Interference Alignment for Time-Varying Channel with Channel and Weight Predictions Based on Auto Regressive Model

In interference alignment (IA), interference signals are aligned in a certain signal subspace of each receiver using transmit weights and eliminated using receive weights. The transmit and receive weights are typically calculated at transmitters with an iterative optimization based on estimated channel state information (CSI) from receivers. However, the estimated CSI is different from the channel when a signal is transmitted by the transmitter, because the channel varies during a CSI feedback and the iterative weight calculation. In IA with the weights based on the estimated CSI, the interference signals are not perfectly aligned and remains at the receiver. As a result, a rate decreases due to the remaining interference signals. In this paper, we propose three IA methods having the capability of adapting to time-varying channel using auto regressive (AR) model by which next state is predicted with some past states. In the first method, the channel when a signal is transmitted at the transmitter is predicted from the estimated CSI at the receiver based on AR model. Transmit and receive weights are calculated with the predicted CSI so that the interference signals are aligned and eliminated. In the second method, transmit and receive weights are predicted with past transmit and receive weights based on AR model, respectively. In the third method, only the transmit weight is predicted with past transmit weights based on AR model and receive weight is calculated with the estimated CSI. The prediction methods are able to calculate transmit and receive weights with low amount of calculation compared with the iterative optimization. Through computer simulation, compared with general IA methods that do not use prediction process, the proposed ones are shown to improve the rate irrespective of signal to noise power ratio (SNR) and decrease the calculation amount.