I was just about to post the same thing as bcwang.
The differences in signal attenuation along the resistance track of the potentiometer are not constant. As you turn down the volume (more attenuation), the resistive differences for a single channel accumulate along the track but also potentially average out. Some portions of the track may be more or less resistive than other parts. The averaging comes from the fact that resistance is additive as you move along the resistive track.
The difference in volume between channels and also along the resistive track for a single channel are not linear or constant because we cannot predict where this mismatch is occurring. As you turn up the volume, the audio signal will come to a point where it is no longer being attenuated by the mismatched section of resistive track. As a result, it's not that the mismatch is still there and you can't hear it. The mismatch is no longer in the signal path and thus, it really is not there anymore. At full volume (unattenuated), there should be little to no resistive track in the signal path and thus no mismatch between channels. As you turn up the volume, the mismatched sections start to "decumulate" (if there is such a word). On the other hand, as you turn down the volume, there is more resistive track and thus the defects accumulate but start to average out for a single channel (think of the left channel as dollar cost averaging a volatile stock and the right channel as buying a big blue chip and the length of the resistive track as Time. After 10 years both stocks are matched and have the same gains and are at the same point, but one took a bumpy ride to get there while the other had smooth consistent gains. From an audio standpoint, we don't care about the bumps, just that we are at the same point at a usable point on the track. ). Because pots used in audio are on a log scale, tolerancing them between channels and along every point on the pot's travel can be impossibly difficult and costly.
For example:
You hear a 3 db mismatch difference when the pot is almost fully attenuated (lowest volume). That doesn't mean the 3 db mismatch will be at the 50% point as well. If that resistive mismatch happens to occur due to mismatch at the 75% point on the track, then you will need to move the volume pot past the 75% mark to not have it in the signal path and thus not "hear" it anymore unless of course the resistive differences average out at various points along the track at which point there would be really be zero mismatch. However, it's difficult to tell where the mismatch is on the track without equipment.
How do you maximize the likelihood that your audio signal will not pass through a mismatched section of resistive track?
1) Turn the volume on the DAC to full volume (the signal is passing through a minimum of resistive track).
2) Adjust your external amp to an appropriate volume.
3) Turn the volume on the DAC down until you do not hear clipping.
If you are using high sensitivity, low-Z headphones, you will need a better toleranced potentiometer. That's where stepped attenuators/pots come in, where different whole resistors are used for each "step" in the mechanism.
That was repetitive, long, and really hard for me to word and will probably contain a few false parses or errors, but I hope I was clear and that it helps someone.
Thanks.
