Title:A Constrained Deep Reinforcement Learning Optimization for Reliable Network Slicing in a Blockchain-Secured Low-Latency Wireless Network

Abstract:Network slicing (NS) is a promising technology that supports diverse requirements for next-generation low-latency wireless communication networks. However, the tampering attack is a rising issue of jeopardizing NS service-provisioning. To resist tampering attacks in NS networks, we propose a novel optimization framework for reliable NS resource allocation in a blockchain-secured low-latency wireless network, where trusted base stations (BSs) with high reputations are selected for blockchain management and NS service-provisioning. For such a blockchain-secured network, we consider that the latency is measured by the summation of blockchain management and NS service-provisioning, whilst the NS reliability is evaluated by the BS denial-of-service (DoS) probability. To satisfy the requirements of both the latency and reliability, we formulate a constrained computing resource allocation optimization problem to minimize the total processing latency subject to the BS DoS probability. To efficiently solve the optimization, we design a constrained deep reinforcement learning (DRL) algorithm, which satisfies both latency and DoS probability requirements by introducing an additional critic neural network. The proposed constrained DRL further solves the issue of high input dimension by incorporating feature engineering technology. Simulation results validate the effectiveness of our approach in achieving reliable and low-latency NS service-provisioning in the considered blockchain-secured wireless network.