Thank you for your reply.
Is this related to the voltage divider that output impedance of the Mojo forms with the (very frequency dependent) IEM impedance or is there something else happening?
Regarding distorsion, I was refering to a post of rob watts saying:
Reducing the output load only starts to increase distortion with 33 ohms - at this level it is very much lower than other headphone amps.
But I may have misunderstood if it actually applies the Mojo and in which scale.
I have not covered this before, but it's fairly simple (well ok maybe not so simple). The output stage is the same as Hugo in that it uses the same discrete transistors. Now the output stage must be short circuit proof, and this is done normally by adding a current limit. The problem with current limits is that they start to function even with moderate OP stage currents, as the limiter starts to apply the brakes. And you can easily measure and hear this effect.
So to overcome this I use a series resistor instead. So when the OP stage is shorted, it will now see the series resistor, and it is safe. But you can't simply connect a series resistor to an amp OP stage; it will add to the output impedance, and it will mess up the damping of the headphone, giving you a soft bass. Also, adding a resistor in the signal path will directly degrade transparency. So what happens is the feedback is taken
after this resistor; so the single feedback loop eliminates the consequences of the resistor being in place. But it means at RF frequencies, when the global feedback is not providing any effect, the resistor is functioning, and its there when the OP is shorted, so the OP stage is still safe. Because the resistor is present at RF, I also add a capacitor to ground, and this gives me another RF filter.
Now conventional DAC's have very complex out of band and RF filters, because DAC's create huge amounts of noise which must be filtered out (if you don't you get more noise floor modulation, and then things sound hard and bright). But adding these filters degrade transparency, simply because you have a lot more passive components in the signal path. Now with pulse array, I run the noise shapers at 104 MHz; and this means that the OP noise is much lower than conventional - but it still needs to be filtered out. So the adding of the OP resistor and the capacitor gives me another RF filter, and I get it for free, as there is no price to pay for it in terms of transparency, as the filter is inside the DAC's global feedback loop.
So this resistor and cap is very valuable arrangement; I get an RF filter with no cost in transparency; I get a an OP stage that is safe under shorts but does not require a limiter which would add distortion. But getting back to Mojo - why does adding a low impedance load subtly change the frequency response at 20 kHz? Its because I wanted to use the same OP stage as Hugo; but the transistors I use are way too big physically for Mojo, so I use the same transistors but packaged in a tiny package. But these have less power dissipation, and that forces me to use a safety resistor that is three times larger in value; and the larger value gives a bigger change in frequency response with low impedance loads. But we are talking about frankly trivial changes; at 33 ohms it is only -0.1dB down at 15 kHz; with 16 ohms it is -0.3 dB at 15 kHz.
But in terms of distortion, because there is not a current limit used for shorting, the current delivery is extremely linear and is the same as Hugo. With IEM's the impedance is low, but the actual voltage is trivial, and you certainly won't be able to measure or hear any distortion with such low voltages.
I hope this gives you a flavor of the complexities involved in designing Mojo's output stage; successful design for optimum sound quality is actually very complex, often involving competing pressures.
Rob