From: tcmay@got.net (Timothy C. May)
Date: Tue, 15 Aug 95 20:50:56 PDT
To: cypherpunks@toad.com
Subject: Purple Boxes vs. Native Signal Processing
Message-ID: <ac56b476090210042171@[205.199.118.202]>
MIME-Version: 1.0
Content-Type: text/plain


At 4:12 AM 8/16/95, Lucky Green wrote:

>I agree. You can easily do encryption and compression on the same chip.
>There is no need for a separate DES/RSA chip. I only included some possible
>DES chip manufactors, because the original poster asked for them.

The trend is away from having two chips when one will suffice. Thus, the
Macintosh 840av and 660av had a Motorola 68040 _and_ a Motorola 56000 DSP
chip for speech processing and recognition, sound processing, etc....they
were dropped and replaced by the PowerPC machines, which dispensed with the
separate DSP chips. (The DSPs were never fully supported by software,
especially from third party vendors....)

And Intel is pushing "native signal processing," wherein DSP functions are
pushed back into the CPU. If the CPU is fast enough, as the fast Pentiums
are, this can work. This may be partly to sell more and faster Pentiums and
partly because Intel has no effective DSP products at this time.

(Intel has generally missed out on the DSP market, despite arguably having
invented the first DSP chip. In 1977 I worked on the Intel 2920 signal
processor, the first general purpose chip to do signal processing. Invented
by Ted Hoff, the same guy who invented the microprocessor, it was abandoned
a few years later. Then came the success of TI's TMS320 (or similar) DSP
chips, the Motorola 56000 series, and Intel was out of the game.)

Pushing DSP functions into the CPU can be taken too far. Apple, for
example, had/has a "Geo Modem," or something like this, which does modem
funtions in the CPU of some machines. Last I heard it was stuck at being
too slow, with no software updates, and effectively is being abandoned.
Modems have gotten so cheap that using the CPU makes little sense if it
also complicates software.

As in everything, the choice in partitioning is crucial.

Of relevance to this list, I see no hope whatsoever that people will buy
gizmos to do encryption if a software-only ("native signal processing")
solution is within a factor of several in performance. After all, people
complain that RSADSI wants "exorbitant" prices ($125) for public key
ecryption and demand that "free" products are needed, so I can't see them
spending $300 or even $100 for a hardware solution that does encryption a
bit faster.

If the Pentium + Soundblaster can do VoicePGP or PGPFone or Nautilus, then
what's the incentive to buy additional hardware? Last I heard about CELP,
it could handle "Pretty Good Voice" on a mere 66 MHz 486, and that's about
2-4x slower than what people are routinely buying today.

(The situation may be slightly different for a fully-productized and
consumerized "bump in the wire" secure phones, where the finished product
will be sold to a different sort of customer than those clamoring for cheap
crypto.)

In conclusion, I'd always look to a software solution first. Anything that
requires chips is automatically harder to build and to sell.

Most importantly, a solution which runs on standard hardware available
around the world will be trivially exportable (technologically) and will
spread within days of availability, whereas a hardware-dependent solution
will likely remain obscure and hard to export. Sofware solutions rule!

--Tim May



---------:---------:---------:---------:---------:---------:---------:----
Timothy C. May            | Crypto Anarchy: encryption, digital money,
tcmay@got.net (Got net?)  | anonymous networks, digital pseudonyms, zero
408-728-0152              | knowledge, reputations, information markets,
Corralitos, CA            | black markets, collapse of governments.
Higher Power: 2^756839    | Public Key: PGP and MailSafe available.
"National borders are just speed bumps on the information superhighway."