TPA6120A2 Headphone Amp Offset Problems

[cynric]

Hi!
I have recently designed and built a TPA6120A2 based headphone amp. It is designed for both unbalanced and balanced inputs, selectable through a switch. Another Opamp then does the balanced->unbalanced conversion.
I initially had DC problems when working in unbalanced mode with no source connected (as the datasheet/evm manual warned me of). However I did not have these extreme DC problems (250mv+) when using the amp in balanced mode (with the other OPamp before the TPA). So I made both inputs use the 2nd Opamp as a buffer by placing the switch in front of it. My volume pot is between the two opamps. Now I do not have the high DC offset when in unbalanced mode anymore, but unfortunately I still have a rather high, gain-dependent (and Pot-dependent) offset left.
I made the TPAs gain selectable with a jumper, so one can easily switch between 3.2 and 6.4. When working with a gain of 3.2 I have only little DC offset of mostly below 20mV. But when I turn my pot from 12 to about 2 o clock the offset quickly rises to ~50-60mV and lowers again to >20 once i pass 4 o clock. With the gain set to 6.4 all DC values double giving me a dangerous offset of max ~110mV.

As a workaraound for this I placed switchable Caps before the headphone output; this works but it is only a bad fix, since I wanted to keep the signal path cap-free :/

What I thought of now is increasing the gain of the buffer stage and locking the TPA gain at 3.2, but if there are any nicer solutions for this problem, I would be very thankful to read them
Thanks for all help!


Schematic:

[jcx]

for an indication of likely dc specs, see the THS6012 datasheet - I believe the TPA6120 is the same chip, talk to a TI application engineer to see if the TPA have better/worse DC specs due to sorting

1st I wouldn't have put the jumper gain select option in - the datasheets and eval board layouts are very emphatic about keeping parasitic C at a minimum on the -in node

the THS6012 DC specs are not great: 5 mVos, 10 uA +in bias could give the numbers you observe

since the +in bias current works into the varying R of the volume pot you cannot easily trim

2 trim pots per channel on the TPA (one injecting a bias canceling current into the +in through a high Ohm R and the other injecting a Vos into the -in would be my approach) and possibly another Vos trim for each input buffer would be necessary to completely trim output Vos - and then you would have to see if temp stability was adequate

I use the TPA in a multiloop arrangement where the higher DC spec outer loop op amp controls the Vos

[jcx]

[continued from above - 30 min edit limit is too short!]

also with Vos adj described above: if you run from both battery power and a different V wall adapter then the Vos adjust will vary, and again as the battery discharges the V varies - as much as 30% depending on chemistry


also your active ground splitter is very likely inadequate - the ground has to "soak" up as much current as the both channels of the TPA can put out when driving your heaviest load: remember all currents flow in closed loops, identify them, design and layout with them in mind and you will be way ahead in the signal quality department

the classic 3 term TLE2426 has very asymmetric and small current capability: less than 30 mA on one side as I recall - you would be much better off with those 4x 2200 uF caps AC bypassing the active ground - and isolating the TLE2426 from the lo Z AC ground with ~10 Ohm R relying on the bulk C between battery +,- terminals and the AC gnd to carry the load current (calculate smallest R that keeps TLE current <20 mA with AC gnd ripple of lowest audio frequency, lowest Ohm load, max amplitude output, both channels simultaneously)

[cynric]

thank you very much, jcx!
as for the offset i have decided to remove the gain jumpers - a gain of 3.2 is more than enough to drive all my headphones at very high levels - and accept the resulting volume-pot dependend offset of upto 55mV. Also I noticed that I can not hear any difference at all, when using the Caps at the output.

as for the ground circuity: I do understand the problem of the TLE2426; however I have been a little unsure about this.
I have read somewhere (cant find the link right now) that when using the TLE2426 in an headphone amp it would be better to place big Cs only between + and - and not 'behind' the TLE beacause of the then halved capacity. However I just found a very interesting site about all this (http://www.tangentsoft.net/elec/vgrounds.html) which suggests exactly the same thing as you. So I will now start with placing 2 of the caps between +/- and gnd (with my current board, i luckily can do this very easy).
I will go next week and measure my +/- levels at my university while driving a heavy load at high volume to see, if the TLE gets imbalanced at some point.

Thanks again!

[cynric]

just to keep this uptodate:
Bypassing the ground with 2 of the 2200µ caps lowered my distortion factor at low R loads by about 4 (at 0.0031 with 30R, <0.0012 above 150R), so this is really a plus! Thanks a lot for the tip, jcx! Additionally I placed two more TLE2426s into my circuit and placed them next to each other, giving me the possibility to tripple the possible GND current (just using one atm).

Also I tried building a DC servo for the offset, but I didn't quite succeed; I will continue working on this after my exams (~2 weeks). I built a 2nd order multiple feedback lowpass from out to +in, but yet I do a positive output voltage from a positive input (since multiple feedback inverts phase, there must be some bug in my circuit)... however, i will care about this in some time..

I did now recalculate resistor values to increase the buffer's gain and thus reduced the TPAs gain to 2, limiting my theoretical offset to a max. of ~60mV (real values should be much lower). I can change the resistors easily with my current prototype, which I will do once I have some spare time again.

Would it be possible to reduce the offset by inducing the 12µA via a large resistor (500k @ +-6V) from negative supply? I tried this at home (with an old, unconnected board) and it seemed to work - however I am not sure, if this has any negative effects...

These are my current circuit and layout; should anybody have any more comments or suggestions about them, please don't hesitate



Thanks!