Ah, the $64,000 question 
Many folk prefer smaller caps and attribute this to the higher slew rate (or lower rise time) compared to larger electrolytics.
Personally, I've had more success with sikorels than any other large cap. Sikorels are supposed to have a much higher slew rate than standard caps, which would seem consistent with this explanation.
But I really don't know (I haven't the scientific training to speak with authority) so maybe I've got it completely wrong.
Either way, the sikorels kick ***
Many folk prefer smaller caps and attribute this to the higher slew rate (or lower rise time) compared to larger electrolytics.
Personally, I've had more success with sikorels than any other large cap. Sikorels are supposed to have a much higher slew rate than standard caps, which would seem consistent with this explanation.
But I really don't know (I haven't the scientific training to speak with authority) so maybe I've got it completely wrong.
Either way, the sikorels kick ***
I like to use small caps as possible, and CHEAP ones, I think that is a WASTE use those very expensive caps. For input, if you have a 22k of resistor to the ground, a 1uF cap is MUUUCCHHH more than enought, the low frequency roll of is 7,23Hz. So, a 470nF cap would be MORE than enough, giving a roll of frequency of 15Hz. Can you hear 15Hz? I don't think so.
With the 470nF cap + 22k resistor, you have a 6Db/oct high pass filter for subsonic frequencies, it will help the amp to not waste power, and also be better for the driver.
470nF cap, you can get a CHEAP WIMA and you'll be VERY happy with this.
So, WHY use expensive and large caps for the input?
I many things that I design, when I'm dealing with impedances such 100K, I use 100nF caps, I have no problems with this, and 100nF caps you can get WIMA or EPCOS inexpensive caps...
So, I think that many people should reconsider the usage of large and expensive capacitors.
With the 470nF cap + 22k resistor, you have a 6Db/oct high pass filter for subsonic frequencies, it will help the amp to not waste power, and also be better for the driver.
470nF cap, you can get a CHEAP WIMA and you'll be VERY happy with this.
So, WHY use expensive and large caps for the input?
I many things that I design, when I'm dealing with impedances such 100K, I use 100nF caps, I have no problems with this, and 100nF caps you can get WIMA or EPCOS inexpensive caps...
So, I think that many people should reconsider the usage of large and expensive capacitors.
-_nando-_ said:
I don't know about the sound of capacitor, but with those kind of -3dB points.... what about the phase distortion? I wont pretend to know exactly how it affects things etc, I have just read that it exists
LTSPICE tells me that 470nF and 22k and at 20Hz the phase will be... 37 degrees... whereas with say 3.3uF and 22k at 20Hz the phase will be... just 6 degrees. I use 3.3uF and 47k, which gives .... 3 degrees.
A quote from TNT-Audio:
"By way of example, a typical commercial value capacitor, rated at 10,000uF/63V and costing some Euro 8-9, will have a speed of 30-40V/uS at best. An equivalent Elna for Audio series black, costing some Euro 15-25, will have a speed in the range of 80-90V/uS, i.e. at the very worst, double the speed of the best case in commercial cap land. A Siemens Sikorel cap, costing some Euro 20-30, will slew at over 100V/uS - but at a price."
http://www.tnt-audio.com/clinica/ssps1_e.html
As I said before, I can't say if this is primarily responsible for my preference for these caps in gainclones.
"By way of example, a typical commercial value capacitor, rated at 10,000uF/63V and costing some Euro 8-9, will have a speed of 30-40V/uS at best. An equivalent Elna for Audio series black, costing some Euro 15-25, will have a speed in the range of 80-90V/uS, i.e. at the very worst, double the speed of the best case in commercial cap land. A Siemens Sikorel cap, costing some Euro 20-30, will slew at over 100V/uS - but at a price."
http://www.tnt-audio.com/clinica/ssps1_e.html
As I said before, I can't say if this is primarily responsible for my preference for these caps in gainclones.
