[sdiy] Adding audio channels, to increase S/N

[Ethan Duni]

I'd add that these sorts of schemes depend heavily on the assumption that
the noise in each channel is uncorrelated. The obvious contrast to this
situation is the use of balanced input/outputs. In the latter case, the
noise in each channel is assumed to be identical, and so can be cancelled
entirely by subtraction, although this requires some "preprocessing" (i.e.,
inverting the input signal to one channel). More generally, one can consider
rather arbitrary multiple input/multiple output systems with differing
models of how the noise in each channel is related, and can then formulate
optimal pre-/post-processing strategies to maximize the signal to noise
ratio (this is a hot topic in modern communications systems research).

I wonder if the original poster might expand on his reason for considering
this stuff? To me, it seems that the effectiveness of any of these schemes
depends directly on how you plan to handle the different signal paths. For
instance, simply running multiple identical signals into cables that are all
stuck together in a snake will not improve your signal to noise ratio.

Ethan

----- Original Message ----- 
From: "Magnus Danielson" <cfmd at swipnet.se>
To: <mclilith at charter.net>
Cc: <synth-diy at dropmix.xs4all.nl>
Sent: Thursday, September 04, 2003 3:40 PM
Subject: Re: [sdiy] Adding audio channels, to increase S/N


> From: Glen <mclilith at charter.net>
> Subject: [sdiy] Adding audio channels, to increase S/N
> Date: Thu, 04 Sep 2003 03:05:04 -0400
>
> Glen,
>
> > Please, help refresh my memory.
> >
> > If the same signal is fed to multiple audio channels, and their outputs
are
> > later summed, the S/N ratio is higher than if only using a single
channel.
> > For example, some people would send the same signal to several channels
of
> > audio tape, in order to increase the S/N ratio of the tape recording.
The
> > idea being, the signal would reinforce itself when the various copies
were
> > added back together, but the random background hiss would not directly
add
> > up, and be somewhat muted in comparison to the signal.
> >
> > I just can't remember what the improvement is, measured in dB, when you
> > double the number of channels used. Is it 3dB? I know it wasn't much for
a
> > simple doubling, but if you had several channels available, the
improvement
> > was supposed to be dramatic.
>
> It is 3 dB for each doubling... if the worst noise is internal to the
signal
> chain and not in the signal to start with...
>
> > I should remember the answer to this, but my mind must be distracted at
the
> > moment.
>
> It works like this... if you have signal of strength S and internal noise
of
> level N, then the output from the sum for the signal is S+S = 2*S since
the
> two output signals is in phase. For the noise phase is random, with random
> phase the two noise variants is orthogonal and the sum output is the
vector
> sum of two vectors being 90 degrees from each other... this ends up as
being
> sqrt(N²+N²) = N*sqrt(2) due to Phytagoras... so real signals doubles (+6
dB)
> when noise only rises with square root of 2 (+3 dB). When looking at the
S/N
> this new S'/N' is +6-3 = +3 dB higher than original S/N.
>
> Cheers,
> Magnus
>