Robust Joint Optimization for Multi-Cell Backscatter Communication System: A Deep Reinforcement Learning Approach

Backscatter communication(BackCom) has emerged as an ultra-low power technology that enables battery-free devices for various Internet of Things scenarios. To improve the network performance, the resource allocation is important; however, most of prior works developed the strategy under the ideal case, i.e., the channel state information (CSI) is perfect, which can worsen the performance. Hence, in this paper, we investigate the robust resource allocation for a multi-cell BackCom system, where the imperfect CSI is considered. To maximize the sum rate of the considered system, a robust deep reinforcement learning-based algorithm is developed by jointly optimizing spectrum allocation, the reflection coefficients of tags, and the transmit power of readers. Specifically, the original non-convex optimization problem is transformed into a Markov decision process. Then, to effectively handle the discrete-continuous hybrid action space, we propose a robust multi-agent parametrized deep Q-network (MAPDQN) algorithm, where the artificial distorted CSI is introduced during the training phase to compensate for the error of imperfect CSI. The numerical results demonstrate the effectiveness and robustness of our proposed algorithm.