Hierarchical Digital Twin for Efficient 6G Network Orchestration via Adaptive Attribute Selection and Scalable Network Modeling

TL;DR

The paper tackles the challenge of efficiently orchestrating highly heterogeneous 6G HetNets under data and computation constraints. It introduces a hierarchical digital twin framework with HDT for rapid problem identification and LDT for fine-grained, scenario-specific modeling, enabled by adaptive attribute selection. Key contributions include a value-oriented attribute differentiation scheme using $SampEn$ and correlation/causality analyses, a multi-level virtual-physical domain synchronization scheme, and STL-NARX-based fine-grained modeling for selected target areas. Simulations demonstrate that the proposed approach achieves higher orchestration efficiency and timely decisions with reduced data transmission compared to all-inclusive digital twin methods, underscoring its practical impact for scalable 6G network management.

Abstract

Achieving a holistic and long-term understanding through accurate network modeling is essential for orchestrating future networks with increasing service diversity and infrastructure complexities. However, due to unselective data collection and uniform processing, traditional modeling approaches undermine the efficacy and timeliness of network orchestration. Additionally, temporal disparities arising from various modeling delays further impair the centralized decision-making with distributed models. In this paper, we propose a new hierarchical digital twin paradigm adapting to real-time network situations for problem-centered model construction. Specifically, we introduce an adaptive attribute selection mechanism that evaluates the distinct modeling values of diverse network attributes, considering their relevance to current network scenarios and inherent modeling complexity. By prioritizing critical attributes at higher layers, an efficient evaluation of network situations is achieved to identify target areas. Subsequently, scalable network modeling facilitates the inclusion of all identified elements at the lower layers, where more fine-grained digital twins are developed to generate targeted solutions for user association and power allocation. Furthermore, virtual-physical domain synchronization is implemented to maintain accurate temporal alignment between the digital twins and their physical counterparts, spanning from the construction to the utilization of the proposed paradigm. Extensive simulations validate the proposed approach, demonstrating its effectiveness in efficiently identifying pressing issues and delivering network orchestration solutions in complex 6G HetNets.

Figures (8)

• Figure 1: The overall architecture of 6G HetNets orchestrated by the proposed hierarchical digital twin paradigm: a). The overall 6G HetNets, where macro-cell and small-cell BSs serve the distributed users to accommodate their QoS requirements. Unsatisfied QoS is caused by unbalanced traffic load distribution and insufficient network capacity for local cells. b). Hierarchical digital twins for efficient network orchestration, comprising higher layers for network problem identification and lower layers for problem-focused network management.
• Figure 2: Higher-layered digital twins for network situation evaluation: a). Differentiation and selection of various network attributes based on the cost and benefit of constructing digital twins. b). The overall situation in terms of the network capacity from the perspectives of BSs and user activities. c). Network segmentation according to the historical user activity and potential connections. d). Target areas in each network segment based on the expected network orchestration efficiency.
• Figure 3: Illustration of the accurate digital twin construction in lower layers to support network orchestration: (a). Device-level data integration consists of data normalization, synchronization, and resampling to unify all data in the temporal domain. (b). Fine-grained modeling based on the integrated data using STL-NARX for individual or fused digital twins. (c). Cross-domain synchronization for all digital twins to address the model misalignment induced by multi-source modeling delays.
• Figure 4: Identification of the target area in higher-layered digital twins based on differentiated network attributes and network situation analysis: (a). Modeling cost evaluated by sample entropy for different users and attributes. (b). Modeling benefit evaluated by correlation analysis and causality inference compared to the network objective. (c). Four-level attribute differentiation based on modeling value. (d). Segmentation of the HetNets into four sub-networks based on user activity. (e). Real-time network situation analysis of one sub-network, including traffic load of BSs and user QoS satisfaction. (f). Target area identification based on orchestration outcome for the selected sub-network.
• Figure 5: The analysis of the four attributes with STL, which decomposes data into trend, seasonality, and residual.