From: Lucky Green <shamrock@netcom.com>
Date: Thu, 21 Nov 1996 17:23:33 -0800 (PST)
To: Mullen Patrick <Mullen.Patrick@mail.ndhm.gtegsc.com>
Subject: RE: Mass-market crypto phones
In-Reply-To: <n1363530815.90062@mail.ndhm.gtegsc.com>
Message-ID: <Pine.3.89.9611211658.A10580-0100000@netcom14>
MIME-Version: 1.0
Content-Type: text/plain


On 21 Nov 1996, Mullen Patrick wrote:

> _______________________________________________________________________________
> From: Clay Olbon II on Thu, Nov 21, 1996 15:20
> 
> >A while back, Eric Blossom posted a URL for a mass-market, phone encyrption
> >device (http://www.comsec.com/).  The point of this post is to posit a
> 
> The above mentioned crypto phone was incredibly expensive, if I remember
> correctly.  What it basically was (Please keep in mind I may be thinking
> of a different product) a box containing a modem and a crypto-crunching
> CPU.  Phone in one end, processed by said CPU, and send out other end.  I
> have/had a similar idea, but I want to try make it a totally software 
> product (assuming you have the necessary hw requirements on your computer)
> where you talk in through a microphone (can something be wired so you can
> use a regular phone plugged into your computer/modem/soundcard???) your
> CPU crunches code, and spits out the encrypted data.  Obviously, the
> complementing steps would be done on the receiving end.  
> Has anyone ever tried anything like this?  

Sure. There are Nautilus and PGPFone. The problem is that software-only
solutions are unlikely to provide a sound quality that is acceptable for
everyday use over analog lines. At least not unless you have a very fast
machine and don't intend to use it for anything else than telephony. Here
is why: 

Computer telephony codecs are subject to a tradeoff amongst three main 
factors:

o Voice quality (you want to maximize this)
o Processing power required (you want to minimize this)
o Bandwidth (you definitely want to minimize this)

Imagine the three factors to form the dimensions of a cube. You can get
the choose two values, but not all three. Example: you can get good voice
quality on a machine with a (relatively) slow CPU, if you have a lot of
bandwidth. Or you can get good voice quality on a very fast machine with
little bandwidth. Of the three factors, band with is the most limiting. 
The best codec in the word on the fastest machine can't give you decent
voice quality on a very slow connection. 

So how much bandwidth do you need? An ISDN B-channel, which is designed to
carry one digital telephone conversation, is 64kbps. But a regular POTS
telephone, using an analog line, doesn't have nearly as much bandwidth to
work with. You should not expect an analog modem to link up at a speed of
more than 14.4kpbs. Some lines may link up at only 9600bps.  Forget about
the less than 10% of US phone lines that allow the use of 28.8kbps. You
can't rely on your customers to use your product only over such lines. 

The first problem is therefore to find a software codec that compresses 
your voice down to 14.4kbps. They exist, but they require a lot of 
computations and therefore a fast CPU.

But that's only half the problem. Next, you need to encrypt the 
data steam. The rule of thumb here is: the stronger the crypto, the more 
cycles you need. Since we are talking about strong crypto, such as 
3DES, you will need _a lot_ of cycles.

Once you add the cycles needed by the codec and the cycles needed by the 
crypto, you will find that you, more likely than not, don't get enough 
cycles from your CPU. The obvious solution to this problem is to use a 
codec that uses lossier compression and therefore less cycles, resulting in 
lower speech quality.

Even after sacrificing voice quality, telephony is most likely the only 
task your computer will be able to do while you are on the phone. This 
may work for you, but it won't work for many others. After all, not
many people have a spare 200MHz Pentium sitting around that they don't 
need for anything else.

The alternative is to use dedicated hardware. Remember that a general
purpose CPU is just that: general purpose. If you want to use the box only
for one purpose, there is often a cheaper, better solution available in
special purpose hardware. For the price of the Pentium box, you can build
a much smaller bump-in-the-cord device that delivers great sound, over
slow links, *and* uses strong crypto.

Which is what Eric Blossom <http://www.comsec.com/> did. And he did 
it extremely well, I'd like to add.

I don't want to discourage you. Try the software-only solutions and see 
if they work you. You can find them at 
<ftp://ftp.hacktic.nl/pub/replay/pub/voice/>

--Lucky