Head-Fi.org › Forums › Misc.-Category Forums › DIY (Do-It-Yourself) Discussions › doobooloo's TPA6120-based balanced headphone amp
New Posts  All Forums:Forum Nav:

doobooloo's TPA6120-based balanced headphone amp - Page 2

post #16 of 199
6/22/06 at 10:49am
Well worth a read:

http://www.tentlabs.com/Info/Article...decoupling.pdf

Specifically applicable for digital designs, but the principles still remain
Reply
post #17 of 199
6/23/06 at 1:13am
another good one for my library thanks.
Reply
post #18 of 199
6/23/06 at 9:05am
here's an idea.

The OPA1632 differential opamp can be viewed as two inverting voltage feedback opamps. What about using a TPA6120a2 in its feedback loops ? You get differential or single ended inputs, both section of the TPA6120a2 work now non inverting, you get SUSY (see the X-Gainclone thread on diyaudio), you can fit the 10R resistors inside a feedback loop and reduce their impact.
Reply
post #19 of 199
6/26/06 at 5:24pm
Thread Starter 

UPDATE!!!

First, thanks again for all your replies!

Over the past few days and all this weekend I've been working on porting this design from Eagle Light to Pad2Pad. Two reasons for this move were:

1. Pad2Pad is free and has no limitations of Eagle Lite
2. Instant price quotes and reasonable prices

Anyway, with this I've lengthened the board by 1" - for a total board size of 2.51" x 5". This still fits the smallest Lansing Micropak cases.

Here is a preview of what's cooking:



Click here for a high-resolution picture.

So... as you can see it is very incomplete with all the wrong part names. I'll get those fixed and post a BOM with all the part values and numbers soon, I'm working on it right now as I move towards completing the board layout first. A few changes from the previous version are:

1. Isolated analog and digital grounds with a ferrite underneath the DS1802 chip
2. "Separated" power supplies - a TLE2426 for the volume control, and a TLE2426 followed by a BUF634 for the amp sections. Plenty of room for capacitance both prior and after the rail splitters. This allows for the use of any single-pole power supply, and allows for a more compact standard 2.1mm/2.5mm power jack onboard. Actually, the height restriction of the large DIN-5 bipolar power jack was the big issue (the Micropak case would not have acommodated the jack), with this solution hopefully a better "channel separation" between the two inverting and non-inverting sections will be achieved.

Why did I not go for some onboard regulation? That way I am more limited with regards to power supply choice - I was thinking of a +/-10V virtual ground with two positive and negative regulators as well but realized that this would limit the possibility of using a battery pack for portable use or other lower voltage power supplies.

Anyway... I also added a power LED between the output jacks, and am planning to add a 1/8" input jack between the power and RCA jacks. Oh, and I replaced the crappy RCA jacks from before and put in a much nicer dual RCA jack.

In defense of all that excess capacitor space... I am planning to use a slimline case so the max capacitor height is going to be around 15mm. The diameter is 10mm. Hence, especially with boutique caps, I won't be able to get much capacitance in there without having so many - but if anyone feels strongly about reducing the rail capacitance and putting something more interesting in there (I was thinking onboard USB DAC or just cutting back board size to 4" length)... please let me know!

Anyway, at this size the board is expected to cost about $8~$9 when I order 20. Not too shabby.

Any criticism or suggestions welcome!
Reply
post #20 of 199
6/26/06 at 5:29pm
Thread Starter 
Oh, forgot to add:

1. The groundplane is now larger for each channel, extending all the way to the rail capacitors.

2. All SMD components are now 1206 size, before they were 0603. The output resistors for the TPA6120s have also been changed to SMD 1206 resistors.

3. Changed the voltage reference diode package to the more common TO-92, allowing the use of a greater selection of diodes.

4. All traces are now 20mils, except for a few short sections (between IC pads, for example) where they are 10mils.

5. The jacks have been laid out for a 3mm front panel and 1.5mm rear panel. I am planning to use FPE for the panels.
Reply
post #21 of 199
6/26/06 at 11:21pm
a couple things i want to ask:

1: why not dump the "multiple" tle 2426's im favor of 1. the virtual ground is buffered seperately for each chip, so why not? the possibility of one of the railsplitters "drifting" at some time is eliminated, and if nothing else the parts count would drop.

