"unfolded diff pair" headphone amp

[kwantam]

Hey all,

It's been a long time since I posted, but I thought I'd stop by and mention a funny little amplifier that y'all might enjoy.

http://web.jfet.org/~kwantam/tda-nh.pdf

Think of Q0 and Q1 as an unfolded diff pair. If beta is large, then they're running at the same current, giving the same transconductance. Obviously, you need a DC differential voltage between them for biasing, so this circuit can't be applied as generally as a true diff pair, but you can use fewer transistors for the same total current and the same input stage gm.

Fortunately, I don't want a DC response from my headphone amplifier, so I can wrap a DC feedback loop around the amplifier to (for example) hold the output at ground DC-wise (this is the function of the op-amp on the right). We want the gain of this feedback loop to roll off well before the audio band (otherwise it'll fight the audio signal) hence the ridiculous time constant of the integrator. Don't need much of an op-amp here, because it doesn't process any audio. The important qualities are high input impedance and relatively low offset; 1/2 an LF353 is perfect.

AC-wise, the whole thing just looks like a kind of funky op-amp. The DC bias point of the amplifier stages is directly related to the output DC voltage; in other words, DC biasing is held by the op-amp. The voltage across R6 wants to be slightly less than 15V if vo is at ground; this means the second stage is biased at about 1mA, so Q2's emitter is about Vcc - 4.3V, and R1 thus has about 5V across it, biasing the input stage at 1mA as well.

At (inverting) gain of 5, closed-loop gain is flat from single-digit Hz up to a few hundred kHz. If we make R31 a rheostat and crank it all the way down to 0, the inverting AC gain is 0, and it's still stable, but now with a crossover frequency a bit over 1MHz.

I'll probably build up a few of these for friends. Might be convinced to make a few extra boards and send them to interested parties for board cost + shipping as long as this offer isn't contrary to the board's rules.

EDIT: mods, I suppose it might have been better to put this in the headphone section. If you'd move it there, I'd appreciate it. Thanks!

[Elvee]

Originally Posted by kwantam

Hey all,

It's been a long time since I posted, but I thought I'd stop by and mention a funny little amplifier that y'all might enjoy.

http://web.jfet.org/~kwantam/tda-nh.pdf

Think of Q0 and Q1 as an unfolded diff pair. If beta is large, then they're running at the same current, giving the same transconductance. Obviously, you need a DC differential voltage between them for biasing, so this circuit can't be applied as generally as a true diff pair, but you can use fewer transistors for the same total current and the same input stage gm.

Yeah, the official name is a "Rush Cascode".

Here are some more tricks of folding/unfolding:
The Tringlinator: a MOS-based Tringlotron amplifier

[kenpeter]

"running at the same current, giving the same transconductance"
Transconductance is 1/2. each BE sees only half the drive signal...

[kwantam]

Originally Posted by kenpeter

"running at the same current, giving the same transconductance"
Transconductance is 1/2. each BE sees only half the drive signal...

Hmm, you seem slightly mistaken. Transconductance is a bias parameter; for a bipolar transistor it's given by collector current divided by the thermal voltage.

Regarding drive signal, depends how you define it. If you define the drive signal as the difference between Vp and Vm, then sure, each one sees half. Another way to define the drive is via superposition: AC ground one input, and drive the other input. In any case, it's clear that whichever input is driven produces an AC current gm*v; in other words, the differential signal current is gm*(vp-vm).

In a diff pair running at the same total current ic, each device has current ic/2, thus gm/2; with the standard current mirror load you get the two currents added together with proper signs, and you once again have gm*(vp-vm). The difference is that you've spent 4 transistors to do this rather than 2.

Elvee, thanks for the pointer on the name. As always, there's nothing new under the sun, just new people to re-discover the same old stuff

[kwantam]

A quick update on this project. I've assembled a board and tested it on the Audio Precision sys 2722 at work. Bandwidth is as predicted, -1dB bandwidth about 1 Hz to 50 kHz or better. THD at 1 kHz measures at 0.001%.

RMS noise at the output in the audio band is about 6 uV, which means the dynamic range is a full 120dB (can easily hit 6V RMS).

Transient intermod comes in at -94dB or better depending on which test variant and which combination of frequencies I pick.

Qualitative data agree with the quantitative: I've listened to it pretty extensively on both Grado GS-1000 and a smaller amount using Beyerdynamic DT-880, and the sound is very clean and detailed.

I'm calling this one a success.

[AKSA]

Named after Christopher Rush in 1964. Used creatively by Tom Holman in his PhD design for NAD of the phono in the 3020.

Not much used since, but taken up recently by KeanToken.

Hugh