[porthillsbomber]

Well that is interesting. Could you post a schematic of your current set up?

 

[Armaegis]

Huh weird, I tried setting it up again just now and it doesn't work... I'm in a rush though, so I didn't get a chance to double check everything.
 
I did try it with your method plus the extra resistors and it works really well though.
 
I did finally figure out the math tonight though (I think). Something I noticed in your calculations (although perhaps you intended it this way), is that your OG component within the OL/OR circuit never changes despite the gain.

 

[Armaegis]

Argh, almost bedtime but I finally got a method that works which doesn't feed OG back into any opamps. I will try to post up a schematic in the next day or so.
 
OG: generated the same as always
OL: feed IL directly into + (resistor not necessary*), feed IR through a resistor into -, feedback with resistor into -
OR: as above but flipped
 
- to meet Meier's electric field conditions for 4 wires, set input resistors equal to twice the feedback resistors
- the OL/OR are essentially simplified differential circuits (derived from wiki), and once you set Rg arbitrarily high then you find R2 doesn't matter anymore.
 
*not necessary, but I might experiment with resistors later to see if this reduces (or increases) crosstalk... (it might just be distortion I'm hearing... not sure), and I like the idea that all input resistors are the same for the sake of symmetry

 

[porthillsbomber]

OG should change with gain. OG = G*(IR + IL)/4, gain G.
 
I think what your are describing is an inverting opamp with a IL ground reference so
 
OL = IL -G*(IR-IL) where G= Rf/Rin. So you want G = 1/2 giving
 
OL = 3/2*IL - 1/2*IR so you would need
 
OG = -1/2*(IL+IR)
 
which gives an overall gain of 2
 
OL - OG = 2*IL

 

[Armaegis]

Wait, we might be talking about different things again... I was talking about the differential circuit you used to generate your OL and OR. Let's use the OL circuit as an example. Your output OL is determined by supositioning the three cases where we take each input separately and assume the others are zero.
 
OL = IL*[R7/(R5+R7)][(R6+R4)/R4] + OG*[R5/(R5+R7)][(R6+R4)/R4] - zero[R6/R4]    (the last component is ground, but if it were non-zero then it would be added in)
 
If R4=R5 and R6=R7, then OL = IL*R6/R4 + OG
so your IL has gain, but the OG component remains the same
 
I did a simulation through TinaTI which confirmed my formulas.
 
I'm doing this from memory and it's 4am, apologies if I goofed... plus it looks all screwed up written in a long line like that. I can barely read it right now and might now have written it improperly.

 

[Armaegis]

Real quick image post as I'm heading out the door... (ignore the opamp listed; it was just the default one selected in the program)
 

 
I've found that using input buffers is absolutely necessary in reducing crosstalk (unless I'm feeding from a lineout). I've also tried putting extra resistors in the path of the + for the left and right channels which seems to help a bit.

 

[Armaegis]

The most noticeable difference tapping headphones directly off the circuit is that the bass end becomes much more powerful. Not "louder" per se, but it gains increased presence and definition. Unfortunately, it also starts to clip/distort, likely because I have no buffers or anything else in there. I can't want to finish my b22 and/or a20 and plug everything together.
 
On a tangent, would it be possible to hack the M^3 to emulate the active balanced ground circuit? How much overkill would it be to then use that to feed a b22?

 

[porthillsbomber]

Ok yes that circuit is the one I was trying to describe in my last post. I'm not sure on what adding the resistor on + does but I wonder if putting one there corrects DC offset?
 
In theory it would be possible to hack a M^3 to do ABG, probably won't be too hard either. I'm pretty sure that would be overkill and maybe even worse than if you just used the same opamps, the mosfet output stage isn't needed.

 

[Armaegis]

The audible difference from putting the resistors into the + path could have been completely imaginary. I figured it was simply due to symmetry; so the signal from my input buffers which gets split three ways all push into the same value resistors. I don't really know if this has any effect going into opamps though, so I'm taking that with a grain of salt.
 
My OL and OG circuits are basically differential circuits but with infinite Rg, at which point the + resistor can go to zero...  I've tried it with the + resistor at 4k and Rg = 1M and I think it gets a little bit grainier; ideally I would use a bigger value of Rg or smaller + resistor but I'm working with what I've got. I think I read somewhere that having the Rg there helps correct for input bias, but then again if you have any sort of noise in the ground then that will probably show up...
 
