In response to my July column I heard from many of you about the efforts that manufacturers are making under the ATSC 8-VSB improvement effort. Philips and NxtWave sent me additional background on their proposals, some of which I will talk about here.
PHILIPS' PROPOSAL FOR 8-VSB IMPROVEMENTS
Philips described its proposal for 8-VSB improvements at an NAB breakfast. Unfortunately, I missed it. But Dave Bryan, one of the authors of the Philips Research proposal, saw my July article and sent me some information about his company's plan.
Philips proposes using symbol mapping to encode data in more robust 2-VSB or 4-VSB modulation modes. These can be mixed with a standard 8-VSB stream for compatibility. A hierarchical mode, HVSB, – like the DVB-T hierarchical mode – does not significantly reduce the overall data capacity of the channel.
A packet formatter is used to create a robust datastream that is then multiplexed with the standard stream and passed through the ATSC convolutional interleaver and trellis encoder. At that point, the robust data is mapped into a limited number of symbols corresponding to either 2-VSB or 4-VSB modulation. The packet formatter and Reed-Solomon encoder generate valid ATSC bytes from the robust datastream that – when processed in the modulator – creates what looks like either a 2-VSB or 4-VSB signal that can be decoded by an ATSC standard trellis decoder.
The ATSC data is formatted so that when it is transmitted, it looks like it is modulated with two (2-VSB) or four (4-VSB) modulation states. Less complex (fewer states) modulation results in a more robust signal that requires a lower C/N (carrier to noise ratio) and is easier to equalize.
Things become more complicated, however, when hierarchical modulation is used. In DVB-T, a more robust signal can be encoded on top of the normal signal without sacrificing the overall data rate. Philips achieves a similar effect with its proposed hierarchical VSB technique. In addition to the robust datastream and the normal datastream, Philips’ method adds an embedded stream that has a threshold of visibility (TOV) somewhat worse than that of conventional 8-VSB.
For a 20/80 ratio of 2-VSB to 8-VSB packets, corresponding to a robust datastream with a rate of 1.7 to 1.9 Mbps and normal datastream at 15.4 Mbps, the total data rate drops to about 17 Mbps. However, with HVSB, it is possible to add an embedded stream at a data rate the same speed as that of the robust datastream, increasing the throughput to 18.8 - 19.3 Mbps. In this mix, the additive white Gaussian noise (AWGN) TOV performance is 15.4 dB for the normal datastream, 12.2 dB for the robust stream and 16.4 dB for the HVSB embedded datastream.
Philips found that with a 10/90 mix of 2-VSB to 8-VSB packets, the 2-VSB packets could be received with a C/N of only 7.5 dB, while the 8-VSB C/N was essentially unchanged at 15 dB. If the packets sent with 2-VSB use 4-VSB instead, the 4-VSB packets can be received at a 13.0 dB C/N and 8-VSB C/N improved to 14.6 dB.
Although 1.5 to 1.9 Mbps may seem like a very low data rate, it should be sufficient for small screen TVs and as a fallback mode when reception isn't possible at the higher data rate. Analog TV viewers endure snowy pictures with ghost and noise in poor reception areas. The lower-resolution low bit-rate signal should be seen as an improvement over noisy, unstable analog pictures.
NXTWAVE'S SCHEME
I heard from NxtWave as well, and the information I received helped fill in gaps in my notes from the company's NAB presentation. My July report implied a single mode with various data rate trade-offs, but NxtWave is actually proposing three enhancement modes: "Training," "Robust Data" and a combined "Robust Data and Training."
The "Training" enhancement mode inserts the training signals at the symbol level, thus removing the ambiguity in the pre-coder for the ATSC trellis encoder. At the transmitter the ATSC signal is pre-processed by a VSB frame-synchronized training re-multiplexer that inserts training bytes to provide placeholders for defined VSB packets inserted by a complex ATSC training compatibility processor. Because the training signals are inserted at specified symbol positions, the complexity of the receiver's training reference and sync generator is significantly reduced.
The "Robust Data" mode is similar to the "Training" enhancement mode, but rather than sending a pseudo-random noise sequence, robustly encoded data (one-third rate trellis coding) is sent. "Robust data" packets carry a null header so that receivers without the circuitry to decode them ignore them. On the transmitter side, the robust and normal data has to be multiplexed in a pre-processor. On the receiver end, a packet separator is needed to de-multiplex the conventional MPEG packets from the robust data packets.
As the name implies, the "Robust Data and Training" mode combines the two modes by inserting both training signals and robust data symbols. Additional multiplexing circuitry is required at the transmitter. In the receiver, circuitry is needed to provide the symbol training reference to the receiver's equalizer.
NxtWave conducted tests in Philadelphia last October to see how datastreams with the training packets performed in the field. The tests found that when the training packets were added, reception continued to improve until 17 percent of the datastream was used for training data.
NxtWave provided some additional information on the tradeoff between data-rate and robust data. At a robust rate of 17.6 percent, the normal service data rate is 15.9 Mbps and the robust data rate is 1.484 Mbps. Other valid combinations are a robust rate of 35 percent – normal and robust data rates of 12.55 and 2.94 Mbps, respectively; and a robust rate of 50 percent – normal and robust rates of 9.65 and 4.21 Mbps, respectively. If 100 percent of the data is robust, the maximum data rate is 8.418 Mbps.
Although the robust mode data rate is limited when the normal service data rate is set high enough to carry HD programs, it should be sufficient for transmitting reduced resolution for transmitting a reduced resolution (SDTV) program stream to small-screen sets using indoor antennas. The robust data packets, which have a C/N advantage of 5.5 dB over normal packets, can be used as training data to improve reception of the nonrobust data packets carrying the HDTV program stream.
COMPARING THE PROPOSALS
Even though the NxtWave and Philips enhancements appear similar, there are differences. NxtWave uses very strong trellis coding to achieve a robust datastream while maintaining an 8-VSB signal. Philips’ achieves its robust data performance by encoding data as false 2-VSB or 4-VSB symbols.
There is concern that advanced blind equalizers (constant modulus equalizers, for example) that depend on VSB symbols occurring with a certain statistical probability will be confused by a signal with its histogram skewed by the addition of extra 2-VSB symbols.
Philips said this is not likely to be a problem because broadcasters will use a low ratio of robust to normal data packets to avoid losing data bandwidth and because the modulus of 2-VSB and 8-VSB is similar.
What may not be apparent is that these are not necessarily contradictory solutions. NxtWave’s approach could be applied to 8-VSB packets in the Philips proposal, co-existing with the embedded 2-VSB packets. If receiver manufacturers were willing to consider a variety of approaches, we would have a compatible system with tremendous flexibility compared with the current ATSC standard. What’s best is that with the changes suggested in these proposals or even with a combination of the two, existing receivers would still be able to receive the "normal" packet stream.
Comments and suggestions are always welcome. E-mail me at dlung@transmitter.com.
