SVM multiregression for nonlinear channel estimation in multiple-input multiple-output systems

Matilde Sánchez Fernández, Mario de Prado-Cumplido, Jerónimo Arenas-García, Fernando Pérez-Cruz

Abstract: This paper addresses the problem of multiple-input multiple-output (MIMO) frequency nonselective channel estimation. We develop a new method for multiple variable regression estimation based on Support Vector Machines (SVMs): a state-of-the-art technique within the machine learning community for regression estimation. We show how this new method, which we call M-SVR, can be efficiently applied. The proposed regression method is evaluated in a MIMO system under a channel estimation scenario, showing its benefits in comparison to previous proposals when nonlinearities are present in either the transmitter or the receiver sides of the MIMO system.