Ensuring Reliable 5G Networks with DRL

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Jul 12 2024

In a paper published in the journal Scientific Reports, researchers addressed the reliability mapping of 5G low orbit constellation network slices to ensure stable link network communication. They tackled state space explosion using deep reinforcement learning (DRL), considering both resource requirements and constraints of virtual network functions (VNF).

Node and link backup strategies were implemented to achieve VNF/link reliability. Experiments showed that this approach improved network throughput, reduced packet loss, and repaired network faults within 0.3 seconds. The method maintained high reliability and kept network delay low.

Background

Past work on 5G low orbit constellation network slices includes methods to optimize network performance but faces scalability and computational complexity challenges. Network slicing, which divides a physical network into virtual ones, is crucial for 5G low-orbit constellation networks, ensuring service reliability through backup and recovery mechanisms.

The dynamic nature of these networks presents challenges and opportunities, with DRL algorithms enhancing resource allocation and routing strategies. A delay-driven slicing method, relying on real-time network data, highlights the need for accurate status information. By learning from data and interactions, DRL improves network slicing reliability, automates network management, and reduces manual intervention.

5G Network Reliability

The integrated network architecture of the 5 G low-orbit constellation combines terminal equipment, satellite network functions, satellite baseband gateway (S-GW), mobility management entity (MME), core network resources, and a management system using software-defined networking (SDN) and network function virtualization (NFV) technologies. It separates the core network into processing and forwarding clouds, providing flexibility and programmability. The MME handles signaling, business control, and user mobility management, while the S-GW manages request routing, security control, and protocol conversion to ensure network reliability and security.

In the European Telecommunications Standards Institute Industry Specification Group on NFV (ETSI ISG NFV) terminology, network slices consist of VNFs and physical NF (PNFs) linked to form a service function chain (SFC). Service choreographers generate service function chains (SFCs) based on service requests, utilizing mapping algorithms for network slice instantiation. These slices create isolated virtual networks on satellite infrastructure. RL faces challenges like data collection, real-time demands, balancing network performance, and ensuring security and stability.

Service orchestration in network architecture manages and optimizes slicing resources, typically deployed in central control nodes like network management system (NMS), network function virtualization orchestrator (NFVO), or network slice manager (NSM). NMS provides a global view and manages network slice resources, while NFVO handles the life cycle of network services. NSM directly controls and manages slicing resources. The deployment location affects transmission delay, with NMS offering low delay for core network slices, NFVO for virtualized functions, and NSM for rapid slice resource management.

The physical network supports 5G low-orbit constellation network slices using weighted undirected graphs of nodes and links. Virtual elements in network slice requests have specific computational and bandwidth requirements mapped to physical nodes through directed subgraph models.

DRL optimizes resource allocation and routing based on variables like load, delay, packet loss rate, and satellite positions, enhancing reliability. This approach ensures robustness by employing deep Q-network (DQN) -based methods with neural networks and experience replay to manage state-space complexities, complemented by adaptive node and link backup schemes for resilience against failures.

Validation of Network Reliability

The experimental validation of the reliability mapping method for 5G low orbit constellation network slices utilized a simulated environment, leveraging the cloud simulation for software-defined networking (CloudSimSDN) platform. This platform created a realistic network topology with 45 interconnected nodes arranged in a tree structure.

Various failure scenarios, including network equipment failures, satellite communication issues, and network congestion, were explored through simulations. The team monitored key metrics such as throughput, packet loss rate, and intra-slice traffic to assess the method's efficacy in swiftly recovering from failures and maintaining operational efficiency. The results demonstrated that the proposed method achieved rapid recovery times of less than 0.3 seconds with a 100% success rate in restoring normal network operations post-failure.

Comparative analyses with alternative methods underscored the superior performance of the proposed approach in managing increasing network slicing requests and accommodating SFCs of varying lengths (10 to 30) and up to 10 types of VNFs. These evaluations highlighted the method's capability to optimize network reliability and minimize delay under dynamic operational conditions. Overall, the findings suggest significant potential for enhancing the reliability and performance of 5G low orbit constellation networks, thereby contributing to advancements in future network architecture designs and deployments.

Conclusion

To sum up, the study addressed shortcomings in traditional network slice mapping by introducing a DRL approach tailored to the reliability needs of VNFs in 5G low orbit constellation networks. This method significantly improved network throughput, reduced traffic delay and packet loss rates, and optimized resource allocation for enhanced reliability and load balance. As network scale increased, the approach efficiently managed larger state spaces with high-performance computing resources like clusters or cloud platforms, ensuring scalability and effective strategy implementation.