Data‐driven approach to design energy‐efficient joint precoders at source and relay using deep learning in MIMO‐CRNs

Abstract

AbstractThis article studies the problem of designing energy‐efficient joint precoder at source and relay for multiple‐input multiple‐output cognitive relay networks (MIMO‐CRNs). Existing optimization methods typically suffer from high computational complexity of finding the optimal solution for such nonconvex fractional programming problems. In contrast to prior works, this article considers a data‐driven approach to design the joint precoders using a deep neural network (DNN). The proposed DNN learns the optimal precoder weights on a set of different channel matrices during the offline training phase and allows the computational cost reduction in the online deployment phase. The numerical results demonstrate that this approach provides a comparable performance at significantly lower computational complexity in comparison with the conventional optimization‐based algorithm. Furthermore, it is shown that the proposed approach is quite robust against the variations in the channel statistics, which makes it suitable for real‐time implementation.