Learning about classD

[lumanauw]

I just made an experiment classD, and failed. It's quite different than building analog amps. I think I have to learn from basic again. The aim is good sounding offcourse
What is the difference (technically and sonically) between

1. ClassD made by full discrete (like UCD) and classD made by IC's (like IRFaudio)?

2. ClassD that takes feedback before LC filter and classD taking feedback after LC filter?

3. How bad is phase shifts made by LC filter? Why manufacturers like IRF and LT takes feedback before LC, while majority here likes to take feedback after LC filter? Pro's and Con's?

[subwo1]

Hi lumanauw, I am just going to ramble off some thoughts quickly as your questions do not have simple answers. Feedback after the filter tends to give lower distortion since filter nonlinearities are compensated for by feedback. A good discrete class D amp can give lower distortion if it is done so that it can almost operate cleanly in linear mode.

I think an amp made largely of IC chips can be done to give clean output as well, but it tends to be harder to eliminate discontinuities in the signal flow. Whenever the signal has to jump between on-off thresholds, distortion increases--the greater and/or slower the jump, the more distortion there tends to be. But, consistency or smoothness in signal flow may be more useful than pure propagation speed of the signal through the amp. Those are some quick thoughts.

I am still working on a P/N design. I am having trouble getting my pspice models for the IRF9540n and IRF540n to give me accurate simulations, though.

[MikeB]

Hi subwo1 !
Welcome to the club, i also can't simulate with irf540n/9540n,
but wanted to use them for my P/N-ClassD (fully discrete).
The models from IRF don't really work in pspice...

Lumanauw, if you want to design dicrete ClassD, you will need
simulations. It's too hard to predict deadtimes/crossconductions,
rise/fall-times. I don't really trust sims at these speeds, but they
give a good idea.

Mike

[subwo1]

MikeB,

[lumanauw]

Hi, Subwo1,

Feedback after the filter tends to give lower distortion since filter nonlinearities are compensated for by feedback

How far is the phase shift made by LC filter? Will it making the front end/comparator/differential receiving the correct signal for NFB correction?

I reallly wonder why large manufacturer with deep knowledge in switching like IRF takes the feedback before the LC, assuming they know that they can take the feedback after LC.

Hi, Mike,

Lumanauw, if you want to design dicrete ClassD, you will need simulations. It's too hard to predict deadtimes/crossconductions, rise/fall-times. I don't really trust sims at these speeds, but they give a good idea.

Yes, in this case, SIM is a must. With analog amps, we can get without SIM, but not in this field, I think.

[classd4sure]

Hi,

I just made an experiment classD, and failed.

Naaaaah, sounds like you got something out of it, right? That's not a failure.

I would say start with the class d reference design thread, lots of good stuff in there for ya.

Sims are a total pain but if you get good with one you can do some neat things, and lots of experimentation rather cheaply.

IR just arent' in the amp design business so much as mosfets + drivers.

Regards,
Chris

[subwo1]

Originally posted by lumanauw
How far is the phase shift made by LC filter? Will it making the front end/comparator/differential receiving the correct signal for NFB correction?

Hi, it may appear that way at first, and to a point, if the delay is really great by using, say, a 100uH inductor, then maybe it will cause problems with filter resonance getting too close to the audio range.
From the opposite perspective, the filter tends to limit how many harmonics can rechurn through the amp. So the sound may actually gain smoothness and fidelity.

An advantage of feedback before the filter would seem to be that it is easier to get really high switching frequencies without producing more high frequency switching content at the output. I don't really see why it should not also benefit from fewer harmonics as well. I may be missing something with these thoughts on harmonics. Best Regards.

[subwo1]

Hi again, indeed, I was overlooking at least one thing about after filter feedback. The phase lead (compensation) can keep the input and output pretty much in net phase over a full switching cycle. The input gets ahead for half a cycle and then behind for the other half. So I doubt with a switching frequency of a few hundred kilohertz the ear has any chance of perceiving the back and forth time difference.

