Bob Cordell Interview: Error Correction

Jump to Latest
mikeks

mikeks

Disabled Account
Joined 2004
en.wikipedia.org/wiki/Censorship
andy_c said:


I was thinking in terms of MOSFETs, so I'm not sure what would happen with bipolars. In my referenced schematic, the DC Vcb of each diff pair will be about 8V. If you set up a sim and try it, you'll see that's way more than needed. This also leaves 8V for the current sources and that's plenty too. If the output stage transconductance is large enough, the voltage across the diff amps' collector resistors will consist (approximately) of only DC plus distortion, so the dynamic range requirement is less than one might think. In the sim I'm looking at now, the diff amps can only swing +/- 4V, and this is more than enough to accommodate an output that's swinging +/-80V, even with a 4 Ohm load.

Yikes! I'm not sure how I'd adapt this to bipolars to tell the truth. Bipolar output stage with MOSFET drivers maybe? That's what John does in his big amps apparently. I'm not sure why you'd want a resistor between EF emitters though. If you bias the diff amps hot so that the collector resistors are small, you may find them pulling too much current through the EFs as is. Neglecting base current, that current is constant even as the error correction current varies. But I may be thinking of something different from what you're describing.
Click to expand...


I'll look into it more carefully, but without the MOSFETs. :)

Cheers.
 
moamps

moamps

diyAudio Member
Joined 2002
moxtone.com
Error correction implemented

Hi guys:

I'd like to share my experience with the practical implementation of error correction and hopefully add more valuable insight to this most interesting discussion. For some time now I have been entertaining the idea that Bob's output stage might make an excellent output stage of a hybrid power amplifier (low distortion, high damping factor). I've just finished building the output stage (picture below) with the following transistors:

Q20, 21 - 2SC2240/2SA970
Q22, 23 - 2SC2240/2SA970
Q24, 25 - 2SC3467/2SA1370
Q26, 27 - 2SC4793/2SA1837
Q28, 29 - IRFP240/9240 .

I'm using Lundahl LL1660/18 transformers and LM317/337 current sources to drive the output stage and adjust bias. The output stage is now connected directly to my CD player (Tascam CD701) and what I'm hearing sounds extremely pleasing to my ears (and I'm a tube guy). According to my preliminary measurements, the follower, as it is now, has a linear frequency characteristic, low THD distortion and the DF over 50!
I'm sure those among you who are familiar with the transformers I'm using already have a pretty good idea about what kind of gain stage I'm planning to build for this project. :) If there's enough interest, I might consider publishing the finished amp on the forum. Code name: Academia. :)

Regards,
Milan
 

Attachments

  • academia_output.jpg
    academia_output.jpg
    54.1 KB · Views: 617
jan.didden

jan.didden

AX tech editor
Joined 2002
linearaudio.net
Paid Member
Re: Re: The bam challenge

mikeks said:
[snip]With the usual single-ended NFB, loop-transmission does NOT increase with the quantity (''error'') we wish to eliminate, while with error-cancellation-by-NFB, loop transmission is directly proportional to extracted error.
Click to expand...

Indeed. Another way to look at the contrast. With se nfb, the forward loop equation always contains the forward gain block gain 'A', while the nfb factor always is linear.

In the circuits we discuss here, the forward loop equation (in the idealised case) is perfectly linear, while the nfb factor equation contains 'A'. The feedback FACTOR varies with the non-linearity of the 'A'gain block.
I called it ODF - on-demand feedback.


Jan Didden
 
Nixie

Nixie

Banned
Joined 2006
Re: Re: Give SPICE a try (& HF Ingress)

gootee said:
Note that the book that I mentioned does indicate that amplifier OUTPUTS _are_ also vulnerable to EMI/RFI, especially if they drive lengths of cable (e.g. speaker wires), "...which act as antennas. RF signals [sic] received on an output line can couple back into the amplifier input where it is rectified, and appears again on the output as an offset shift."

For amplifier outputs, they recommend adding series resistance or ferrite beads, or an RCR "T" filter.
Click to expand...
Can you recommend specific ferrites for power amplifier outputs driving the usual 4 or 8 ohm loads? I also want to add such a filter at the output of my electrostatic headphone amplifier, so the load is a capacitor in that case; how do I estimate the appropriate ferrite for that as well?
 
jan.didden

jan.didden

AX tech editor
Joined 2002
linearaudio.net
Paid Member
Re: Re: Re: Re: The bam challenge

ingrast said:


Extremely insightfull.

While A behaves, the loop sits quietly there unobstrusively.
The least deviation, it wields a big :mallet: to put things back in short order.
Click to expand...


