DYNAMIC SLICE NETWORK MANAGEMENT METHOD IN 5G FOR QOS ENHANCEMENT IN VOIP

Abstract

Modern IP telephony networks face numerous challenges related to ensuring guaranteed Quality of Service (QoS) for voice traffic. Dynamic changes in load, latency, jitter, and packet loss significantly impact the quality of voice calls, which is particularly critical for VoIP systems widely used in corporate networks, operator infrastructures, and 5G technologies. Traditional QoS mechanisms, such as DiffServ and MPLS, provide basic traffic routing and prioritization but are unable to quickly adapt to changing network conditions. This leads to degraded VoIP service performance, especially in high-load networks and scenarios involving Network Slicing in 5G.

In this regard, the application of machine learning (ML) methods for QoS state prediction and dynamic slice network management is becoming increasingly relevant. This study proposes a novel method utilizing LSTM networks to predict QoS parameters and Reinforcement Learning for optimal slice selection. This approach enables latency minimization through predictive traffic redirection, reduces jitter and packet loss to ensure voice connection stability, and automates network management, decreasing the need for manual QoS policy configuration.

The key research results confirm the effectiveness of the proposed method. A predictive algorithm for key QoS metrics was developed based on machine learning techniques, enabling the early detection of potential issues. The proposed adaptive network management algorithm optimizes bandwidth allocation and traffic prioritization, significantly improving service quality for users. Experimental analysis demonstrated that the proposed approach outperforms traditional methods such as DiffServ and MPLS, reducing average latency by 25–30% and enhancing voice communication stability.

The findings of this study can be used for further optimization of IP telephony networks and improving real-time resource management mechanisms. Future research directions include integrating the proposed method with cloud-based VoIP solutions and extending the model using deep learning techniques to enhance the accuracy of QoS parameter prediction.