Whats wrong with large filter caps for Gainclone?

[millwood]

On the idea of using a low-value resistor in place of the inductor in a pi filter. I have seen it mentioned 20 or 30 years ago (I am quite sure that some tube amps actually use this set-up for filament).

But I am not quite sure how well it works with amps as you essentially increased the output impedence of a PS. during the 80% of the time when the cap is discharging, that's like increasing the ESR of that cap substantially. It may induce parasitic oscillation as well.

[millwood]

the PS circuitry here.

I varied the inductor from (almost) zero H to 100mh (the upper limit of what one might use in real life).

[millwood]

this is the output for no inductor (I think I set the inductor value to 0.00001mh).

the green curve is the charge current going through the diode (right scale), and the blue one is the voltage over the load (right scale).

as you can see, voltage fluctuats from 32 - 35v, or 10%.

note that the "fuzzy" voltage line is due to driving a 20khz current source.

[millwood]

here is the case with a 100mh inductor.

the voltage fluctuates from 32v to 33 v (3% variation). the thick voltage line, again, is a 20khz signal on top of the basic 50hz carrier.

it does seem to be that case that using an inductor helps smooth out the 50hz carrier but unfortunately, the 20khz variation gets a low worse: 0.2v variation vs. almost nothing in the prior case.

conclusion? inductors are better for PSs that deal with more steady load.

[millwood]

sorry, I attached the wrong chart. here is the one for a 100mh inductor.

[kropf]

I promised some measurement results, so here they are. They are probably not as extensive as you had hoped.

First, the amp vitals. LM3875 inverted configuration, dual EI 24VAC transformers in a separate box, vanilla 400V, 35A diode bridges, 2200uF caps next to the chips.

The main test I ran was to place an 8ohm resistor across the +/- terminals of one of the diode bridges. IIRC, the positive rail, although with hindsight the negative rail would have been a better choice given the 30 dB worse PSRR at 1 kHz. Qualitatively, with my ear very near the speaker, I could hear little difference in hum/hiss even though one rail had been dropped asymmetrically to 22.7V from 35V and had a huge 5V(p-p) 120Hz ripple.

Unfortunately, the dynamic range of the scope was only about 80 dB, so it was not possible to resolve really low level detail.

Freq. Level
------ --------
0 Hz 0 dB (22.7V)
60 Hz below noise floor at -60 dB
120 Hz -22 dB (1.8Vrms)
240 Hz -47 (0.1Vrms)
360 -48
480 -69
600 -63
720 -70
840 -79
960 -76

So, given the much larger current drain, these values match Pedja's sims fairly well. I didn't try a measurementat high frequency to see diode RF noise.

I then removed the 8ohm load resistor and played a 1 kHz test tone. Significant harmonics of 1 Khz appeared on the supply rail due to the split supplies, although at a lower level than the -50dB of the 1Khz signal (output about 1 Watt). 4 kHz was below the noise floor. These harmonics could conceivably give the chip PSRR problems at high frequencies and high output levels.

Jeremy

[apassgear]

Originally posted by janneman


With the choke filter, it's the last cap that furnishes the "slug of current". Since the diodes only conduct 20% or less of the time, the xformer couldn't do it for 80% of the time anyway, even if there is no choke.

I have good experiance using input chokes on class A amps, they get as clean as you ever thought and it's a test that I would do with these class B chip amps provided I had a test bed for it and extra secondary voltage needed for this setup.

Looking for a good sound I would test different cap sizes and types, and use something like 10 to15 mH gapped chokes

Thing is I have not built this chip amp yet to do some testings, when I do I'll report back.

Keep on with these nice ideas, pedja's modellings are very enlighting even for my humble understanding.

[millwood]

Originally posted by kropf
Freq. Level
------ --------
0 Hz 0 dB (22.7V)
60 Hz below noise floor at -60 dB
120 Hz -22 dB (1.8Vrms)
240 Hz -47 (0.1Vrms)
360 -48
480 -69
600 -63
720 -70
840 -79
960 -76
Jeremy

so essentially having huge amps doesn't help much (diminishing rate of return).

I got similar results on a discrete amp I am working on. Injecting 5v/50hz ripples to the supply rail (otherwise 36vdc) induced an increase of 0.004% in THD on an 8ohm load, with just some simple RC decoupling for the input and vas stages.

I basically concluded that 6600uf per rail will be sufficient for my application, even for driving 4ohm loads.

I did a quick simulation of the "audioamp" circuitry in switchercad. with a 2vac/50hz ripple in the positive supply rail (10vdc), thd is 0.2759%, vs. 0.2708% without the ripple.

this is a pretty typical simple discrete amp, without any current source or curren mirrors. so real amps should perform a lot better than this.

[Pedja]

Hi Jeremy,

Thanks for the results. One thing I forgot to post earlier: all the graphs with log Y axis I posted are relative to 1V (0dB) and not to the carrier. So, the absolute levels in your measurements are showing better figures then they would be if scaled as those in my simulations. However, since the current drawn in your case was notably higher (fluctuates between 3A and some 4.3A if I am not mistaken) the things still could fit to each other.

I ran shortly that 2200uF/8 Ohms load case and the spectral content of the harmonics relates very well to your measurements, though cap’s ESR and parasitic series resistance could change this content somewhat. But what puzzles me is that you get 5V p-p swing around the 22.7V. I see swing between some 23V (or 22.7V you mentioned) and 34V (it can’t reach 35V anymore even at the peak) so it seems to me I am missing something about your circuit.

(Generally I also believe that our results are likely in agreement, just need to synchronize them.)

I then removed the 8ohm load resistor and played a 1 kHz test tone. Significant harmonics of 1 Khz appeared on the supply rail due to the split supplies, although at a lower level than the -50dB of the 1Khz signal (output about 1 Watt). 4 kHz was below the noise floor. These harmonics could conceivably give the chip PSRR problems at high frequencies and high output levels.

Yup, something like harmonic distortion at PSU lines. And seems to me like there is an IMD here also.

Pedja

Ps: Two more notices about the previous graphs that I forgot to give along the road. All FFTs are done for 10th cycle (90ms-100ms period). If some ask why not the logarithmic freq axis - MicroCap can’t show the logarithmic X axis to display a negative values.

[Pedja]

Originally posted by IanHarvey
It might be interesting to repeat the simulations with a low-value resistor (say 1 or 2 ohms) between the diodes and the filter caps.

Here they are.

1000uF, 0.1 Ohm ESR, 350mA drawn




10000uF, 0.1 Ohm ESR, 350mA drawn




So I should correct my previous answer to Jan. According to what I saw looking at more cases now, series resistance contributes to noise if used with a higher capacitances. With smaller cap it could be beneficial. It seems this should be tuned.

Pedja