[ygg-it]

My output stage amplifier originally was equipped with two Ad826 in parallel (so 4 stages in parallel)
 
Now I did some tests:
 
1) Original AD826//AD826
2) LT1364//LT1364
3) AD826//LT1364

 
Why do you think that my amplifier now sounds better in the third configuration?

 
Do you think because you may "mix and balance" the tonality performance of each single amplifier ?
 
Or may be you "cancel" each others fatiguing or noising harmonics?
The only good things I found on the web is that someone else already did...

 
http://www.barberelectronics.com/parallelopamps.htm
 
(but I didn't so brutal soldering: I inserted a 47 ohm at the output of each parallel stage...)
 
Is there any literature that may technically explain this empiric terrific results?

 

[amb]

Paralleling opamps without output load-sharing resistors is a bad idea -- the opamps may "fight" each other.

 

[ygg-it]

I already put the external resistor, as I said.

 
I'm asking why two different amplifiers sound better than two identical opamps.
 
Probably because I put the fastest one (LT1364) in the "slowest" loop. This is the only reason I can see....
 
 
http://www.intersil.com/data/an/an1111.pdf
 
A1 = AD826 (350 V/ms Slew Rate)
A2 = LT1364 (1000 V/ms Slew Rate)
 
"At higher speeds, the output
of A1 will drive the load significantly before A2 and the output
drive will never reach twice the current."
 

 

[jcx]

even with load sharing R it still doesn't work for multiple op amps to share the same feedback - op amps have huge open loop gain and vary in gain and input V offset by too much for them work by literally paralleling  the pins
 
each op amp must have its own feedback to control the open loop gain
 
only buffer amps with internal unity gain feedback can sometimes be paralleled
 
I would expect that if you looked a the op amps output pins you would see only one actually moving with the signal and the other(s) saturated - stuck near either the positive or negative supply V

 

[ygg-it]

I'm not parallelling the pins... I'm using config in figure 1
 
http://www.intersil.com/data/an/an1111.pdf
 

 

[jcx]

the link in your 1st post shows the "wrong" way
 
fig 1 in the intersil link is known here as the "A47" cirucit from the Headwize article and works, fig 2 in that app note has problems
 
I have showed that you can Class A bias op amp outputs with a variation of the paralllel "master/slave", output current copying scheme
 
http://www.head-fi.org/forum/thread/188758/ad8397-class-a/15#post_2264079
 
the 47 Ohm current sense/share resistor values can be less - which gives better stability with cable capacitance loading the output
 
I also point out a way around the op amp common mode input limits of many op amps using the "A47" scheme in my posts in the above linked thread

 

[ygg-it]

Ok, thanks very interesting.
 
But what about using different op-amp ?
 
(also in your scheme...)

 

[jcx]

I build composit circuits related to Jung's Multiloop - one op amp is chosen for input characterisics and controls the outer feedback loop around a higher output current, faster op amp which drives the 'phones
 
 
you could unbalance the load sharing resistors in the "A47" scheme to get more current from the "slave" op amp and use one with better input for the "master"
 
for Class A push-pull output bias in my posts the two op amp's output current capability should be better matched - each op amp supplies the full current at the peak of one half cycle
 
you don't have to have bias Class A for the full range - say for only the 10-20% of the range near 0, then both op amp's output currents add near the peaks - this would be a "heavy" Class AB bias in accordance with Nelson Pass' First Watt site's opening comment:
"Dick Olsher famously remarked that “The first watt is the most important watt."
 
if you don't want the Class A push-pull, rather than using the parallel scheme I'd just pick a output op amp with enough current drive on its own - like the TPA6120, LT1210 or AD8397