Summary

Engineer Al Limberg argues that inserting repetitive PN training sequences into specific ATSC data segments could dramatically improve adaptive equalizer performance under dynamic multipath conditions. The proposed modifications would sacrifice less than 1% of channel capacity while extending coverage to previously unreachable receiver locations, comparing favorably to COFDM alternatives.

Source document circa 1994 preserved as-is

Last Updated Saturday, February 19, 2000

 

Substantial Improvements in ATSC Multipath Performance

By Al Limberg

There are some substantial improvements in ATSC multipath performance that are possible by modifying the standard slightly, which modifications should not interfere with the operation of ATSC receivers already in the field.

The basic problem with the current ATSC standard is the slow initialization of the adaptive equalizer weighting coefficients if one relies on blind equalization techniques. So re-initialization of weighting coeffficients is too slow to be practical (or in many cases possible), if tracking is lost in the blind equalization algorithm used to compensate for dynamic multipath.

There are people out there, particularly David Koo of Philips and Charlie Dietrich of Sarnoff Labs (if theyíre lurking), who have had experience with de-ghosting NTSC and know how effective a good training signal can be in combating multipath. The repetitive long PN sequences described by Dietrich and Goldberg in U. S. patent No. 5,065,242 titled "DEGHOSTING APPARATUS USING PSEUDORANDOM SEQUENCES" issued 23 August 1994 are easily adapted for use in DTV. The training signals can inserted into 314th (or 314th and 315th) data segments of data fields, with precoding at the end of the first data segment of each field still repeating the last twelve symbols of the 313th data segment of the preceding data field. This doesnít screw up the ATSC trellis decoder or the ATSC convolutional de-interleaver. Some data-field comb-filtering circuitry could be adversely affected, if any DTV manufacturer is actually using such circuitry.

You lose less than 1% of the data carrying capacity of the ATSC channel and extend coverage to most receiver locations where reception was previously impossible. This is less than the reduction of the data carrying capacity for COFDM, so the economics are better for the broadcaster than with COFDM. The data recovery portion of the receiver is considerably simpler than for COFDM.

DTV presents a far simpler de-ghosting problem than NTSC. One is only trying to distinguish between eight bins, rather trying to accurately reproduce 256 or so JNDs in an analog luma signal.

There are no horizontal sync signals, color bursts, porches or pedestals that have to be gotten rid of by comb filtering. There is no 1H limitation on the length of the training signal. So re-initialization of the weighting coefficients of the adaptive filter can be done within a few thousand symbol epochs at those 24 millisecond intervals re-initialization is supported by the current availability of training signal.

Blind equalization can adapt the weighting coefficients after re-initialization rapidly enough to follow most dynamic multipath conditions encountered during reception with a stationary receiver. Unless youíre unlucky enough to have signal phase inversions in every data field you will only lose portions of those data fields in which signal phase inversions actually occur. You wonít lose signal completely because the blind equalization algorithm takes so much time to re-initialize the weighting coefficients of the adaptive filter that re-establish tracking of the filter is nigh impossible even with relatively slow changes in dynamic multipath. (For succinctness I write ìsignal phase inversionsî but read ìany horrible wrench in multipath reception conditionî.)

Adaptive filters that employ blind-equalization are a type of narrow bandwidth degenerative feedback loop. Pull-in of the loop is the initial problem. Once the loop is locked, hold-in is relatively easy. Just as you jamload a degenerative feedback loop to lock it quickly, you use training signal to initialize the weighting coefficients of an adaptive filter quickly.

Simply fixing the ATSC standard so that a stationary single-carrier receiver can be made to work under usual multipath conditions might be judged by some to be preferable to reducing channel carrying capacity by 14% to allow reception with indoor antennas at maybe one or two percent greater coverage area.

With fond regards,
Al Limberg
|Home| |E-MAIL|