My attempt to design of high performance composite headphone amplifier

[solderdude]

When a source has an output capacitor of say 2.2uF the -3dB point will shift from 7.2Hz (10k) to 14Hz (5k) for instance when the input R varies.
Of course DC coupled sources or ones with much higher coupling capacitors won't be affected.
Bye bye (±0.1dB): 12Hz – >80kHz in this case.
The amp may well have that response but in practice (depending on the source) the end result may not be.

That was my point regarding the varying (10k at min vol setting and 5k at max vol setting in this particular design) input resistance.
I agree that sources should not have problems driving 10k loads and one could say that if the source has a 2.2uF or 3.3uF output capacitor it could be considered a poor design.

I too would put a large non-polar cap in front of the volpot. would change C17 and C18 to a resistor (probably end up somewhere between 100 Ohm and 1k) and R4 and R12 be a 100k or 1MOhm resistor.
Maybe noise levels would be slightly higher but any DC on the input would not result in a scratchy sound during volume adjustment.

It makes me wonder why most designs using a TPA6120 all adhere to at least 10 Ohm series resistance (or even higher) and not simply use an inductor in parallel. I have no practical experience with the TPA though.

 

[KSTR]

It makes me wonder why most designs using a TPA6120 all adhere to at least 10 Ohm series resistance (or even higher) and not simply use an inductor in parallel.

It needs careful selection of both the inductor and the resistor, plus assuming a reasonable range of load capacitance. With large load capacitance there will be ringing in the step response when the L//R still is on its inductive slope. Also, the shunting stray capacitance of the inductor is critical, it must have a self-resonant frequency way above the highest frequency of interest which easily is around 10MHz.

 

[theatrus]

Whats the reasoning behind rejecting the LME49600 buffers? I have a personal design with them which listens well (though I have not measured it).

 

[Bamboszek]

@theatrus
You can get two TPA6120A2 in price of one LME49600, and first one is stereo where LME is single channel.
Great results can be achieved with LME49600, as JDS Labs Atom or The Wire shows. But TPA6120A2 is both cheaper and potentially better performing. Out of loop distortion of LME49600 is pretty bad while TPA6120A2 is very good even standalone. That means feedback loop has easier work. That may (but not necessarily) lead to better performance in composite amplifier.

 

[Bamboszek]

@Dave Zan suggested me some interesting stability simulation method with so called Tian Probe. Technical details could be found here and complete step by step guide for multiloop feedback analysis could be found here.
If I'm right my amplifier should have 39 dB of gain margin and 51° phase margin with 300 Ω || 10 nF load (Figure 2). With 300 Ω || 1 nF load PM is 46° and GM is infinite (Figure 3). Last graph presents Bode plot with closed loop (Figure 4).