Bob Cordell Interview: Error Correction

KSTR said:
[snip]Maybe the use of a non-opamp circuit symbol would have done the trick even better ... the next time you publish this trick 🙂

- Klaus


Yes... Like this, perhaps:

Jan Didden
 

Attachments

  • elt p1 f5.jpg
    elt p1 f5.jpg
    91.3 KB · Views: 730
bias

KSTR said:
Another great circuit idea of you, Edmond. Have to digest it slowly, I guess.

Thanks Klaus and take your time.

In fact this circuit looks quite close to what I have in mind, one difference being that I wanted to use the true minimum instead of average of the output current (which is more or less identical only while in class-A), compare that to a reference, feeding a servo in turn. I feel that sensing the minimum might be essential to keep things in order, under large class B operation. It's intended for a D2S stage where, when used without e.c./feedback, keeping the correct bias will be essential (and this optimum might still be itself dependent on actual die temp, something I would need to investigate). There still is the problem then whether the minimum detector is fast enough to catch the sometimes very quick transition through the minimum, same problem that you see, I read.

That's precisely the problem. I doubt whether it's feasibly to design a sufficiently fast, accurate and reliable minimum detector. Therefore, I've chosen for a circuit that "spread out" this minimum to a much larger range of about 1A of output current. IOW, when both N- and P-channel MOSFETs are still conducting a measurable/detectable amount of current, yet reproducing a signal that is closely related to the bias current at quiescent condition. (see the middle curve of fig. 3)
Regrettably, I don't think that my circuit performs equally well if applied to a D2S amp. :sad:


That's why I think of having regular "slow" temp sensing and only adjust its op point with the quicker servo over a limited range.

- Klaus

Hmm..........:scratch:

Cheers, Edmond.
 
KSTR said:
Regarding the temporal bias drift under changing load conditions, has anyone ever thought about a servo style circuit which measures the actual bias current and adjusts it? With the cascoded circlotron approach it is quite easy to track the sum of the transistor currents without the use of degeneration/current sensing resistors in the main pair, and the minimum of it (occuring during output current zero crossings), which is the actual bias, could be captured with a peak detector (with droop, of course), smoothed, processed and summed into the thermally derived bias. Which would be especially useful if the thermal loop tracking and time constant is large and unreliable, when using devices without on-die temp sensing means but a traditional heatsink-mounted Vbe-mult instead.

- Klaus


In this regard, the LT1166 is a fascinating part that, I believe, largely solves the problem you are concerned about here. It uses a translinear circuit to effectively set the right bias spread under dynamic, real-time signal swing in a Class-AB MOSFET output stage based on the ever-changing voltages across the output source resistors. A nice side effect of this is that the "other" output transistor in Class-AB operation in theory never quite cuts off.

Alas, this IC has never really caught on with designers.

Cheers,
Bob
 
LT1166

Bob Cordell said:
In this regard, the LT1166 is a fascinating part that, I believe, largely solves the problem you are concerned about here. It uses a translinear circuit to effectively set the right bias spread under dynamic, real-time signal swing in a Class-AB MOSFET output stage based on the ever-changing voltages across the output source resistors. A nice side effect of this is that the "other" output transistor in Class-AB operation in theory never quite cuts off.

Alas, this IC has never really caught on with designers.

Cheers,
Bob

At first glance, I was also fascinated by this IC, BUT my simulation results were rather disappointing. Moreover, for a D2S output stage, this IC spoils the whole concept (of D2S).

Cheers, Edmond.
 
Re: LT1166

Edmond Stuart said:


At first glance, I was also fascinated by this IC, BUT my simulation results were rather disappointing. Moreover, for a D2S output stage, this IC spoils the whole concept (of D2S).

Cheers, Edmond.

I have asked you and many other people before, WHAT IS YOUR EXACT PROBLEM WITH THIS IC. Everybody, including yourself, came up with "seems like" or "I don't remember, but I had problems", type of answers.

