Abstract: A low rate CSI (Channel State Information) feedback scheme based on linear prediction is proposed for closed-loop multi-user MIMO-OFDM systems. The subcarriers in OFDM system are divided into several sub-bands according to the coherence bandwidth. MS (Mobile Station) predicts CSI of each sub-band, then feedbacks the quantized and coded prediction error to BS (Base Station). BS recovers CSI from feedback information with the same linear predictor, and performs multi-user spatial multiplexing by zero-forcing beamforming with recovered CSI. The downlink capacities are analyzed in theory, where water-filling power allocation method is assumed. Numerical simulations verify that, BS can recover CSI precisely even if the feedback data (real part or imaginary part) is quantized only by one bit. Compared with the 3GPP LTEs feedback scheme based on codebook, the proposed scheme can suppress CCI better, and achieve much larger capacity. Meanwhile the feedback payload is same as that of 3GPP LTEs scheme.