- Jun 15, 2023
- Wireless, Network
Toward Achieving Next Generation Mobile Networks
#AI-RAN, #Core Network
How Mobile Network Works
Do you know how the social networking services and phone calls you make on your smartphone are made possible? You may only be familiar with your smartphone and the antenna of the wireless base station. However, there is a vast mobile network spread across the country constructed by communication service providers, which is further connected to the internet and telephone networks. This article explains the basic configuration and mechanism of mobile networks. Additionally, it outlines SoftBank's cutting-edge research and development efforts toward achieving next-generation mobile networks.
Mobile networks are primarily composed of UE, RAN, CN, and TN. Through cooperation among these components, your smartphone can connect to the internet and the phone network. This article briefly explains the mechanisms of each component as follows.
UE (User Equipment) is a general term for devices such as smartphones and tablets. UE connects to mobile networks using radio waves to transmit and receive data and voice. It is responsible for controlling authentication and communication encryption when connecting to a mobile network. RAN (Radio Access Network) is a generic term for wireless base stations that communicate using radio waves through UE. RAN is responsible for efficient management of wireless resources, management of area coverage for communication while on the move, and processing signals, including sending and receiving radio waves through antennas. CN (Core Network) is a generic term for a set of functions ("Network Function") that manages the actual transfer of user data that UE communicates, as well as processing communication protocols. The CN is connected to multiple RAN and is responsible for routing user data and connecting to other networks such as the internet and telephone networks. It also manages communication protocol processing, such as user authentication, billing management, and security management. TN (Transport Network) is a generic term for the network that connects RAN to CN. TN is composed of optical fibers that spread throughout Japan, and network equipment connected by them, and is responsible for the path that RAN installed throughout Japan reaches the internet and telephone networks through CN.
Terms such as 4G and 5G, often heard in mobile networks, indicate the generation in the process of advancing UE, RAN, and CN. These generations are renewed about once every ten years, with 5G currently being the latest standard used for commercial deployment by telecommunication providers worldwide. Research and standardization toward 6G is expected to be done in the near future with commercial deployment expected to begin in 2030, while 5G is expected to be used in parallel.
Stateless Core Network
Background (Expectations and Challenges toward Beyond 5G/6G)
One of the important use cases for mobile networks toward Beyond 5G and 6G is digital twinning. Digital twinning is an environment where real-world information is collected and digitized, forming a virtual space for simulation and optimization. To achieve digital twinning, data must be collected 24/7 from numerous sensors installed everywhere. Therefore, scalability and robustness are critical roles that future mobile networks should play.
Currently, CN centrally manages tens of millions of UEs, including smartphones and tablets, and hundreds of thousands of RANs. However, for 6G, the number of IoT devices, such as sensors, is expected to increase to hundreds of millions. To provide scalability and 24/7 services and address this influx of UE, there are challenges associated with the configuration of CN standardized for 5G. The most significant challenge is that CN consists of multiple NFs (Network Function), each separately maintaining and managing the states of numerous UE. While there are no major issues as long as all NFs are functioning correctly, if a failure occurs in any NF, it puts an unexpected load on all associated NFs. This can lead to a congested CN, resulting in malfunction and large-scale service disruptions for all UEs across the country.
Our Initiatives (Proc5GC)
To address these challenges, SoftBank is developing a brand new CN (named as Proc5GC) that leverages web and cloud technology. Our attention was drawn to a significant purchasing event hosted by a leading e-commerce site, which demonstrated exceptional scalability and robustness by efficiently processing over one billion payment transactions per second, without experiencing any delays.The characteristics of the EC site is that a single function is responsible for processing and individual functions are stateless, which maximizes cloud computing capabilities. We applied these features to mobile networks and reconstructed the structure of CN. The specific actions we took are as follows:
Disaggregation of the CN functionality into Procedure Functions
We split groups of procedures that were realized by multiple NFs cooperating in existing CN into small software (Procedure Functions) that are responsible for only a single procedure. Stateless Procedure Functions
We separated the state of UE and RAN owned by each procedure and managed them commonly in external databases. Reactive Procedure Functions
While existing CN had NFs waiting for procedures to be accepted, Proc5GC introduced a method to execute Procedure Functions when procedures are initiated.
Through these efforts, we were able to change the structure of CN to be very similar to that of the leading EC site. As a result, we can maximize cloud computing capabilities and enable the connection of UE and RAN without limitations on their scalability as long as there are computer resources. Even if there is a problem with Procedure Functions, it does not result in a widespread impact on the entire CN like existing CN, which makes it possible to localize malfunction.
Achievements and Future Plans
We presented the concept, design, and basic performance evaluation of this research and development at the Technical Committee on NS (Network Systems) of the Institute of Electronics, Information and Communication Engineers (IEICE), a Japanese academic society.
link1:https://ken.ieice.org/ken/paper/20221215XCPU/ link2:https://ken.ieice.org/ken/paper/20230303fCSS/ link3:https://ken.ieice.org/ken/paper/202303023CSP/
We also submitted a paper summarizing the same content to the IEEE International Black Sea Conference on Communication and Networking, and it has been accepted（ https://blackseacom2023.ieee-blackseacom.org/ ）.
Moving forward, we plan to conduct mutual interoperability verification with commercial RAN and UE products, evaluate scalability using public cloud, and evaluate practical operation using an experimental operational environment. Additionally, we are considering providing feedback to standardization bodies based on insights from these verification tests.
