Class D feedback and stability

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hi Guys

I've been searching for this and cannot find it.

For all of us that dont hold a PhD in system engineering and control theory, it is a bit hard to distinguish the different class D typologies and feedback-schemes. - and especially what pros and cons the different schemes have.

So far I know and (sort of... ) understand the difference between the clocked and self-osc designs. But within the self-osc group we have post- and pre-filter feedback, first and second order integrator, UcD and much more.

Does anyone know of reference/doc/pdf with an (simple, non-math-heavy) overview?

I have build and simulated a smallish classD (feedback-scheme like iraudamp) and I get some OK results, but then I got a bit more brave and started testing clipping behavior, squarewave-response, no-load, and different start-up scenarios. - Thats when things start to fall apart...

Any inputs are welcome. I can share schematic and results later.

Attachment: first, single-sided prototype: IR2184, IRF640N, +/-50V, 82% eff, 0.1% dist. Not impressive, and yes the 2184 is not a prime candidate for this.

Kind regards TroelsM
 

Attachments

  • 1536178307003.jpg
    1536178307003.jpg
    50.3 KB · Views: 833
Last edited:
Hej Troels,

I noticed you did not get much response on your question. It is not because it is a trivial question, on the contrary it is a very relevant and challenging query.
Most of us use class D amplifier chips more or less according to the datasheet and application notes. Then, we do not need to challenge our brains with this complex issue. I notice, only few manufacturers recommend post filter feedback. The 30-60 KHz output filter cut-off frequency will limit the feedback effect. It doesn't mean post-filter feedback should be discarded without reconsideration.

The reason why you do not have such detailed descriptions of how to make successful feedback loops for class D amplifiers is probably that this knowledge is the lifeblood of the manufacturers.
All know today how to design a good output push-pull stage and the corresponding driver stage. The only real difference in sound can be implemented by a better analog→PWM conversion with the right feed-back (or eventually feed-forward) to achieve that. The manufacturers are most likely not going to tell their competitors how they do.
There is one source of information that may be useful – patents. To obtain a patent, you need to disclose your idea. But, a patent disclosure is often not so easy to understand for practical use. I once glanced at what is supposed to be one of the central patents for “class T” modulation (Dr. Triphati, founder of Tripath). It was about relevant DSP and I could not use that information for much. Perhaps other patents are easier to understand. Analog feedback (or -forward) is quite easy to understand qualitatively. DSP very quickly gets mathematical and a PhD is useful.

On UcD, Philips or Hypex may have some information. It may not be impartial (balanced) information as they have own economical interests.

If you will disclose your findings on the topic, I am looking forward to learn.

FF
 
Last edited:
Hi Guys

I've been searching for this and cannot find it.

For all of us that dont hold a PhD in system engineering and control theory, it is a bit hard to distinguish the different class D typologies and feedback-schemes. - and especially what pros and cons the different schemes have.

So far I know and (sort of... ) understand the difference between the clocked and self-osc designs. But within the self-osc group we have post- and pre-filter feedback, first and second order integrator, UcD and much more.

Does anyone know of reference/doc/pdf with an (simple, non-math-heavy) overview?

I have build and simulated a smallish classD (feedback-scheme like iraudamp) and I get some OK results, but then I got a bit more brave and started testing clipping behavior, squarewave-response, no-load, and different start-up scenarios. - Thats when things start to fall apart...

Any inputs are welcome. I can share schematic and results later.

Attachment: first, single-sided prototype: IR2184, IRF640N, +/-50V, 82% eff, 0.1% dist. Not impressive, and yes the 2184 is not a prime candidate for this.

Kind regards TroelsM

The fact that you have a prototype is a step in the right direction, I'm keen to learn more in this area but don't have the time, however I suggest you look at Chocoholic's Class-Dlite amplifier and read the early stage development of the thread (lots of neat tips and tricks) specifically around the modulator not too detailed but enough to get you somewhere, there you will find what few have done here, build an amplifier from start to finish with serious R&D and help from others (and out performs the iraudamp), forum members phase_accurate (Charles Lehman) has supplied hes approach to the modulator scheme.

see attachment for the simple post feedback topology with some examples on the calculations you maybe need for your research.

