John Shutt wrote: > A-VSB plays tricks with the data interleaver and > the trellis encoder to create pseudo training > signals that occur much more often than the > original ATSC training sequence does. Doug and > C.B. have long said that if ATSC had more > training signals, it could better cope with > dynamic multipath, but the existing training > signals were so few and far between that most > receivers ignore them altogether and instead go > for blind adaptation. Again, if you use the data symbols for your "training sequence," you have a virtually continuous training sequence. Even if data symbols aren't as "perfectly" randomized as a PN sequence, you can afford to listen for many, many more symbols to do your training, because doing so does not impact on spectral efficiency in any way. Nor does this change the standard in any way. In other words, if a given PN1023 sequence is very, very good for training, then use 2000 data symbols instead. What's the big deal? Use 3000 symbols if it makes you happy. Additionally, each segment in n-VSB starts with a 4-symbol data segment sync, which is very robust. It can be decoded down to 0 dB of C/N. So if A-VSB syncs up their trellis magic to some new sync signal, you should be able to do the same thing using the existing data segment sync sequences, along with the previous 2000 (or whatever you choose) symbols of data. What's nice about this is that it's all up to the receiver design. You can make the receiver as robust as you please, without affecting the standard in any way and without reducing spectral efficiency. > The problem is that if you add additional training > sequences, you break existing receivers. Which is why I specifically did not propose using a new training sequence. The old training segment is used *only* as a time tick, to set the clock in the portable receivers. > Even with more frequent training signals, adapting > for severe changes in multipath is a recursive process > that will span several training sequences. Doug says it can be done otherwise. But even if you make it recursive, you have an infinite supply of randomized data symbols at your beck and call. > And the amount of data payload you eat up in the extra > sequences brings your total to something lower than the > 1999 Sinclair Baltimore tests used for DVB-T, and that > data rate was deemed "unacceptably low" by many ATSC > advocates on this list. Using the data symbols, the spectral efficiency stays at 3.3 b/s/Hz. But it's possible that the info in the time slots allocated to portable devices might be encoded with additional FEC, both convolutional and block, as explained in E-VSB. This works to lower the C/N margin. > E-VSB did absolutely nothing about dealing with multipath, > and relied on lowering the C/N threshold to give more > margin to aggressive blind equalizers. With 5th gen receivers, E-VSB mainly lowers the C/N margin. With older receivers, it also helps with multipath. But with this new scheme, the multipath is attacked by the training-on-data-symbols idea, and the extra FEC of E-VSB may be used in addition, to further lower the C/N margin. 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.