Virtualization with GPUs
To meet the ever-increasing needs for communication, radio base stations have been evolving. However, as base stations were traditionally built using dedicated hardware, their evolution depended on the hardware vendors. In contrast, virtual Radio Access Network (vRAN) technology separates the functionality traditionally implemented on dedicated hardware into hardware and software, allowing for the implementation of base stations as software on general-purpose computers. This makes it possible to add functionalities through software updates alone, allowing for quick and easy response to communication needs.
However, when it comes to digital processing of wireless signals, the processing speed on a general-purpose computer is not enough. It is essential to combine a hardware module (an accelerator) specialized for this processing with a general-purpose computer. There are various types of accelerators, such as ASIC/FPGA/DSP/GPU, and there is a trade-off between flexibility and performance, and each has different characteristics in terms of balance. If performance is prioritized, it may become difficult to evolve the functions of the base station, while if flexibility is prioritized, power consumption may increase.
To address such challenges, SoftBank is working on using GPUs as accelerators to resolve the trade-off between flexibility and performance. Unlike other accelerators, GPUs can efficiently use applications other than vRAN, such as AI inference, learning, and 3D graphics. This makes it possible to use the infrastructure of mobile networks not only for vRAN but for Mobile Edge Computing (MEC) sites for various applications as well.
SoftBank has been working on using GPUs for vRAN since 2019 and has conducted performance verification as an accelerator, impact verification with MEC/AI, and verification of the coexistence of vRAN and MEC/AI. The specific efforts that have been made are as follows.
L1 Accelerator performance verification
We conducted numerical simulations to measure the data transmission performance both from the UE to RAN (uplink) and from RAN to UE (downlink), and confirmed that the processing speed and power consumption are low enough to meet the low-latency processing time required for 5G communication performance. MEC/AI (Maxine) impact verification experiment
We have performed an impact verification experiment for MEC/AI with "NVIDIA Maxine" to transmit low-resolution (equivalent to 180p) videos and perform "super-resolution" processing by AI on the MEC server, resulting in the generation of high-resolution (equivalent to 720p) videos. We demonstrated that the same quality of video delivery can be achieved with a smaller network bandwidth compared to the case when the MEC servers are not utilized. This experiment also confirmed the effectiveness of AI and MEC for uplink communication in 5G. Establishment of AI-on-5G Lab
”AI-on-5G Lab" is a research facility where various solutions, including AI technology, can be verified and considered for application to business areas in a virtualized infrastructure environment that combines 5G vRAN and MEC Verification of the coexistence of vRAN and MEC AI
We have successfully verified the End-to-End connection between vRAN and image processing MEC applications in actual machines. Additionally, we have achieved real-time person detection by AI on the same hardware configuration as vRAN.
In May 2023, we announced a collaboration with NVIDIA to build a next-generation platform for generative AI and 5G/6G. We plan to develop and verify functionalities that dynamically and efficiently allocate computer resources to meet the demands of various applications such as vRAN and AI, and work toward further optimization of wireless resources and power consumption in vRAN.
The previous achievements
The technology verification of 5G virtual base stations utilizing NVIDIA GPUs has been conducted. Successfully conducted a verification experiment using 5G and MEC to perform "super-resolution" processing on videos with "NVIDIA Maxine." SoftBank to open “AI-on-5G Lab.” with NVIDIA for commercialization of fully virtualized Private 5G SoftBank Corp. Successfully Verifies GPU-based vRAN on Actual Machines NVIDIA Collaborates With SoftBank Corp. to Power SoftBank's Next-Gen Data Centers Using Grace Hopper Superchip for Generative AI and 5G/6G
Research on RAN Control Functions
In mobile communications, the diversity of connected devices has increased with each generation, enabling various services to be provided. New frequencies have been assigned to respond to these changes, resulting in improvements in communication quality and speed.
However, there are also challenges. Low frequency bands suitable for building wide area coverage are already being used for various purposes, including mobile communications, leaving no available frequencies to be easily acquired in the future. Furthermore, as wireless communication is being used for various purposes and communication volume is increasing due to device evolution, there are concerns about decreased communication speed caused by congestion in wireless infrastructure. To guarantee communication quality within limited frequency resources and improve the investment efficiency of wireless infrastructure, it is crucial to improve spectrum efficiency.
To realize wireless infrastructure capable of efficiently and appropriately allocating wireless resources at any given moment and place in response to varying communication demands, SoftBank is focusing on research and development on inter-base station coordination technology.
Single Frequency Network (SFN) is a form of inter-base station coordination. Normally, signals transmitted by surrounding base stations become noise and lower the quality of received signals. However, with SFN, adjacent base stations cooperate and coordinate with each other to operate as a single, large base station, suppressing noise signals and improving the quality of received signals. In addition, there are no handovers between base stations, which are caused by the movement of UEs, resulting in improvements not only in user experience but also in the utilization of wireless and computer resources by reducing signal processing due to handovers.
The inter-base station coordination function, which manages and controls many base stations, will need to perform statistics/analysis/training, not only on RAN information but also on user demands that change with location, season, time, and weather conditions. This will be essential to realize a flexible and efficient wireless infrastructure that not only responds to user demand. By processing this information through AI, the optimal RAN control profile can be determined. This allows the flexible combination of base stations and flexible processing resource allocation within base stations, and enables more efficient allocation of wireless resources.
SoftBank is conducting research through various approaches, such as the ones mentioned above, to build next-generation wireless infrastructure that is satisfying for both operators and users.