I've got parts for a 3-channel a20 amp incoming right now. I need to order some more opamps too so I can actually use the same opamp for all three channels (right now I have a different one creating my OG, which I suspect will cause a bit of distortion due to slightly different response). I plan on testing the ABG circuit plugged into the a20 amp and fiddling with configurations and resistor values until I find the best sound, then I will finish building the b22. I might even go really crazy and build two b22's, one with the ABG circuit and one without (then sell one, because I've got bills to pay

).
 
I wish I could somehow test the outputs with RMAA, but I would need a dummy head and microphones... and that's a couple steps more than I'm willing to go for this venture which is really just a proof of concept more than anything.
 
The crazy idea with the M3 was to build it as some sort of modular unit that could be used independently (like a transportable unit), or for home dock it into the beefier b22 and and feed it the ABG signals (and power the M3 with the 4th output from the s22). Did I mention the crazy? Mwuhahah...

 

[Armaegis]

Some more experimenting... using both sides of my dual opamps (AD8656):
- use one opamp to generate OL and OR
- use one side of the second opamp to generate OG, then use the other side in line as a voltage follower to double current
 
I think I've almost got it. Cancellation is excellent, although the last bit that I can hear is grainy. If I fiddle a few things here and there I can improve cancellation but increase the distortion, or vice versa. Hmm... I'm going to chalk up some of it to crosstalk in the breadboard and opamps and improperly matched resistors, but I'll have to keep experimenting.

 

[Armaegis]

Also, adding resistors between the OL/OR and headphone taps is still a significant factor to reducing crosstalk. Using 120 was good, using 400 is much better and to my ears about the same as plugging headphones directly into the lineout of my iBasso D10 (regarding crosstalk only; the volumes and performance will change obviously). I will try adding some output buffers later.

 

[Armaegis]

I got a sample set of 5 RC4560 dual channel opamps.
 
#1: input buffer, left signal from my DAC goes to both sides of the chip to generate two outputs: IL1 and IL2
#2: as above, using right... generates IR1 and IR2
#3: use one side only, generate OG using IL1 and IR1
#4: use one side only, generate OL using IL2 and IR2
#5: use one side only, generate OR using IL2 and IR2
 
The way I figure, the OG circuit connects IL1 and IR1 together directly, so those should be isolated from the others. OL and OR share the set of IL2 and IR2 and those two never touch directly (although they do touch a bit through the feedback loop). Does it make an audible difference in practice? I think so... but it's such a minute difference that it could be placebo.
 
Anyhow, now that I'm using the same chip for everything my distortion is gone and my crosstalk is way down. I'm quite happy with this. I need to try this with some single channel opamps (sigh, more money).
 
Now I just have to wait for my transformers to arrive and I can build my a20 and see how it all goes... *knock on wood*
 
I guess I should get some perf/protoboard since I'm sure the breadboard I'm using is another big source of noise.

 

[porthillsbomber]

You are absolutely correct about the separate input buffers. It is crucial to stop the cross talk related to a non zero source impedance don't know why I didn't think of it. In theory you actually only need the buffer on the ground channel inputs but it can't hurt on the others.

 

[Armaegis]

I figure if you do it to one, you should do it to the others. In practice I don't think it makes a huge difference...
 
From some brief testing, in order of worse to best (in terms of crosstalk)
no input buffers, fed from headphone out on laptop
no input buffer, fed from headphone out on iBasso D10
no input bufer, fed from lineout on iBasso D10
input buffer, hp out from D10
input buffer, lineout from D10 (barely any difference from above)
dual input buffer, lineout from D10 (placebo difference from above)
 
I say once you're at the point of using lineout with an input buffer, any extra circuit shenanigans aren't worth the hassle. Keep it simple and spend the money on better matched components.
 
Then again, I haven't tried plugging this all into a full sized amp yet. Once I get stuff built and running I will repeat some of my experiments.

 

[Armaegis]

For my input buffers, would it make a difference if they were inverting vs non-inverting? I think I had read somewhere that inverting was more stable, so I've just been using that. Also, in my pseudo-science-placebo head I want as few things to touch the ground line as possible.
 
I'm also thinking that if I want to implement a variable gain switch somewhere, the input buffer stage would be the place to do it; just change the resistors in the feedback loops.