Channel prediction algorithm based on high order singular value decomposition (HOSVD) denoising is sensitive to the small number of antennas and cannot cope with the maximum Doppler shift, which may result in a sharp drop in the channel estimation and prediction performance and the loss of channel capacity. To solve this problem, this paper proposes an improved channel prediction algorithm based on HOSVD denoising. The proposed algorithm first smoothes the sampled channel state information (CSI) by using the space-time correlation inherent in the multiple-input multiple-output channel, and then uses HOSVD to reduce the influence of noise and reconstruct the channel matrix based on the multi-dimensional structure characteristics of the channel. Finally, the recursive least squares filter is used to predict future CSI. Simulation results show that the channel estimation error and the prediction error of the proposed algorithm were better than those of the comparison algorithm. This is because the proposed algorithm uses matrix rearrangement and space-time smoothing to increase the number of antennas virtually, which can effectively reduce the influence of the Doppler shift factor and the rank missing problem caused by the small number of antennas on the prediction accuracy, thus compensating the loss of channel capacity. At the same time, by comparing the influence of antenna number and the Doppler frequency shift on the performance of different algorithms, it can be found that the proposed algorithm could provide better prediction performance and channel capacity under unfavorable conditions such as Doppler frequency shift and few antennas, which has certain advantages.