Goal-Directed Service Provisioning and Value-Oriented Multi-Dimensional Resource Orchestration for Future Wireless Networks

#Multi-dimensional-multiple-access-(MDMA)-technique #multi-dimensional-resource-orchestration #quality-of-service-(QoS)-provisioning #resource-utilization-cost #multi-access-point-(AP)-coordination #multi-agent-deep-reinforcement-learning-(MADRL)

The rapid evolution of wireless communication toward beyond-6G (B6G) and next generation Wi-Fi networks is driving a fundamental transformation in how radio resources are perceived, managed, and optimized. As future networks are envisioned not only as communication infrastructures but as intelligent, multi-functional ecosystems, such paradigm shift introduces unprecedented challenges of multi-dimensional resource management including spectrum scarcity, severe spatial-temporal interference coupling, device heterogeneity and dynamically evolving service requirements. Therefore, this thesis establishes unified theoretical and algorithmic frameworks for goal-directed, value-oriented, and learning-empowered resource orchestration and service provisioning to support future intelligent wireless networks.

    To accommodate heterogeneous device capabilities and diverse quality-of-service (QoS) requirements, a user-centric multi-dimensional multiple access (MDMA) framework is proposed by introducing a novel concept of resource utilization cost to quantify the energy, processing, and computational complexity associated with non-orthogonal interference cancellation procedure. By formulating user-specific situation-dependent cost-gain utilities, a max-min optimized solution is derived, achieving finer service provisioning granularity and higher inter-user fairness than state-of-the-art schemes.
    
    Building upon this foundation, a novel demand-aware prioritization mechanism and diversity, equity, inclusion (DEI)-based resource allocation scheme are proposed. By incorporating demand-aware priorities for individual users, the DEI-based metric evaluates the weighted mean-variance tradeoff of network-wide user utilities. The proposed algorithms can inclusively support differentiated service categories with higher average utility and smaller inter-user disparity verified extensive simulations. The DEI method can also adaptively accommodate and prioritize diverse QoS demands based on individualized service requirements and dynamic resource conditions.

    To address the growing complexity of coupled spatial-temporal interference and distributed resource reusing in ultra-dense Wi-Fi networks, a novel multi-access point (AP) multi-dimensional resource coordination mechanism is developed. The proposed method amalgamates coordinated spatial reuse (Co-SR) and coordinated orthogonal frequency division multiple access (Co-OFDMA) by dynamic channel access sensitivity configuration, resource unit (RU)-level interference assessment and reusability evaluation. Simulations validate that the proposed multi-agent deep reinforcement learning (MADRL) method can opportunistically coordinate multi-dimensional resources of each AP for diverse QoS provisioning, outperforming baselines by effective interference mitigation with fairness guarantees.

    Finally, a goal-directed, contextually adaptive multi-AP coordination is developed using an hierarchical learning framework. The meta-level performs coarse-timescale coordination goal adaptation using semi-Markov proximal policy optimization (PPO) algorithm, while the execution level solves fine-timescale coordinated resource decisions via a goal-conditioned MADRL model. An alternating joint-training strategy is introduced to reconcile timescale disparity and coupled non-stationarity, enabling adaptive and context-consistent network intelligence that aligns local actions with evolving global intents.