This is a guest blog by Shamik Mishra and Altran Research & Innovation team.
5G is more than just smartphones with fast connectivity powering great innovative applications. GSMA estimates that by 2025, there will be nearly 1.2 billion 5G connected devices covering nearly one-third of the global population.
While it is expected and widely perceived that 5G will provide faster broadband access, it will also drive the adoption of other prominent technologies like disaggregated radio access network, industrial IoT, computer vision, artificial intelligence (AI) inference and hardware acceleration to support use cases such as private 5G networks for enterprises, automated warehouses and factories (Industry 4.0). Another use case can be the deployment of 5G connected Industrial-IoT (IIoT) devices, which may find their way into smart city infrastructure, robotic assembly-lines, warehouses and offices which would connect to these networks.
Computer vision in manufacturing is expected to radically improve the productivity of its processes. On the shop floor, computer vision can aid in predictive maintenance, inspect packaging in the assembly line, automate barcode reading, safety and tracking logistics. However, computer vision implementation can be costly if the processing is done on premise or it can prove ineffective if processing is exclusively done in the cloud. Instead, the use of low-cost cameras to stream videos over a reliable, low latency 5G network to aggregate and execute machine vision algorithms in a localized data center via edge compute will be better approach.
In this scenario, the 5G network functions enable the edge compute and can provide powerful capabilities including a functional private network (5G RAN and Core network) along with computing capabilities at the edge of the network. When developed on open infrastructure with well-defined application programming interfaces (APIs), such architecture can help application developers and device manufacturers (of smart cameras, for instance) to easily add content and create innovative services.
In the past, 3GPP worked on defining the network architecture and protocols with 4G narrowband IoT (NB-IoT) and enhanced machine type communication (eMTC). Today, 5G has a suite of enabling technologies which improve these requirements through Ultra Low-Latency (uRLLC) and Massive Machine-Type Communications (mMTC). Low latency is built into the 5G stack where protocol / messaging aspects like Grant-Free scheduling, PDCP duplication etc. delivers lower latency than 4G and improves reliability for mission critical communications. 3GPP is also specifying network and edge awareness to 5G applications that can enable application developers to differentiate and choose based on network capabilities that are exposed to them through APIs.
Another key change 3GPP and O-RAN alliance (an industry governance on open RAN initiatives of which both Altran and Arm are active members) brought in with 5G is a completely disaggregated architecture, with standardized interfaces. This addresses data privacy concerns for private 5G networks as the entire subscriber to network data-path can be terminated in an enterprise data-center.
The previous diagram provides an overview of the different components that constitute a 5G private operational network for industries. This blog explores the 5G access and edge communication components in the context of Arm technologies.
The key considerations for private 5G networks and Industry 4.0 applications are:
A disaggregated network that can in parallel secure a data path to the local edge compute while scaling the network independently, can address the data privacy and security concerns. Depending on the use case, reliability and low latency have to be factored into the network topology and compute architectures while building the disaggregated network.
Arm and Altran have been conducting research on integrated 5G and edge infrastructure which leverages multi-core processors based on Arm Cortex A72. The A72 based multi-core processor integrates with smart network interface cards (NICs) and technologies for data path acceleration architecture (DPAA2) for faster network processing. This helps achieve lower processing latency for the layer2/3 data paths.
The on-board crypto accelerator processing allows us to offload packet ciphering / de-ciphering. This also lowers the packet processing latency and increases the throughput capacity.
PCIe 3.0 integration and accelerated NIC fast paths allow interconnection of multiple baseband cards for hosting localized 5G RAN, 5G core network (user plane) and edge applications.
Arm platforms have a rich set of features built into their integrated DPDK libraries for example, vector processing to increase data path throughput. Lockless algorithms are used in libraries to provide scalability across large number of cores in a system on chip (SoC). Arm memory models (e.g. C11 memory model) are used wherever possible to leverage relaxed memory models for Arm architecture. Memory barriers have been fine tuned to be optimal for Arm platforms.
Altran is building an integrated private 5G framework on an Arm based platform deployable in enterprises. The 16-core A72 Arm processors can run a disaggregated element of the 5G RAN – a central unit (CU) or a distributed unit (DU), or the 5G core network’s user plane function (UPF). Multiple cards can be connected over ethernet based accelerated NICs, allowing a distributed deployment which can scale up across multiple radio sites.
Each hardware unit can be connected to an enterprise network on a secure link, with the network encryption functions offloaded to on-board hardware crypto co-processors. The subscriber data can terminate in a local data center or be connected over a secure link to an offsite data center. Accelerated data paths use the DPAA2 to offload switching elements, leaving most of the Arm cores free to run applications. The architecture is based on containers and orchestrated by a micro Kubernetes through a cluster architecture. This cluster architecture can also run edge compute applications including AI inference.
Altran is working with NXP and Qualcomm on the Arm based architectures for 5G RAN in the areas of macro, industrial along with enterprise / residential small cells.
Low total cost of ownership (TCO) based private data center Infrastructure market is gaining momentum thanks to private 5G. This is fueled by the need enterprises have for compute, storage, and networking solutions. The recent Arm Neoverse V1 and N2 platforms created for intelligent edge infrastructure is an important evolutionary step for holistic private 5G infrastructure. This will give further impetus to the private 5G networking market, complementing it with low-cost data center solutions.
These are powerful multi socket Arm Neoverse cluster units providing integrated 5G Core, 5G access and Edge compute platform. These integrated systems can deploy latency sensitive applications from smart manufacturing, factory automation and ware houses. Arm’s rich ecosystem initiatives for edge like the Project Cassini and open source frameworks will be a key enabler to build virtualized 5G services and applications which pave the way for a whole new world of possibilities.