Quote:

"
A feedback topology shall be proposed that allows the straightforward inclusion of the output filter into the feedback loop. It can be adapted to high- or low- Q output filters equally well. It is not restricted to either self-oscillating or carrier – based modulator topologies. One advantage is that the slew-rate requirements on the operational-amplifiers used for the proposed topology are quite relaxed since they are not forced to handle fast transients. Another advantage is that the feedback branch is based on resistors only.
"
 

Attachments

  • rad50BB0.pdf
    177.6 KB · Views: 1,262
Last edited:
Disabled Account
Joined 2005
Good feedback = less load dependant
From there then of course post filter feedback is better
But why less design use it?
The answer is because it need very good design compared with prefilter feedback because post filter feedback deals with a mV signal comparator which is not easy to create a good shape of triangle signal at very low voltage.

Post filter is straight forward take square output changed to triangle which can be very easy to get a good triangle shape. This is the good advantage. But it has less achurate because this feedback does not include LPF filter.

I am not PHD and this is just my opinion.
 
I think there is a lot of information scattered on this forum alone, but to remember where each information was is difficult to say the least. Chocoholic bigger amp, systemD_2k4 have lot of information. but in general, first you select your "type", how amp will work. Here you have clocked, ucd, hysteresis,... design. For each you need to know, what their +'s and -'s are.
But in the end, the single most important thing is feedback. If you have perfect or imperfect pcb layout, feedback will have to do all the work. But even feedback can do only so much, if stages before it are sloppy. Feedback can do only so much.

I've seen many things can be done, but are far above average diy user, let that be clocked, ucd, hysteresis or any other. Some people look at waveforms and say, what do I need to do, to improve the design. Like have multi-slope ramp to which you compare your sine wave. This will impact how amp will work above some level of signal (line near clipping).

But as I see it, if you would have perfect layout or no component, pcb, wire inductances, capacitances and fets would be perfect switches and there was no input to output delays (or be constant), then your only worry would be (PID) feedback. But even feedbacks are complex, there is not one for everything, you can get lost in only this topic alone
 
I would also recommend to take 50min to listen to:
YouTube

Ncore amp, as it was said in this video, needs only ~200kHz worth of bandwidth, to have 53dB of gain in feedback loop @ 20kHz, in other terms, this would the as much control as ~9MHz bandwidth linear amp

Now am not sure, but does Ncore have two output inductors?
 
Last edited:
OK, lunchbreak update:

I have semi-OK results with the "standard" pre-filter feedback. Distortion is high (Thd:0.05, thd-noise:0.15), but it appears to be stable: no start-up problems, clipping is semi-clean. -but I wanted some post-filter feedback to tame the output if the load is removed. Also as a purely intellectual exercise for my foggy brain.

I did not have much luck with post-filter feedback and either it would have start-up problems or behave badly to no-load or clipping. Most likely due to my lack of knowledge, but it didn't work for me.

I read somewhere that a company use the output-cap-current as feedback. Without much theory I put together a rough simulation. I thought it was less than ideal to sense the actual cap-currrent so instead I put a smaller cap in parallel and sensed the current with a 100mOhm series-resistor. The current in the 2 caps should be the same with gain-difference?

It did not behave very well until i split the org feedback resistor in two and put a cap over the bigger half ( someone: that must have a fancy name in regulation-theory). Anywhos. I guess the cap lowers gain at high freq?

in LTSpice it really helps to dampen the unloaded filter-response. No impact on distortion (approx 70dB down).

I do not claim to be the first to do this, but how wacky is it and what are the downsides?

Rough sketch attached

Kind regards TroelsM
 

Attachments

  • DSC_0763.jpg
    DSC_0763.jpg
    995.3 KB · Views: 1,193
I think you might want to switch lm311 to something else, like lm361, even lm319 would be better.
Also your feedback is not what I would expect to see, you might want to change that.
And I have to say, most designs I've seen, they brought feedback to + input of comparator
 

Attachments

  • ucd.PNG
    ucd.PNG
    18.5 KB · Views: 540
Last edited:
if the sum of phase shifts don't add up to 360, no oscillator, doesn't matter what comparator is used. loop gain, system response, low distortion modulator, favorite comparator, PCB layout is a secondary conquest to explore. system analysis, propagation delay needs to be analysed, effort using a simulator helps a lot (I don't see any), the forum is fragmented with knowledge for a killer amplifier better than Hypex designs :D.

The forum just needs more injection knowledge around digital supervision.
 
@Luka: I believe that the sketch you show is a UCD. The one I showed is directly inspired/stolen from Infinion (iraudamp). I'm not sure if its called sigmaDelta or hysteresis-somthing. Apart from the strange cap-current-feedback its very standard.

