PSU filter capacitor value vs. output voltage swing

[johnr66]

I set up a simple experiment to see how changing the supply filter capacitor values influenced the maximum amplifier output voltage before clipping. I had to use the parts I had on hand and this was a 9-0-9v 2a transformer and a TDA2040 audio power amp IC. The supply was setup to be a full wave split design with the cap values shown filtering each rail. The amplifier's output was connected to a non inductive 4 Ohm load. I measured the maximum output RMS voltage just before clipping at 40 and 120Hz (actually slightly off 120Hz so the signal would not be in sync with the 120Hz supply ripple).

Here are the results of the test
[IMGDEAD]http://www.jracrylic.com/pix/captest.jpg[/IMGDEAD]

The non clipped output voltage did not change much until I went smaller than 2,200uf filter caps. After this, the supply ripple was large enough to start affecting the output swing considerably.

Of course, this is only a 8 watt amplifier. Two channels and higher power demand much more current from the supply.

 

[TheGabbleDuck]

Quite interesting.

Do you have any data on the ESR of the caps? If you're adding 1000uF caps one at a time for those results, then I'd be interested in seeing those figures.

It looks like you've exchanged caps for:
470uF
1000uF
2200uF
3300uF
4700uF and finally
8000uF
I take it these are singular across the rails? I'm pairing up two 10,000uF samwha for each rail on a PSU design at present, so it looks like I might need some more. Info on whether to use 2 x 4700 or 1 x 10,000 would be interesting

Thanks for this though!

 

[MarianB]

The PSU filter caps cannot influence the amplifier max output swing, that would imply changing the voltage witch is impossible, as long as the total capacitance is enough to get a small voltage ripple then the output swing is about the same no matter how much caps would be add to that minimum needed capacitance, the test comes just to prove that.

PS: On 50/60Hz linear PSU the caps ESR makes no difference, the freq is small enough.

 

[DF96]

The PSU filter caps are a significant factor in determining the maximum undistorted output swing. The data shows this. Calculations and modelling show this too. Cap ESR has a small effect.

The reason is that the cap value determines the extent of voltage dip just before the next charging pulse. The voltage dip determines the peak output voltage before clipping takes place. This is all gone into in another thread on here.

 

[MarianB]

I completely agree @DF96, and i think that is just what i sayd, maybe not in so many words, but the fact is as long as you have enough capacitance to get a small enough voltage ripple then the amplifier max output swing cannot be influenced just by adding more capacitance. And to make myself better understud, let say just for the sake of argument that for some X current and Y voltage ripple accepted, you need 22mF, if that ripple is small enough then you have made sure the amplifier's swing is at it's max potential, so it would make no difference if adding let say another 22mF of capacitance. The voltage ripple would be in the desired range either way, that is what i wanted to underline.

 

[DF96]

You seem to agree with me, then disagree, in the same sentence!

Cap size affects voltage swing, but diminishing returns set in once the cap is sufficiently large. Adding more cap makes a difference, but only a small difference.

 

[MarianB]

That is just semantics... yes theoretically it makes some difference but it is too small to take into account, ok now?

 

[DF96]

The precise meanings of sentences always involves semantics.

 

[Fast Eddie D]

Very good John. I obsserved an even greater increase in power into a 6 ohm dummy load when prototyping my "Zero Budget" amplifier. I took the power amp board out of a tabletop "hi-fi" "50+50 watts RMS" . I observed 12 watts RMS both channels driven. I increased psu caps from 3300 uF to 20,000 uF, no other changes, and observed 21 watts RMS before clipping. Then I used two transformers instead of one with 10,000 uF psu caps and obserbed 29 watts RMS before clipping. At this point the performance and subjective sound clarity was so good that I slapped it into an enclosure as a power amp for my second TV.

It was so mediocre when I started, and now it sounds better than my buddy's "mid-fi" Sony reciever (although not as powerful). Total cost? Under $30. Most of the stuff came right out of my electronic junkbox.

