitably chosen foil caps in parallel turn this rise in impedance into a fairly flat impedance after the turnover, which is certainly better for next to no extra cost, even though it may be sorting out the symptom rather than the illness.
But, if you look at it as a pi-filter made up of two filter saps with a resistance inbetween, then a foil cap in parallel with the large cap on the rectifier side becomes largely un-necessary, and the foil cap on the amplifier side is actually the local decoupler cap on the amp.
Thinner wires may actually help here, damping the series resonant circuit that may arise out of the foil caps and ESL of the electrolytic. There is an important exception to this, where local decoupling has to be much more carefully considered, and this would be a power supply for a class D amplifier. In this case one may get 'surprising' effects which can be traced to skin effect in the wiring.
On another note, simply using thinner wires to connect filter caps in parallel can already provide significant PI filtering. I will remind readers here that even though we hear on a log scale, some important physical phenomena pertinent to this discussion occurs at a linear scale. Using 10-20mOhm of wire between two caps may look insignificant but it's on the order of magnitude of the cap's impedance. In other words, the second cap in such a 'PI' filter will see ~1/3 of the total ripple current, which, while being only a ~9 dB improvement, may actually make it's life far easier, and the resulting performance better.
limzn,
considering the ripple as a voltage is much more relevant; that is the magnitude of fluctuation in DC output voltage at a specific output current, the difference between the minimum and peak supply voltage. The voltage on the capacitor, at any moment depends on the charge it holds. The ripple voltage increases with capacitor discharge as the voltage drops below the peak value. A longer filter time constant gives better smoothing, a complete discharge means no smoothing.
The value of the ripple voltage is dependent on load current, power supply frequency (period of the ripple signal) and capacitor value:
Ripple voltage = load current x diode off time / filter capacitor.
considering the ripple as a voltage is much more relevant; that is the magnitude of fluctuation in DC output voltage at a specific output current, the difference between the minimum and peak supply voltage. The voltage on the capacitor, at any moment depends on the charge it holds. The ripple voltage increases with capacitor discharge as the voltage drops below the peak value. A longer filter time constant gives better smoothing, a complete discharge means no smoothing.
The value of the ripple voltage is dependent on load current, power supply frequency (period of the ripple signal) and capacitor value:
Ripple voltage = load current x diode off time / filter capacitor.
Last edited:
I think you are losing sight of what I am trying to say...
So many believe that taking for example a 4700uF cap and using it directly in series with a speaker (as on a single ended rail design where the cap is used for DC blocking) is similar to a DC coupled amp and speaker with 4700uF across the rails.
In the first example, the cap can be obviously and intuitively seen to have a direct influence on the output across the load.
In the second example the PSU caps do not have that same relationship.
That's what I am trying to get across
My definition of "in series with the load" would be that whatever flows in the PSU caps would also flow in the load. That clearly doesn't happen as the PSU caps carry a mixture of left plus right currents, noise, spikes, 50/60 hz ripple etc.
The low frequency roll of point of an amplifier is also not determined by the size of the PSU caps... and if they met the definition of "in series with the load" the caps would influence that.
If you took an ordinary DC coupled amp, for example a chip amp and added a 470uF cap in series with the speaker you would certainly notice a lack of LF response. If you took the same amp and replaced the PSU caps with 470uF the amp would maintain it's LF response.
It depends on where you put the NFB sampling point. A 470uF cap in series with the load will not reduce LF frequency response if the feedback is taken after the capacitor, although it may limit LF power. "In series with the load" is, like "signal path", not an entirely helpful concept so arguing about it might not prove fruitful.
This is certainly true, however, the ripple voltage is much more relevant WRT PSRR of the amplifier, however, increased ripple voltage also translates to a rather different shape and increase of the ripple current on the charge side of things, rather than discharge. Most of the ripple voltage is caused by the actual capacitance doing it's work, being sharged and discharged. However, the current waveform is largely responsible for most of the aggravation due to parasitic elements. Current and voltage ripple are both very relevant for the electro-chemistry inside the electrolytic cap, while the interplay of various currents, especially the charging current which is limited by the available diode conduction angle having to cover the required energy transfer, and further complicated by diode reverse recovery, is problematic for various reasons, some of which are heat generation, resonant effects, stray field and induction problems from the transformer. This is quite difficult to see eg. on a scope, superimposed to the sawtooth-like ripple voltage, unless the caps are in really bad shape. Even so, with a good scope it is possible to see a 'step' in the waveform when the diodes start conducting, and some damped oscillation when they stop. A current probe will, however, reveal lots more information.
