Cyclic Interference Alignment and Cancellation in 3-User X-Networks with Minimal Backhaul

We consider the problem of Cyclic Interference Alignment (IA) on the 3-user X-network and show that it is infeasible to exactly achieve the upper bound of $\frac{K^2}{2K-1}=\frac{9}{5}$ degrees of freedom for the lower bound of n=5 signalling dimensions and K=3 user-pairs. This infeasibility goes beyond the problem of common eigenvectors in invariant subspaces within spatial IA. In order to gain non-asymptotic feasibility with minimal intervention, we first investigate an alignment strategy that enables IA by feedforwarding a subset of messages with minimal rate. In a second step, we replace the proposed feedforward strategy by an analogous Cyclic Interference Alignment and Cancellation scheme with a backhaul network on the receiver side and also by a dual Cyclic Interference Neutralization scheme with a backhaul network on the transmitter side.