Christer wrote:
Again, I find Hawksford is EQUIVALENT to ordinary feedback. There is no difference at all in concept. In implemenation, and there are many variants, there seems to be one thing that stands out: the use of a PFB loop to generate loop gain.
If you suggested to the linear control systems community any control system that uses a PFB to generate loop gain should be known as an "error correction" system and is somehow special, you'd be laughed at. If you further claimed that it is special because there is only feedback loop gain when there is an error signal, you'd be gently helped into a straitjacket.
For the sake of minimizing thrash in this thread, please express your arguments objectively.
This "sounds" good and it is the subjective argument I keep hearing from Jan and Rodolfo, but it is not supported by the objective analysis. When I do the maths I don't get this result. So I am asking everyone to support their subjective hypothesis with objective argument. Otherwise we are in danger of drowning in a sea of wishful conjecture.
Again, I find Hawksford is EQUIVALENT to ordinary feedback. There is no difference at all in concept. In implemenation, and there are many variants, there seems to be one thing that stands out: the use of a PFB loop to generate loop gain.
If you suggested to the linear control systems community any control system that uses a PFB to generate loop gain should be known as an "error correction" system and is somehow special, you'd be laughed at. If you further claimed that it is special because there is only feedback loop gain when there is an error signal, you'd be gently helped into a straitjacket.
For the sake of minimizing thrash in this thread, please express your arguments objectively.
traderbam said:...it is the subjective argument I keep hearing from Jan and Rodolfo, but it is not supported by the objective analysis. ....
.......
For the sake of minimizing thrash in this thread, please express your arguments objectively.
If looking at the same phenomena under different perspectives must be dismissed as subjective, we are making creative thinking no favor at all.
The colophon in offensive.
Rodolfo
traderbam said:
Untrue:
http://www.diyaudio.com/forums/showthread.php?postid=1069128#post1069128
traderbam said:
I disagree-profoundly.
The PFB is incidental, as the loop gain (in the major loop) is equal to error extracted at balance.
Proof!
In post #1128 Milan wrote:
I was thinking that I would characterise Bob's output stage in simulation and publish what it actually does. Then invite everybody to suggest designs ideas that improve upon those characeristics and evaluate them objectively, and then incorporate the proven ideas into one or two new designs that perform better.
How would that be?
Hopefully this will all end in an improved design that is sound, both academically and otherwise.
I was thinking that I would characterise Bob's output stage in simulation and publish what it actually does. Then invite everybody to suggest designs ideas that improve upon those characeristics and evaluate them objectively, and then incorporate the proven ideas into one or two new designs that perform better.
How would that be?
In post #1135 Andy_c wrote:
The use of a shunt regulator at the output of a series regulator has a similar impact. Without the shunt reg the output V might be unchanged, but what the shunt is doing is applying feedback to minimize the output Z and reduce noise.
With Bob's output stage the overall gain will appear to be close to that with the "EC" feedback enganged into an 8-ohm load. Into a low Z it will vary a lot. If you measure the output Z of the amp it will be changed by the feedback factor as you'd expect.
Application of feedback doesn't have to change the voltage gain of the system.
One comparison might be insightful...it's just a vague comparion and I hesistate to mention it in case it sets off a whole new slurry of "creativity" in this thread.
The use of a shunt regulator at the output of a series regulator has a similar impact. Without the shunt reg the output V might be unchanged, but what the shunt is doing is applying feedback to minimize the output Z and reduce noise.With Bob's output stage the overall gain will appear to be close to that with the "EC" feedback enganged into an 8-ohm load. Into a low Z it will vary a lot. If you measure the output Z of the amp it will be changed by the feedback factor as you'd expect.
Application of feedback doesn't have to change the voltage gain of the system.
traderbam said:
Furthermore, nowhere have we claimed loop gain depends on the presence of error. What has been highlighted as distictive is the fact the corrective action is taken only when the error signal develops.
When an error appears, then - yes I am not afraid of saying it - an infinite gain takes care of it.
I do not think we should be obliged to append allways a discaimer on the lines that when we say "infinite gain" in fact we are acknlowledging there is no such thing in the real world. Sensible people do not need such clarifications.
Rodolfo
ingrast said:
Actually, i have; here.
And no one, absolutely no one, has yet produced a disproof.
And you know what Rodolfo?
Further proof that i am right may be adduced by considering that your closed loop transfer function is identical to mine here, with your 1/A' term equated to unity.
