🚀 go-pugleaf

In article <1992Apr16.030623.29152@uc.msc.edu> eab@uh.msc.edu (Edward Bertsch) writes:
>The talk about the professional 'secure' telephones (MIL-std devices
>that cost big bucks) made it sound like the basic system was
>
> voice -> mic -> A/D converter -> encrypt circuit or algorithm -> modem ->
> telephone line -> modem -> decrypt circuit or algorithm -> D/A converter ->
> speaker
>
>Certainly this can be done with an IBM PC, Macintosh, Amiga, Sparcstation,
>or other computer that many already have.

You left out a vital component, at least in secure telephones that are
to operate over conventional telephone lines: the vocoder.

To gain reasonably good speech quality, simple waveform coding of
speech (i.e.  sending the output of an A/D converter) requires a much
greater transmission rate than can be supported by the current
generation of telephone modems. Indeed, if you are to provide standard
telephone quality speech, you'd use what the phone companies use, a
"mu-law" PCM codec running at 64 kb/s. Unfortunately there aren't too
many 64kb/s modems that can operate over conventional telephone lines,
Telebit's recent announcement of a "70,000 b/s" modem notwithstanding.

So you need to reduce the data rate somehow. You could sample at a
lower rate, or with fewer bits per sample, but that would inevitably
degrade speech quality. It's better to remove only the redundant
information, i.e., to compress it. Various "delta modulation" schemes
that send only the differences between adjacent samples exist.
Motorola has a line of two-way secure voice radios that use an
adaptive delta modulation scheme, which I believe operates at 16 kb/s.
But that's still uncomfortably fast for an ordinary phone line, and
voice quality does suffer.

The best way to transmit low bit rate voice is to compress it with a
"lossy" compression scheme that deliberately discards information that
the listener cannot hear. That's what a vocoder does. It actually
models the human vocal tract as a time-varying 'excitation' feeding a
series of time-varying audio filters that mimic the changing
resonances of the human vocal tract as the speaker moves his or her
tongue, lips and jaw. The system transmits these excitation and filter
parameters instead of the actual audio signals.  The data rates
typically vary from 1200 bps to 9600, with better speech quality at
the higher rates, of course.

Unfortunately, vocoder algorithms tend to be very compute-intensive,
so at present they are practical only on DSP chips that are not yet
widespread in personal computers (the NEXT workstation is a notable
exception). And until now they've been useful mainly in secure
telephones, which is traditionally a small and closed market. But the
development of digital cellular telephones is pushing the vocoder into
mass production. Both the TDMA digital "standard" and the competing
CDMA system that we are developing here at Qualcomm use vocoders. So
no matter which system wins in the market, vocoders will become a mass
market item in the next few years. And sooner or later, they should
find their way into commodity PC hardware. Then all you'll need to add
is software to build a secure phone.

Phil