[proid]

  Which op amp are you using? 
 
Did you take the advice (given everywhere, actually) to run the amp with a cheap but known good op amp during the initial setup? 

 
I don't have any spare cheap opamp so i just use the 1641 and it's no so expansive so it's fine for me. Today, i power on the amp by 2 9v battery and get dc at output about 6V and i measure the voltage at output of opamp, it's 8.6v (the same with power supply), is that ok?
Edit: I have just disconnect the voltage gain stage and just power output stage and still get 6.8V dc at output

 

[Avro_Arrow]

Ideally, the DC output of the diamond buffer will be close to zero
even when not in a feedback loop.
Remove the op amp and try the buffer on it's own.
Just connect the input of the buffer to ground so it's not
just floating.
Measure all the voltage points. They should be pretty close
from one side of the circuit to the other.
If you post the schematic with the values you used, I can
put it in the simulator and give you approximate operating
voltages to look for.

 

[Avro_Arrow]

 
You can use this diagram and just tell me what the values are.

 

 

[proid]

   
You can use this diagram and just tell me what the values are.
 
 

 
Thanks for your help, Arrow. I found out the problem: the output transistor is flying (not properly solder in the board) so it doesn't work, now both channel is fine with dc offset < 20mV. 
Then i connect the voltage gain stage to output stage and power amp on and once again, i get 7V dc at output so problem must be from the gain stage. I use all value exactly like the original ppav2

 

[Avro_Arrow]

 
Sounds like it's time for some new op amps...

 

[proid]

Finally finished, all problem come from not proper wiring. Right now i bias buffer at 20ma, opamp at 2ma and well, it sound quite good but not as good as my expectation, i always want to try a discrete diamond buffer and i think this PPAv2 must be superior to my O2 but its not, it's nearly the same

 

[Avro_Arrow]

 
Try a higher bias on the buffer...as much as your heat sinks can stand.

60mA would be a good target.

 

[proid]

   
Try a higher bias on the buffer...as much as your heat sinks can stand.
 
60mA would be a good target.

 
But Tangent stated that go over 30ma won't get audible improvement? I think may be i should try the original 3 channel design.

 

[tomb]

   
But Tangent stated that go over 30ma won't get audible improvement? I think may be i should try the original 3 channel design.

 
Yes.  That's a common comment, but 20ma is probably not even biasing them into Class A.  At least in the BJT diamond buffers I've built, the output transistors probably measure about 12-15ma (~25mV) with no bias current at all.

 
 
EDIT: You may be leaving no headroom at all - a rudimentary headphone load may throw the buffer out of Class A, causing the transistors to switch on and off, with its perceived noise/lack of clean signal/etc.