Deep Learning LMMSE Joint Channel, PN, and IQ Imbalance Estimator for Multicarrier MIMO Full-Duplex Systems

This letter investigates joint estimation of the channel, phase noise (PN), and in-phase (I) and quadrature-phase (Q) imbalance in multicarrier MIMO full-duplex wireless systems. We approximate the time-varying channels with a basis expansion model (BEM) to reduce the number of unknowns. We then propose a pilot-aided linear minimum mean-squared error (LMMSE) estimator for the BEM coefficients. To improve its performance, we develop a deep learning (DL) network. The DL network is trained offline by using simulation data and then deployed for online estimation. The numerical results illustrate that the proposed DL-LMMSE estimator outperforms conventional estimators, such as maximum-a-posteriori (MAP) in terms of the mean-squared error (MSE).