Power Supply Resevoir Size

[abraxalito]

Originally Posted by CharlieLaub

Increase the 0.22R resistor to 1 ohm and use a wire-wound 5W or larger ceramic/sandstone power resistor. When used in conjunction with the 10,000uF cap, this gives you the 15Hz corner frequency I mentioned.

Have you tried a suitably sized inductor in place of the resistor? I can't help thinking that the resistor is a bit of a waste of power and we can get better HF filtering with a choke. Recently I've played a bit with CLCLC filtering using iron dust toroidal chokes (values of a few hundred uH) and my rudimentary measurements show they do get rid of ripple nicely above a kHz or so. However I've yet to discover what happens when a real amp is used as a load, rather than my test resistor.

[CharlieLaub]

Originally Posted by gootee

Actually, almost always, the power supply capacitor current is exactly the signal that you hear. For very-convincing "proof", see the image from an LT-Spice simulation that I attached to post # 372 of this thread:

Power Supply Resevoir Size

Cheers,

Tom

Tom,

From what I can tell, you show in post 372 is that the PS reservoir capacitors are delivering current to the load via the output devices, which are following the input signal. That's pretty normal stuff, exactly what should be expected.

I believe that what you are inferring by the post above is that voltage fluctuations that are on the rails will show up at the load in the same way. That is at a minimum a very misleading statement (if that is indeed what you were saying... ). I am sure that you have heard of power supply ripple rejection ratio... it's the ability of the amplifier to reject voltage fluctuations on the rails. For my favorite chip amp, the LM3886, this starts out around 100 dB, falling to 75 dB at 1k Hz (worst rail). So the 1V ripple at 120 Hz becomes an 0.00001 V ripple on the output (assuming 100 dB PSRR). Hardly of any concern and entirely negligible.

I see how what you are trying to say above, but the reversed logic is faulty.

-Charlie

[abraxalito]

Originally Posted by CharlieLaub

For my favorite chip amp, the LM3886, this starts out around 100 dB, falling to 75 dB at 1k Hz (worst rail). So the 1V ripple at 120 Hz becomes an 0.00001 V ripple on the output (assuming 100 dB PSRR). Hardly of any concern and entirely negligible.

Speaking of faulty, this has an error in the calculation The PSRR is referred to the input, not the output. So with a gain of X20 its 26dB worse than your figure. The PSRR at 20kHz for the LM3886's negative rail falls to a fairly dismal 50dB - what's on that supply rail therefore just gets 24dB attenuation on the output in a typical implementation.

[CharlieLaub]

Originally Posted by abraxalito

Have you tried a suitably sized inductor in place of the resistor? I can't help thinking that the resistor is a bit of a waste of power and we can get better HF filtering with a choke. Recently I've played a bit with CLCLC filtering using iron dust toroidal chokes (values of a few hundred uH) and my rudimentary measurements show they do get rid of ripple nicely above a kHz or so. However I've yet to discover what happens when a real amp is used as a load, rather than my test resistor.

LC filtering can be done, however, you may inadvertently create a resonance above the audio band (or so I read) because of interaction with other components (I forgot what exactly, sorry!). I do use wirewound power resistors, which have a small inductive component, but most of the attenuation is purely RC.

Let's run some numbers to see what the power loss might be:
I suggested 1R + 10,000 uF to give a corner frequency of 15 Hz. I mentioned that the cap size was appropriate for about 20W+20W of power. At full power this would be about 4A RMS current, so 4V loss. But this would only really happen if the amp was operated long enough at "full sine wave power" to draw that much current long enough to drain the caps down 4V. Mostly there will be brief high power peaks with the average being much less (like 10-20dB less, e.g. my crest factor value typical for music), so the average rail voltage loss would only be around 0.1 - 0.3 V, which really is not much of a loss considering that with this one component (the 1R resistor) you clean up the rails significantly.