[lumanauw]

Hi, Subwo1,

I see majority of self-oscilating designs here are using after LC feedback point (with some exceptions, like IRF amp). From the IRF paper, fig 7 of iraudamp1.pdf, the red curve is peaking. They write that peaking is due to resonance of LC filter.
Will the amp which take feedback after LC do not have this?

How about sonics and stability comparison of feedback after LC and before LC. Any member here have made both?

This is my experimental classD. The feedback is taken after LC. I use IR2111 driver, I have clean square drive now (but in higher frequency, the square have oscilation/ringing in the edges)

Without R1,C2,R3,C3, the oscilation frequency is about 24khz. When I put 10k for R1, the oscilation frequency rises alot, but the square is ringing in edges, so I lowered the frequency by putting 1k for R3 and 220pf for C3.

I cannot put R3 and C3 in parrarel with feedback resistor (C2 place) like UCD patent. If I put it there (or just a single C2=100pf), all the mosfets blown.

There are things that I dont understand about L1 and C1 value.
If I put L1=200uH and C1=220uF (nonpolar Elko), the amp looks normal, but cannot drive high volumes. The sound is wrecked. In low volume, seems fine.

If I put L1=30uH and C1=2.2uF (WIMA MKP), the mosfets blows instantly (upper and down mosfets, all blown)

If I put L1=30uH and C1=47uF (nonpolar elko), or C1=220uF(elko), the elkos and L are heating up very quickly, and the current draw at idle rises alot.

If I put L1=200uH and C1=47uF (nonpolar elko), both the L and C heating up.

I wonder how come the designs here uses L only 40uH and C only 1uF without problem for self oscilating classD?

Seems I cannot have that low values of L and C. How to do that? Is this value of 40uH and 1uF valid only on switching frequency >300khz? My current is only 40khz, none of these low value can work. I've tried to raise the frequency by not using R3 and C3, but the idle current is very high, I shut down the arrangement imediately.

I've blown many-many IRF640 today.

[subwo1]

Hi lumanauw,

...peaking. They write that peaking is due to resonance of LC filter.

With suitable phase compensation and a sufficiently high switching frequency, and so long as the output does not clip and the input signal does not change too much too quickly, then post filter feedback can enable the amp to damp output filter resonance. There are many ifs; its true.

I have clean square drive now (but in higher frequency, the square have oscilation/ringing in the edges)

I'd say the turn-on time is too fast, and there may also be a lot of gate ringing. The turn-on portion of each MOSFET driver buffer needs some resistance. I'd try separating the emitters, placing a 10 to 20 ohm or so resistor between them. Connect the intersecting point between the PNP emitter and the resistor to the gate.

There are things that I dont understand about L1 and C1 value

I can only try to give some general guesses about the types of things that are happening since I can't tinker with the actual circuit. If you can, try to set up a simulation model of your circuit in LTspice so that you can compare the two and maybe get some ideas.

In some filter combinations, I suspect one or two things could be happening. One is that the filter could be going into heavy resonance, and the other is that the inductor is saturating. It would seem to be most likely with the 30uH unless it is air core, but especially then it could be inducing interference into your circuit. But a saturating core will overload and overheat your MOSFETs and cause them to blow.

I wonder how come the designs here uses L only 40uH and C only 1uF without problem for self oscilating classD?

Seems I cannot have that low values of L and C. How to do that? Is this value of 40uH and 1uF valid only on switching frequency >300khz? My current is only 40khz, none of these low value can work. I've tried to raise the frequency by not using R3 and C3, but the idle current is very high, I shut down the arrangement imediately.

One reason is that a high switching frequency permits the inductor impedance to remain high enough. The general answer to excessive power consumption during higher switching frequencies is switching losses. These can be causes by switching on one MOSFET too quickly after the other has turned off. Maybe the added gate resistor will help that problem. Sorry about the string of dead MOSFETs.

If you are going to get a functional amp, the IR2111 IC is a relatively easy path to that goal. I am confident you can get it to work! Best regards.