A recursive identification approach with gradient subspace tracking for MIMO-OFDM channels

In this paper, we investigate a recursive state-space modeling approach to represent multi-input multi-output orthogonal frequency division multiplexing (MIMO-OFDM) channels, which are considered as space-time finite impulse responses. The approach is proposed according to two assumptions. Firstly, the order of the channel system is considered as priori known. Secondly, the input-output data for identification are based on embedded pilot sequences. The approach is used to directly estimate of the subspace spanned by the column vectors of the extended observability matrix without performing singular value decomposition. In comparison with usual AR and ARMA models of MIMO-OFDM channels, the approach is immune to measurement noise so as to obtain the model precisely enough. Simulations also demonstrate the method can identify the channel response well when training data are sufficient.