Whats wrong with large filter caps for Gainclone?

[jan.didden]

well adding to the previous post the voltage will be depleated more rapidly on the smaller cap, but that does not make much difference recharging the cap, but will sure make diference on music as you listen to it.

I tried to interprete the graphs by Pedja, and I come to the following conclusions:

Extra ESR (or extra wiring resistance in the xformer, which leads to the same) makes the current pulse "flatter" and thus reduces hf content;

All other things being the same, increased load increases the rippel (of course) and from the fft's I think this increases the hf contents. In the graphs the carrier level drops but the harmonics stay about the same which means relatively more harmonic content.

Do you agree Pedja? How about those graphs with log vertical scales?

Jan Didden

 

[Pedja]

Here they come, logarithmic Y axis. Cap’s ESR is 0.1 Ohm.


1000uF and 100mA:

An externally hosted image should be here but it was not working when we last tested it.

1000uF and 350mA (certain troubles begin):

An externally hosted image should be here but it was not working when we last tested it.

And yes, finally that interesting fact can be seen below: higher capacitance can make higher noise floor at HF. Explanation goes like “it is made by those short duration pulses (which have wide bandwidth in the freq domain)”.

10000uF and 100mA:

An externally hosted image should be here but it was not working when we last tested it.

10000uF and 350mA:

An externally hosted image should be here but it was not working when we last tested it.

So, as seen above, the second interesting thing is, excluding the bottom end, this 10000uF likes better 350mA than 100mA load.


Again this might not be the technical answer on “why low capacitance supply works?” question. Oh yes, it works. But IMHO it is not the answer. At least, not the whole answer. YAMV*.

Pedja

* your answer may vary

 

[Christer]

Pedja,

did you make you simulation model detailed enough to also
get RF noise from the diodes? That is something that is very
dependent on capacitor size, among other things.

 

[kropf]

time to measure

Pedja,

Thanks for the additional plots.

The next question is what happens when we have a 1kHz or 20kHz signal superimposed on the 120 Hz charging pulses.

Now it's time to measure; I've got a digital scope with FFT at work, so measuring is easier than simulating (for me). My IGC has 2200uF per rail and dual 24VAC transformers per channel. It does have a couping capacitor, so it won't be possible to measure a DC PS drain for a direct comparison to the previous simulations.

Jeremy

 

[Pedja]

Jan,

Extra ESR of the cap makes current pulses lower in amplitude and longer in time. Difference is bigger as it is bigger the drawn current. As you pointed out, added series resistance has similar (though not exactly the same) effect. But generally, noise floor is higher with higher ESR or added series resistance (I am not talking about the noise suppression, it is still only about the noise generation).

I think we all can agree that the career has lower amplitude in the case of the bigger caps and less drawn current. Generally speaking you were right claiming that the excessive large caps can cause HF problems. But as with all the problems in this world, it is the question how serious they are.

Pedja

 

[Pedja]

Christer,

No, that overcomes my current exercise since, if we want to go along that road, it should be modeled transformer as well and possibly should be taken into account some more parasitic phenomenon. I might be wrong, but I restrained here from fast and Schottky diodes and used conventional 1N4007 believing it will minimally mess the thing.

Pedja

 

[Squalish]

pedja: think you could model some parallelled film/lytic caps, based on realistic specs?

 

[Pedja]

Well, if you have specs... It could be modeled surely better than it is done by the capacitor as primitive passive element. Good reading here:
http://www.aeng.com/pdf/capacitor.pdf

Pedja

 

[Christer]

Pedja,

I didn't think you modelled diode noise, but wanted to check.
It is actually not difficult at all to get such phenomenae in a
simulation. Just adding some inductance to the AC source will
make things happen. However, I really have no idea what are
reasonable values for R and L of the secondaries, so I can't
say how close to reality my experiments are. So far I have only
bothered to see if the phenomenon can be captured in
simulations at all, which is obvoiusly can.

 

[apassgear]

Extra ESR (or extra wiring resistance in the xformer, which leads to the same) makes the current pulse "flatter" and thus reduces hf content;

All other things being the same, increased load increases the rippel (of course) and from the fft's I think this increases the hf contents. In the graphs the carrier level drops but the harmonics stay about the same which means relatively more harmonic content.

Well said Jan,

From my point of view, i agree with every word, and not taking in consideration pedja's modeling or testing myself, just as an understanding of this efects on my humble mind.

 

[Pedja]

Christer, I actually made these graphs having the transformer (.subckt) in circuit but values used shouldn’t relate to any real transformer. As about the transformers, generally their R is pretty low, L is pretty high but AFAIK you always have some pitfall. As about the diodes, I tried some others in this circuit and yes they do interfere into the look of these graphs. Gen Semi’s models of SB560 Schottky which I in fact use make these graphs looking very nervous (like the left half of the two later graphs). Hence the simulations I posted here should be understood more like the orientation of what happens, no more than that. My impression is that a lot will depend on the actual implementation and used parts. I like SPICE to model the filters, to determine operating points, to estimate the stability and even to estimate the distortion (yes) and some other things as well, but this is the area where I don’t believe entirely that the right thing can be easily made. Finally, my original intention wasn’t to get the insight about the whole circuit but about one partial phenomenon – influence of the capacitance value on the noise generation. If I had success trying to isolate that phenomenon then I done something.

