"Tommi" wrote in message


I may well be suffering the myth, but my understanding is that it
matters whether you sample a sine wave 2 or 8 times. Tests have been
made where subjects had to determine which sound came first from
their headphones. The same signal was fed to both L and R channels,
only the other one was delayed by 5-15 _micro_seconds.


Some of the people were able to "localize" the sound source even when
it was delayed only by 5 microseconds. This implies that a sampling
rate of 192kHz(which results in 5.2 microsecond's sample intervals),
for example, is not only pushing the nyquist rate to the ultrasonic
range, but also presents better channel separation on multichannel
systems.

For the purpose of discussion, I'll stipulate that your facts are correct to
this point. I really don't know that, but it would help me make an important
point if we don't argue over that part of your comments.

So, it doesn't necessarily matter if you sample a sine wave
2 or 8 times on a mono system, but on a multichannel system higher
sample rates result in better localization.

The myth here is that signals in a digital system can have interchannel
timing differences that are only integer numbers of sample periods. IOW this
myth as applied to 44,100 Hz sampling is that interchannel timing
differences can only be multiples of 22.675736961451247165532879818594
microseconds. I agree that this seems to be intuitively clear. But it is
also quite wrong.

The myth comes from the idea that two signals in different channels that are
displaced in time can only be expressed as the same set of sample values,
but time-shifted. This is not the case. Two signals in different channels
that are displaced in time can be expressed as different sample values.

For example, if two slowly-increasing (ramp) signals are displaced in time,
one signal might have a set of sample values that starts out 0, 10, 20,
30... This is a ramp that starts at t = 0. The time-delayed version of this
signal in another channel could have a set of values that is 0 at t = 0, but
is 5, 15, 25... for successive samples. If you looked at these two signals
over time, you'd say that the second signal is time-shifted by an amount of
time equal to half a sample period. And, that is how it would sound.

The correct time resolution of sampled signals is the sample period divided
by the number of distinct amplitude levels. In the case of 16/44 this would
be 5.1418904674492623958124444033093e-10 seconds or
514.18904674492623958124444033093 picoseconds. This is a tiny, tiny number.
In reality, it is lost in the noise.