SCAR: State-Space Compression for Scalable AI-Based Network Management of Vehicular Services

TL;DR

SCAR tackles the scalability bottleneck in AI-driven network management for high-rate vehicular CQI data by introducing a two-stage, edge-assisted compression pipeline. An offline SAST-enhanced K-means clustering defines CQI pattern centers, while an online RBFN classifier maps incoming CQI reports to compact center classes, enabling a fixed-dimensional state representation for a compression-aware RL manager guided by NGMN fairness. Key findings show that SCAR increases feasible-management-time by about 14% and reduces unfair allocation time by about 15% relative to RL with uncompressed state, with SAST reducing compression distortion by approximately 10%; the approach remains technology-agnostic and scalable across edge and cloud domains. Together, these results indicate that edge-based state abstraction coupled with compression-aware learning yields substantial gains in fairness, efficiency, and responsiveness for AI-assisted management in dynamic vehicular networks.

Abstract

The increasing demand for connected vehicular services poses significant challenges for AI-based network and service management due to the high volume and rapid variability of network state information. Traditional management and control mechanisms struggle to scale when processing fine-grained metrics such as Channel Quality Indicators (CQIs) in dynamic vehicular environments. To address this challenge, we propose SCAR (State-Space Compression for AI-Based Network Management), an edge-assisted framework that improves scalability and fairness in vehicular services through network state abstraction. SCAR employs machine-learning (ML)-based compression techniques, including clustering and radial basis function (RBF) networks, to reduce the dimensionality of CQI-derived state information while preserving essential features relevant to management decisions. The resulting compressed states are used to train reinforcement learning (RL)-based management policies that aim to maximize network efficiency while satisfying service-level fairness objectives defined by the NGMN. Simulation results show that SCAR increases the time spent in feasible management regions by 14% and reduces unfair service allocation time by 15% compared to reinforcement learning baselines operating on uncompressed state information. Furthermore, simulated annealing with stochastic tunneling (SAST)-based clustering reduces state compression distortion by 10%, confirming the effectiveness of the proposed approach. These results demonstrate that SCAR enables scalable and fair AI-assisted network and service management in dynamic vehicular systems.

Figures (5)

• Figure 1: Proposed SCAR Framework
• Figure 2: RL-Based Management System.
• Figure 3: The filtering concept of CQI centers through the: a) $N=2$ dimensional representation; b) KN-tree representation.
• Figure 4: The Error Distribution for the Gaussian Parameter ($\sigma$) when processing top a) $M=3$, b) $M=4$, c) $M=5$ CQI features. The Error Distribution for the Learning Rate ($\eta$) when processing top d) $M=3$, e) $M=4$, f) $M=5$ CQI features.
• Figure 5: Ablation Study on CQI Compression for CACLA2 and Performance Comparison With State-of-the-Art Resource Management Methods