Cloud and 5G are the biggest trends in the tech industry today. These are feeding on each other, resulting in the tremendous proliferation of both across many industries. The most profound and long-lasting impact of that will be in disrupting the decades-old practice of building cellular networks, especially Radio Access Network (RAN). In this article, I explore the idea of what it takes to build a modern, cloud-native, fully virtualized, multi-vendor Open RAN, and specifically the role of specialized accelerators in it.
What Is A Modern 5G Open RAN?
For the last four decades, despite five generations of technologies and numerous connectivity innovations, cellular operators have relied on the same architecture and very similar ecosystem to build their RAN. At the same time, the other part of the cellular network called Core Network (CN) has been fully modernized with cloud-native and virtualized architecture.
The biggest reason for RAN continuing with legacy architecture is the need to support extremely low latency. So far, this was only possible with fully integrated proprietary hardware and software coming from the same company. That means vendor lock-in, lack of diversity and, generally, higher costs. But thanks to the developments in computing, it is now possible to disaggregate hardware and software, use a mix of commercial off-the-shelf (COTS) hardware and hardware accelerators to achieve the same or better performance. This will end vendor lock-in, offer diversity and lower costs.
The modem 5G RAN is fully virtualized, cloud-native and features open interfaces. Virtualization converts various RAN functions into software. Cloud-native means RAN utilizes universal cloud architectures, such as Kubernetes and hypervisor, making scaling easy and cost-effective. Open interfaces enable interoperability across multiple software and hardware vendors. These concepts were introduced and perfected in the cloud market for some time, and it’s natural for the 5G ecosystem to utilize them for the modern RAN. The overarching objective is to lower the entry barrier, significantly expand the RAN ecosystem and cost-optimize 5G deployments.
What Is Needed, And Who Are The Players?
The legacy RAN ecosystem is tiny and shrinking, mainly because of the huge entry barrier and geopolitical issues. With the Open RAN, there will be various players from almost every part of the technology ecosystem. These include cloud giants like Google, Amazon, Microsoft and Facebook; players that offer COTS compute such as Intel, AMD, Arm and Marvell; hardware accelerator providers such as Qualcomm; and a plethora of software providers and systems integrators. Same players could be providing multiple versions of these things. For example, many legacy players such as Samsung, Nokia and Ericsson are transforming and looking to be key providers of Open RAN. There will likely be lots of realignment, including mergers, acquisitions and collaborations across players in this field in the near future.
One of the key things many are realizing is that the initial assumption about COTS being able to manage all the RAN workload can’t be further from the truth. It cannot provide the needed low latency—the very reason it took so long to transform RAN. That means a mix of dedicated hardware accelerators that perform crucial RAN functions, working seamlessly with the generic COTS hardware.
Hardware Accelerator For Real-Time Functions
As can be expected, there is a lot of processing involved in running the RAN workload. A part of it is non-real-time, such as radio resource management, ciphering, retransmissions, etc. These can be efficiently run on x86 or Arm-based general-purpose COTS processors (aka host). And then there are real-time functions such as demodulation, beamforming, channel coding, Forward Error Correction (FEC), etc.—sometimes referred to as “High-L1” functions—that are extremely latency-sensitive and would bring down the whole system if not executed properly. Purpose-built, dedicated hardware accelerators can efficiently offload these functions from the host processor.
Hardware accelerators can be used in two configurations: look-aside and in-line. In the look-aside configuration, accelerators only communicate with the host processor. This only works for non-crucial functions. In the in-line configuration, the accelerator sits between radio units and the host and communicates with both in real time. This configuration is a must for time-critical High-L1 functions like demodulation, beamforming, channel coding, etc.
These in-line accelerators have an even more prominent role when implementing advanced features, such as massive MIMO, beamforming, carrier aggregation and others. The requirements will become more challenging as networks grow even more complex.
From the user perspective, 5G is all about enabling new experiences, such as extended reality (XR), the metaverse, industrial IoT and others. All require ultra-reliable low-latency connectivity (URLLC). It is impossible to provide URLLC in Open RAN without well-designed, finely tuned, advanced in-line hardware accelerators.
The use of accelerators is a concept extensively used in cloud infrastructure, which bodes well for 5G RAN cloudification. In the modern virtualized Open RAN, it’s not just supporting these features but offering performance to meet or exceed that of legacy RAN that will be a critical factor for success. Further, the performance of accelerators will be a tool for vendor differentiation. As a proof point, HPE and Qualcomm announced an Open RAN platform with in-line hardware accelerators, starting the trend. I am sure many will follow.
After decades of status quo, the 5G industry is ready for disruption. The virtualized, cloud-native Open RAN is set to revolutionize the 5G infrastructure business with the expanded ecosystem, vendor diversity and lower costs. One key component that makes modern Open RAN possible is the in-line hardware accelerators. They are essential to managing the extremely latency-sensitive parts of the RAN workload and are crucial to enabling advanced network features and new experiences.
The accelerator performance will also be a differentiation tool for vendors. Many vendors are vying to lead in this space. But only the ones with proven technology, extensive technical expertise and the right collaborations will win. The race is on, and we will have to wait to know the outcome.