State Aware Traffic Generation for Real-Time Network Digital Twins

2025-09-16

AI Summary

To address the digital twin’s (DT) demand for real-time, low-overhead network data provisioning, this paper proposes a lightweight traffic generator tailored for mobile network digital twins. The method integrates a Hidden Markov Model with a feed-forward Mixture Density Network to accurately capture dynamic transitions among buffering, streaming, and idle states, thereby synthesizing packet payloads and inter-arrival times that preserve statistical fidelity, diversity, and temporal dependencies. Innovatively, it introduces a state-aware compact architecture enabling millisecond-scale in-twin fine-tuning—achieving real-time synchronization with live network traffic at zero additional overhead. Experiments demonstrate sub-second training on GPU; synthesized traffic closely matches ground-truth traffic across multiple metrics—including distributional accuracy, burstiness, and autocorrelation—thereby significantly enhancing the DT’s data-driven modeling capability and self-optimization performance.

Technology Category

Natural Language Processing: GenerationMachine Learning: Deep Generative Models & AutoencodersSearch and Optimization: Sampling/Simulation-based Search

Application Category

Web Mining and Content Analysis: Web data generation and simulationSystems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applicationsSocial Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Digital twins (DTs) enable smarter, self-optimizing mobile networks, but they rely on a steady supply of real world data. Collecting and transferring complete traces in real time is a significant challenge. We present a compact traffic generator that combines hidden Markov model, capturing the broad rhythms of buffering, streaming and idle periods, with a small feed forward mixture density network that generates realistic payload sizes and inter-arrival times to be fed to the DT. This traffic generator trains in seconds on a server GPU, runs in real time and can be fine tuned inside the DT whenever the statistics of the generated data do not match the actual traffic. This enables operators to keep their DT up to date without causing overhead to the operational network. The results show that the traffic generator presented is able to derive realistic packet traces of payload length and inter-arrival time across various metrics that assess distributional fidelity, diversity, and temporal correlation of the synthetic trace.

Problem

• Generating real-time synthetic network traffic data
• Reducing overhead from complete trace collection
• Ensuring digital twins receive accurate traffic updates

Innovation

• Hidden Markov model for traffic rhythms
• Mixture density network for payload sizes
• Real-time fine-tuning without network overhead