 

[MarianB]

Maybe so @DF96 but sometimes there is such a thing like overanalising stuff. theoretically yes, you are verry right, but practically the differences are too small, what difference could really make a few hundreds of mV more swing to a couple hundreds W power amplifier?

Anyway, i think this cannot help anyone so do not see any point in continuing this, if you want it so, then it is exactly as you say and i am sorry to have sayd anything.

regards.

 

[Fast Eddie D]

Maybe so @DF96 but sometimes there is such a thing like overanalising stuff. theoretically yes, you are verry right, but practically the differences are too small, what difference could really make a few hundreds of mV more swing to a couple hundreds W power amplifier?

Anyway, i think this cannot help anyone so do not see any point in continuing this, if you want it so, then it is exactly as you say and i am sorry to have sayd anything.

regards.

I don't think you get the point. My own research has yielded even more dramatic results. DF is playing with a tiny amp and a small power supply. Just increasing capacitance 5x in a small crappy amp I gained over 1 dB per channel, and subjectively better sound (especially bass).

 

[nigelwright7557]

In commercial amps they supply just enough capacitance to stop drop out at full power.

Adding more capacitance is a DIY or top end thing.

Sometimes too much capacitance can cause problems. I had a class d amp that once I got above 20,000uF per rail start screeching on power down ! This was due to the slow discharge of the smoothing capacitors. I had to add a PSU monitor and hold off the 2092 until power supply volts were in range. It also held the 2092 in reset on power down.
I told IR about the problem and they just said add a PSU monitor like I had.

 

[johnr66]

Quite interesting.

Do you have any data on the ESR of the caps? If you're adding 1000uF caps one at a time for those results, then I'd be interested in seeing those figures.

It looks like you've exchanged caps for:
470uF
1000uF
2200uF
3300uF
4700uF and finally
8000uF
I take it these are singular across the rails? I'm pairing up two 10,000uF samwha for each rail on a PSU design at present, so it looks like I might need some more. Info on whether to use 2 x 4700 or 1 x 10,000 would be interesting

Thanks for this though!

I used a single capacitor (for each rail) of given value. They are caps I had lying around. No special low esr versions and such. In your case, I'd doubt there would be a noticeable difference either way you go.

BTW, the lowest cap value used was only 100uf! Also interesting is how the lower frequency (40Hz) suffered in the 1000-3300uf range.

 

[MarianB]

I don't think you get the point...

Don't i? well the let go to some explanatory calculations, let say we have a power amp requiring a 1A current at max output swing, now let say that we want the voltage ripple to be a max of 1V, and we have a linear PSU composed of a 50Hz power transformer, a bridge rectifier and smoothing caps, how do we assure tha 1V of max ripple? simply by making surw we have enough capacitance, how much? wel it is simple: C=(I*T)/U, and so C=(1*0,01)/1=10mF or 10000uF, just for that 1V and 1A. Any other voltage ripple or current needed can be added and the result easy obtained.

In your case 3,3mF was too small of capacitance, increasing it is only normal to get way better results since you got much smaller voltage ripple.

 

[Fast Eddie D]

In your case 3,3mF was too small of capacitance, increasing it is only normal to get way better results since you got much smaller voltage ripple.

Well then you do sort of get it.

3300 uF was the value chosen by the designers of the consumer product. It must have been "adequate" according to the conventional math that they used.

Of course I got much lower ripple. Ripple was reduced from 25% @ "full power" to less than 10%. Adequate is in the eye of the beholder I suppose.

Subjectively. increasing capacitance (to a point) improves performance; bass "punch" and clarity, "soundstage", "dynamics." Maybe it can't be measured, but these old nearly deaf ears can hear the difference.

 

[Fast Eddie D]

In commercial amps they supply just enough capacitance to stop drop out at full power.

Adding more capacitance is a DIY or top end thing.

Indeed. It sure works though.

Sometimes too much capacitance can cause problems. I had a class d amp that once I got above 20,000uF per rail start screeching on power down ! This was due to the slow discharge of the smoothing capacitors. I had to add a PSU monitor and hold off the 2092 until power supply volts were in range. It also held the 2092 in reset on power down.