Another attribute of ec. You cut the ec loop, and the amp happily continues to work as if nothing happened, albeit with less linearity. Try that with a convential global nfb amp!

Jan Didden
 
traderbam

traderbam

Member
Joined 2002
Jan wrote:
Another attribute of ec. You cut the ec loop, and the amp happily continues to work as if nothing happened, albeit with less linearity. Try that with a convential global nfb amp!
Click to expand...
Cut the all the NFB and it will go nuts. Bob's output stage will fry itself. Bob's circuit is different from the pure Hawksford model - it two halves are cross-connected so there is common-mode NFB that sets the output stage bias. The linearity degrades when the differential feedback path is disconnected.

Believe what you want. But my question is...if you were convinced it was just NFB in disguise, would you choose to configure your output stage like this? Is this the best way you could devise for applying 30dB of NFB to an output stage?

The trouble is, if you really believe there is something more than NFB at work, you'll restrict your thinking. You'll insist on two feedback paths rather than one, you'll insist on having an adjustment you don't need and you'll insist on using a positive feedback loop to generate gain, which has the drawback of amplifying its own distortion.

I've demonstrated by commonly accepted maths that its theoretical function is that of a NFB system, no more, no less. No one has yet shown fault in my maths. Bob agrees with it too and it's his circuit.

I am seeking to be persuaded by practical evidence, simulation result or rational argument that implementing a 30dB NFB loop around a FET output stage is BEST implemented by this "EC topology" but at the moment I have my doubts.

I am eager to see any evidence from you or Rodolfo as to why you think a PFB loop around a unity gain buffer makes a superior gain stage in a real circuit. I think that would be a really useful thing to share.

Brian
 
jan.didden

jan.didden

AX tech editor
Joined 2002
linearaudio.net
Paid Member
traderbam said:
[snip]I am eager to see any evidence from you or Rodolfo as to why you think a PFB loop around a unity gain buffer makes a superior gain stage in a real circuit. I think that would be a really useful thing to share.

Brian
Click to expand...


Brian,

I will share my circuit, but not yet. I have my reasons, and I hope you will accept that.

In the mean time, I think Bob has mentioned that the ec ONLY on his output stage did dramatically decrease distortion. And that is with using two transistors that were there anyway for the bias. Or at least one of them.

Is that not worthwhile?

Jan Didden
 
traderbam

traderbam

Member
Joined 2002
I will share my cicuit, but not yet. I have my reasons, and I hope you will accept that.
Click to expand...
Of course.
In the mean time, I think Bob has mentioned that the ec ONLY on his output stage did dramatically decrease distortion.
Click to expand...
By about 30dB.
Is that not worthwhile?
Click to expand...
Sure it is. :up:
Is this the best way to implement NFB around a FET output stage? Maybe it was in 1984. I don't think it is now. I could be wrong, but I want to find out. :magnify:

NFB around an output stage is not a new idea. Bob's version of Hawksford's circuit is interesting to me not least because it uses positve feedback and it doubles up the function of the bias transistors. What I don't really like on the face of it is that it needs 10 transistors, it needs adjusting, it isn't very symmetrical (electrically), and at its nominal feedback setting it has poor stability into capacitive loads - so it needs an output inductor.
 
J

jcx

Member
Joined 2003
The counterargument:

Hawksford/Cordell error correction is a 2 degree of freedom linear feedback control system,

All expressions of 2 degree of freedom control systems are equivalent.

Therefore Hawksford/Cordell error correction is equivalent to conventional error feedback with a command prefilter

Therefore the output linearizing disturbance rejection effect is exactly equivalent to error feedback and has the same gain-bandwidth/stability limits of ordinary 1 degree of freedom error feedback

A little reading: (slow to load, ~ 500K )

cntrl002.png


cntrl003.png


[I've referenced this a few time before in this thread I.M. Horowitz, "Synthesis of Feedback Systems" 1963 - that mkes this a 40+ year old "controversy"]

I like Bob's circuit, I do think understanding it as a local feedback loop and tailoring its loop gain and frequency response is likely to be a good way forward
 
traderbam

traderbam

Member
Joined 2002
Hawksford is a negative feedback loop without any positive feedback. This was demonstrated above with the bare minimum of intuitive Control theory.
Click to expand...
I think it is your maths that has made your diagrams incorrect. Your are manipulating non-linear elements as if they were linear.
I think if jcx wishes to correct any of my reasoning he is capable of doing so himself and I warmly invite it. I don't see any conflict with what I have said and what jcx has said.
 
You must log in or register to reply here.
Top Bottom