Hi,
I set the input high pass filter to between 1Hz and 2Hz. (RC~=90mS)
I then set the NFB filter to 0.7Hz to 1.4Hz (RC~=130mS)
Then the PSU filter should be set to 0.5Hz to 1Hz. (RC~=180mS).
for 8ohm speakers this requires +-20mF per channel and for 4ohm speakers +-40mF per channel.
I can hear the difference between input RC=90mS and RC=20mS.
I don't know if I am hearing phase improvement or frequency extension. What I do know is that even with small speakers they sound more realistic with the F-3db<=2Hz.
They also sound as though they go deeper and also sound stronger in the very low bass. I do not experience any reduction in bass quality when setting the high pass filter as suggested.
To me, it appears that all the filters are complementary and MUST be set low to get the best out of the signal and out of the speakers.
I set the input high pass filter to between 1Hz and 2Hz. (RC~=90mS)
I then set the NFB filter to 0.7Hz to 1.4Hz (RC~=130mS)
Then the PSU filter should be set to 0.5Hz to 1Hz. (RC~=180mS).
for 8ohm speakers this requires +-20mF per channel and for 4ohm speakers +-40mF per channel.
I can hear the difference between input RC=90mS and RC=20mS.
I don't know if I am hearing phase improvement or frequency extension. What I do know is that even with small speakers they sound more realistic with the F-3db<=2Hz.
They also sound as though they go deeper and also sound stronger in the very low bass. I do not experience any reduction in bass quality when setting the high pass filter as suggested.
To me, it appears that all the filters are complementary and MUST be set low to get the best out of the signal and out of the speakers.
sharpi31 said:
If your amps drives 50W into an 8 Ohm load, it can hit the speaker with a peak voltage of 28V. Assume for a moment your amp needs to reproduce a transient that is within the 20 kHz audio BW, and it happens that that transient occurs when the output is crossing at one of the peak voltage limits (i.e. it is at -28V) and it suddenly needs to swing all the way to the opposite rail (i.e. to +28V). It will have to swing 56 volts in 1/2 a cycle at 20 kHz, so it will happen in 25 us. That means the amp needs to slew at 2.24 V/us.
ANY cap can keep up with that.
In selecting filter caps you want to make sure you have enough capacitance to provide power to the load at the lowest frequencies. Low freq reproduction usually requires large voltage swings. If the power supply voltage sags under the load because the power line doesn't "top-off" the charge in the caps quickly enough (such as when reproducing a sound with a fundamental frequency lower than the power line frequency), the power output will cease to increase as the input signal tries to drive the output closer to the rail voltage. That is the very definition of clipping.
Use big caps- they are cheap at these low voltages. Bypass them with a couple smaller film caps, and stop worrying.
I_F
I_F,
Seems you forgot to account for the half wave rectifying action of a class a/b amp, from the view of the PSU. This creates lots of harmonics on the rails and is the reason why excellent low inductive bypassing -- at the point of load: the chip -- is crucial. These LM3886 etc have poor power supply rejection which makes them very sensitive to HF noise one the rails, plus they run out of loop gain for feedback at higher frequencies with the usual gainclone gain setting. The typical gritty sound in the trebles with insufficent bypassing reflects this pretty well.
If one uses very big cans paralled with film caps one can quickly end up with a "hole" in the power supply impedance just in the range where it is needed most. Also this can form a tank circuit and provoke HF ringing. With wiring inductances and doubled HF bypassing (at the PSU and at the chip) other ringing effects are quite common (which may be "cured" with snubbers at the right place). Proper bypassing of chip power-op-amps is sort of an art. The trick is to make your design way backwards, from the chip to the xformer.
- Klaus
Seems you forgot to account for the half wave rectifying action of a class a/b amp, from the view of the PSU. This creates lots of harmonics on the rails and is the reason why excellent low inductive bypassing -- at the point of load: the chip -- is crucial. These LM3886 etc have poor power supply rejection which makes them very sensitive to HF noise one the rails, plus they run out of loop gain for feedback at higher frequencies with the usual gainclone gain setting. The typical gritty sound in the trebles with insufficent bypassing reflects this pretty well.
