Re: BER, FEC, channel access and packet sizes

[Phil Karn <karn@unix.ka9q.ampr.org>]

>  Current wisdom among some of the wireless communications professionals
>seems to be that expecting requiring significantly better than .1% - 1%
>*bit* errors from the physical layer for mobile links is not effective

That's a high error rate all right. Let's see...at .1% BER, the
probability of getting a 256-byte packet through would be

(.999)^(256*8) = .129 or 12.9%. So I guess we're already doing pretty good.

The usual rule of thumb for ARQ protocols is that the packet loss rate
should be kept below 1%, i.e., 99% of the packets should get
through. For a 256-byte packet, that means the bit success rate must
exceed

        (.99) ^ (1/2048) = .99999509

which corresponds to a bit error rate < 4.9e-6.

>From one of the references I have, a (64,40) Reed-Solomon code over
GF(64) could provide this performance on a hard-decision channel with
1% BER. That is, even at this high error rate you'd need only a third more
bits for error correcting overhead. That's a lot better than hammering
away with uncoded frames, hoping you'll get lucky.

Phil