[opendtv] Re: E-8VSB
- From: "Manfredi, Albert E" <albert.e.manfredi@xxxxxxxxxx>
- To: "OpenDTV (E-mail)" <opendtv@xxxxxxxxxxxxx>
- Date: Fri, 22 Oct 2004 18:46:34 -0400
Al Limberg wrote:
> IEEE publication CH2538-7/88/0000-0468, a 1988
> paper by A. Chouly & H. Sari titled "APPLICATION OF
> TRELLIS CODING TO DIGITAL MICROWAVE RADIO" indicated
> that both trellis-coded 32QAM and trellis-coded
> 64QAM are inferior to uncoded 16QAM as fade depth
> increases beyond a limit such as 6 or 7 dB. Using
> noiseless system signatures, uncoded 16QAM was found
> superior to that both trellis-coded 32QAM and
> trellis-coded 64QAM in strong multipath situations.
This is also true for COFDM, though, to a certain
extent. I think the intriguing question is just what
the tradeoff is in the two schemes, between simpler
symbols and more robust convolutional FEC. In other
words, I think we're talking matters of degree here.
It could very well be true that COFDM responds
more to increased FEC depth than 8-VSB does, in
severe multipath. It would be nice to know.
Fortunately, for COFDM, we have tables that describe
the tradeoff, at least in theory. (The FEC mentioned
here is the Viterbi convolutional scheme only.) The
numbers are out of ETSI EN 300 744.
In Gaussian fading,
16-QAM with 7/8 FEC (i.e. next to nothing) ~ 64-QAM
with 1/2 FEC (i.e. the most robust in DVB-T).
But the bit rate available with 16-QAM is 15.4 Mb/s,
while the 64-QAM cum heavy FEC gives only 13.1 Mb/s.
So under LOW multipath, you come out ahead with
simpler symbols, with COFDM, for a given level of
robustness.
=20
In Rayleigh fading,
16-QAM with 7/8 FEC ~ 64-QAM with 3/4 FEC.
The bit rate for 16-QAM is still 15.4 Mb/s, but
the 64-QAM with lighter FEC gives 19.7 Mb/s.
But using 64-QAM with 1/2 FEC ~ 16-QAM with
3/4 FEC, and the bit rate in both is 13.2 Mb/s.
So in more severe multipath, you seem to gain
less with simpler symbols in COFDM, and you
come out even or better if you diddle with
the FEC rather than the symbols.
(In these examples, the C/N margin required in
Gaussian fading was much better than the Rayleigh
case, of course. I'm just showing the comparison
between symbol differentiation and FEC
prediction in compensating for multipath.)
I'll bet that similar effects would be seen
with 8-VSB, but it's possible that under
severe multipath, you do better with simpler
symbols than with more FEC, as Al says. I don't
know. But the fine-tuning advantage still goes
to adjusting the FEC rather than the symbols.
Bert
----------------------------------------------------------------------
You can UNSUBSCRIBE from the OpenDTV list in two ways:
- Using the UNSUBSCRIBE command in your user configuration settings at
FreeLists.org
- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word
unsubscribe in the subject line.
Other related posts:
