Optimization of training-based channel estimation for OFDM/OQAM systems operating on highly time-frequency dispersive channels

Offset QAM orthogonal frequency division multiplexing (OFDM/OQAM) is a promising technique for high data rate next generation cellular systems, operating on highly time-frequency dispersive channels. To combat dispersive channel interference, OFDM/OQAM systems allow very well time and frequency localized prototype functions. The good localization of the prototype function results from the separation of the transmission of the in-phase and quadrature phase components. Unfortunately, this separation introduces a crosstalk between these components when imperfect channel estimation is used to demodulate transmitted data. The main goal of this paper is the enhancement of channel estimation quality for OFDM/OQAM systems through the minimization of this crosstalk. Our approach consists in introducing a number of optimum linear constraints on neighbors of pilot symbols used to estimate the channel. The optimality of these components is related to a minimization of the residual interference mean power. In our study, for the representation of the signal space of our OFDM/OQAM system, we consider both square and hexagonal lattices in the time-frequency plane.