From: Phil Karn <karn@qualcomm.com> (Phil Karn)
Date: Mon, 8 Feb 93 13:24:26 PST
To: Phiber Optik <phiber@eff.org>
Subject: Re:  Compressed/Encrypted Voice using Modems
Message-ID: <9302082123.AA00695@servo>
MIME-Version: 1.0
Content-Type: text/plain



>Shannon didn't consider intersymbol interference.

Sure he did. That's why the bandwidth term appears in his channel capacity
equation.

Shannon built on and generalized Nyquist's earlier work. Shannon's law
says only that it is theoretically *possible* to signal over a bandlimited
AGWN channel with specific bandwidth and S/N ratio with arbitrarily low
error as long as the signalling speed is less than the channel capacity
as given by his formula. What it doesn't show is *how* to do it. In
this respect, Shannon's law is much like the 2nd law of thermodynamics;
it tells you how efficient you can make a heat engine in theory, but it
doesn't tell a turbine designer how to shape his blades.

By the way, just to bring this back to crypto, everyone should be aware
that not only did Claude Shannon establish modern information theory,
he also wrote a seminal paper that established much of modern cryptography.
It was originally written during WWII and classified at that time, but
it was declassified soon after the war and appeared in the BSTJ in 1948,
I believe. Probably the most well known aspect of this paper is his
discussion of "product ciphers", whereby you can combine different
ciphers that are by themselves relatively weak (substitution and
permutation, which he calls "confusion" and "diffusion") and produce
a far more powerful cipher. DES is based on this principle, as are other
modern ciphers.

I bet this is one paper that the NSA wishes had never been declassified.
Far more important, in my opinion, than anything by Friedman.

Phil