Impact of CSI on distributed space-time coding in wireless relay networks

We study the impact of channel state information at transmitter (CSIT) in a wireless relay network in which one transmitter-receiver pair communicates via M relays. Applying a distributed linear dispersion code together with amplifyand-forward protocol to relays, we aim to find the amplifier coefficients such that pairwise error probability conditioned on CSIT is minimized. Under perfect CSIT, we propose an on-off gradient algorithm that finds efficiently the set of relays to switch on. For M = 2 we provide an explicit condition for which each relay is activated. Under partial CSIT, i.e. the transmitter knows the transmitter-relay channels perfectly but the relay-receiver channels statistically, a simple waterfilling solution is derived. Our main finding is that as relays get closer to the transmitter the usefulness of CSIT decreases, while as relays get closer to the receiver the best strategy becomes relay selection that enables to reduce the noise amplified at the relays.