IEEE Symposium: Improving DTV Reception
December 11, 2002
This month I'll report on this year's IEEE Broadcast Technical Symposium in Washington, D.C. This is one of the best conferences for learning about new technology in RF for broadcast. This year there was also an interesting paper on Ka-band satellite links.
LINX DTV RECEIVER
The Linx Electronics DTV receiver technology attracted a lot of interest at NAB, but few details were given on how it operates. A common question was, how could the Linx demodulator handle multipath that other demodulators couldn't? Richard Citta and Yiyan Wu presented more information on the techniques the Linx receiver uses to improve DTV reception. There was a fairly simple mathematical analysis in the paper, they presented at the conference, but I'll avoid it for this report. As a result, please excuse the simplifications.
DTV demodulators use an equalizer to minimize distortions on the received signal. Distortions include multipath, interference, frequency response variations, etc. The performance of the equalizer is critical in determining if a DTV receiver will succeed in receiving signals in a difficult environment.
A modern equalizer consists of a channel-matched filter, a feed-forward equalizer and a decision feedback equalizer. The channel-matched filter should provide the maximum signal to noise ratio (SNR) at its output. The decision feedback equalizer removes delayed ("post-cursor") echo multipath distortions while the feed-forward equalizer minimizes leading ("pre-cursor") ghosts. The design (hardware and software) of the equalizer determines the time range and amplitudes over which echoes can removed.
Dr. Yu explained that while the decision feedback circuitry operates at a rate of once per received symbol, better performance was obtained taking samples for the feed-forward equalizer at a multiple of the symbol rate. Filter coefficients are chosen to match the channel response (distortions on the incoming signal) as close as possible as fast as possible. While this sounds simple, filter coefficients can be difficult to impossible to calculate on a channel with strong (equal) ghosts and pre-ghosts.
One option discussed in the paper is to use a pre-filter to convert pre-ghosts into post ghosts. This can be done using discrete match or mirror filter to create a mirror image of the channel response. The combination of the mirror filter and the channel impulse response give a total response with maximized SNR and, in the example shown in Fig. 1, one dominant echo. Fig. 1 shows the channel impulse response (top plot, in red), the mirror filter response (middle plot, in blue) and the combined response (lower plot, in violet). As you can see from the combined plot, combining the responses does result in better SNR but more taps are needed to handle it. In the example, the response is now twice as long.
If you have looked at the tap energy plots from over-the-air DTV signals, it is easy to see that doubling the number of ghosts greatly increases equalizer complexity. The good news is that there is little penalty for truncating the response at 99 percent of the channel energy. The strong error correction in the ATSC VSB system makes it possible to ignore echo levels greater than 20 dB below the main signal path, leaving less echoes to correct.
| Fig. 1|
The signal with the response shown in the bottom trace in Fig. 1 clearly needs more correction before it can be demodulated. The output of this discrete-matched channel filter goes to the feed-forward equalizer, which removes most of the pre-ghost energy. Residual error is handled by further adaptation of the filter coefficients in the alogrithm used to minimize distortion. Refer to the slides from Dr. Wu's presentation at IEEE on the Web at http://www.Linxelectronics.com/pdf/IEEEsympo1.PDF for graphs showing how the feed-forward equalizer takes the symmetrical response from the channel filter and eliminates the pre-ghosts and the decision feedback equalizer removes the post-echoes.
As mentioned before, using fractional taps often allows the equalizer to outperform symbol-based equalizers. If a limited number of taps are used, performance suffers. However, a large number of taps increases hardware complexity and can hinder the ability of the equalizer to repsond to rapidly changing multipath. Finding the right compromise is essential.
On paper and in the presentation Yiyan Wu described this technique as one way to improve equalizer performance. He did not go into details on the Linx receiver, which also excels in synchronization and carrier recovery, a crucial element of receiver performance. Additional details on how this was achieved was not discussed. From this equalizer discussion, it is clear the improvements in the Linx receiver come at the cost of increased hardware complexity. Fortunately hardware technology continues to improve at a rapid rate. Designs that are too expensive for widespread use today will probably be quite affordable tomorrow!
Of particular interest was the result of field tests conducted at a location east of Chicago where multipath is quite bad. Pictures of the test site and spectrum analyzer plots of the received signals are now available on the Linx Web site http://www.Linxelectronics.com/pdf/IEEEsympo2.PDF) .
How did it work? Richard Citta looked at all available DTV signals in Chicago - eleven in total. Using the reference receiver and a horizontal dipole looking out the SW window, six had good reception, one had reception with burst errors and four did not lock. With the Linx receiver, all had good reception, with only one of the stations having a SNR less than 20 dB. Using a vertical monopole, only four of the stations had good reception on the monopole while seven did not lock. On the reference receiver, only one station did not lock, Channel 53. With this test configuration, the Channel 52 DTV signal was about 25 dB stronger than the channel 53 DTV. Citta pointed that in the real world, if the undesired-to-desired ratio is greater than 25 dB, reception will fail. Stations planning low-power DTV operation need to consider how much power stations on adjacent channels are using. Channel 45 had a S/N of 22.5 dB but due to a failure in the ATSC encoding stream it was not transmitting ATSC data when the test was done.
Take time to look at the spectrum analyzer plots on the Linx Web site and compare them with what you've observed at your problem locations. If you've seen worse, Linx would like to know about it!
Next month I'll take a closer look at extending DTV coverage through translators, single frequency networks (SFN) and on-channel boosters. There was a lot of discussion about this at the IEEE Broadcast Symposium. While significantly more complex than a single transmitter system, single frequency networks may be the only way to provide reliable DTV reception in some areas. Comments are always welcome. Drop me a note at firstname.lastname@example.org