Performance Optimization for Interference Alignment Based on MMSE Prediction to Time-Variant Channels

—Interference Alignment (IA) is a precoding technique that achieves the maximum multiplexing gain over an interference channel when perfect Channel State Information (CSI) is available at transmitters. Most of IA researches assume channels remain static for a period but vary independently from block to block, which neglects the temporal correlation of timevariant channels. In this paper, we propose a novel scheme that transmitters utilize a number of samples to predict CSI instead of obtaining CSI through feedback all the time. By making full use of the correlation of time-variant channels, our proposed scheme is able to reduce overhead and compensate for the feedback error due to low feedback Signal-Noise Ratio (SNR). Furthermore, we find an optimized prediction horizon achieving the maximum sum rate of our system, which is the best tradeoff between prediction error and overhead length. Simulation results verify that our scheme outperforms the traditional nonpredictive feedback scheme.