AMSA: An Adaptive ML-Based Synchronization Arbiter for Resilient Time Distribution in 5G and Beyond Networks

Abstract

Fifth-generation (5G) and emerging 6G mobile networks impose stringent timing requirements (|TE| ≤ ±1.5 μs per ITU-T G.8271.1) that the static Best Master Clock Algorithm (BMCA) of IEEE 1588 cannot reliably guarantee under real-world conditions involving packet delay variation (PDV), Global Navigation Satellite System (GNSS) outages, and signal spoofing. This paper presents AMSA (Adaptive ML-based Synchronization Arbiter), a four-stage intelligent software overlay that replaces the deterministic BMCA with a data-driven, context-aware time source selection mechanism. The proposed architecture integrates Gamma-distributed PDV filtering, polynomial-regression-based holdover prediction (termed Virtual Rubidium), trust-score fault detection, and a Random Forest classifier trained on a hybrid dataset of operational telemetry and synthetic anomalies for adaptive source arbitration. AMSA is validated through a Monte Carlo simulation framework across four distinct operational scenarios and corroborated with field measurement data from an operational 5G transport network. Results demonstrate that AMSA reduces the 95th-percentile time error by 55% compared to standard BMCA, eliminates all 209 ITU-T G.8271.1 mask violations observed under high-PDV conditions, extends holdover autonomy by over 30%, and detects covert GNSS spoofing within 250 seconds. Field validation across three scenarios achieves 100% agreement with independent expert assessments, including correct classification of out-of-distribution anomalies.