Distributed gain matrix optimization in non-regenerative MIMO relay networks

We consider a half-duplex multiple-input multiple-output relay network with multiple source and relay nodes and a set of collocated destination antennas. Each relay is equipped with multiple antennas and is constrained to map its receive signal linearly to its transmit signal. We devise a gradient based scheme that enables each relay to locally optimize its gain matrix with respect to the achievable sum-rate without requiring explicit knowledge of all fading coefficients characterizing the channel state. In particular, we demonstrate by numerical evidence that the overhead due to exchange of channel state information required by the scheme does either not scale at all or, if so, very slowly with the number of relay nodes or antennas therein.