To: WSFAlist at KeithLynch.net
Date: Sat, 23 Dec 2006 03:24:00 -0500
Subject: [WSFA] Re: Modems, and barbed wire fences
From: ronkean at juno.com
Reply-To: WSFA members <WSFAlist at KeithLynch.net>

On Fri, 22 Dec 2006 09:41:11 -0500 "Mike B." <omni at omniphile.com> writes:
> At 12/21/2006 11:14 PM, Keith F. Lynch wrote:
>
...> The old 300 baud modems used 4 frequencies, two for originator (mark

>
> and space) and two for receiver, all of which were well within human
>
> hearing range (1070 to 2225 hz).

Yes, 1070 and 1270 for the originating channel and 2025 and 2225 for the
answering channel.  A 200Hz peak-to-peak frequency shift in each case.

They were just simple tones, and
> didn't have to be all that accurate so long as they were close
> enough
> to be differentiated from each other by the receiving equipment (I'd
>
> guess that a 10% error wouldn't be a problem given the tolerances
> for
> most electrical components used for the filters).

The center frequencies (threshold frequencies) for each channel were 1170
and 2125 respectively, so frequencies below the threshold would be read
as 'mark', and above, as 'space'.  Obviously, if a transmitter which was
intended to send 2025 and 2225 tones, instead sent frequencies 10%
higher, it would send tones which would always be read as 'space', so a
10% transmit frequency error would be unacceptable.  In practice, it was
easy to make the transmitted frequencies highly accurate, since they were
generated by digitally dividing down from a much higher frequency crystal
oscillator, the crystal being something like +/- .005% accurate.  If the
transmit frequencies were, say, 3% off, it would probably still work, at
the cost of the system being somewhat more susceptible to degradation in
the presence of noise.

The analog filters in 300 bps full-duplex modems were actually quite
elaborate, especially the receive filters.  In the early to mid-1970s,
the filters used op-amps, and resistors and capacitors of 1% tolerance,
and they were epoxy-potted to prevent subsequent drift of component
values due to humidity.  The transmit filters had to block the DC
component of the digitally generated waves, and smooth out the 'staircase
steps' of the raw waveforms, so they were passband filters, designed to
have a 'flat top' response with symmetrical skirts, so the delays for the
'mark' and 'space' frequencies would be equal.  Same flat top and
symmetrical skirts for the receive passband filters, but the receive end
also needed band-stop filters to null out the other channel, as well as
more generally having good out-of-band attenuation to minimize the effect
of line noise.

...To
> go faster than that over voice lines they started doing things like
>
> breaking the available spectrum (about 6 khz) up into smaller bits
> and running multiple parallel channels at different frequencies, and
>
> avoiding those that didn't have enough response over a given
> connection (some of the early 9600 baud modems), phase encoding of
> data, and basic data compression techniques.

The old analog telephone system was modem-unfriendly.  A hundred years
ago, they began putting 'loading coils' on long distance copper lines to
equalize the amplitude frequency response, but that made for a steep
delay versus frequency response - O.K. for voice, but bad for modems.
Hence the multiple parallel narrow band approach that was tried for some
early 'high speed' modems.  The beauty of the 0-300 bps 'Bell 103' type
modems was that they were bullet-proof; they could work on just about any
telephone connection, including transoceanic calls.  I remember one old
modem guy in the 70s telling me that a Bell 103 modem could work 'on a
barbed wire fence'.  Some old rural telephone lines used fence wire (on
insulators) instead of copper telephone wire, because it was cheaper.

In the early 1970s, it was widely thought to be practically impossible,
though not theoretically impossible, to send data faster than 2400 bps
one-way (half duplex) on a long distance telephone line having the
standard 300 Hz to 3400 Hz long distance analog passband.  In 1975, a
2400 bps half duplex telephone line modem cost about $2,000, equivalent
to perhaps $10,000 today, with inflation.  But these days, so-called 56K
modems can send data at up to 53,000 bps over a telephone line.  Apart
from data compression, there is little room for further improvement in
dialup modems, because the dialup voice connections are processed in the
phone system at 64kbps.

Ron Kean

.