As open-source software makes inroads into networking environments, the radio access network (RAN) space has become ripe for innovation. Service providers, telecom operators and enterprises are beginning to envisage market opportunities for cloud-native open RAN technologies across various use cases, including 5G and the network edge.
The evolution of 5G networks to cloud-native architectures with open RAN is taking advantage of IT and cloud capabilities that have been perfected over many years by the hyperscalers. Marvell, a maker of data infrastructure semiconductors, sits in the middle of this critical transition as a supplier of optimized silicon to both RAN and cloud providers.
At this year's Mobile World Congress, Marvell launched a new addition to its OCTEON 10 Fusion family of baseband processors for 5G base stations based on 5nm process technology. With OCTEON 10 Fusion, Marvell starts with the OCTEON 10 DPU, limits the number of CPU cores and uses the real estate for DSP cores, accelerators for massive MIMO and other base station-centric technology. The end result is a chip that both leverages core technology and is optimized for a big market. The OCTEON 10 Fusion baseband platform will support use cases for the traditional RAN architecture as well as open, virtualized RAN. Optimization has its fans. Nokia’s ReefShark chipset includes OCTEON 10 Fusion and OCTEON 10 technology. Samsung and Dell have similarly adopted it for their 5G offerings.
The evolution of RAN
A traditional RAN is a cell tower with an antenna on top and a base station below. The base station has two main components: the remote radio unit (RRU) and the baseband unit (BBU). Incoming signals from end-user devices are converted into data by the base station and transmitted to the core network.
A traditional RAN comprises closed and proprietary RRU and BBU hardware architectures, resulting in vendor lock-in. Virtualized RAN (vRAN) is a move away from proprietary equipment to commercial off-the-shelf (COTS) hardware that carries out its tasks in software using virtual network functions (VNF).
Open RAN has a set of standardized and interoperable interfaces, giving operators their choice of the best options for them, be that COTS, accelerators or cloud providers. Open RAN allows operators to decouple software from hardware, facilitating the migration to a cloud-native model.
Virtualization of the core network has existed for some time, but the critical and demanding nature of RAN workloads has held back vRAN and Open RAN. The complexity lies in Layer-1 (L1; the physical layer or PHY) processing, the only layer in the OSI model that interacts with actual hardware, transmission and signaling mechanisms.
The BBU has two functional units: a distributed unit (DU) for real-time L1 and L2 scheduling functions and a central unit (CU) for non-real-time, higher L2 and L3 functions. The radio unit (RU) handles the digital front end (DFE) parts of Layer 1, as well as the digital beamforming functionality (a technique that focuses a wireless signal toward a specific receiving device using multiple antennas).
Operators can virtualize CU functions on general-purpose hardware in the cloud. Given the demands of large deployments and L1 processing, DU and RU functions require low latency, low cost and low power consumption.
Marvell’s OCTEON 10 Fusion baseband platform supports use cases from the RU and the DU for both traditional RAN architectures and open, virtualized RAN. In the same way that SmartNIC accelerators improve efficiency and flexibility for networking, security and storage virtualization within the data center, the OCTEON 10 Fusion creates performance benefits by offloading L1 functions from the host server.
The most effective option for operators to save energy is in the RAN because they must typically provision so many base stations. Indeed, limiting power consumption is essential when operators evaluate vRAN or Open RAN solutions. Marvell has implemented advanced power management in the OCTEON 10 Fusion platform, which delivers best-in-class RAN performance per watt across a wide range of 5G use cases.
A large ecosystem with vendor choices
The O-RAN Alliance, in which Marvell is an active participant, is tasked to eliminate vendor lock-in by providing standards-based interfaces between RAN components, namely the RU, DU and CU. Kubernetes and a service management and orchestration (SMO) platform handle the cloud-native containerized software and the automation of RAN-as-a-service.
Marvell works closely with many cloud and carrier ecosystem partners to bring Open RAN to market. The collaboration enables operators to combine best-in-class L1 hardware and software from different vendors in an open, standards-based platform that integrates into the PCIe slot of any x86 or Arm-based host server in a carrier or cloud data center.
For example, the OCTEON 10 Fusion includes specialized accelerators optimized for processing complex 5G massive MIMO beamforming algorithms. (MIMO stands for multiple-input, multiple-output). Marvell partnered with Analog Devices to develop an RU reference design with the OCTEON 10 Fusion chip that supports 32T32R antennas with 400 megahertz of operational bandwidth and 300 megahertz of instantaneous bandwidth, enabling up to 40% savings in power consumption.
Overcoming resistance to cloud-native Open RAN
It is beyond dispute that RAN needs to evolve to keep up with industry growth and deliver on the promise of new 5G-based services. Disaggregating RAN software from hardware and making the software available and deployable in the cloud is a natural first step.
Keep in mind that RAN, although typically an underutilized resource, is the most significant area of CapEx and OpEx for telecom operators. Moving RAN into the cloud brings all the benefits of cloud computing, pooling resources with higher utilization rates in a shared cloud infrastructure, resulting in potentially huge savings. Marvell’s new Octeon 10 Fusion platform fits into this vision beautifully because it enables cloud scalability, integrating the largest cloud orchestration and management software platforms with features, performance and power efficiency on par with today's most advanced 5G networks. This makes for a much easier migration based on the same proven silicon and software.
We are in the early stages of the transformation to cloud-based RAN, however. There is a real risk that open systems could lead to fragmentation and lack of interoperability; in that scenario, operators would have much to lose from poorly implemented RAN deployments. To the risk-averse, using a single vendor for end-to-end RAN architecture could still appear much more straightforward, quicker and safer.
There are also performance issues to be considered. General-purpose processors cannot handle the demands of a RAN or a massive MIMO. It’s possible to augment a general-purpose CPU with look-aside accelerators—whether FPGAs, GPUs or ASICs—to offload the most demanding tasks. Yet for now that approach remains cost-prohibitive and power-hungry.
That said, consensus does exist on the virtues of dedicated, optimized silicon inline acceleration to handle complete data flow and functions. The Marvell Octeon 10 Fusion is an excellent example of an optimized inline accelerator in this vein. It is CPU-agnostic and enables RAN software based on open-source, industry-standard interfaces. Most importantly, the inline hardware acceleration delivers feature, performance and power parity compared to existing 5G networks.
The computing industry’s journey from closed to open systems took decades and saw many failed attempts before Linux and cloud technologies came along. We see good progress of that type from the O-RAN Alliance, which includes 300 operators, vendors and research and academic institutions that are focused on interoperability and standardization of RAN elements, including interconnection standards for hardware and open-source software elements across vendors. Marvell is an important contributor to the Alliance, and its new platform offers a taste of what will be possible when the wireless industry as a whole embraces open systems.