@Reactance: Thank you for the input. I'm sorry but I don't fully understand what you are trying to say. I have been simulating a lot, but I find it a bit tricky to get all relevant parameters of the circuit included in the model without over-complicating it. I do have a fully working sim-model that correlates pretty good to my actual build.

Kind regards TroelsM
 
This is the switching residue on the output. A bit high, but it will be in an active speaker with 75cm of speaker-wire, so I'm not worried. Approx 300Khz, 3Vpp, 50Mhz, scope, 10Mhz probe.

Based on that, i guess the switching stage is OK with reagards to EMI. At least my radio did not blow up when I tested the amp....

Kind Regards
 

Attachments

  • DSC_0775.jpg
    DSC_0775.jpg
    977.1 KB · Views: 385
Hi Troels,

Would it be possible to give us a link to the Iraudamp amplifier schematics that has inspired you?

My assumption, from your hand-drawn overview schematics, is the “LM311” (shown as a buffer) is a PWM modulator operated from a fixed clock oscillator? If so, I am surprised you have a startup problem. From what you say to luka, I understand it is not a self-oscillating design?

The “POWER” in your hand-drawn schematics is the IR2184?

Your expectation is a decent control of the output in case you disconnect the load. “ddapkus” from TI has told us that the load is an essential element damping the output filter and removal of the load may even destroy components in the output. You want to remove it and even control the output without load. Without a load, the output filter is substantially undamped. Even if your carrier frequency is above the resonance frequency and your interest is the dynamic behavior below the resonance frequency, an undamped resonance circuit is probably very unpleasant in a regulation loop. Do you really need to disconnect the load?

My experience from the past is that high frequency feed-back noise can paralyze a regulator. At the output of of your filter, you still have an important ripple residual at the carrier frequency. That residual is filtered further by the speaker cables and the speaker inertia. You feed the noisy signal back to you regulator op-amp. Could it be that the high frequency noise (ripple) disturb your regulator op-amp such that you need to low-pass filter the feedback signal?

I would start with only one feedback loop and make that one work for a start. Probably the voltage feedback and remove the lead (22pF) element. Experimenting with two feedbacks at the same time, without knowing exactly how each contribute, may make things worse.
 
Hi

Short version: The handdrawn schematic is wrong. The primary feedback is taken before the filter. Sorry for the confusion. It is self-osc.

Link to the origin: http://www.irf.com/technical-info/refdesigns/iraudamp1.pdf

Yes, some of these designs have a low-pass filter that cuts above the switch-freq to remove noise.

With regards to the residue-photo I posted: its close to no-load, so maybe not the best test.

Kind regards
 
Disabled Account
Joined 2010
OK, lunchbreak update:

I have semi-OK results with the "standard" pre-filter feedback. Distortion is high (Thd:0.05, thd-noise:0.15), but it appears to be stable: no start-up problems, clipping is semi-clean. -but I wanted some post-filter feedback to tame the output if the load is removed. Also as a purely intellectual exercise for my foggy brain.

I did not have much luck with post-filter feedback and either it would have start-up problems or behave badly to no-load or clipping. Most likely due to my lack of knowledge, but it didn't work for me.

I read somewhere that a company use the output-cap-current as feedback. Without much theory I put together a rough simulation. I thought it was less than ideal to sense the actual cap-currrent so instead I put a smaller cap in parallel and sensed the current with a 100mOhm series-resistor. The current in the 2 caps should be the same with gain-difference?

It did not behave very well until i split the org feedback resistor in two and put a cap over the bigger half ( someone: that must have a fancy name in regulation-theory). Anywhos. I guess the cap lowers gain at high freq?

in LTSpice it really helps to dampen the unloaded filter-response. No impact on distortion (approx 70dB down).

I do not claim to be the first to do this, but how wacky is it and what are the downsides?

Rough sketch attached

Kind regards TroelsM
You are on a similar track that I have followed with pffb around TPA325x.

As dampening snubber might get quite lossy, a PFFB is preferable to achieve a stable output with or w/o load.

You can do the following:
Insert a series resistor of 1~2kOhm before both symmetrical inputs.

Connect a small capacitor (start with 100pF) from each post-filter output to its analogue input.

Now you have a differentiating pffb that is effective mostly close to the output filter self resonance.

Test with a sharp, unfiltered rectangle signal connectet to input with or w/o load.

Adjust cap value to best output rectangle.
This works on my TPA3251 pcbs like a charme.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.