2: any thoughts on the possibility of a switch for "turning off" the "second" tpa chip? this would save a little power when used in single ended mode. probably just a pipe dream though.

added:
any thoughts on thickening the power suply traces where possible? again, it may not be necesary, but it probbaly wont hurt.
Reply
post #22 of 199
6/27/06 at 1:09am
Quote:
Originally Posted by doobooloo
Two dual opamps take in signal, acts as buffer into TPA6120 providing low impedance output. One TPA6120 is set at positive gain and the other at negative gain. Gain magnitude should both be equal, of course, just inversed.
Congratulations! you might want to know that you have found all of the ways I can readily imagine to give yourself the worst possible power supply distribution arrangement

in a balanced amp there is no external driver gnd connection, the +/- drive currents are equal and opposite in one channel (R or L), this strongly suggests putting the +/- polarity amplifiers for one channel in the same dual op amp chip - one chip for R and the other for Left

since the +/- ps currents can be equal and opposite the least external magnetic field coupling can be had by putting the +/- supply traces as close together as possible – you have done the opposite by putting them as far apart as you can, this gives large loop areas for the rectified drive current (when operating Class B at higher levels) in each supply to spray this nasty nonlinear magnetic field over all of the signal traces in your layout

By putting both + channels in one dual op amp and both – in the other you actually force the output currents for both channels to complete their loops through your gnd traces to make up the current balance otherwise inherent in balanced drive

By forcing output current to unnecessarily complete through the gnd you are using many large caps between the +/- rails and ground, putting them in series relative to the traces that should be delivering the drive current, reducing the cap’s effectiveness by half

By forcing +/- supply currents to complete through a pair of buffered active gnds that aren’t even connected together you have both driver’s currents flowing through your gnd buffers – adding cost, complexity, and potentially distortion

More comments in this general application are in another tpa layout thread:

http://www6.head-fi.org/forums/showthread.php?t=127473

Since you expressed a need to avoid boredom you will I hope welcome this opportunity to ripup your layout and start over
Reply
post #23 of 199
6/27/06 at 12:13pm
Thread Starter 
Thanks for your comments, jcx!

First, regarding power supply line arrangement... what about something along the lines of this...



And with only one buffer driver and shared ground?

Also, regarding this comment...

Quote:
in a balanced amp there is no external driver gnd connection, the +/- drive currents are equal and opposite in one channel (R or L), this strongly suggests putting the +/- polarity amplifiers for one channel in the same dual op amp chip - one chip for R and the other for Left

By putting both + channels in one dual op amp and both – in the other you actually force the output currents for both channels to complete their loops through your gnd traces to make up the current balance otherwise inherent in balanced drive
The only way to do this in my design (single ended -> balanced) as far as I can see is to add an additional buffer to provide a low-impedance output to two pairs of dual opamps, one inverting and the other not, and then switch the signals around so that each TPA gets either both +s or both -s as you mention. So, there's a tradeoff - of adding an extra opamp in the signal path, but having the TPAs not "fight" for output current.

On the flip side, in that new configuration, if I ever want to use the amp in single-ended mode, the two chips would be fighting for current in that configuration instead, right?

Quote:
By forcing +/- supply currents to complete through a pair of buffered active gnds that aren’t even connected together you have both driver’s currents flowing through your gnd buffers – adding cost, complexity, and potentially distortion
So, if I use the new layout above using a single buffered active ground for both, this problem is (at least slightly) alleviated?

Also thanks for the link - I've read through the thread once but I'm not quite sure I get everything you and peranders talk about. I'll have to read it through a few more times and study the board layouts and see what more I can do to improve mine. Meanwhile, please feel free to shoot down whatever I'm doing as much as possible!
Reply
post #24 of 199
6/27/06 at 4:00pm
Thread Starter 
I've also been creating a groundplane (shared between the two "channels") with polygons - rather painstaking job but it's coming together rather nice. Also I'm going to dramatically reduce the number of caps and make the board shorter again to 2.51" x 4".

