Fake Components

[Mark Hennessy]

If you look at figure 4 and 6 you will be able to see that the distortion lower with an 8 ohm load compared to a 4 ohm load.
It could be projected down to a higher impedance that the amp will see with a headphone load on it.

No, you simply can't make that assumption - you'd have to run some tests to prove it. There are many other sources of distortion in a power amplifier. With a blameless amplifier, going to 16 ohms rarely causes a reduction in distortion.

You might also note that the LM3886 is pretty much free from LSN - when you get into the fine detail, it's a much better chip than the TDA7294.

For what I bought the set for the distortion figure at 5W is easily good enough.

Of course. Needless to say, 0.005% is orders of magnitude better than anyone can hear.

In fairness, ST did quite well to get the distortion down to that level, given that they used a MOSFET output stage - presumably for marketing reasons, as there's no rational technical reason to build an audio output stage using them. I wonder if that explains the peculiar slow rise in THD as the power rises from 0.005% to 1% over the power output range of 55 to 70 watts - compare that with the LM3886, where the distortion stays low until clipping - the rise from <0.005% to 1% takes place over just a couple of watts, which is exactly as you'd expect for well designed solid state amplifier

But that is the raw distortion of the IC in an ideally-implemented test setup. Whether JVC managed to get the same results is another question.

When I build a design with the LM3886, I have no problem getting 0.003% at 1kHz. A friend of mine designed a PCB for a pair of them, and got nearly 1% on one channel and around 0.1% of the other. He didn't believe that the PCB layout could be a problem, but to humour me, he implemented some of the changes I recommended. The new PCB produced much lower distortion - around 0.07% IIRC (the notebook with the results in is not in this office), but still with differences between channels, which is a pretty big clue that the PCB layout is still not optimal. Finally, I cut tracks and added jumper wires to implement the last of the changes I suggested, and the distortion fell down to very close to the expected 0.003%.

All of this was purely PCB layout - none of the component values were changed during this. There are many ways to mess up even a simple circuit. And even if there is almost certainly no subjective difference between the first PCB and the final version, it still shows the value of measuring THD to prove that you haven't screwed something up.