Once again, I will add that what I have seen suggests that pointing the finger only at the capacitors is misguided. It is true that there are mechanisms at play that can be traced to the capacitors, but they never really act alone, it's always an interplay of components in the power supply, wiring, magnetic fields from magnetic components. It is also true that often changing the capacitor(s) will have an audible effect, but thinking it's solely due to the capacitors is like saying that the last straw broke the camel's back. Simply put, I think the hoopla over caps is largely exaggerated, but that's no reason not to look at them carefully when designing things with them.
I disagree with your definition.
"whatever flows in the load must also flow through/from the PSU caps" could be a definition of in series with the load.
Take the example of the mains disconnected ClassAB amplifier.
When the re-charging circuit is disconnected from mains, we all know and agree that the amplifier continues to work for a finite period of time.
While the amp is continuing to work, by sending a variable AC signal through the load, that current to the load comes directly from the charge stored in the rail capacitors.
The load current flows from the capacitors. There is no other source of energy to power the amplifier.
A dual polarity PSU is simply a re-location of the output cap to the other side of the load, except that there are two of them. One to supply the +ve half cycles and a separate one to supply the -ve half cycles.
Member
Joined 2009
Paid Member
Mooly, The negative feedback point is key - because it determines the PSRR to a large extent. With large PSRR the power supply caps influence is largely 'corrected' for by the gnf. If you have a cap in series with the output and the gnf take off point is before it, then the influence of the cap is greater.
I agree that the caps in the psu generally have less influence because of this.
WuYit - arguing about it could well be fruitful if we all take something useful away from the discussion, me included !
I agree that the caps in the psu generally have less influence because of this.
WuYit - arguing about it could well be fruitful if we all take something useful away from the discussion, me included !
I don't agree...
The amp with a series connected cap to the load can not reproduce a DC output level across the load. We all must agree on that. Whether the feedback includes the cap or not the amp has no response to DC as measured at the load.
If you imply that relocating the cap to the other side of the load (with the cap becoming the two reservoir caps in a split PSU) is then same as the series connected cap and load (and that is what you are saying) then can you explain why the amp can now happily sustain a DC output voltage and current across and into the load when fed entirely by the discharging reservoir caps.
Member
Joined 2009
Paid Member
I don't think Andrew was claiming anything about dc current, he's continuing the conversation here about signal current.
When using a directly coupled amplifier output it is the dc integrity of the speaker ground that determines the ability to send dc through the load. If the speaker ground is a virtual earth, decoupled from the supply rails with capacitors then indeed no dc can flow and this can be nice for speaker protection. If the speaker ground has a dc return path to the power rails then the it is possible to send a dc current through the load - and such a dc return path may be accomplished using a virtual earth with resistors/active devices coupled to the power rails, or by using a centre tap on the psu transformer. In all cases, the signal current will flow from the speaker ground back to the supply rails through the power supply. If the lowest impedance path is through the caps (the most common scenario) then most of the signal current will flow through these caps.
When using a directly coupled amplifier output it is the dc integrity of the speaker ground that determines the ability to send dc through the load. If the speaker ground is a virtual earth, decoupled from the supply rails with capacitors then indeed no dc can flow and this can be nice for speaker protection. If the speaker ground has a dc return path to the power rails then the it is possible to send a dc current through the load - and such a dc return path may be accomplished using a virtual earth with resistors/active devices coupled to the power rails, or by using a centre tap on the psu transformer. In all cases, the signal current will flow from the speaker ground back to the supply rails through the power supply. If the lowest impedance path is through the caps (the most common scenario) then most of the signal current will flow through these caps.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.