The momentum behind 5G deployments is continuing to grow rapidly in multiple geographies. Operators are challenged by the huge increase in radio bandwidth and additional network complexity they need to support while maintaining high reliability and meeting fixed OpEx and CapEx budgets. Implementing new designs for macro and small cell base stations as well as new radios is a key element in meeting these challenges.
The designs for the 5G Radio Access Network (RAN) will take on new functionality to support the broader set of use cases. These use cases cover the need for multiple classes of service from Enhanced mobile broadband (eMBB), Ultra Reliable Low Latency (URLLC) to Massive Machine type communications (mMTC). Connectivity from the client device through to the cloud data center is likely to require lower latency response at gigabit speeds. Across the spectrum of 5G stakeholders, architects are exploring ways to meet wireless network KPIs and to manage costs by employing new techniques like software defined workloads, and balancing general purposes compute capabilities with appropriate acceleration offload to manage power and latency. In addition, new Open RAN platform alternatives are being investigated in an attempt to broaden choice and increase flexibility in 5G designs.
The pace of innovation is increasing rapidly. The Open RAN architecture enables a new paradigm shift where cellular networks are composed of hardware and software components from multiple vendors operating over standard network interfaces.
The public cloud operators leveraged software defined principles to disaggregate hardware from software layers and drove innovation through virtualization technologies. Operators see the same opportunity with 5G RAN with disaggregation of base station and allowing multiple functional split options so that power and cost optimized RAN solutions can be built for diverse network traffic conditions. These tasks (as shown in the previous image) are split over many different network elements – from Radio Unit, Distributed Unit and a Centralized Unit all way to the cloud.
These new elements and the combination of all these technologies are helping transform support for new use cases required in 5G rollouts. For example, the Radio Unit which is extremely power sensitive, needs enhanced capabilities to handle beamforming and beam steering functionalities. The Distributed Unit will need to handle higher throughputs and more complex workloads, while being able to utilize new capabilities like machine learning and managing radio resource and user scheduling. In the Centralized Unit, we see a mix of high performance and efficient CPU architectures providing the raw compute performance needed to run cloud native applications and packet processing environments with Smart NIC offload to handle some of the real time functions that can be efficiently processed in more heterogeneous compute environments (functions like cloud-based L1 processing, network interface and security processing for example).
To maximize functionality, power efficiency and lower the overall system costs, a software-based accelerator approach can disaggregate conventional, siloed tasks in RAN and process data more efficiently.
Historically, digital signal processing (DSP) cores were used for the Physical Layer (PHY) processing in the Radio Access Networks. The addition of performant SIMD/Vector engines to the Neoverse Arm CPU cores gives the designer the ability to develop Vector processing capability within a scalar processing environment to develop a mix of the two on a single instruction set architecture.
This offers the flexibility to architect use case driven hardware and software partitioning in the time critical math functions especially in the lower layers of the radio access network.
The vector engines in Neoverse Arm cores enable “software acceleration” for a more programmable environment and can further be complemented with functional block accelerators or small footprint FPGAs for efficient system designs.
We have been working in close collaboration with our ecosystem partners to develop and fine-tune signal processing software routines for optimized performance. Today we are announcing the Arm RAN Acceleration Library (Arm RAL). This library helps with implementing common signal processing functions in 5G Radio Access Network using Arm Neoverse cores.
Arm RAL renders several substantial benefits to the design of the NG wireless base stations. It allows generally available software developer talent pool to be used for development of 5G PHY Layer and reduce dependency on very specialized DSP software skills. Getting the PHY Layer running on general purpose Neoverse cores, makes this workload cloud-friendly, and paves way for it to be deployed in Open RAN deployments on general purpose Arm servers. This approach protects the base station designer’s software investments by facilitating the easy porting of the base station software allowing it to benefit from the generational performance gains Neoverse line of processors are delivering.
The first public release of Arm RAL is starting with R 20.10, and it constitutes math kernels and signal processing primitives required for PHY High (Layer-1 Upper) and PHY Low (Layer-1 Lower) implementation. In widely accepted O-RAN Split 7.2 these functions would reside on DU and RU respectively. Major classes of math kernels and signal processing primitives include:
Arm RAN Acceleration Library is distributed in source code form and is available for anyone to download it from developer.arm.com. You can access the code by obtaining and signing an EULA.
Download Arm RAL