Xref: utzoo rec.audio:8743 sci.electronics:4155
Path: utzoo!utgpu!water!watmath!clyde!att!ihwpt!jgk
From: jgk@ihwpt.ATT.COM (joe klinger)
Newsgroups: rec.audio,sci.electronics
Subject: Re: Extracting "Center Channel" from Stereo Image
Summary: Trying to undo a "pan" pot
Keywords: stereo center channel
Message-ID: <2718@ihwpt.ATT.COM>
Date: 27 Oct 88 21:02:40 GMT
References: <355@ivucsb.UUCP>
Organization: AT&T Bell Laboratories - Naperville, Illinois
Lines: 61

In article <355@ivucsb.UUCP>, todd@ivucsb.UUCP (Todd Day) writes:
> 
> What I propose is that we remove Johny from the "center channel",
> if you will, and put his voice in a speaker that sits on top
> of the TV.  In this manner, you will remove the dependence of
> having to sit in the exact center to hear the center channel
> properly.  You will be able to sit right next to the left
> speaker, and Johny will still sound like he's talking from
> the center (during his monologue, he is the only person
> on camera).
	<stuff deleted>
> Therefore, the left channel will be  L+0.5C.
> This means the right channel will be R+0.5C.


The difference between left, right and all locations in between
in conventional stereo is one of amplitude differences.
At an audio mixing console you simply take an input and
route it to the master stereo bus through either a direct patch,
(either left, right, or both) or more commonly, through a 
"pan" (panorama) pot which does a crossfade of the signal between
the left and right channels.  Signals placed in the center of
the room are represented equally on the left and right channels
but signals off center are represented higher on one side
than the other.

To this point I believe we are in aggrement.  To simply sum
the left and right channels together which would certainly yield
L+R, but would also destroy your stereo imaging as signals located
on the left and right speakers would have a stronger
representation from the center speaker.  

Digital..... I"ve used a Bruel and Kjaer dual channel analyzer
which had the capabality of displaying the cross correlating spectrum
of both channels, that is the signal that both channels have in common.
I belive what they are displaying is the residual of a difference 
between the channels.  

I don't think you have to try to model the ear but if you can, 
write a paper on it, many people would like to know how.  
The mechanism which provides image localization however, is
psychological, aided by physiological transducers (ears).  True 
the ear has a narrowband filter-like composition, but that is
only the nature of "sampling" (aargh... I didn't mean that).

You can probably do something like L, R, C separation digitally,
but it is going to cost you in processing power; bandwidth * 2.4 (Nylquist)
* 16 or so bits of resolution * two A/D channels and one to three D/A channels.
Not considering DSP programming time.  Can you really process two channels
of data on one DSP?  
I'd really like to see an analog solution, it must exist in the form of 
a black box somewhere, perhaps in Bob Carvers lab or over at Aphex or
Omnisonics.

Good Luck

Joe Klinger
att!iexist!jgk

Disclaimer -	These opinions are my own and may or may not be supported
		by my employer.