[opendtv] Re: Koetter-Vardy RS decoder (Mark's memo)

Al Limberg wrote

> If the R-S code doesn't have to locate byte errors,
> twice as many byte errors per R-S codeword can be
> corrected using an alternative decoding algorithm.
> The rumor is that work is being done on using soft
> trellis decoding for locating byte errors.  I
> believe this practice is mentioned in the revised
> ATSC document A/54.

I saw that mentioned in A/54 before. Not sure it's
still in there now.

Maybe we're talking about two different things, though.
What you're describing is that while Reed-Solomon can
detect, locate, and correct errors, it can correct more
errors if some other algorithm is doing the error
localization function. Often RS is used in conjunction
with a CRC, and the CRC is given the task of detecting
and localizing the errors in the data block. Then the
RS algorithm can be more successful at correcting these
errors.

For RS(n,k) coding, up to (n-k)/2 symbols, or bytes
in this case, can be corrected when RS locates the
errors, but twice as many if something else locates
errors. For ATSC, (208-188)/2 =3D 10 consecutive bytes
can be corrected as of now, but 20 consecutive bytes
if something else localizes the errors.

So the idea is that the convolutional FEC is capable
of identifying which bytes have been received with
errors, and if this FEC can communicate with the RS
FEC, the RS algorithm will be able to correct 20
consecutive corrupted bytes.

What I'm wondering is whether the trellis code can't
somehow benefit from knowing what kind of distortion
is affecting reception at any given time. Perhaps,
using the training sequence, the receiver can determine
what kind of fading is present. I would expect that
different types of echo will affect the 8-VSB symbols
in different but predictable ways. The 8-VSB symbol
maps to a consistent set of 3 output bits for each of
8 states. The likelihood of erroneous 3-bit patterns,
i.e. ones which don't map into valid bytes with from
four symbols, ought to be a function of what kind of
symbol distortion is present right now.

Perhaps this can be used to establish the weights
assigned to each arc (or edge) of the trellis. If
this first FEC decode algorithm is more successful
at correcting errors, the RS FEC might not have to
be exercised, or might have an easier time of
correcting remaining errors. Or if this soft trellis
algorithm determines that correcting errors here is
too hard for a given type of distortion, it can merely
flag errors and let RS do its thing.

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: