Deep Reinforcement Learning-Based Intelligent Network Slicing for Dynamic Resource Allocation in Beyond-5G and 6G Networks

Abstract

Beyond fifth-generation and sixth-generation mobile systems must support heterogeneous services over shared radio, transport,
edge, and core resources. Static slicing cannot respond reliably to bursty demand, changing radio channels, mobility, and
service-level agreements. This study develops a constraint-aware deep reinforcement learning framework that jointly adapts
resource shares, admission decisions, and migration priorities for enhanced mobile broadband, ultra-reliable low-latency
communication, and massive machine-type communication. A centralized orchestration agent observes traffic forecasts, queue
states, link capacity, energy costs, and slice-level service deficits; a proximal policy optimization controller produces candidate
allocations that are projected onto a feasible resource region. The methodology specifies a Markov decision process, normalized
multi-objective reward, safety filter, training protocol, and ablation study. To avoid reporting unperformed experiments as
findings, numerical results are explicitly labeled illustrative scenario calculations and constitute a reporting template rather
than measured performance. The proposed framework offers a testable design for investigating service compliance, spectral
utilization, energy efficiency, and operational stability in 6G-ready networks.