Enabling Combined Relay Selection in Stochastic Wireless Networks by Recurrent Neural Computing

Multi-carrier relay selection is of particular interest and challenge due to the spatio-frequency coupling and the dynamics of available spectral and relay resources. Among a number of promising relay selection schemes, combined relay selection stands out as an equilibrium between system complexity and reliability. Recent research progress has witnessed the capability of neural computing as a powerful tool to efficiently realize combined relay selection for a given network topology where the number of relays and their locations are fixed and known. However, for contemporary wireless networks that are highly dynamic, the classic neural computing methods can hardly help out because of the scale drifts of input and output matrices. To enable multi-carrier combined relay selection in stochastic wireless networks (SWNs) where the number of available relays for selection could vary, we propose a recurrent neural network (RNN) based framework and devise several training methods suited for various application scenarios. In addition, we conduct a set of computer experiments to verify the effectiveness and efficiency of the proposed RNN-based framework compared with several baselines. With the obtained experimental results, we also evaluate and discuss the proposed framework's reliability, generalization ability, and the robustness against imperfect channel state information (CSI).