Resource Allocation and Computation Offloading in Ultra-reliable Low-latency Communication Systems via Deep Reinforcement Learning

In this work, we study a joint resource allocation and computation offloading optimization scheme in the uplink ultra-reliable low-latency communication (URLLC) system. To guarantee the reliability performance, the average error probability minimization optimization problem is formulated under the delay and power constraints. Due to the non-convex optimization problem, a multi-agent deep Q-network-based scheme is proposed to optimize the blocklength, transmit power, and offloading scheme jointly, in which the agent performs the offloading scheme to process data via the computation offloading or the local computing. The proposed scheme constructs the reward function depending on the accumulated error probability under the delay and maximum error probability constraint which is applied to the high-dimension action space. Simulation results show that the proposed scheme achieves better performance and guarantees the targeted error probability compared with the benchmark schemes.