Achievable Rate Based Training Optimization for Time Varying Relay Networks

In time varying relay networks, imperfect channel estimation deteriorates the performance and must be accounted for system design. By doing this, this paper derives the achievable rate expressions first, and then studies the training optimization problem based on those rates for both Amplify-and-Forward (AF) and Decode-and-Forward (DF) relay networks. There are two kinds of training designs for relay networks. One is remaining the design just optimal for source-destination (SD) channel as if there is no relay node (we name it SD design). Another is optimizing over the whole relay network. The former way is simple but is shown to have a great degradation, while the latter one is optimal but too complex to get closed-form optimal solutions. In this paper, we simplify the latter way and propose new sub-optimal training designs in closed-forms. Simulations show that our proposed designs for both AF and DF can achieve considerable gains compared with SD design and have negligible performance degradation compared to the optimal design.