I just happened to be working on a chip amp last night with almost this exact same PS. After the bridge rectifier I have one pair of 10k uF caps, then a 1R resistor on each rail, then another pair of 10k uF caps. Attached are two full power (about 25W+25W) measurements that I took, one with the 1R resistor and one with it replaced by a wire short (0R). The mains and ripple harmonics have all but disappeared - the 120Hz ripple is reduced by 20dB. With the 0R connection between pairs of caps, the ripple amplitude was almost as much as the second order distortion peak!

-Charlie

Edit: Also of note is that the IMD is reduced/eliminated. You can see the small side bands (e.g. 1k Hz plus 120Hz, and 1k Hz minus 120 Hz) which are present for the 0R case disappear in the 1R case.

[CharlieLaub]

Originally Posted by abraxalito

Speaking of faulty, this has an error in the calculation The PSRR is referred to the input, not the output. So with a gain of X20 its 26dB worse than your figure. The PSRR at 20kHz for the LM3886's negative rail falls to a fairly dismal 50dB - what's on that supply rail therefore just gets 24dB attenuation on the output in a typical implementation.

Hmm, interesting, thanks for pointing that out. But I think most junk on the rails is ripple harmonics (at least below 20k Hz) and those die out by 1k or 2k Hz even in the worst case. At least that is what I see in my measurements...

-Charlie

[abraxalito]

Originally Posted by CharlieLaub

LC filtering can be done, however, you may inadvertently create a resonance above the audio band (or so I read) because of interaction with other components (I forgot what exactly, sorry!). I do use wirewound power resistors, which have a small inductive component, but most of the attenuation is purely RC.

I agree its something to watch out for and not something that will show up accurately in LTSpice (being as it knows nothing about frequency dependent loss in inductors). The chokes I'm starting to play with are by deliberate choice lossy into the 100's of kHz so I doubt whether there's really a problem in practice. I'll have to suck it and see though.

Let's run some numbers to see what the power loss might be:
I suggested 1R + 10,000 uF to give a corner frequency of 15 Hz. I mentioned that the cap size was appropriate for about 20W+20W of power. At full power this would be about 4A RMS current, so 4V loss. But this would only really happen if the amp was operated long enough at "full sine wave power" to draw that much current long enough to drain the caps down 4V. Mostly there will be brief high power peaks with the average being much less (like 10-20dB less, e.g. my crest factor value typical for music), so the average rail voltage loss would only be around 0.1 - 0.3 V, which really is not much of a loss considering that with this one component (the 1R resistor) you clean up the rails significantly.

I agree that for such a low BOM cost the series resistor is great bang for the buck in terms of cleaning the rails. However I think your estimation of the voltage drop is too optimistic for a couple of reasons. Firstly the crest factor of the resistor current in the schematic you've drawn will be less favourable than a sine because of the lowish capacitance you have prior (2.2mF). And secondly the amp gets much less efficient at lower power levels.

A quick way of estimating this is that the load itself is 8R, there are two of them effectively in parallel, so 4R. With 14dB crest factor the average power for a 20+20W stereo amp in the two speaker loads (combined) is 1.6W. I'd expect an average 0.4W in each resistor so roughly 0.6V drop. This ignores the crest factor issue I mentioned and ignores RMS/average conversion so in practice it'd be somewhat worse by a few 10s of percent points but it gets in the ballpark.

[CharlieLaub]

Regarding the value of the series resistor, you are correct (abraxalito) in being a little concerned about the 1R value. IMHO it is about the most I would put there, but in return you get a great deal of attenuation/rejection. As you showed, the voltage drop that will result is pretty minor.

Here are my calcs:

10,000uF + 1R

Frequency     Attenuation   Harmonic
60            -11.8 dB      
120           -17.6 dB      2
180           -21.1 dB      3
240           -23.6 dB      4
300           -25.5 dB      5
360           -27.1 dB      6
420           -28.4 dB      7
480           -29.6 dB      8
540           -30.6 dB      9
600           -31.5 dB      10

If the capacitance goes up, the resistance can decrease for the same amount of attenuation. For instance using 45,500 uF and 0R22 (0.22 ohms) give exactly the same attenuation in dB at each frequency that is shown above, however, there will be about 4.5 times less voltage drop in the series resistor (if that is a concern).

