Resource orchestration in network slicing using GAN-based distributional deep Q-network for industrial applications

The Industrial Internet of Things (IIoT) is an emerging and promising concept that allows intelligent manufacturing through the connectivity of 5G/6G and the interaction of industrial production units. The introduction of network slicing in 5G and beyond has made it possible to manage and allocate resources to various applications according to their requirements. In this paper, we study network slicing within a radio access network containing IIoT devices which include base stations that share the same physical infrastructure. We use deep reinforcement learning-based resource orchestration technique to achieve variable service demands of environment state-value and resource allocation as environment action-value. We describe the cognitive decision objectives to maximise the optimal policy for IIoT reward by achieving higher system throughput, spectral efficiency (SE), service level agreement (SLA), transmission packet rate with low power consumption and transmission delay. We use generative adversarial network-based deep distributional noisy Q-networks (GAN-NoisyNet) to learn the action-value distribution. Furthermore, we introduce dueling GAN-NoisyNet, which employs a duel generator that estimates the action advantage function and state-value distribution. Finally, we conduct extensive simulations to verify the performance of the proposed GAN-NoisyNet and dueling GAN-NoisyNet.