I'll post some pics later today when I've cleaned the layout up a bit more.

Any more suggestions/criticism welcome in the meanwhile...
Reply
post #25 of 199
6/27/06 at 8:00pm
Hi, you could also consider the following app-notes as supplied with the
TPA6120A2 EVM board (slou169.pdf) fromTI.

Please note their concerns and solutions regarding the ground plane's proximity
to the input pins (The input is very sensitive to stray capacitance )

Interestingly enough they opted for a ferrite bead in the supply line's
330R @ 100MHz.

Hope you get it working
Reply
post #26 of 199
6/27/06 at 10:56pm
If you want to build something really good it is better to be very specific about its purpose, balanced output can be a good approach to some problems, single ended is OK too but trying to make the same hardware do both will likely compromise the performance. Likewise you say you want to use a dual ps and a single as a secondary option – if you really intend to build a dedicated supply I would go with a dual supply only, have a “real“ ground and not play with active ground buffers at all, this choice makes the 1st option of balanced/single ended operation far easier

A slight twist that may let you have some options in one layout is to use stuffing options – 2 different circuits can be built up on the common pcb, but you can only have one choice for each assembled pcb – only practical when most of the circuit functions are common

But in any approach it really is helpful to use the schematic drawing function and keep circuit and layout in sync and avoid gross errors of miswiring – and of course keep straight what it is you are trying to build as well as communicate with others

The ps wiring in the previous post is much better if you use balanced load, putting the +/- from one channel in the same dual is still likely to be a little better. But if you really want to have a good layout for single ended load too you have to put gnd plane under the ps traces and get much lower Z gnd out to the output jack – even layout is influenced by operating mode

I can't imagine having the extra op amps you are using to buffer the digital pot and not using them to make multiloop amps servoing the tpa's - even a simple multiloop with a +2 gain tpa greatly reduces the distortion compared to your use of the buffers to drive the low Z of the inverting tpa input R

Try reading Walt’s chapter on audio amps

http://www.analog.com/library/analog...h6_final_I.pdf

pdf page 50 + 6.48 in book, heavy duty current boosted multiloop stuff starts ~ 6.60
Reply
post #27 of 199
6/28/06 at 2:35pm
Thread Starter 

UPDATE!

Thanks again for all your interest!

Quote:
If you want to build something really good it is better to be very specific about its purpose
Well - to get one thing straight... I'm not aiming to create the ultimate high performance balanced amp here. I'm trying to create a very small amp capable of driving both single ended and balanced headphones from a single ended source and standard 24V wallwart power supply or battery pack. I wouldn't really mind sacrificing some of the design "optimality" to get the above... Of course, a clean true biploar power supply would be better, but if I can get away with a buffered active ground then that's good enough for me. Running two TPA chips with one doing + and the other - may not be optimal for running balanced phones but if the audible differences are merely subtle compared to both +/- on one chip then I really wouldn't mind.

Size and convenience are really the key drivers here, I guess.

The thing is - I have a prototype of this rough design on a perfboard with all through-hole components and the TPA6120 mounted on adapters (gasp!) and in sockets (gasp!!!). Grounding is a horror, it's all over the place. I'll be the first one to say that it's a horrible way to make use of these fine chips, but guess what - they sound great even so. So, I am sure that doing a slightly more optimized PCB layout is only going to improve things and I'd be very happy even if it's not 100% optimal.

At the same time - that said, I'd definately like to see the last drop of performance squeezed out of this configuration given the design criteria limitations.

Quote:
But in any approach it really is helpful to use the schematic drawing function and keep circuit and layout in sync and avoid gross errors of miswiring – and of course keep straight what it is you are trying to build as well as communicate with others
I would love to draw a schematic and link it to the board layout but the Pad2Pad program has no such function. I'll at least re-label all the parts once the design is more or less solid. Maybe once I think things are done I'll try to draw the schematic by hand to make sure I've got everything correct.

Meanwhile then... any good simple free schematic drawing programs out there?

Anyway, the newest revision...

All layers:


Top only:


Bottom only:


So yeah... I drew that groundplane by hand by stitching numerous polygons... Hopefully it's worth that effort. Groundplane clearance is 20 mils. Since the DS1802 chip groundplane is not the -2.5V ground but the same 0V virtual ground set up by the TLE.

