Whats wrong with large filter caps for Gainclone?

[Pedja]

I took some steps toward the real world. One result of it is attached. Supply is made of 400W, 2x22V toroidal transformer, double bridge formed by SB560 and 2x4700uF per voltage, then I have the regulation. Drawn current is 80mA. Absolute level is not scaled either to DC or to 1V. I think it is anyway good for better understanding of what comprises the waveform regularly called a 100Hz ripple.





(Interesting thing: the higher noise floor you see below 300Hz went down for 10-15dB when I added 1 Ohm series resistor after the rectifier. So, the RC filter on work or something else…)

Short experiments with some parameters told me that the situation might not change exactly as predicted in the simulations but the simulations were far from being totally wrong either. Once again, take them as the orientation of how some parameters influence the things but remember that the transformer and diodes actually used as well as the parasitic properties will determine the final results.

I might do one day more extensive and systematic measurements of this kind, varying as much of parameters as possible, but it could last. So, no comparison from me for now but someone else might send a classic 1000uF GC supply FFTed (at the moment I don’t have any such GC at hand and I could only improvise) so we can compare it (of course “generally” and not “directly”) with the graph above.

Pedja

[millwood]

I think my take away from this is that 1) there is little high frequency content in voltage. and 2) paired that with an amp having decent psrr, PS ripple shouldn't be a major problem.

However, i would add that when large caps are used, current going through the diodes / bridge may carry a lot of HF content. Thankfully, that gets BETTER under heavier load.

[carlosfm]

Amen.

Oh, yes, a small value resistor after the bridge helps.

[millwood]

Originally posted by carlosfm
Oh, yes, a small value resistor after the bridge helps.

a 50w rms / 8ohm amp can drop upwards of 3amps. For a 1-ohm resistor, that means a drop of 3v peak per rail.

I am not sure sure if that is worth a 10 - 15db drop in the <300hz range, where psrr usually is pretty good.

[carlosfm]

Originally posted by millwood

a 50w rms / 8ohm amp can drop upwards of 3amps. For a 1-ohm resistor, that means a drop of 3v peak per rail.

I am not sure sure if that is worth a 10 - 15db drop in the <300hz range, where psrr usually is pretty good.

Yes, I agree, 1 ohm is too much, but 0.1~0.22 should be OK.
I was talking PSUs generically.
As for op-amps (even power op-amps in GCs) I always put a 0.1 cap between + and - pins on the chip.
That makes PSRR at high frequencies a little better.

[janneman]

Originally posted by carlosfm



Yes, I agree, 1 ohm is too much, but 0.1~0.22 should be OK.
I was talking PSUs generically.
As for op-amps (even power op-amps in GCs) I always put a 0.1 cap between + and - pins on the chip.
That makes PSRR at high frequencies a little better.

It doesn't do diddly for PSRR. PSRR is an attribute of the amplifier itself.

It MAY do a little to reduce hf ripple, but with .1uF it will be minimal.


Jan Didden

[Joe Rasmussen]

Hi Guys

Pedja sent me an email earlier in the week. When I mentioned that I had done FFT measurements of power supplies, he asked, as I recall it, if I could supply one of the JLTi or gainclone, using 1000uF caps etc.

Can't supply exactly that at the moment, but will as soon as possible and post it here.

What I will furnish is an FFT measurement of a hybrid amp (tube front end & Mosfet outputs) capable of 115W into 8 Ohm and 180W into 4 Ohm. This is an amp called the JR-100 that I designed for a company called Audio Fidelity about 12 years ago.

A few details so that the measurement can be seen in correct light. Power supply is +&- 50V DC, from 500VA tx, uses standard 35A bridge (nothing special) and 15.000uF/63V electros. The bias current is 300mA. The measurement was taken across the plus 50V rail, so it is single ended (I could do it across both rails - but needs to configure for balanced input on my equipment - trickier to wire up - but could be illuminating to do).

A 'true' AC measurement shows that total ripple is 55mV. Because it is a 'true' measurement, this includes the 100 Hertz ripple and all harmonics in total.

The measurement is calibrated. The 0dBu line at the top is 0.775V (following the normal convention). That means we can roughly calculate the AC value of every harmonic. For example, the 9th harmonic at 900 Hertz is -60dB, that means we can calculate its AC value as -60dB relative to 0.775V RMS. Your calculator will show that to be 0.775/1000 = 0.775mV out of 55mV total.

Please feel free to make your own comments and observations. Certainly there are similarities with Pedja's, but I don't think his is calibrated in msg #131?

Joe R.

[janneman]

Joe,

This is no load, just the 300mA bias?
Did you do any measurements under (sine or music) load?
(I know, you give us the world, and immediately we want another one. Sorry)

Jan Didden

[carlosfm]

Originally posted by janneman

It doesn't do diddly for PSRR. PSRR is an attribute of the amplifier itself.
It MAY do a little to reduce hf ripple, but with .1uF it will be minimal.
Jan Didden

Hi Jan,
Yes, you're right, it reduces ripple, nothing to do with the chip by itself.
Anyway, what value would you suggest?
Thorsten suggests 1uf, but that's a big cap to put under a circuit between + and - pins of an op-amp most of the times (at least if it's a polyester cap).

[Joe Rasmussen]

Originally posted by janneman
Joe,

This is no load, just the 300mA bias?
Did you do any measurements under (sine or music) load?
(I know, you give us the world, and immediately we want another one. Sorry)

Jan Didden

If I was to do that, there would be a near infinite set of parameters to define. If under load, what fequency or frequencies, single or multiple? What ouput level? 1W? 10W? 100W? What load, 8 Ohm, 4 Ohm or some reactive load simulating a speaker As for using music, how do you do FFT measurments under music/dynamic conditions when what you see above is averaged out over 1000 samples? The very nature, and shortcoming (?), of FFT measurements precludes it.

I do believe, at least hope, that one day a meaningful standardised (or set of) dynamic tests can be devised. With the increase of computing power, such type of tests may become a reality? That will get away from the severe limitations of static measurements.

That may be the 'another world' you wanted? Make that both of us!

Joe R.