Which is why Traderbums musings are quite...quite nonsensical!
Re: Error correction implemented
Nice work if you can get it. My soldering iron's hot and ready. I have about 10 kilos of transistors lying about. All systems go!
Academia's output stage is still working nicely. My sonic impressions are extraordinarily positive and there is no doubt in my mind that the project is worth pursuing to completion.
As regards the output stage circuit, I still need to compensate the interstage transformers at around 50kHz. Also, there is a small problem with DC drift so I'm planning on adding a servo.
This is how the output stage looks at the moment.
Regards,
Milan
traderbam said:
Nice work if you can get it.
My soldering iron's hot and ready. I have about 10 kilos of transistors lying about. All systems go! Academia's output stage is still working nicely. My sonic impressions are extraordinarily positive and there is no doubt in my mind that the project is worth pursuing to completion.
As regards the output stage circuit, I still need to compensate the interstage transformers at around 50kHz. Also, there is a small problem with DC drift so I'm planning on adding a servo.
This is how the output stage looks at the moment.
Regards,
Milan
Attachments
In post #1185 Rodolfo wrote:
I like the NFB loop with an integrator ahead of the "plant" with a stable GBP frequency. This is the op-amp strategy, of course (I don't like other aspects of op-amps, though). Examining the performance of PFB loop to make an integrator block is something I'm interesting in.
Rodolfo wrote:
My only hesitation is that the distortion of the practical PFB unity-gain amplifier will be amplified by the PFB around it. So for this circuit to be better than a conventional integrator it must be shown that the net distortion is less. That would be of great interest to me.
I'd say this is exactly the same as with a simple NFB loop. The error "correction" signal is proportional to the error. The gain exists all the time.
Agree. The loop behaves as a 1 pole (or more) integrator in practice, with the first pole pretty close to dc, as you have objectively demonstrated.
My only hesitation is that the distortion of the practical PFB unity-gain amplifier will be amplified by the PFB around it. So for this circuit to be better than a conventional integrator it must be shown that the net distortion is less. That would be of great interest to me.
mikeks said:
Well Michael, let's see.
The first link points to an equation where for example there are factors of the form K+e. This is dimensionally incorrect for it implies sum of a dimensionless term (K) and a voltage/current or whatever physical magnitude.
The second link points to a gain expression coincident with mine, where the error term does not appear as a variable affecting the result. This is correct and implies the presence of error or lack of it does not affect system gain. This is of course to be expected of a linear system as is represented in this example, the error being treated as an added signal.
Rodolfo
traderbam said:
I can relate to this, again as the saying goes, the proof is in the pudding.
So far I have consistently been able to adjust for minimum distortion, and to achieve unmeasurable levels within my measurement capabilities.
I have remarked before it remains to be seen whether under the same conditions of device capabilities and circuit complexity, other alternatives are similar in performance.
Rodolfo
traderbam said:
Brian,
I think it is clear from Horowitz that if you add gain to the plant (which could be done with a PFB loop or any other kind of gain block ahead of the plant) you have a 2DoF system of type a, and it is furthermore equivalent to a 2DoF system of type f, which as far as I can see is topologically equivalent to HEC. Both these topologies are thus different from ordinary feedback, which is a 1DoF system. I am however relying on Horowitz here, you should consult him, or other similar literature, for proofs.
However, isn't the problem and confusion rather about what I wrote earlier, that while control theory treats the plant as given, in audio we actually design the plant itself along with its control system. That makes the distinction between 1DoF and 2DoF fuzzy since adding gain to the plant can be viewed as modifying the design of the plant itself. We can, in principle, increase the OLG of the amp and get the same result with ordinary feedback as with HEC. However, since we modified the plant itself we loose the clear distinction between 1DoF and 2DoF.
Christer said:
On second thought, maybe it is more confusing than enlightening to view it this way. In the cases we typically look at in audio, there is no OLG in the plant in the first place (that is, the OLG is usually slightly less than 1). It is not very helpful then to say that we use all the available OLG to correct the error. My point was just to try finding a simple pedagogical way of looking at things without digging into the maths, but maybe it is unavoidable here. It is probably necessary to realize that we add gain (in disuise) in the form of the PFB loop. On the other hand, since the PFB loob can be collapsed to a gain block, it just makes it even more obvious that Horowitz' cases a and f are equivalent.