I have simulated, used and measured extensively this IC, in both discrete and IC based (LM4702) power amps, both for biasing MOSFETs and bipolar power devices. My conclusion is that LT1166 is an excellent solution, as long as the internal transconductance stage is not used (there is a LT application note about how to do this). The only problem I encountered was the clipping behaviour but nothing that a Baker clamp couldn't solve. And BTW, it measures much better than it simulates.

Perhaps I wouldn't recommend this solution for the ultimate performance (where cost and complexity are non-issues), but simply dismissing this $2 IC, and instead choosing a circuit that takes a s***load of board space and a handful of components, sounds silly to me.

Here's the model I was using in simulations. The subckt pins are numbered identical to the circuit pinout.

*
.subckt LT1166 1 2 3 4 5 6 7 8
Q4 N001 N005 N002 PN
Q3 N049 N005 3 PN 5
Q1 N003 N007 N004 NP
Q2 N007 N013 3 NP 5
R1 2 N004 200
R2 N002 2 200
Q5 3 N001 4 NP 40
Q6 N002 N001 4 NP
Q7 N005 N008 N006 NP
R3 4 N006 1k
Q8 N008 N008 N009 NP 2
R4 4 N009 500
R5 N008 N012 425
Q9 N010 N008 N011 NP
R6 4 N011 1k
Q10 1 N024 4 NP 20
Q40 3 N003 1 PN 40
Q39 N004 N003 1 PN
Q11 N022 N014 N023 NP
Q12 N017 N014 4 NP
Q13 N014 N015 4 NP
Q14 N015 N015 4 NP
Q15 N019 N008 N021 NP
Q16 1 N048 N001 NP
Q17 N048 N020 N019 PN
R7 4 N024 1.2k
R8 4 N023 160
R9 N022 N025 120
R10 4 N021 1k
R11 N018 5 5k
R12 N019 3 6k
R13 N017 3 1k
Q18 N015 N017 N016 PN
Q19 N014 N018 N016 PN
R28 N020 6 200
Q20 N025 N022 3 PN
Q21 4 N025 N026 PN 10
Q22 N037 N027 N026 NP 10
Q23 N027 N029 3 NP
C3 N018 N014 5p
Q24 N046 N028 N010 NP
Q25 N047 N031 N010 NP
Q26 N032 N034 N033 NP
Q27 4 N032 N003 PN
Q28 N033 N040 N035 PN
R14 3 N031 1k
R15 3 N033 6k
R16 N028 8 5k
R17 N035 1 1k
R18 N036 1 160
R19 N027 N029 120
R20 N030 1 1k
R21 N038 1 10
R22 N045 1 1k
R23 N044 1 1k
R24 N040 1 2k
R25 N041 1 1k
R26 N039 1 1k
R27 N034 7 200
Q29 N047 N047 1 PN
Q30 N046 N047 1 PN
Q31 N031 N046 1 PN
Q32 N029 N046 N036 PN
Q33 N037 N040 N030 PN
Q34 N024 N037 N038 PN
Q35 N016 N040 N045 PN
Q36 N043 N040 N044 PN
Q37 N042 N040 N041 PN
Q38 N013 N040 N039 PN
Q41 4 N042 N040 PN
Q42 N042 N043 N012 NP 10
Q43 N043 N043 N008 NP
C2 N046 N028 5p
C1 N037 N024 20p
R29 N007 N013 200
R30 N005 N049 200
D1 3 N050 DZ
D2 4 N051 DZ
D3 1 N050 DD
D4 3 N051 DD
R31 1 N032 200k
R32 N048 4 200k
R33 N042 N008 10Meg
.model DD D(Is=2.52n Rs=0.568 N=1.752 Cjo=4p M=.4 Tt=20n)
.model DZ D(Is=0.6u Rs=.5 Cjo=150p nbv=5 bv=12 Ibv=1m)
.model NP NPN(BF=125 Cje=.5p Cjc=.5p Rb=500)
.model PN PNP(BF=125 Cje=.5p Cjc=.5p Rb=500)
.ends LT1166
 
