The Sound and Testing of Class D

[classd4sure]

Hi,

http://www.eetasia.com/ARTP_8800376594_499486.HTM

Discuss.

[sx881663]

Designing a Class D amplifier solely for good THD test results does not mean it sounds good, let alone qualifies it as "audiophile."
The above statement from the “linked” article is about the only thing in all of it I agree with. As we know there is class d then there is class D as in the UcD design. There is an inherent limitation due to the basic switching frequency and the time between samples. Basically what we end up with is a switching amp of undetermined bit length that establishes the direction and amount of current in the output filter network. This current flows for the sample time length and this is the ultimate limit to linearity. With the output filters connecting the dots and filling in this between time most of the other digital crap really has little bearing on the final result. Of course there is the dead time issue and this does matter as it affects this current flow.
I think the charts he did are for a non global/open loop feedback type design. (Would this still be class d?) I see no reason the UcD type design should or does fall apart at low level. Of course if the comparator doesn’t have high enough gain that is a whole other thing that could show up this way. In other words the low level stuff would get buried in the noise. What he didn’t show was a typical a/b amp in the same test. Would it even do that good?
Roger

[classd4sure]

Hi,

I know it has little bearing on the class d that we know and love so well. His writings are geared towards the "true digital" ... which I would bet his IC isn't even.

Why did I link to it? I bet you're scratching your head about that right about now.

It's controversial, just look at the title he used. I wasn't aware class d amps were even known to measure well, not until recent times anyway, but this does seem like a recent paper.

I think with self oscillating amps the time between samples is 1=1, 1 "sample"=1 switch= 1 bit amp, which is supposedly 100% linear by that definition.

In that sense I do agree with something else that he wrote, when he said the output stage = the resolution, and agree with you in that it is somewhat undetermined, just throw it in the feedback loop and forget it.

If you build that delay into the control loop, it's not so much an error source as it is a bottleneck to increasing Fs. We're again at a point where the limiting factors are the available devices. Although, it's _not_ a bottleneck to a good sounding amp, because we just don't need more, not that more wouldn't be nice.

All the same, I think linearity is what it's all about.

Thanks for the reply, I believe there's much more to this article to be picked apart yet.

Regards,
Chris

[classd4sure]

What he didn’t show was a typical a/b amp in the same test. Would it even do that good?

Well, we have to realize he's writing that paper to push his product. It's an advertisement.

I guess the only smart question here is "which a/b amp?"

I think I'd like to see more graphs like that.

Regards,
Chris

[sx881663]

Originally posted by classd4sure
I think I'd like to see more graphs like that.

I wonder what he used to generate those graphs, Audio precision?
It does look like real data but has a lot of range shown. I do admit it is an interesting test and probably significant for all amplifiers equally.
On another mater the absolute linearity for low level information is directly related to the gain and noise floor of the comparator. Where the inputs cross is the event marker for a reversal of polarity. The linearity of the comparator only relates to its common mode response as the cross over point may be at a few volts + or – so this becomes the real deciding factor. The gain and any hysteresis must be uniform across the entire input voltage range. One wants this common mode range to be large for good signal to noise but not at the expense of nonlinear response at the extremes. The problem is that the real important low level information will likely as not be riding on much larger signals and it does require gain to pull this information out. The gain must not swamp out this information or it all falls apart. It makes one appreciate all the care that has gone into common mode rejection in the UcD units. What is the common mode signal? It is everything but the differences between the input and output corrected for gain and timing. As there will be significant current from the output and into the grounds this makes them even more important. Remembering the grounds are connected to both comparator inputs as the reference so they must track or there will be errors introduced. One big juggling act and accomplished quite nicely in the UcD. I think the comparator performance is one of the major points to the excellence of the UcD.
Roger

[phase_accurate]

I do admit it is an interesting test and probably significant for all amplifiers equally.

I think the best part is that it could be carried out with much less effort than any THD or IMD measurement if purpose-built test equipment is used.

Regards

Charles

[soongsc]

I think the linearity issue is related with the PWM portion which used to be associated with Harshness of these amps at higher frequencies. I remember this was a similar issue back in the days when it was debated whether SMPS should be used in audio equipment or not.

If linearity is an issue, and it's linearity problem at small signal levels, then using very small signals should show the results more significantly. If it's a full range linearity issue, then if the tendency is compression, it would be identifiable through listening, otherwise, I wouldn't mind some dynamic range expansion.

[sx881663]

Charles,
Do you think we could enlist a good programmer and do this with our computer/sound card? I feel this could be some important data.
Soongsc,
It is all connected together as the comparator is the front end for the amp and creates the actual pulse to pulse timing in accordance with the feedback.
Roger

[phase_accurate]

Do you think we could enlist a good programmer and do this with our computer/sound card? I feel this could be some important data.

This would be nice indeed from the user's point of view. Any ADC however has his weaknesses exactly where we want to be able to find the irregularities of the DUT (i.e. linearity at low levels).

I thought of an analog circuit that is comparing input and output signal directly.

Regards

Charles

[classd4sure]

Originally posted by phase_accurate


This would be nice indeed from the user's point of view. Any ADC however has his weaknesses exactly where we want to be able to find the irregularities of the DUT (i.e. linearity at low levels).

I thought of an analog circuit that is comparing input and output signal directly.

Regards

Charles

An analog computer, couple op amps.... yeah, nothing's faster

Are you guys seeing the possibilities like I am here?

It's more than just simplicity, it's the strong visual aspect of it, everyone knows what a straight line should look like, so that's all you need to compare it to.

I think a multitude of tests could be still be done while expecting to see the same straight line output.

Sweep the frequency, step the power, multi tone tests.. can you think of others?

I was impressed that Mueta gave a linearity graph on their recent update, but I think they swept frequency and kept the gain constant. Why leave it at that?

Regards,
Chris