Archived from groups: comp.dcom.lans.ethernet (
More info?)
Albert Manfredi wrote:
> "glen herrmannsfeldt" <gah@ugcs.caltech.edu> wrote:
>
>> For a phone line, Shannon's limit is pretty strict, especially
>> as most are digitized at 64000b/s somewhere along the way.
>>
>> For UTP cable, the limit is not so sharp. Attenuation increases
>> pretty fast with increasing frequency, but that can be corrected.
>> Some noise sources can also be corrected, such as using echo
>> cancellation techniques. With increasing work on transceivers
>> there may still be some distance to go.
>
> Shannon's law is the limit, after you've taken whatever measures you're
> going to implement into account. It does not state how you achieve that
> limiting usable bit rate, it just tells you what the usable bit rate
> will be. And it depends on the bandwidth (in Hz) of the transmission
> medium and the S/N ratio on that medium. S/N ratio, not the ratio of
> (for example) the main signal to noise + echo (S/[N + E]). The S/N
> ratio, in turn, would be affected by the length of the medium.
>
> Capacity (in b/s) = bandwidth (Hz) * logbase2(1 + S/N)
>
> Note: S/N is a ratio, *not* expressed in dB in the equation above.
>
> In practice, if you hope to approach the Shannon limit, you would make
> constructive use of echo energy and you will also implement a good error
> correction code. By "constructive use," I'm saying that you would
> equalize the channel so that echo energy is added to the energy of the
> main signal, so the S/N ratio will actually consist of ([S + E] / N).
>
> Let'e say for example that the Shannon limit is 30 Mb/s for a particular
> link. You are free to pump 1 Gb/s through that link, but that will
> create lots of errors. So if you want to keep throwing 1 Gb/s at the
> link, you'll want to introduce error correction techniques, which will
> use up some of that bit rate. Shannon's law predicts that the very best
> you can do, whether you try 1 Gb/s and then add error correction, or
> whether you simply lower the transmitter's raw bit rate, or a
> combination of such techniques, will be to achieve a usable 30 Mb/s.
>
> I guess what I'm saying is that the practical manifestation of Shannon's
> law might not appear sharp, but that doesn't mean that you'll violate
> the limit. It just means that as you approach the limit, you will have
> to endlessly tweak all your error correction, echo cancelling, and any
> other trick you're trying.
>
> For a regular POTS phone line, limited deliberately to 4000 Hz by your
> baby Bell, Shannon's law says that you can achieve 56 Kb/s only if your
> S/N ratio is a whopping 42.1 dB.
>
> On the other hand, if your baby bell eliminates the 4 KHz filters from
> your POTS telephone line, things would be different. Assume the voice
> grade UTP has a bandwidth more like 50 KHz, just for the sake of
> argument. Now you can achieve 64 Kb/s with just 1.55 dB of S/N. In fact,
> if you can actually achieve an S/N ratio of 42.1 dB as you had above, by
> eliminating those 4 KHz filters your voice grade cable will now be able
> to carry 700 Kb/s.
This is all well and good, but for purposes of Shannon's Law calculations,
what is the "bandwidth" of CAT5E cable? And don't say "100 MHz"---that's
what it's _tested_ to, not any kind of hard upper limit.
> Bert
--
--John
Reply to jclarke at ae tee tee global dot net
(was jclarke at eye bee em dot net)