International audienceThis paper proposes two new methods for identifying multiple-input-multiple-output (MIMO) nonlinear systems. Specifically, we are interested in radio over fiber (ROF) uplink communications channels. The electrical optical (E/O) conversion in such systems introduces relevant nonlinear distortion and, consequently, the overall channel can be viewed as a convolutive post nonlinear (CPNL) mixture. The proposed techniques for system identification exploit the use of orthonormal basis functions to accelerate the convergence of the least mean square (LMS) algorithm and to decouple the minimum mean squared error (MMSE) estimation in a block-processing scheme