On linear precoding of non-regenerative MIMO relays for QAM inputs

Recent works have established that MIMO systems optimized for Gaussian source signals may suffer unexpected performance loss when practical inputs are in fact discrete QAM sources. There is a practical need in the optimization of MIMO related systems of various networking scenarios to specifically target source signals of finite QAM alphabet. In this work, we investigate the precoding optimization of wireless two-hop non-regenerative three-node MIMO relay networks driven by finite-alphabet inputs. Exploiting a known optimal structure for the precoder at relay and a special convexity property, we propose an iterative two-step numerical optimization algorithm. This algorithm is a general solution, not only for arbitrary source signals but also for cooperative networks with or without direct link. Simulation results demonstrate substantial performance improvement by the new precoder over precoders optimized under the Gaussian input assumption.