Distributed orthogonalization in large interference relay networks

We study fading interference relay networks where M single-antenna source-destination terminal pairs communicate through a set of K relays using half-duplex two-hop relaying. Two specific protocols are considered, P1 introduced in H. Bolcskei, et al. (2004), H. Bolcskei and R.U. Nabar (2004) and P2 introduced in A.F. Dana and B. Hassibi (2003). P1 relies on the idea of relay partitioning and requires each relay terminal to know one backward and one forward fading coefficient only. P2 requires each relay terminal to know all M backward and M forward fading coefficients and does not need relay partitioning. We prove that in the large-M limit the minimum rate of growth of K for P1 to achieve a strictly positive per source-destination terminal pair capacity is K infin M³ whereas in P2 it is K infin M². The protocols P1 and P2 are thus found to trade off the number of relay terminals for channel state information (CSI) at the relays; more CSI at the relays reduces the total number of relays needed to achieve a strictly positive per source-destination terminal pair capacity in the large-M limit