Reliable, low-latency communication in AI-empowered (private) 5G networks with shared infrastructure

This presentation will explore the challenges and innovations behind the 5GECO project, which aims to improve the management and optimization of 5G mobile networks. We will cover the integration of intelligent radio and transport edge networks to enhance performance and efficiency. Additionally, we will discuss the architectural advancements and practical implementations that enable seamless coordination between the radio access network (RAN) and transport networks, showcasing the potential for improved service quality and operational agility. Finally, we will highlight the project’s vision for creating a highly adaptive and intelligent 5G ecosystem, demonstrating how these innovations can meet the diverse and demanding requirements of next-generation mobile networks.

In the 5G Standalone (SA) era, Network Slicing enables diverse industrial applications, categorized into enhanced Mobile Broadband (eMBB), Ultra-Reliable Low-Latency Communication (URLLC), and massive Machine-Type Communications (mMTC). Despite 5G SA’s advancements, real-world trials reveal significant signal coverage limitations, affecting these services. While densifying base stations could help, it’s not sustainable. Network Sharing among operators offers a more viable solution to ensure service quality without escalating infrastructure costs. The 3rd Generation Partnership Project (3GPP) introduced initial network sharing concepts in Release 15, with further development in Release 19. We propose a new Network Sharing method that creates isolated virtual networks (slices) on shared infrastructure, ensuring resource isolation in a multi-operator environment. This approach is validated by a Proof of Concept (PoC) using Network Slicing and the Open Radio Access Network (ORAN) paradigm, demonstrating effective resource management by a neutral host across multiple telco operators.

The development in neutral hosting is a promising opportunity for network operators and industry. Private 5G deployments raise opportunities in the sense that owners can have big cost reduction both in CapEX and OpEX with shared infrastructure, maintenance and operation of the necessary equipment and licenses required to operate such networks. Citymesh explored deploying a test network in a dense environment, highlighting the possibilities created by this approach, and examined shared spectrum applications, challenges, and opportunities in Belgium. This talk will present a first hand view of this experience, its achievements and lessons learned.

In the era of fifth-generation (5G) cellular networks and beyond, network sharing has emerged as a promising approach to address the escalating demand for spectrum and infrastructure resources. Intelligent Neutral Host (INH) is an advanced network-sharing method facilitated by Open Radio Access Network (O-RAN) capabilities. This presentation will address the challenge of radio resource management in a multi-operator, multi-slice scenario. We will introduce AI-empowered algorithms using Q-learning and deep Q-learning for optimally allocating physical resource blocks (PRBs) while meeting diverse operator requirements. These algorithms are implemented as an xApp on the Colosseum platform, the world’s largest wireless research testbed for radio experiments. The algorithms introduce a dynamic resource allocation strategy that adheres to Service Level Agreement (SLA) constraints and optimizes real-time Key Performance Metrics (KPMs). We will assess the performance and efficacy of the algorithms in a complex traffic scenario to demonstrate how effectively they allocate resources among operators’ slices to satisfy their respective SLA while ensuring optimal resource utilization. Compared to traditional algorithms (round robin and proportional fair), the proposed algorithms significantly minimize SLA violations for eMBB and URLLC slices.

• Demo 1 – imec-IDLab-UGent:
  Multiple Operator Core Network (MOCN) Scenario Radio Resource Allocation xApp in Accelleran’s dRAX system
• Demo 2 – Accelleran:
  Efficient RAN Management and Optimization through a Cloud-Native Open RAN Platform (poster presentation)
  
• Demo 3 – imec-IDLab-UAntwerp:
  Optimizing Radio Resource Allocation in 5G using Transport-Aware Intelligent RAN
• Demo 4 – Nokia Bell Labs:
  RT-Prague Interactive 360-degree Video over Low-Latency Low-Loss Scalable Throughput (L4S)
• Demo 5 – imec-IDLab-UAntwerp:
  ML Pipelines to support lifecycle management of Virtual Network Functions

The main challenge for Intelligent neutral host is to control and ensure capacity per PLMN and per Slice. In 5G ECO Accelleran provide solution based on disaggregated RAN components  (RIC / CU / DU / RU- 7.2 split). In 5GECO we introduce bottom-up approach to fit INH requirement. Accelleran developed E2RC service model to ensure Dynamic radio resource allocation between slices. This solution can help neutral host owner to control and ensure capacity per tenant and ensure SLA requirement per user. Accelleran delivered setup with can allocate PRB dynamically per slice (Dedicated, Prioritized and Shared resources). With 5G ECO project Accelleran start development of our own commercial DU with capabilities to satisfy 5G ECO requirement. Accelleran will talk about experience and lessons we learned.

The advent of 5G necessitates a fundamental transformation in network and service management to address the increased complexity and operational agility required for new business opportunities like network slicing. Achieving end-to-end automation is crucial for timely service delivery and economic viability. Our goal is to develop largely autonomous networks that operate independently based on high-level policies, executing self-management operations such as self-configuration, self-diagnosis, and self-optimization. Enabled by AI and ML, these networks can efficiently manage resources and services without human intervention. In this presentation, we will show discuss how AI and ML empower mobile networks to self-manage and dynamically allocate resources for ultra-reliable, low-latency communications. We will also discuss integrating MLOps for AI model lifecycle management and outline the phases of our intelligence orchestration system, from planning to continuous maintenance and termination, showcasing the future of 5G network management.

In the past, there was a misperception that low-latency and high-throughput services were not compatible, leading to current low-latency solutions being deployed in a non-scalable manner with a limited number of use cases. Such an effect inhibits the development of real-time applications and slows down the evolution of congestion control, which could be used for traffic adaptation. With the recent advent of L4S (Low Latency, Low Loss, and Scalable Throughput) technology, applications can now seek the benefits of low-latency and high-throughput based on their particular needs, and this can be deployed scalable in public networks. Additionally, networks and applications can develop mutually beneficial relationships to make for a better user experience without incurring unnecessary control overhead.