OPA2227 + NE5532 Composite Headphone Amp
 

This is my first junkbox attempt at composite amps, and any feedback is greatly appreciated.
http://www.diyaudio.com/forums/attac...1&d=1320541196
Pot is intentionally placed before coupling caps so that if any DC develops at the source, I'd know that there is a fault.

OPA2227 was chosen for to its automatic input bias current cancellation, relatively low GBW, and good phase margin.

I happened to have enough NE5532 to make a poweramp, so naturally this is the buffer of choise. With 4x gain at the global feedback loop, it should slow the effective bandwidth at OPA2227 down to 2Mhz, although I am not certain whether a 2Mhz - 10Mhz difference is a good enough safety margin.

Likewise, I've considered parallel OPA2604 on the output, being 20Mhz GBW these should make a good candidate but at increased cost. Douglas Self wasn't too thrilled about these, but I have a few lying around and they shouldn't be too bad inside the feedback loop. If all else fails, OPA2227 will have to operate on local feedback.

Current sharing resistors are at 5 ohms, but these could probably go down to 1 ohm in practice.

Now a couple of questions.

Is a single 100 Ohm resistor be good enough at isolating the gain stage & buffers? I recall seeing a design utilizing isolating resistors before every buffer, another without but using a defined input impedance with a resistor to ground, or perhaps it wouldn't be necessary in the first place?

Another question would be towards the output network.

I picked 30R + 20nF zobel values based on headphone impedance, but couldn't find any specific guidelines on how "heavy" the high frequency load should look like, and 1uH || 5 Ohms was used to isolate the cable capacitance.

I believe there was a specific composite headphone amp sample circuit by Walt Jung that that had only 5uH || 50 ohms on the output without any zobel shunt, would there be any reason not to use a zobel? I find it strange why it's not more often seen in practice on headphone amps, even though everyone recommends them on gainclones.

You most probably wouldn't need any kind of compensation for the "local feedback" configuration. '5532s are unity gain stable and all, and the 5 ohms would keep off some capacitive loading already.

As for the "global feedback" config, consider using a configuration à la Apheared-47 but with inequal current sharing resistors (your 5532 array already has them, the OPA2227 might get 47 to 100 ohms). In any case I'd simulate the affair in order to make sure it's stable.

The A47 topology looks like a form of feedforward to me, please correct me if I am wrong. What is the advantage of this other than that some of opa2227's current drive can also be utilized?

http://www.intersil.com/data/an/an1111.pdf

According to this app note, the A47 style would have some delay between the 2 opamps, and I would think this might introduce phase shift and result in less stability?

I've only just dipped my toe into the simulation territory, and hopefully I would be able to transfer this into TINA.

I gave it a bit more thought, it seems like the A47 topology which allows direct drive should avoid the outer opamp from going nuts if the output array somehow drops dead (ex. thermal shutdown, not that 5532 has this feature), but this can be useful if the power stage was a OPA541.

paralleling 20-30 mA output current op amps seems lame when you can get op amp/buffers with much higher current ratings, up to .4-1 A

the other probelm is that its much easier if the output op amp/buffer is lots faster than the input op amp for multiloop feedback stability

and if your'e really working from your junk box then small signal Q are likely output buffer parts fro headphone amp power levels

A buf634 would have given me 250mA on a single chip, 10x NE5532 gives me roughly double of that for about the same cost, and I happened to have lots. Factor in extra decoupling caps for each opamp however, they are back even.

I intend to build this for practice, the next would likely be something like an OPA277 wrapped around OPA541 biased into class a w/ LM317 current sink.

And I've got some other stupid projects are on the back burner... like the CMOS inverter array headphone driver:rolleyes:

Looks like this:- 5532 parallel headphone amp

That one sure would be fun, but it seems like kind of a moot exercise when TDA2822Ms are dirt cheap (over here at least... €0.26 for one) and can be gain-hacked easily.

Very much alike, except that I happen to be on frugal-fi budget, so whatever I already have would have to make do.

The other good thing about the 5532 array is that it's already thermally spread-out, so it could probably even take some current sink into the negative rail.

If I used 6 duals in each array, it should have no trouble with 100mA class a bias on each channel!

You may want to check whether output stage A biasing really is beneficial before actually going ahead with something like this though. AFAIR, it commonly seems to be detrimental instead, but I guess the effect would heavily depend on opamp type. A 60 dB noise gain circuit like Samuel Groner used should be well-suited to researching the affair.