ITR: Superimposed Training for Wireless Fading Channels

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

G. Tong Zhou Georgia Tech Research Corp A major challenge in wireless communications is that the channel changes, sometimes rapidly; thus channel state information needs to be acquired frequently before the symbol recovery stage. Two prevalent approaches for channel estimation are training based and self-recovering (i.e., blind) ones. Pilot symbol assisted modulation (PSAM) is a popular training method, in which known pilots are periodically inserted into the symbol stream prior to transmission. These pilot symbols occupy time slots and impose a recurring transmission overhead. Blind approaches on the other hand, are completely data-driven with no loss of information rate but a significant increase in complexity. For fast fading channels, blind algorithms may require too many data to converge or converge too slowly to be practical. This research investigates a class of superimposed training algorithms that combine the best of both worlds. Instead of inserting pilots as in PSAM, known periodic pilots are superimposed onto the symbol stream, prior to waveform modulation. The channel estimation algorithm may appear to be blind, but is superior to blind approaches because simple first-order, and no more than second-order statistical methods, are used for channel estimation. The superimposed training approach can also be viewed as a watermarking technique with the pilot sequence serving as the marker. It is shown to be applicable to a variety of wireless channels: frequency selective, time selective, and doubly selective fading channels, as well as nonlinear channels. Issues investigated include trade-offs among bit error rate, transmission power, information rate and capacity, as well as optimization of pilot strength, transmission power allocation, and the effect of the added pilots on nonlinear power amp! lifiers.