New self oscillating post filter feedback topology...

[Eva]

You will need op-amps for implementing active poles, the trick is to wire them in such a way that the circuit does not rely on them much above 20khz.

[81bas]

Originally Posted by Eva

You will need op-amps for implementing active poles, the trick is to wire them in such a way that the circuit does not rely on them much above 20khz.

May I rephrase your sentence? I would say "the trick is to wire them in such a way that the circuit does not rely on them near Fo of the output LC filter". I am right?

Also, is the usual integrator an example of the active pole filter? I have tried to read about it somewhere in Internet/Wikipedia, but it is usually connected with too complex mathematics for me...

[ViennaTom]

@EVA: Why would I need active poles?

[81bas]

Originally Posted by ViennaTom

@EVA: Why would I need active poles?

Explanation in attached image...

[ViennaTom]

Oh I see, the idea of introducing an active pole is to have loop gain rising towards lower frequencies, right? Interesting idea...But what is this good for? Output impedance at low frequencies? Isn't it better to have constant loop gain throughout the audio band?

[81bas]

Originally Posted by ViennaTom

But what is this good for? Output impedance at low frequencies? Isn't it better to have constant loop gain throughout the audio band?

I would say, in this case each decides on its own, what is better...
But again, 25dB loop gain is too low for me...

[ViennaTom]

@81bas: After some calculation and simulation, I came to the conclusion that the loop gain in my circuit is about 20 (equals 26dB) So you are quite right in assessing my circuit! You must be quite an expert! How did you calculate? I divided output pk pk amplitude by (comparator input peak peak ripple times audio frequency voltage gain). e.g. 360 / (0.32 x 56) = 20

[81bas]

Originally Posted by ViennaTom

@81bas: After some calculation and simulation, I came to the conclusion that the loop gain in my circuit is about 20 (equals 26dB) So you are quite right in assessing my circuit! You must be quite an expert! How did you calculate? I divided output pk pk amplitude by (comparator input peak peak ripple times audio frequency voltage gain). e.g. 360 / (0.32 x 56) = 20

oh, I am not an expert at all I have seen this value in previously attached image, and supposed that most of UCD designs will have the same loop gain too...
However, I completely agree with your way to calculate the loop gain in UCD. The key point here is the amount of switching ripple at the comparator input, as you mentioned... I think also, that increasing the switching frequency will increase the loop gain of UCD, since amount of the residual ripple will be lower!

[ViennaTom]

So I implemented a pole @ ~20kHz (C4, R6). Of course w/o any opamp!!
Now the calculated loop gain is ~ 52dB from DC to ~1kHz, then falling linearly to ~ 26dB @20k. I think that is quite ok for ultra-high bass and lower-mid-area damping factor.
Step response: You can see from the small signal response: it is - of course - nearly perfect. Big signal step response looks like slight instability for the first glance, but it is not: If higher slew rate is demanded than output filter can deliver, the regulator loop is "overloaded" and the high loop gain "pole loaded" regulator will "remember" missed volt-second product during ascent or descent of output voltage and then answer with the overshoot. This is NOT instability. It will occur at high slew rates and can be more or less avoided by placing a bandwith limiting input filter in front of the amplifier (like 5Hz - 25kHz 2nd order, which I recommend for PA amplifier....)

[81bas]

Originally Posted by ViennaTom

So I implemented a pole @ ~20kHz (C4, R6). Of course w/o any opamp!!
Now the calculated loop gain is ~ 52dB from DC to ~1kHz, then falling linearly to ~ 26dB @20k. I think that is quite ok for ultra-high bass and lower-mid-area damping factor.

Cool! I am not sure, whether it will work for real world comparator, but anyway...

Could you please say, how big is the difference in switching frequency between "zero" and "near clipped" output state in your simulations? Usually there is a 2:1 frequency drop, as Eva said, but it is for "normal" UCD...