Retrospective Interference Alignment for the $K$ -User $M\times N$ MIMO Interference Channel

Abstract

In this paper, we consider a $K$ -user $M\times N$ MIMO interference channel, with outdated channel state information at the transmitters (CSIT). Based on the technique introduced by Cadambe and Jafar, for the IC with perfect and up-to-date CSIT, we propose and analyze a scheme that guarantees interference alignment with outdated CSIT. The interference is aligned by the transmitters, and removed by the receivers using zero-forcing by Cadambe and Jafar. However, this is not feasible for the IC with outdated CSIT, since there is no up-to-date CSIT to compute the interference alignment precoders. To overcome this limitation, we propose that the interference should be aligned by the receivers instead. With this approach, the interference caused by a transmitter spans the same subspace at all receivers, i.e., the interference at receiver $k \neq m$ may be obtained through a linear map from the interference at receiver $m$ . Therefore, one interference retransmission per transmitter is enough to cancel all the interference. This reduces the overhead required for interference removal from $K(K-1)$ to $K$ interference terms.