[chuao]

Here's an amp I'm designing for sennheiser 580/600/650 headphones; I'm curious what people think about it. I have to say first that I'd rather not get any "that's too complicated, it's absurd, you're a moron" replies...they make me feel bad. It is absurd. I know.

Here's a simplified schematic with some functions abstracted (diamonds are controlled current sources):

The gist is that in the input stage, the complementary feedback pair (CFP) keeps the current in Q2 relatively constant, improving its linearity; on the output side, local feedback sets the current in Q14 to generate a negative-impedance load that cancels the effect of the actual load, eliminating VBE distortion and nonlinear loading on the previous stage due to non-constant beta.

My motivation is to build an amplifier with the best possible linearity yet with the fewest transistors in the signal path, and a "natural" distortion spectrum with both even and odd harmonics. Many of the implementation ideas I'm borrowing from other people (esp. this guy: http://peufeu.free.fr/audio/). In a nod to him, all the "important" transistors (by coincidence) have constant power dissipation, though I'm curious how much that actually matters.

Detailed description (sorry if I go overboard!): Power supplies are probably +-40 volts to accommodate all those current sources and cascodes; plan is to implement that as just a bank of emitter followers buffering a filtered voltage source. Extreme low noise is not a design goal, though R4 will probably have its own separate power supply since it's coupled to the output with no attenuation and I don't want feedback.

Q2, Q3, and R1 essentially form one transistor with very high transconductance...Q2 passes a current just enough to turn on Q3 via the voltage drop on R1 (around .6 volts), so the VBE of Q2 is roughly constant regardless of the current through the whole mess and the emitter degeneration resistors (R2 and R3) see a voltage drop that very tightly tracks the input signal. Using a current source for I1 allows independent control over the bias and signal currents.

The JFET Q4 is a "floating cascode" which causes Q2 to see a constant collector-emitter voltage drop...this eliminates base-width modulation as another potential source of distortion. The current source up top then "flips around" the signal, and Q6 cascodes the current source and keeps the input transistors in the active range. Q4 and Q6 must both be FETs to prevent the signal from being corrupted by base current.

Because JFETs have relatively low transconductance, Q6 makes a poor cascode and it may be necessary to either add an additional cascode transistor to the current source or use a trick like on the input stage to effectively raise the transconductance of Q6...only experience will tell.

Moving on, R4 converts the signal back to a voltage, and Q9 and Q13 buffer it to the output. R6 (labeled as 1 Ohm on the simplified schematic) senses the current in the headphone, and all the stuff on the right attempts to subtract it from some constant bias voltage and present that to the base of Q14. Properly adjusted, this will keep the current in Q13 constant regardless of signal...effectively, it sees infinite load impedance.

That alone would keep the base-emitter voltage relatively constant and avoid the nonlinear loading effects on the previous stages but for the Early effect (base-width modulation)...changes in VCE with signal force changes in the base-emitter voltage to compensate the effect of a widening and shrinking the collector-base depletion region, as well as changes in the rate of hole conduction and carrier recombination in the base thus changing the base current. Whew...so, Q8 and Q12 are another set of "floating cascodes" to eliminate THAT effect. JFET Q7 is needed to avoid corrupting the signal with the base current of Q8.

My one nagging issue is current source I1...it may have an output impedance of 1 meg, and R1+R2 is probably going to be about 2K and of course, but I suspect that its nonlinearity will have a small but measurable effect on the distortion. So it probably needs a cascode, too.

But man, after all that work I predict a very clean distortion spectrum, so I think that I'll just not bother with any global feedback, and here, HOPEFULLY, is an amplifier that satisfies just about everyone (except grado fans...Kevin Gilmore would not be happy about that current sensing resistor).

Please do let me know if anyone thinks of a significant distortion soruce I've failed to consider, or if you think it just won't work outright.

 

[Francis_Vaughan]

A couple of thoughts. I guess the DC adjust might be a source of noise injection depending upon how you drive it.

But a more philosophical criticism. With all the distain of feedback, and desire to avoid it, it seems rather strange to see a negative impedance output amplifer - which is nothing more than a simple feedback amplifier - right in the signal path. Any criticism of the issues of feedback in the amplifier must apply equally to this part of the design.

 

[chuao]

I've got no philosophical criticism of feedback...the point is that, if you want the lowest possible distortion, you always must minimzie the distortion before adding global feedback. which is the main goal of this design.

the neat thing about the feedback loop at the output is that it's in the current domain, and its effect on the voltage transfer function to the output is very limited, being expressed through a log function. (same deal with the DC adjust point.) it's very far from being a global feedback loop around the entire amplifier, not that I would have a problem with a global feedback loop either.

I mean, everybody says (feedback advocates included) that one should make the distortion as low as possible before global feedback, which makes sense, so why not take it to the extreme? the low power dissipation required of headphones makes it easily possible.

the whole discussion of pro- versus anti-feedback is silly, in my opinion, as other issues like slew rate are way more important. maximizing slew rate, however, does require either low feedback factors (e.g. dynahi, which uses no compensation capacitor) or clever tricks, at least as complicated as the amplifier I propose. my hope is that this amp will have distortion and slew rate comparable to the dynahi as drawn, and with global feedback could be a bit lower distortion.

the only "philosophical" issue here is that I don't like the idea of removing distortion by balancing the amp, as that kills the even-order distortion only and the odd-only spectrum is perhaps unnatural sounding. this isn't too outrageous a claim...you can see it in, for instance, Jung's papers on slew rate distortion.

 

[chuao]

oh, I should point out that when I say, "I don't want feedback" at the end of the third paragraph under the schematics, what I mean is I don't want feedback there. That is, I don't trust the current-to-voltage transfer function of the power supplies to sound particularly good, so I'll try to avoid including it in the output by separating the high-current power supply from the one which couples directly to the output voltage via R4.

I should also point out that I'm far from certain that this amp will sound better for its complicated design...it's going to be a big experiment for me.

 

[Francis_Vaughan]

OK, I think we are much on the same page. The "no feedback" is a bit of a red rag on occasion.

I do rather suspect that the negative impedance amp will be difficult. I would really want to see some sort of stability analysis. Since you have the headphones in the loop I do worry about stability. It may be that you find that a much more conventional approach to a single ended output is actually better. But if you don't try the new stuff, you won't ever know.

I would be interested to see how you go with the cascoded complementary feedback pair. Msr Perrot's ideas are at least worthy of interest.

If I were building this, I would really want to build each innovation up one by one. See how each affects the design.