A Channel Prediction Scheme Under The Condition of Incomplete History

In massive multiple-input multiple-output (MIMO) systems, when user movement causes serious channel aging, it requires the use of channel prediction to solve the problem that channel estimation cannot obtain real-time channel state information (CSI). However, the existing work is based on the assumption of obtaining a perfect historical sequence, without considering the problem of incomplete history caused by pilot signal spacing. The method of filling in the incomplete value with the predicted value will cause the prediction error to accumulate and affect the final performance. In this paper, we propose an interpolation scheme based on the variational autoencoder (VAE) to generate incomplete value that conforms to the distribution of historical sequence data, thereby restoring the channel dynamic information in historical sequences. In addition, we design a prediction module based on complex residual network, which effectively captures the correlation between the real and imaginary parts of the CSI sequence as well as the spatiotemporal correlation. Experimental results verify the superiority of the proposed framework in prediction performance.