LAUSR

Backscatter-assisted wireless-powered relaying has emerged as a promising approach to extend the transmission range of energy-constrained devices, e.g., sensors, with improved energy efficiency. Mode selection adapting to the dynamic channel conditions has a great impact on the performance of this hybrid relaying paradigm. However, designing an effective mode selection is challenging as the instantaneous channel state information (CSI) may not be available and is subject to errors and delays. Furthermore, instant CSI estimation incurs continual communication and computation overhead that burdens the operation of energy-constrained devices. This work considers the mode selection problem in a downlink cooperative scenario, wherein a hybrid relay aids the finite blocklength non-orthogonal transmissions from a source to a pair of destinations. To eliminate the need for instantaneous CSI estimation, we exploit the statistical system information to indicate the mode selection. In particular, we first derive the analytical expressions of average end-to-end block error rates in both active and passive relaying modes, based on which we can obtain the reward gap of the system. We then propose a bandit learning-based mode selection that exploits the reward gap to assess the upper confidence bound. The proposed mode selection method is light-weight and requires no knowledge of instantaneous CSI. Simulation results are presented to validate our analytical results and expose the efficiency of the proposed mode selection over conventional approaches. Besides, we demonstrate that the proposed approach is robust to estimation errors of the reward gap.