Re: Re: LT1166

syn08 said:
I have asked you and many other people before, WHAT IS YOUR EXACT PROBLEM WITH THIS IC. Everybody, including yourself, came up with "seems like" or "I don't remember, but I had problems", type of answers.
[snip]
First, this IC is useless for a D2S OPS.
Second, a traditional (and well trimmed) bias generator gives better results (i.e. less distortion and nicer step response) in a class-AB MOSFET amp. Besides, preventing completely turnoff of the OP-devices, as this IC does, is pointless in case off MOSFETs, as they don't exhibit the 'hole storage' effect.

PS: And there are stability issues too!
 
Re: Re: Re: LT1166

Edmond Stuart said:

First, this IC is useless for a D2S OPS.
Second, a traditional (and well trimmed) bias generator gives better results (i.e. less distortion and nicer step response) in a class-AB MOSFET amp. Besides, preventing completely turnoff of the OP-devices, as this IC does, is pointless in case off MOSFETs, as they don't exhibit the 'hole storage' effect.

PS: And there are stability issues too!


Maybe I'm slow today, but remind me again what a D2S output stage is?

Thanks,
Bob
 
Re: Re: Re: LT1166

Edmond Stuart said:

First, this IC is useless for a D2S OPS.
Second, a traditional (and well trimmed) bias generator gives better results (i.e. less distortion and nicer step response) in a class-AB MOSFET amp. Besides, preventing completely turnoff of the OP-devices, as this IC does, is pointless in case off MOSFETs, as they don't exhibit the 'hole storage' effect.

PS: And there are stability issues too!

Agreed with the D2S. Now, please define "less distortion" and "nicer step response".

On the same note, I'm sure one can build for $50 in parts and pcb space a discrete circuit that has 0.1ppm distortion, better than a $2 lme49710 opamp at 0.7ppm.
 
Re: Re: Re: Re: Re: Re: LT1166

Bob Cordell said:



Why don't you just tell me what it is? I would certainly extend you the same courtesy. Life is too short for this sort of game-playing.

Bob


"Difference of 2 Squares". A concept developed (for audio) by (as far as I know) Ian Hegglun. An interesting concept but not without it's own (mostly practical) issues. Anyway, it simulates great with perfectly matched parts!

Download this, it probably won't stay up for long.

http://perso.orange.fr/francis.audio2/AmpHegglun.doc
 
Re: Re: Re: Re: Re: Re: Re: LT1166

syn08 said:



"Difference of 2 Squares". A concept developed (for audio) by (as far as I know) Ian Hegglun. An interesting concept but not without it's own (mostly practical) issues. Anyway, it simulates great with perfectly matched parts!

Download this, it probably won't stay up for long.

http://perso.orange.fr/francis.audio2/AmpHegglun.doc


Thanks!

Bob
 
Re: Re: Re: Re: Re: Re: LT1166

Bob Cordell said:
Why don't you just tell me what it is? I would certainly extend you the same courtesy. Life is too short for this sort of game-playing.

Bob

It wasn't my intention to play games. Actually, I thought that it was you who was playing games, as I couldn't believe that you never heard of D2S and did not follow the interesting discussion on the Blowtorch thread. If my assumptions are wrong, I apologize.
Anyhow, here: http://www.diyaudio.com/forums/showthread.php?postid=1477051#post1477051 you can find a (very short) explanation of D2S.
 
Re: Re: Re: Re: Re: Re: LT1166

syn08 said:
How much less? Less compared to what? How did you estimate the distortion contribution of the LT1166?

Less ringing or overshoot compared to what?

Again RTFP, in particular mine. 😀
As I said before, WRT an OPS with a traditional (and well trimmed) bias generator.

Regarding the distortion contribution of the LT1166?, that was zero as I used ideal components for the simulation. This means that the bias generator of a real LT1166 will produce even worse results.

BTW, you said: "it measures much better than it simulates". Due to sloppy MOSFET models?