Subject: Re: [linux-audio-dev] RFC: API for audio across network - inter-host audio routing
From: Lamar Owen (lamar.owen_AT_wgcr.org)
Date: Fri Jun 14 2002 - 19:30:33 EEST
On Friday 14 June 2002 11:35 am, Richard C. Burnett wrote:
> a 20khz audio signal. At 40Khz sampling that would give you 2 points on
> the wave to reconstruct it.
Which is enough to reconstruct the sine wave if the output is through a proper
post-DAC filter.
> What if you're sampling intervals are 180 out
> of phase with that signal? You would lose it completey. There are also
> all sort of phase distortion problems at higher frequencies for the same
> reason.
Hmmm. Wouldn't that be 90 degrees out of phase for the sample to occur at the
zero crossing? However, if the sample occurs at the zero crossing for any
given frequency in the composite, unless the sampled frequency is exactly at
the sampling frequency (above normal audibility) the next sample won't be at
zero crossing.
But the rules for digital sampling have always been that the highest samplable
frequency is less than half the sampling rate (Nyquist's Theorem). Not less
than or equal to.
And at what frequency does the ear's many bandpass filtered detectors lose
frequency resolution? Well, to answer that one, let me tell you about the
radio engineer that I know who can tune CCTV monitors' horizontal
oscillators by ear -- thanks to a steep notch he has at 15.735kHz in his
hearing. He can hear clearly above that notch, and below that notch, but has
almost zero hearing in the null. He was able to get the horizontal oscs to
within a couple or three Hz. That would be an unusal accident of his
auricular structures to get that steep of a notch, but such individuals do
exist. I _wish_ I had a steep notch at 15.735kHz, as singing flybacks give
me a headache!
Now as to phase distortion, it is quite interesting. FM broadcast theory will
teach you what phase modulation does to a signal, producing sidebands related
to the frequency of the modulation at various frequencies and amplitudes
according to Bessel functions. Thus phase jitter shows up as FM sidebands
clustered around the original signal. So what you are really hearing may not
be true phase distortion, but the intermodulation products of the sampling
rate and the sampled frequency. And since the ear is a frequency-domain
detector, it gets really complicated really quickly.
While I understand the stated reasons for 192ksps, I wonder if the effects are
really audible. Double-blind testing would be required.
> Now, let's take signal processing. Have you ever worked with video or
> image files? A few of my professional friends work with this stuff
> everyday. Their experience is to work in the highest resolution possible
> for all image transforms and then scale down to what you need. The reason
> is multiple signal processing techniques suffer the addition of round off
> related distortion. In that I mean there are less points of interest in
> the data and the result can and will be different in the end.
'Quantization noise' -- but the key is 'can you perceive it' -- otherwise the
typical high quality video (I'm thinking DV or DVCAM here) wouldn't already
be compressed with a lossy algorithm. You cannot see the loss of quality
unless you try to capture a still -- and then you wonder why there's all that
pixelation in the 'digital' still. I've done that -- I thought copying from
S-VHS to DV (on a Sony dual-well DV/S-VHS machine), then bringing it over to
my notebook on FireWire would give me great stills. I got better stills
using an analog vidcap card on a much slower PC than with the DV-> Firewire
transfer.
> And to point out something about low-latency, if it really wasn't an
> issue, people wouldn't be working on it for free. It's extremely
> noticable to almost everyone I know.
The biggest thing about low latency for me is the issue of track-syncing
during multitrack punch-ins. But even that can be worked around -- CoolEdit
Pro does it with a latency compensation constant. Which really means that as
long as the latency isn't too long and it is relatively constant it really
doesn't matter -- I've done punchins with CEP on relatively high latency
hardware that were perfectly synced by CEP's built-in latency compensator.
However, everyone has a high latency processor anyway, between their ears.
The brain's latency has been measured before, and it is substantial fractions
of a second. But the brain has built-in compensators for its own latency --
and any good musician who 'feels' the instrument can readily adjust to any
latency by their brain applying its own already time-tested latency
compensators. Between the time the brain tells the finger to pluck the
string and the time the string is plucked is also measured in substantial
fractions of a second. But musicans already adjust to this.
There's a known 3/4 second delay between the eye receiving an image and the
brain processing the image (which is pretty hefty horsepower, in reality).
This is why all driving schools suggest three second following distances, to
allow for the brain's latency (seeing the car's brake lights, understanding
what that means, activating your own brakes, the latency of the brakes being
applied and you stopping. And people are worried about the turn-on latency
of incandescent lamps versus LED's -- it probably makes little real
difference, but it looks good on papaer).
Performers in an orchestra or band automatically and without much thought
compensate for their own delays and latencies, including the one between the
conductor's baton hitting the beat and their own brain seeing the conductor's
baton hitting the beat.
Yet no one is really aware of this high latency, because it is normally below
the level of consciousness. I have actually experimented myself, just now,
in trying to perceive the delay between thinking the word I'm going to type
and seeing the word typed on screen. It is quite long, relative to the ms
latencies we're talking about.
In short, your brain doesn't have to be aware of it if it doesn't want to.
Yet, there are other benfits of low latency.
-- Lamar Owen WGCR Internet Radio 1 Peter 4:11
This archive was generated by hypermail 2b28 : Fri Jun 14 2002 - 19:22:36 EEST