So Jeremy, whatever I posted here, only the measurement will be the right thing. Can’t wait to hear your results.

Pedja

(circuit used for sims below)

 

[Christer]

Pedja,

My point was not to complain about you not taking the diode
noise into account (and I don't think you understood it that
way either). I just meant to point out that there are several
mechanisms here that contribute to the noise spectrum. One
is the shape of the charge current curve, which you investigated,
and the other is noise from diode shut-off, which depends on
the charge currents, of course. Since the latter is conisdered a
real issue, and not only a theoretical exercise, by many, it should
not be entirely left out of the "equation" when discussing the
capacitor size.

 

[millwood]

Re: time to measure

The next question is what happens when we have a 1kHz or 20kHz signal superimposed on the 120 Hz charging pulses.
Jeremy

that's actually fairly easy to estimate using data we have so far. the voltage drop wouldn't be as much, and there wouldn't be much impact on the charging current.

I did that with a transistor driven by a signal source as load, and in pedja's model the current sources can be modulated.

 

[millwood]

i put in a small cap cross the diode (0.01u - 0.1u) and it made no material difference to the current's FFT. Now coming to thinking about it, even with a large cap (10,000uf), energy over 5Khz on the diode current is almost zero. so I suppose the RF emission from rectifiers isn't a big deal.

 

[Pedja]

The next question is what happens when we have a 1kHz or 20kHz signal superimposed on the 120 Hz charging pulses.

First, again the graph without the superimposed signal (as the first graph in the post # 82) but, for the convenience, here showing somewhat higher bandwidth. 1000uF cap, 0.1 Ohm ESR, 100mA

An externally hosted image should be here but it was not working when we last tested it.

AC voltage source 10kHz/10mV put in series after the rectifier

An externally hosted image should be here but it was not working when we last tested it.

With higher drawn current, as the noise level starts to become higher anyway the differences between these two graphs at the HF start to become smaller but, of course, base freqs (10k and first harmonics) of the injected signal remain.

However, note that this is more serious problem than it could seem at the first sight. The simplest answer is: such signal will make problems. But the source of such signal shouldn’t be a voltage source in series after the rectifier, in reality it is rather a current source in parallel with the cap (a second device plugged to the same supply). Anyway, a lot will depend on the cap’s AC properties. And diodes are definitely important here.

Of course, the source of the external noise can be also at the transformer’s primary side, right at the secondary, between the secondary and the rectifier… Seems to me like a huge topic.

Pedja

 

[JOE DIRT®]

try the test out with a random signal (pink noise)

 

[jan.didden]

Well said Jan,

From my point of view, i agree with every word, and not taking in consideration pedja's modeling or testing myself, just as an understanding of this efects on my humble mind.

Still, it is nice if ones mental gymnastic is (sort of) confirmed by measurements. But I was wrong on one account. I said that increased load increases the rippel (true) and thus increases hf noise (not true). This is borne out by Pedja's curves (the one were he says: "this cap likes 350mA better than 100mA". I think what I missed was that if you increase load, and increase rippel, the diodes open up earlier in the cycle. The current pulse is flattened and the net effect seems to be LESS hf noise.

Jan Didden

 

[millwood]

The current pulse is flattened and the net effect seems to be LESS hf noise.

Jan Didden

that is very correct. In a way, the high peak current on the diode is due to the RC constant mismatch on both sides of the diode: RC constant is higher on the load side of the diode than on the source side of the diode. So if the two match exactly, you will not see any ripple current on the cap.

In this particular case, the output voltage (as applied on the load) will be exactly the same as the input voltage (on the other side of the diode).

 

[Pedja]

i put in a small cap cross the diode (0.01u - 0.1u) and it made no material difference to the current's FFT. Now coming to thinking about it, even with a large cap (10,000uf), energy over 5Khz on the diode current is almost zero. so I suppose the RF emission from rectifiers isn't a big deal.

Just a cap across the diode shouldn’t be as effective as you would want. And check the real RF, make a high resolution analysis and see what happens in MHz and GHz range.

… in pedja's model the current sources can be modulated.

… in reality it is rather a current source in parallel with the cap (a second device plugged to the same supply).

Sorry, I overlooked this notice of yours, yes, these two options boil down to the same.

try the test out with a random signal (pink noise)

This can show a general ability of the supply to shunt the noise but this would definitely demand some further (better) modeling of the caps, diodes, other parasitic properties and I don’t have a time to do this now (maybe soon)… Anyway, do you know the model statement for the pink noise?

Pedja

 

[JoeBob]

With all this talk of capacitor sizes, it makes me wonder if anyone has tried a PI filter, I haven't heard any talk about it.

Adding an L would smooth things out, but are there any obvious downsides I don't know about?