Some changes:

1. Obviously, the groundplane. Of course it's not there under the input pins to minimize capacitance as recommended per skilpad and TI.

2. Power supply. Traces put close together between the two "channels" and using only one active buffer for both. Capacitance post-buffer has been reduced to four caps, but cap diameter has been increased to 12.5mm. The DS1802 uses a single TLE2426 and four 6.3mm caps after the virtual ground. There are four 10mm caps before both virtual grounds and the power jack. I forgot to put a protection diode - I'll do so in a later revision.

3. I can't fit a 3.5mm jack at the rear anymore! I can if I remove one of the rear 10mm caps but... but...)

Is there anything I'm still doing horribly wrong? I'm sure the amp will "work" and produce sound fine but given the design limitations above I would like to know if there's anything else I can do to improve performance. Thanks again in advance!
Reply
post #28 of 199
6/28/06 at 3:48pm
Thread Starter 
Quote:
Originally Posted by skilpad
Interestingly enough they opted for a ferrite bead in the supply line's
330R @ 100MHz.
I noticed that too - do you think it's a good idea on this board as well?
Reply
post #29 of 199
6/28/06 at 5:55pm
Maybe,


You should however place it as close to the DC input connector as possible
And it should be the first series component after the DC input proper.
(Please check total current consumption + peak current of total loads so that
the bead don’t saturate, or current starve the circuit)

A potentially better solution would be to use an external Clamp-on Ferrite tube
over the entire DC cable just before it enters the DC input connector
This way people could use it, or not and experiment with different types, and
Amount of turns around the ferrite tube.

Still on the filter subject, one could use a series resistor Say 10 to 50 Ohm in the positive supply to the
Volume control circuit between the closest big capacitor and the rest of them feeding the TPA’s

The idea is to isolate the capacitors decoupling the volume control circuit from the current peaks
Drawn by the TPA’s

All of this is of course to extract the very last bit of performance and should be viewed more
As academic than absolute requirements.
Reply
post #30 of 199
6/29/06 at 1:23pm
I still advocate jumpers, by using twisted or star quad magnet wire for the horizontal run to the tpa's power you can have a nearly solid ground plane, the V+, V- currents in a tightly coupled twisted jumper radiate even less than the present pcb trace, when the jumper is taped down to the backside ground plane it is closer than traces running over solid gnd plane too (a few mils of varnish on magnet wire vs 0.062" of FR4)

I've also put together this LtSpice schematic/sim of the lowest risk multiloop approach that works in your amp - no extra op amps, only a couple of extra resistors in the outer feedback loops



U2,4 are just your buffers recycled to improve amp performance, there is still high Z at U2 input for the pot (LT1022 is a fet op amp input similar to opa134, ths6012 is the industrial version of TPA6120, the LT1022 model is included in free LtSpice Sw CAD III http://www.linear.com/company/software.jsp )

some people object to inverting the amplified + gain circuit's output to get the - side drive on "unequal delay/gain" grounds, complete nonsense for audio frequencies with these op amps - the improved quality of the multiloop and convenience of the circuit implementation more than make up for the "error":

in the sim the +r,-r phase difference is 179.833 degrees => 0.167 degrees error from the ideal 180, this is at 20 KHz !, actual "phase error" across the headphone is 1/2 that value - how much and how reproducable do you think dynamic headphone's phase is at this frequency? (if you've got hair you can't even put the same pair of headphones on twice with this phase accuracy @ 20KHz)

if you use the 1st inverting amp topology in my diyAudio thread:
http://www.diyaudio.com/forums/showt...806#post512806
the +r,-r phase diff reads 179.999 in the 20 KHz sim

its not just Walt pushing multiloops:
http://focus.ti.com/lit/an/sboa002/sboa002.pdf

[rename the attachment, stripping the .txt to run in LtSpice]
Reply
New Posts  All Forums:Forum Nav:
  Return Home
Head-Fi.org › Forums › Misc.-Category Forums › DIY (Do-It-Yourself) Discussions › doobooloo's TPA6120-based balanced headphone amp