|
Reliable Symbols: Decision-Directed Learning with Quality Control
Stuart Schwartz,
Princeton University:
Thursday October 17, 2002, 4:30 PM,
Princeton University, Friend 004
We introduce the concept of reliable symbols (RS), a method which
permits the use of high-order constellation communication (e.g. QAM) in
severe ISI conditions. We show that the RS method enables the
equalization of dynamic channels without the need for training samples,
even when high-order constellation signals are used. We present a blind
equalization system, which is based on an adaptive channel estimator
and a subtractive equalizer. These, in turn, depend on the 'Reliable
Symbols' concept. Our blind algorithm converges at rates significantly
higher than those attainable using any other blind algorithm.
We have successfully implemented the algorithm in two areas of
general interest. In the first, the channel suffers from a high-level
of ISI, closing the binary eye, as often occurs in practical
situations. In the second, typical of radio communication, the channel
varies rapidly and the SNR is at a modest level. We have successfully
equalized static and dynamic severe ISI channels using 4096QAM. As far
as we are aware, no other published technique enables successful blind
channel equalization for high levels of ISI and time-varying channels
when such high-order constellations are utilized.
*Work done in conjunction with Eilon Reiss and the staff at
Verticalband Ltd., London, and Prof. L. F. Turner from Imperial College
of Science and Technology, London
|
 |