Refined-Deep Reinforcement Learning for MIMO Bistatic Backscatter Resource Allocation
Bistatic backscatter communication facilitates ubiquitous, massive connectivity of passive tags for future Internet-of-Things (IoT) networks. The tags communicate with readers by reflecting carrier emitter (CE) signals. This work addresses the joint design of the transmit/receive beamformers at the CE/reader and the reflection coefficient of the tag. A throughput maximization problem is formulated to satisfy the tag requirements. A joint design is developed through a series of trial-and-error interactions within the environment, driven by a predefined reward system in a continuous state and action context. By leveraging recent advances in deep reinforcement learning (DRL), the underlying optimization problem is addressed. Capitalizing on deep deterministic policy gradient (DDPG) and soft actor-critic (SAC), we proposed two new algorithms, namely refined-DDPG for MIMO BiBC (RDMB) and refined-SAC for MIMO BiBC (RSMB). Simulation results show that the proposed algorithms can effectively learn from the environment and progressively improve their performance. They achieve results comparable to two leading benchmarks: alternating optimization (AO) and several DRL methods, including deep Q-network (DQN), double deep Q-network (DDQN), and dueling DQN (DuelDQN). For a system with twelve antennas, RSMB leads with a %26.76 gain over DQN, followed by AO and RSMB at 23.02% and 19.16%, respectively.