Nativeness and Openness: The Dawn of Wireless Network Intelligence

As 5G deployment deepens, the continuous emergence of new services and scenarios drives the evolution of networks, providing a high-quality network foundation for the new era of digital and intelligent living, industries, and society. Meanwhile, wireless networks are becoming increasingly complex, facing challenges such as multi-mode multi-band collaborative networking, differentiated QoS guarantee, bit-watt curve optimization, and operational efficiency improvement. To address these challenges, the introduction of artificial intelligence (AI) to enhance the intelligence and automation capabilities of wireless networks has become widely accepted in the industry.

Wireless Networks Evolving Towards Intelligence

Wireless network devices have witnessed a constant enhancement in computing power in recent years. In addition to fulfilling basic network functions, the collaborative computing power among multiple devices can support increasingly efficient AI algorithms. Moreover, wireless networks have access to massive data generated by all users, allowing for nearby AI model training and near-real-time inference computations, facilitating instant network strategy optimization. Wireless networks now possess the three essential elements of AI, including computing power, algorithms, and data, making them well-prepared for intelligence.

From R15 to R18, 3GPP has defined the RAN intelligence architecture and has been continually pushing for the evolution of native intelligence within RAN. Native intelligence, as one of the core capabilities of 5G-Advanced, supports the intelligent transformation of wireless networks, meeting the requirements for new capabilities, services, and technologies in the future. After years of exploration, wireless network intelligence has evolved from a technical concept to a reality, thanks to the collective efforts of the industry, and is now steadily moving towards commercialization.

ZTE has begun its research on wireless network intelligence and its applications in commercial networks since the pre-research stage of 5G networks. Currently, it has established a comprehensive network intelligence architecture that can support a wide range of intelligent network applications, tailored to different real-time demands, operational frequencies, and complexities.

Intelligent Service Exposure Empowers Multi-Dimensional Applications

ZTE's RAN intelligence architecture consists of three layers: physical network layer, intelligent service layer, and intelligent application layer (Fig. 1). These three layers enable native intelligence within the wireless network and empower operators to achieve cross-domain closed-loop operations, cross-domain collaborative computing, and cross-domain service orchestration.

The physical network layer encompasses all hardware infrastructure and raw data sources that form the basis for native intelligence within RAN. It includes various devices such as base stations, network management systems, intelligent platforms, as well as a diverse range of data resources from terminals or relay nodes such as mobile phones, wearable devices, connected cars, drones, ATG aircraft, and satellites. These resources facilitate interaction with the intelligent service layer through data collection and policy delivery.

The intelligent service layer utilizes the computing and data resources from the physical network layer to model and orchestrate computing capabilities and govern data. It employs various algorithm models to orchestrate intelligent atomic capabilities, and ultimately exposes computing power, data, and algorithms in the form of services to the application layer.

In terms of computing power, the native computing resources and enhanced computing resources within multiple base stations can be pooled and aggregated through transmission, enabling dynamic sharing and collaborative orchestration of computing power across base stations. This achieves load balancing and efficient utilization of computing resources. The combination of computing power at the base station level with the computing power of intelligent modules on the network management side and an independent intelligent platform forms the foundation for distributed native computing capabilities.

In the realm of data, standardized data collection processes and a unified data model are employed for data collection, cleansing, correlation, and labeling within the RAN system. This enables the generation and maintenance of modeled data such as user profiles and base station profiles. Model training and inference for native intelligence within the RAN system is efficiently supported by combining distributed data services with domain knowledge as well as model knowledge.

In the dimension of algorithms, the real-time intelligent engine deployed at the base station side supports extremely lightweight model training and near-real-time model inference. It improves network efficiency and user experience through precise prediction and proactive optimization. On the network management side, the lightweight intelligent engine supports lightweight model training and non-real-time model inference, facilitating efficient and proactive network operations through capabilities like quality insights and intelligent troubleshooting. Additionally, this architecture also supports the deployment of digital twin systems on independent intelligent platforms. Through data interaction with the physical network, it implements a series of atomic capabilities such as dynamic simulation, predictive optimization, intelligent decision-making, and data derivation.

The intelligent service layer can manage and orchestrate the above-mentioned computing power, data, and algorithm atomic capabilities in a microservice architecture. It exposes these capabilities to operators in the form of AI as a service (AIaaS), empowering the application layer. Operators can efficiently and cost-effectively integrate AI capabilities into their own systems through service calls and low-code development. This allows for flexible and on-demand feature expansion as well as resource scaling.

At the intelligent application layer, operators can leverage AI services to achieve highly automated and intelligent processes for network planning, construction, and maintenance. They can access on-demand capabilities for network visualization, modeling and simulation, precise network planning for multiple scenarios, optimization of network policies and parameters, precise network fault diagnosis, and predictive maintenance. In addition, operators can perform on-demand trial runs and performance evaluations for new services, as well as research and low-cost experimentation for new technologies.

ZTE's Achievements in RAN Intelligence

In 2020, ZTE launched the industry's unique NodeEngine computing base station solution to drive the commercial implementation of native intelligence in RAN. By adding computing boards to traditional base stations, it enables elastic expansion and capability exposure of edge computing. The NodeEngine solution can be centrally managed and scheduled by the intelligent network brain, providing ubiquitous computing services.

In 2021, ZTE launched a wireless intelligent orchestration solution based on the above-mentioned native intelligence architecture, including user orchestration and network orchestration. User orchestration focuses on delivering the best possible experience and resource efficiency by intelligently guiding users to optimal frequency bands and cells, thereby enhancing user satisfaction. Network orchestration aims to maximize 5G experience while ensuring the fulfillment of 4G network demands, providing operators with a more beneficial resource sharing mode and improving network efficiency.

In 2022, with the successful implementation of computing base stations and wireless intelligent orchestration, ZTE combined a range of achievements in system architecture, technological innovation, and commercial deployment of wireless network intelligence. This led to the official launch of the comprehensive RAN Composer native intelligence solution. Leveraging the computing power, data, and algorithm services at the intelligent service layer, ZTE developed several intelligent applications at the application layer and rapidly advanced their commercial deployment in the field.

The evolution of wireless networks towards intelligence has begun, and ZTE's RAN Composer native intelligence solution has seen multiple intelligent applications deployed both domestically and overseas, showcasing its value in enhancing experience, efficiency, and network maintenance. Moving forward, ZTE will continue to advance the process of wireless network intelligence, guided by the core principles of surpassing, intelligence, and leadership, to build the most superior network.