Good stuff Nigel.

 

[MarianB]

Audio electronics is verry subjective. There are as many performance standards as are designers i think... i also think you agree with me that most of the "of the self" products tend to be minimised on some issues, i mean producers make compromises for the sake of the final product price.

Now in your case, you have mentioned the amp has 50+50W of power, so let say about 3A needed at full swing for both channels, for 3,3mF caps the voltage ripple would be about 9V, and that on the condition that the power transformer has no voltage drop at full power ( the ideal case ) well that is ofcourse bad for the power amp, and ofcourse the full swing is affected verry much only by too small of a capacitance, then why did they put that low caps? for economics i think. You have also say that you have used 10mF and get better performance, well with 10mF the voltage ripple goes down to 3V for the same current, way better for the power amp, for under 1V it would have need at least 30mF of capacitance... better yet, giving enough capacitance for verry low voltage ripple the power amp performance would be better not only by the full swing but also as you sayd about it;s sound quality, the base especially giving that there is the most current requirement, and if the caps are big enough then they would be able to supply enough energy to the amp to minimise the voltage dropping on full load.

 

[TheGabbleDuck]

I'm usually designing SMPSUs & DC/DC converters for digital systems, but thought I'd add my own thoughts here.

I've seen devices drawing 150uA affect a 3.3V rail by several hundred mV. Ok, 150uA average, but when clocking data out of the device at 10MHz, the clock edges cause spikes on the main PSU, even though it's bypassed / decoupled at the device - an ADXL345 accelerometer. (10uF and 0.1uF as specced in the data sheet.)

The point of the above is that fast transients can cause trouble, and power supplies have to deal with these. Electrolytics can become inductive at high frequency, it's in the nature of their construction. So, 470uF vs 2200 or 22,000uF will make a difference in different ways.

I was speaking about this with a designer at work, and he made the observation that I generally follow anyway, large value at the supply, but anything less than 10s of uF should be at the device, due to track inductance.

I think in this case, ripple current ratings of the electrolytics in question play a part, but definitely size too.

RC time constant of 470uF + 8R= 3.76mS, 4700 + 8R = 37.6mS, 10,000uF + 8R = 80mS. That's time constant - in that time the cap will have discharged by 63%.

RC charge / discharge curves

So, if we want to keep to 90% of supply, we've got to choose a capacitor that will hold 90% of it's charge when discharging between 120Hz rectified peaks, So, 120Hz = 8.3mS, therefore we want a cap with 83mS time constant for the load. Looks like 10,000uF would nearly work, for 8R loads, I'll work backwards though:

Time constant T = RC = 1/2pi(f*C)

83mS = 8R*C
0.083 / 8 = 10.375mF = 10,375uF

For a 4R load it can be seen from this we have to double the capacitance. For us Brits, the time constant would be 10 * 1/100, or 100mS. That's 2,500uF for 8R loads and 25,000uF for 4R loads, preferably. More if you want less ripple - these are for 90% of supply figures, as I said.

Hope this helps some!

 

[Fast Eddie D]

We're on the same page Marian.

That 50+50 watt amplifier in fact only produced 12+12 watts before clipping into its rated 6 ohm load. Beefing it up to the max almost doubled the power with the same anemic transformer. Your calculations are correct but irrelevant to the reality of the actual performance of the amplifier. The unit was basically junk right out of the box, but it was good enough for my shop tunes. Once the CD changer stopped working I immediately cannibalised it. I was surprised at the performance I was able to get out of it with junkbox parts and $20 worth of rat shack capacitors etc.

Do you have any comments on the subjective performance increases available with optimising (really overbuilding) psu capacitance? Woo to you? I've been doing some informal experimenting with capacitors (not just electrolytics but poly bypass caps too) and it's made a believer out of me. I also have observed subjective performance increases by using capacitor arrays vs one big capacitor. My background is electrical engineering and I used to think it was woo too.

 

[Fast Eddie D]

Thanks Gabble.