If one uses very big cans paralled with film caps one can quickly end up with a "hole" in the power supply impedance just in the range where it is needed most. Also this can form a tank circuit and provoke HF ringing. With wiring inductances and doubled HF bypassing (at the PSU and at the chip) other ringing effects are quite common (which may be "cured" with snubbers at the right place). Proper bypassing of chip power-op-amps is sort of an art. The trick is to make your design way backwards, from the chip to the xformer.
- Klaus
I_F,
The psu caps are both smoothing the rectified AC and delivering current as the amp demands it. While the voltage swings required by the amp may not require the full slew rate of the capacitor, a higher slew rate would still be advantageous for rapidly responding to the noise and ripples from the rectifier (which extend way beyond 20KHz).
In other words, slew rate will effect how well the cap can smooth voltage irregularities on the supply rail.
The psu caps are both smoothing the rectified AC and delivering current as the amp demands it. While the voltage swings required by the amp may not require the full slew rate of the capacitor, a higher slew rate would still be advantageous for rapidly responding to the noise and ripples from the rectifier (which extend way beyond 20KHz).
In other words, slew rate will effect how well the cap can smooth voltage irregularities on the supply rail.
Leolabs said:
Hi Leolabs,
I'm not an expert. But here are some things that I THINK I know, about that:
For filtering, if you're thinking of the electrolytics, then the ESR (Equivalent Series Resistance) is a very big factor, and is mentioned on the datasheet for caps for which the figure is better than average.
(Aside: ) Regarding datasheets: Manufacturers, of almost all types of electronic components, often simply don't give data for any parameters that aren't too good, for a particular component. It can be difficult to notice, if you don't already know what ought to be there, but isn't. Bottom line: Missing spec = "big red flag".
Getting back to ESR for (especially electrolytic) caps: In power supply filters, for example, the ripple voltage amplitude is DIRECTLY-related to the ESR. Expensive caps might have lower ESR. But it's often even better to parallel several smaller (and probably cheaper) caps, instead. i.e. Just like paralleling resistances divides their values by how many there are, and makes the total resistance smaller, so it goes with paralleling caps to divide their ESR. e.g. Two identical caps in parallel will have twice the capacitance but only HALF the ESR, compared to just one of them.
Note, too, that ESR varies with frequency, especially for electrolytic capacitors, if we're talking about audio frequencies. For example, if ESR is given at 100 kHz, but your application is for a power supply smoothing cap that will see 100 Hz or 120 Hz, then you should try to see what the ESR might be at the lower frequency. One possibility, for that: Sometimes, Tan(delta) (i.e. the tangent of the "loss angle") is also given on the datasheet, or at least on the manufacturer's website; or maybe DF (dissipation factor), usually as a percentage, is given, which is just the tan(delta) x 100. IF the tan(delta) or DF is given for a DIFFERENT frequency that the ESR is given for, then maybe we are in luck, because tan(delta)@freq = 2 x Pi x freq x ESR@freq. So, we can then solve that for the ESR@freq, at a second frequency (for which tan(delta) or DF is known). And if THAT frequency is also not the one we're interested in, then we can use the two ESR vs freq datapoints for linear interpolation or extrapolation, to try to at least get an estimate of the ESR at our frequency of interest.
The LIFESPAN of PSU electrolytics is also usually directly affected by their ESR, since higher ESR usually means more power dissipated in the capacitor for a given ripple current, which HEATS the capacitor. And heating is probably the worst enemy of an electrolytic capacitor.
