| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.52 MB | HTML |
Autores
Orientador(es)
Resumo(s)
Accurate end-to-end (E2E) delay prediction is critical for optimizing network performance and ensuring quality of service in 5G networks. This paper investigates two complementary qualitative E2E delay prediction methodologies: a Bayesian approach based on a Hidden Markov Model (HMM) and a deep learning approach using Long Short-Term Memory (LSTM) networks. The Bayesian framework exploits the sequential nature of delay data to infer probabilistic relationships and predict future delays via a Bayesian chain. In contrast, the LSTM model directly learns temporal dependencies from the data, capturing complex interrelations in the E2E delay series. A comprehensive evaluation is conducted using the 5G Campus dataset, covering multiple network scenarios. Results show that prediction accuracy depends on both the amount of prior data and the prediction horizon, with the Bayesian method achieving 30–92% accuracy. Across all scenarios, the LSTM consistently outperforms the Bayesian approach, highlighting its effectiveness in learning dynamic patterns in the data. Scenario-dependent factors, such as probing rate and network configuration, are also shown to significantly influence performance. Overall, the comparison demonstrates that while the Bayesian approach provides an interpretable probabilistic framework, the LSTM offers superior predictive performance, making it valuable for proactive management of E2E delay in 5G networks.
Descrição
Funding information:
This work was supported by FCT-Fundação para a Ciência e Tecnologia, I.P. under Project UID/50008/2025–Instituto de Telecomunicações and Project 2022.08786.PTDC.
Publisher Copyright:
© 2013 IEEE.
Palavras-chave
Bayesian networks End-to-end delay estimation range estimation General Computer Science General Materials Science General Engineering