[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: bandwidth in the Internet

>Much of this is in the V.42bis logic.  Well since we are encrypting, modem
>compression does not buy much.  Turn it off and check out the difference.
>Of course if you do this, non-encrypted compressable data suffers
>conciderably (unless you have a PPP level compression).

>Of course the delay in the V.42 logic cannot be helped.  My experience is
>you never want to run without V.42, even if you risk the 12 naks of death.

A few years ago I investigated propagation delay in V.32bis modems and
came to a different conclusion. Turning off V.42bis (no compression
and no error control) only slightly reduced the delay.  It was even
present in synchronous mode where there's a direct connection between
the serial port and the data pump (this was an older and fairly
expensive Motorola/Codex modem that still supported sync mode).

Remote loopback measurements with a Fireberd bit error rate test set
in synchronous mode showed the round trip delay over a local call to
be in the range of 80-90 ms. That's about 1200 bits at 14.4 kb/s.

That's way too many bits to be accounted for by the modulator,
scrambler or the Viterbi (trellis) decoder. The only possible place
were that many bits could be stored is in the receive equalizer. This
is a finite impulse response filter (a tapped delay line) that
corrects for nonflat group delay (aka nonlinear phase response) in the
phone line that causes propagation delay to vary with audio frequency.
This filter is pretty long to allow for worst case group delay
distortion.  An adaptive algorithm sets the taps on this delay line
when the modem trains, so the exact amount of delay can change from
call to call.

Apparently no attempt is made in these algorithms to push the
equalizer taps as far forward as possible, i.e., to use the minimum
delay necessary to correct the actual amount of delay dispersion on
the line. The echo canceller in my modem indicated a far end echo less
than 1 ms away, as expected for a local call, yet the peak equalizer
tap was still being set in the 40 ms range (2x40ms = 80ms).

So the bottom line is that the lion's share of delay in a modern
dialup modem is almost certainly the receive equalizer.  Better
algorithms could be devised to reduce the equalizer delay to the bare
minimum for each call, but in today's modem market it may be difficult
to convince a manufacturer to do something about it unless a whole lot
of people complain.

Phil