Automatic loop gain control - just an idea...

[ViennaTom]

First about me: When other guys prefer to go out meet some girls, I often prefer to sit down at home, draw just another bode plot, and simulate loops that include the output LC filter ;-) And does anybody here know why women are usually not interested in optimizing the class D control circuit? ;-) Since it is only a hobby, I am not as deep into the thing as, e. g. Bruno or Sovadk... time is limited and simulations take usually a loong time because of the complexity involved...
So one of my ideas comes from the problem of finding the right loop gain:
If the load is basically unknown and can typically be [2Ohms < Rload < infinite] there is some great variation to take into the concept. As a result, the design can only be optimized for a certain (most typical) load, e. g. 6 Ohms, while the other loads will also have to work somehow......yeah somehow...
Example: Take an UCD type self-oscillator. If optimization is for 6 Ohms, say, a heavy 2 Ohms load would "like" to see a far lower output impedance to work best, while an 8 or 16 Ohms speaker wouldn't need this soo bad. A high loop gain is always good, and increasing the loop gain would do exactly this (lower outp. impedance), but the no load and hi-ohmic load stability and low overshoot criterion does not allow. The 2 Ohms would on the other hand do a great deal to damp down the LC resonance Q, resulating in slower phase decay within, say, the 30 - 150kHz range. So if the phase decays slower, one could increase loop gain (in this case simply by increasing the impedance of the feedback RC - smaller C bigger R while RC stays constant -> increasing the phase boost onset corner frequency -> lower ripple amplitude at the comparator input), as long as the low-ohmic load is connected. The same thing goes for different phase variations in load, one being more critical than the other...
Got my idea? I propose to (and ask if someone has already thought about this?) "sense" the onset of oscillation in the freq area, where the typical amp/speaker combi is most prone or "expected" to resonate (by a band pass), and somehow use the band pass amplitude to auto-correct the effective RC-impedance (-> disconnecting the RC from the comp input -> amplifying the RC current in a variable way using a special circuit like VCA, then feeding the resultant current into the static voltage divider node = comp input), thus building an outer very slow loop that is designed to "track" the whole system constantly at the very boundary of (but never into severe) oscillation always using an as high-ohmic "effective RC" as possible. If a speaker is disconnected, the system would resonate for several ms, but then automatically cease. When a speaker is connected, it would slowly increase loop gain to best "fit" its own inherent impedance and phase. Of course, it is important to have a steep (6th order?) bw limit in front of the amp for audio input, so that audio can not fall into the "detection" circuitry and therefore modulate itself...

I have far too less time for the loop simulation "hobby" ;-) so this could be why I have not found a good implementation of this concept so far. Or even the concept is bad? What do you think?

[newvirus2008]

With a Zobel network (lossy of course) the amplifier may not oscillate even in the no-load condition and you may be able to include the choke in the feedback loop if the LC resonance is high (e.g. 100 kHz), even with the inductive speaker.
This way, the response doesnt change much with varying load impedance and there wont be the need to change the control law on the fly.

[darkfenriz]

At self oscillation frequency the impedance of the load is mainly determined by filter C. Whatever hapens in the 30kHz to 150kHz area does not matter much, because it is converted to loop gain which is higher or lower depending on the load.
On the other hand an UcD modulator yields best properties at the load which is exactly a square root of L/C and there is not much you can do about that.
Here comes your adaptive feedback scheme if you can build and try it, go for it.
I found a different solution to the same issue, namely current control mode, which turns an LC filter to a cuurent cource driving a load and filter C, so that the small signal tranfer function is always an easy to compensate first order filter with just corner frequency dependant on load and independant HF attenuation.

[ViennaTom]

Originally Posted by darkfenriz

At self oscillation frequency the impedance of the load is mainly determined by filter C. Whatever hapens in the 30kHz to 150kHz area does not matter much, because it is converted to loop gain which is higher or lower depending on the load.
On the other hand an UcD modulator yields best properties at the load which is exactly a square root of L/C and there is not much you can do about that.
Here comes your adaptive feedback scheme if you can build and try it, go for it.
I found a different solution to the same issue, namely current control mode, which turns an LC filter to a cuurent cource driving a load and filter C, so that the small signal tranfer function is always an easy to compensate first order filter with just corner frequency dependant on load and independant HF attenuation.

Yeah, the idea is that for low ohmic load the square root of L/C is too high and must be compensated by as high loop gain as possible. This idea of "stability steering" is also applicable to other topologies, e. g. forced triangle modulation. I'll try and set up a Simetrix file for this when I find time too see if there could be any "benefit". (I don't like the idea of a lossy RC zobel, although it can of course greatly improve the whole situation...)
I ve also thought about "current mode control" so far - a controlled current source (L) working into a RC parallel load, with the outer voltage loop error signal setting the intended current. But this has several other problems, so I ve given up on that one for now..

[newvirus2008]

In the carrier-based topology, the amplitude of the triangle maybe adjusted to change the loop gain, again on the fly.

[ontoaba]

Good boy, Just avoid those girls, except you gonna marry her and get serious.

I have another way to do with self oscillation for dynamic load.
A guided self oscillating converter. It used for tracking rail ans simulated for SMALA. I post it here:
The Composite / Hybrid- Linear + Switching Power Amplifier

Why not playing with its hysteresis range/region to rise its frequency when overloaded?