Abstract: Fast fading of propagating channels degrades the performance of multiple-input multiple-output (MIMO) orthogonal frequency-division multiplexing (OFDM) systems. The theoretical benefits of MIMO -OFDM systems may not be fully achieved in broadband high mobile applications because the channels are both rapidly time-varying and frequency-selective. The theory of compressed sensing (CS) shows that fast fading channels in high-dimensional spaces can be recovered from a relatively small number of random plots. The relatively nonzero channel coefficients are tracked by random pilots at a sampling rate significantly below the Nyquist rate. However, existing CS-based channel estimations require the sparsity as a prior for exact recovery. The sparsity of fast fading channels could not be well-defined. In actual fast fading environments, the MIMO-OFDM channels are often frequency selective and time-varying delay. The sparsity of fast fading channel Impulse responses on MIMO-OFDM the system is unknown. A sparsity adaptive estimation of fast fading MIMO-OFDM channels based on CS is given in this paper. The sparsity adaptive estimation enjoys a potentially higher sparsity level from transmit-receive antennas and multi-symbol processing. The multi-antenna structure and sparse time-frequency basis is constructed. The time-varying channel impulse responses within the multiple antennas and group OFDM symbols have a more sparse nature. The fast fading channel coefficients within the multiple antennas and group OFDM symbols can be represented by a few coefficients, which reduces the number of channel measurements. The fast fading channels are estimated by a sparsity adaptive compressive sensing technique without prior information of the sparsity, when the channel sparsity is rapidly varying and not available in MIMO-OFDM systems. The simulation results show that the new channel estimator can provide a considerable performance improvement in estimating fast fading channels. The proposed estimation method for fast time-varying channels has strong robustness and high spectral efficiency, and small the mean square error.