Also, as the amplifier power increases and larger caps are required to give the same ripple voltage on the rails, the resistance can be reduced while attenuation remains constant. The higher power amp will draw more current, so the voltage drop will probably remain about the same all things being equal. Using larger than average capacitors will allow the resistor value to be decreased even further... this counters the "minimum capacitance" concept that I have seen mentioned in this forum. More is indeed better in this application.

-Charlie

[gootee]

Originally Posted by gootee
Actually, almost always, the power supply capacitor current is exactly the signal that you hear. For very-convincing "proof", see the image from an LT-Spice simulation that I attached to post # 372 of this thread:

Power Supply Resevoir Size

Cheers,

Tom

Originally Posted by CharlieLaub

Tom,

From what I can tell, you show in post 372 is that the PS reservoir capacitors are delivering current to the load via the output devices, which are following the input signal. That's pretty normal stuff, exactly what should be expected.

I believe that what you are inferring by the post above is that voltage fluctuations that are on the rails will show up at the load in the same way. That is at a minimum a very misleading statement (if that is indeed what you were saying... ). I am sure that you have heard of power supply ripple rejection ratio... it's the ability of the amplifier to reject voltage fluctuations on the rails. For my favorite chip amp, the LM3886, this starts out around 100 dB, falling to 75 dB at 1k Hz (worst rail). So the 1V ripple at 120 Hz becomes an 0.00001 V ripple on the output (assuming 100 dB PSRR). Hardly of any concern and entirely negligible.

I see how what you are trying to say above, but the reversed logic is faulty.

-Charlie

What?!! Think again! That was MUSIC, NOT NOISE.

You originally said:

Originally Posted by CharlieLaub
<snipped>
Remember, this is just the power supply. There is a tenuous connection to what you would actually "hear" from the amplifier, so do not do listening tests to "proof" your tweaks!!!
<snipped>
-Charlie

and I was merely pointing out that the connection is not tenuous at all.

The power supply current IS what you hear.

There was no "inferring by", no "reversed logic", no "faulty", and no "misleading", at least not on my part!

Whiskey Tango Foxtrot?!

How could you have thought that I was implying, as you said, "voltage fluctuations that are on the rails will show up at the load in the same way"?

I said NOTHING that could possibly be construed to have implied anything at all like that.

[abraxalito]

@Charlie : In terms of the higher harmonic rejection/filtering of the series R, ISTM there's going to be a limit imposed by the main cap (10mF in your case) ESR. With 10mohm ESR and 1R series R the best attenuation you can hope for is 40dB for example. Splitting those components in half does better - a two stage filter gives better numbers than a single, where the numbers are higher harmonic rejection. With a series L though the attenuation even with a single stage keeps on going, at least until you reach the SRF of the inductor.

The series L is probably costlier, definitely bulkier at lower powers - at higher it won't need heatsinking though so probably wins out in the kinds of applications Nelson Pass designs. In the link you removed from your original post, I see he's using 67mohms between two banks of 40mF. My concern with that is that while this looks good looking in from the rectifier, looking back from the power amp the first set of caps all of a sudden has +67mohm of ESR which seems to me to be something of a waste of decent caps...

[DF96]

Originally Posted by gootee

Actually, almost always, the power supply capacitor current is exactly the signal that you hear.

Be careful of causality here. The current through the cap may be the same (apart from charging pulses) as that through the speaker, but neither is set by the cap; both are set by the feedback loop (if there is one, which is usually the case for SS amps). Now if you go 'non-feedback' things get harder, but that is only for people who take pleasure in making life harder for themselves.

If you really can hear the PSU caps in a conventional SS amp then something is wrong: grounding, PSU caps too small, NFB take-off point or ground reference incorrect etc.