[kriskalish]

I've embarked on an amplifier design journey that is probably going to take me to something very similar to the pimeta. I'm currently thinking I want a 3 channel (active ground) design. One of the issues you seem to have to wrestle with is how to get your ground channel to sink the current sourcing of the other two channels combined. 
 
Buffering immediately comes to mind. The pimeta uses an IC buffer like the BUF634 or LMH6321, but the 634s are expensive and the LMH6321 is somewhat expensive and not available in DIP. Given that, it seems like building a buffer out of discrete components is a reasonable alternative. I've also decided that discrete components are kind of sexy.
 
So this leads me to my question: What are some good discrete buffer amplifier designs?
 
The Jung diamond buffer looks nice but has a lot of parts. Will I be able to get decent audio quality with fewer parts? I've stumbled across a few class A designs, but that's probably not doable even though I'm using 8xAAA batteries.
 
 
I also found something called the Monofied Sijosae Buffer here on the forums.
 

Is this a class AB push-pull buffer? It looks like the JFETs are set up to pull current through the first stage of BJT's, but not really enough to make the whole thing class A. Also, should the capacitor be ceramic?
 
 
I tried to simulate a circuit using this as the output buffer and my amplifier breaks every time I put the buffer inside my op amp's feedback loop like so:

 
I also tried putting a 100pF cap in the feedback loop across the opamps output and inverting pin like Jung's article on op amps suggests, but this only makes a minor improvement. If I bump it up to 1uF I get a sine wave out of my oscilloscope, but the gain is only two. Between the ridiculous cap value and wrong gain I think I'm doing something wrong.
 
 
Any advice for buffering my amp would be appreciated,
-Kris
 
 
 

 

[tomb]

This was Steinchen's original implementation for the Millett Hybrid that we used for the MAX/MiniMAX:
http://www.diamondstar.de/dDB_overview.html
A version of the diamond buffer was also used on the PPAV2.
 
There's also AMB's Jisbos buffers, but I'm not sure if he still has them on a website anywhere.
 
Peranders also sells some SMD-based diamond buffer boards and kits, I believe, known as the Sjöström Super Buffer.

 

[jcx]

I highly recommend against what is known here as "3-channel" gnd systems - the input, feedback and headphone driver gnd should be the same piece of copper, as "close" together electrically as possible - THE Star gnd point of your system
 
with single battery supply like a 9V you need a "supply splitter" and/or AC coupling - if using a battery stack then just use the midpoint - you can take other measures to "balance" discharge or prevent cell reversal
 
if you want DC coupling with a single battery or "wall wart" then an active supply splitter does need to sink/source the peak current of both channels driving the headphones - and incidentally double your current draw, cutting battery life in half
 
with a single active gnd supply splitter the quality of the active gnd can be lower with little consequence - its errors just looks like power supply rail noise/modulation to the R,L channel amps - and op amps have very high PSRR at audio frequency

 

[Avro_Arrow]

VG1 should be referenced to center point (gnd) not V-.
 

 

[kriskalish]

 
 
Quote:

This was Steinchen's original implementation for the Millett Hybrid that we used for the MAX/MiniMAX:

 
Thanks for the link! These designs looked like class A designs which I don't really have the juice to run. I noticed while following the links that Ti had run some benchmarks on the Millet Hybrid with the discrete buffer, the OPA551, and BUF634.  It looks like the diamond buffer didn't perform as well in the benchmarks until it was biased pretty hard. Very interesting... If I drop the "active-ground" thing it might be affordable to get two of these fancy ICs...
 
Quote:

I highly recommend against what is known here as "3-channel" gnd systems - the input, feedback and headphone driver gnd should be the same piece of copper, as "close" together electrically as possible - THE Star gnd point of your system
 
with single battery supply like a 9V you need a "supply splitter" and/or AC coupling - if using a battery stack then just use the midpoint - you can take other measures to "balance" discharge or prevent cell reversal

 
This seems really reasonable and pragmatic. If the battery protection measures are simple, this will dramatically cut my part count down and cost. Are there monolithic IC's out there that will give me automatic shutdown functionality if one side of my pack gets too low? Maybe I just need to Google harder.
 
 
Quote:

with a single active gnd supply splitter the quality of the active gnd can be lower with little consequence - its errors just looks like power supply rail noise/modulation to the R,L channel amps - and op amps have very high PSRR at audio frequency

 
This actually brings to mind another question. It seems like a discrete buffer will have pretty poor PSRR.  If I'm using batteries configured with a star ground configuration will it really matter if my buffer's PSRR is not very good? 
 
 
 
Quote:

VG1 should be referenced to center point (gnd) not V-.

Ah thanks.  I think TI-TINA must be "smart" and just know what to do because I made a simple test scenario with it connected incorrectly (as in my screenshot) and correctly as you said and it works both ways.  That's a good catch though.
 
I was able to fix my circuit in the simulator by replacting the JFET on the bottom with a P-channel one. This led me to realize I had put the resistor on the wrong side, so now I have to N-channel JFETs as the original schematic shows.

 

[jcx]

I don't unreservedly endorse every thing about it but you might look up the "O2" amp project site, associated design blog

 

[kriskalish]

I did some real reading on the merits of real ground. I found a number of sources that suggested it was the right way to go (not just the O2 site). I even encountered some of your older posts here on head-fi which were all pretty informative.
 
However, I still have a question about implementing a "star ground" on one of those protoboards with copper pads.
 
To what extent does every component need its own path to ground?  For example, if I have an opamp and a buffer per channel, each with its own bypass caps from the power supply rail to ground, should I go as far as run a decent guage wire for each cap to ground for a total of 8 wires? Or would it be acceptable to just run 4 wires by tying each the two caps on each component together then sending only one wire to ground from each opamp/buffer?
 
In this example I've assumed that it's already a good idea to give the feedback loops their own path to ground.
 
-Kris

 

[jcx]

star gnd is an ideal - the real world usually involves compromise
 
for headphone amps it can be practical to just use a gnd plane and ignore details - particularly if you use Class A output stage where all pwr, gnd currents are linear and any cross coupling from the shared current paths in the gnd plane only add tiny amounts of gain error and channel crosstalk
 
for Class AB and higher current loads the next practical level is to ask which "gnd" paths have large, nonlinear currents - call them "dirty gnd" and pay attention to connecting them to "clean gnd" - which may just have signal, feedback gnd, then connect the 2 at one "point"
 
the "dirty gnd" concept should be applied in a line powered supply with xfmr, rect and reservoir cap layout, connect with only one "branch" gnd wire from the +/- Cap gnd center point to the "clean" gnd plane
 
I really like pth clearance hole gnd plane PCB for prototyping, but have used perf gnd plane board that I chamfered the gnd plane away from the holes with component leads with a small drill bit twirled by hand
 
additional isolation, splits can be cut with xacto (safety glasses) and the copper peeled off the board, additional low Z pwr bus can be made from Kapton tape insulating Cu tape that you solder to