Magnatec class d amp sulution

[Jan-Peter]

Lars,

Our amps are switching at +/-450kHz.

Just curious;
Why are you claiming feedback before the outputcoil would be better? For what for a reason would this be better?


Your modules looks nice, for shure that you can go so low in impedance.


Regards,

Jan-Peter

www.hypex.nl

[Lars Clausen]

Jan Peter: Thank You My main argument for taking the feedback before the coil (and BTW also to have only a single feedback loop, which results in slightly higher THD measurements) is to keep the time delay in the feedback loop as low as possible. This gives me lower TIM, and better sound quality (my main concern - more important than good THD data). Another thing:
If you take an amp with feedback after the coil, the time delay and also switching frequency will get longer (lower freq) as you add capacitive load on the amplifier. I imagine if you add a few uF across the output, the freq may get low enough to destroy the amp with blind current in the coil .. ? This problem is non existent with amps that take the feedback before the filter.
One last thing is that by letting the filter coil determine the switching frequency, as happens when you take feedback after the coil, you might have trouble syncronizing the module with other modules in a multichannel setup. Please correct me if i'm wrong, there may be some way around this, that i am not aware of

I think both systems have pro's and con's. One solution may be better in some applications, while the other is better in other applications.

BTW: you stuff looks really nice !

All the best

Lars

[Jan-Peter]

Lars,

Nice we like each other products

We use the total time delay of the amplifer, included the delay in the filter to create the selfoscillating. In such a way the whole systems becomes a liniear gainblock. Because of the feedback behind the outputcoil, the impedance of the outputcoil is in a way removed from the output. And will be below 0.010 Ohm. Because of the feedback behind the outputcoil we don't have a peak in frequency response around the LC frequency.

In never tested several uF at the output of the amp, but several 100nF will not be a problem. The amplifier sees already a big capacitor at the output


In a multichannel setup we create slightly different extra delays in the feedbackloop to create a difference of 35kHz in every Class-D amp. We already did this in a 2-way and 3-way active studio monitor.

Regards,

Jan-Peter

www.hypex.nl

[Bruno Putzeys]

Originally posted by Jan-Peter
In a multichannel setup we create slightly different extra delays in the feedbackloop to create a difference of 35kHz in every Class-D amp. We already did this in a 2-way and 3-way active studio monitor.
www.hypex.nl

I never do that... On our 2x250W board, difference tones are well below 10uV (you need to sift them out with an FFT to find them).

Admittedly it takes some practice.

[phase_accurate]

Did you ever build anything in the real life, or just comment on everybody else's work?

I developed a class-d amp 13 years ago, when info on class-d amps AND suitable components were still very scarce. The schematic can be found on this forum.
It didn't use feedback from the filter, so it was a "first-timer" like all the other ones.
Even though it wasn't intended as audio amp it sounded very nice.

If you take an amp with feedback after the coil, the time delay and also switching frequency will get longer (lower freq) as you add capacitive load on the amplifier. I imagine if you add a few uF across the output, the freq may get low enough to destroy the amp with blind current in the coil .. ? This problem is non existent with amps that take the feedback before the filter.

You will not have a deviating switching frequency with carrier-based class-d amps (i.e. PWM) like mine was and the Magnatec also is.
Since you will not have really large capacitive loads in real life (and veeeeeeeery seldom purely capacitive ones !) it will not be a large problem with amps like yours either.

And the delay of the filter is definitely NOT a problem. I made the fatal mistake to use my imagination only over the years in order to find ideas how it could be done. But it is definitely better to use imagination AND maths to come to conclusions for how to do it. In the meantime I do not only know how one could use after-filter feedback takeoff with first-order PWM loops but also high-order delta-sigma loops.
The absolutely easiest solution for feedback takeoff from the filter I came up with, I use to call "the simple tweak". It can be applied to any class-d topology where the feedback signal is going into an inverting integrator. It can be used with other topologies as well but it would then be a little less simple. From the measurements that were made by Stereopile, I assume PS-Audio does something similar.

Regards

Charles

[Bruno Putzeys]

Originally posted by Lars Clausen
My main argument for taking the feedback before the coil (and BTW also to have only a single feedback loop, which results in slightly higher THD measurements) is to keep the time delay in the feedback loop as low as possible. This gives me lower TIM, and better sound quality (my main concern - more important than good THD data).

The problems you quote are typical of control structures where the post-filter feedback is added as an afterthought instead of being an integral part of the solution.

While it isn't obvious at first to solve the "time delay" problem, it's very amenable to the use of lead compensation. The whole UcD concept revolves around uh... an extreme case of lead compensation.

Thus executed, the sonic tables turn. It is my experience that amplifiers without post-filter correction all have a sense of glassiness in the top-end. This is often confused with tube-like warmth, but is a definite detraction from neutrality/transparency.

TIM is simply a restatement of an amplifier's slew rate capability, and its ability to remain linear when brought close to its slew rate limit. In linear amplifiers, distortion often already starts increasing when you're only getting near the slew rate limit. In class D amplifiers, the mechanisms responsible for this is not present. Therefore, as long as power bandwidth exceeds 20kHz, there is no correlation between slew rate and sound quality. In general, I have little sympathy for the still mythological status of TIM. Still today I get people charging at my desk, waving a copy of Otala's paper, proclaiming they know the source of "solid state sound" now. It's only an intermod measurement, nothing more!

[Lars Clausen]

Bruno: I would never throw Otala at you, in fact i have never read his papers. However it seems obvious that the 'older' (in microseconds) your feedback signal is, that you attempt to align with the current input signal, the higher the mess in signal transients. Maybe not of any importance when you are measuring response to nice sinewaves in the lab. But when we are talking music, it's a whole different ballgame. And i think that it is also obvious that if you take this feedback signal after the coil, then it is nessescarily a little bit 'older' and has 90 degrees of phase shift compared with the direct connection to the output stage. This 90 degrees at fc which is example 90 kHz, makes the voltage (which is what you use to feedback) delay 90 degrees at 90 kHz, or roughly 3.5 microseconds more delay after the coil than before. But OK i agree if you are using a high order sigma-delta feedback loop, the coil delay would be insignificant, as the 4-5th order feedback filter will probably have much higher delays. I am not convinced this is a better way to go.

Your remark about 'as long as power bandwidth is higher than 20 kHz, there is no correllation between slew rate and sound quality' i guess it is people like you who put MC4558 opamps (dual uA741) in modern CD players, because they can just meet the 20 kHz, and so there is no reason to go for higher slew rate. I will only add, that i don't agree with this kind of engineering.

But let's not start the old discussion about whether mathematical or intuitive engineering is better. Anybody will claim their own way of doing things is the best, and much better than anybody else's.

Phase Accurate: Thank You - that answered my question.

[mikeks]

Originally posted by Lars Clausen
Bruno: I would never throw Otala at you, in fact i have never read his papers.

You've missed nothing!

[mikeks]

...how do you class-d folks measure loop gain in your systems?

[ashok]

Hi Partyjups,
Could you please email me a copy of the application notes.
My email id is ashokm(at)sify.com . Please change the (at) to (@).
Thanks.
Ashok.