Another thing that most "bigger" caps have more of is parasitic INDUCTANCE (i.e. ESL, or "Equivalent Series Inductance"), which is usually "a bad thing", too, since any time-varying current flow will induce a voltage, across any inductance, that's proportional to the RATE OF CHANGE of the current. And "induced" voltages, from parasitics, are "a bad thing", since they "shouldn't be there". They arithmetically ADD to whatever voltage the cap is in series with, such as... the power supply voltage! ergo: noise/ripple on the power rails, and noise/ripple in the current being drawn by an amplifier, etc etc.
Recapping (no pun intended): The ESR (aka "parasitic series resistance") in a capacitor induces a voltage across the cap that is proportional to the current through the cap. And the parasitic inductance (ESL) in a cap induces a voltage across the cap that is proportional to the rate-of-change of the current through the cap. Neither one is good. Ideally, current would flow where we want it to flow without inducing ANY "extra" voltage due to parasitics. (NOTE that PCB traces and wires ALSO have parasitic resistance and inductance! That's the main reason for using a "star ground": the induced voltages that we don't want to spread everywhere.)
I guess that, usually, or at least if all other things are equal, physically-larger capacitors have more parasitic inductance than smaller ones, just as longer wires or PCB traces have more inductance than shorter ones. So, bigger electrolytics "typically" would have more parasitic inductance than smaller ones. Does paralleling several smaller electrolytics, instead of using one bigger one, help to reduce the parasitic inductance, like it does with ESR? Yes, I think, since the equations for paralleling inductors work the same way as the ones for resistors. So, using two or more smaller capacitors (which typically would have lower inductance, each, than bigger caps, anyway) in parallel, wins again.
Re-iterating: Paralleling several possibly-smaller electrolytic capacitors should tend to make them better, approaching or surpassing the quality of a single more-expensive or larger cap, at least in most of the ways that matter the most.
Another sometimes-important point about electrolytics and their parasitic inductance, as it relates to power supply pin bypassing for chipamps: As the famous CarlosFM once pointed out, here, certain chipamps' power pins "don't seem to like" too much inductance to be between them and their current sources. And, as I think he probably also at least hinted at, IIRC: One way to try to "compensate-out" the unwanted effects of an inductance is..... with an RC snubber, which is usually just a series RC to ground, and looks just like what many people call a "Zobel" network.
Unfortunately, even though the inductance-compensating snubber design equations are very simple, we DO need to know the actual value of the inductance, to use them. And the inductance specs for most commonly-available, low-cost electrolytic capacitors are not published. But, we can maybe just "guess", probably in the lower tens-of-nano-Henries range, to start with, and then maybe resort to using trial-and-error, from there.
Note that the same thing is often done, i.e. using snubbers, to compensate-out the parasitic inductance for the large PSU smoothing capacitors, not to mention the effects of the power transformer's inductances.
Googling for "snubber design" should find more than most people want to know, about it.
Gotta run. Sorry to have blathered-on, for so long, about all of that!
I hope that someone will correct anything that I might have gotten wrong, and fill in anything important that I might have omitted.
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
It seems that no one trusts National's recommendations -- well, they are very well considered, exhaustively researched cookbook designs. As close as possible to the chip 100nF in parallel with 10uF. They have used ceramics and polypro for the 100nF. 1,000 uF can be a little further away.
On the power supply, a minimum of 10,000 uF. If the connection from the power supply to the chip pcb is more than a few inches you should bypass on the supply board as well.
On the power supply, a minimum of 10,000 uF. If the connection from the power supply to the chip pcb is more than a few inches you should bypass on the supply board as well.
You could attempt to determine the "slew rate" of a capacitor by charging it up to its rated voltage then measuring how much current it is capable of dumping when its leads are shorted out -> (dv/dt)=i/C, (Volts/second=Amperes/Farads). The more inductance a cap has the slower (and lower in magnitude) the current pulse will be.
If you think of it that way, just about any cap will have a "slew rate" that is more than acceptable for use in a linear amplifier.
If you think of it that way, just about any cap will have a "slew rate" that is more than acceptable for use in a linear amplifier.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.