One of the biggest challenges for operators looking to migrate to a virtualized RAN (vRAN) is to find the right chipset combinations. The vRAN may be all about software, but the 5G workloads are so intensive that many operators question whether general purpose hardware could ever support them as effectively as dedicated platforms.
Traditionally, macro base station vendors designed or commissioned their own proprietary chipsets, fully optimized for intensive RAN tasks such as high end signal processing. The economics of dedicated, inhouse platforms are hard to justify when margins are tightening and operators are demanding the cost-effectiveness and scalability seen in the cloud. Nokia’s decision to move away from its own FPGA-based base station chips to a more packaged solution late last year just highlighted the dilemma – it had majored on FPGAs (field programmable gate arrays) for their flexibility and performance, but found the costs made it uncompetitive.
But the 5G RAN does not lend itself to general purpose processors (GPP) alone. In theory, the servers which will support baseband virtual network functions (VNFs) will be ‘commoditized’, but for macro base stations, they will in fact need to be very specialized and high performance platforms because of the intense processing demands of 5G.
Top end CPUs will be coupled with specialist accelerators for the most intensive tasks, especially in a distributed RAN architecture where processing is pushed out to ‘distributed units’ (DUs) close to the cell sites, making low latency a priority. Vendors and operators are exploring architectures which will enhance the return on investment for these roll-outs, by integrating edge computing capabilities tightly with the DUs so that a single unit can support 5G network processes along with data processing and storage, both for internal and customer-facing applications.
And of course, there will still be specialized chips, including RF components, in the radio/antenna units at the actual cell sites; and embedded communications processors in many locations (though many of those embedded vendors such as NXP and Texas Instruments have largely migrated from proprietary cores to ARM-based designs).
All this makes the 5G chipset space a complicated one, and very far from the long-ago Cloud-RAN vision of a centralized network that ran entirely on off-the-shelf CPUs (central processing units). The more distributed and edge-focused the RAN, the more low latency is essential; and the more 5G use cases and traffic levels proliferate, the more the goalposts move in terms of signalling and connection volumes, and the challenges of supporting Layer 1 (physical layer) workloads to the same level of performance as a dedicated, integrated architecture.
And the challenge is intensified still further by operators’ push towards far lower cost platforms, enabled by open, multivendor systems. Intel has been in the vanguard of tests of emerging vRANs that are both disaggregated and open, but ARM has now joined the O-RAN Alliance, setting up a new round of the Intel/ARM processor wars in this demanding field of engagement.
And another long-standing Intel rivalry, with Nvidia and its GPU (graphical processor unit) platform, is also moving into the 5G world, thanks to work by Verizon and Ericsson, with Nvidia itself, on using GPUs as accelerators for Layer 1 tasks, and to support an integrated edge compute/distributed vRAN unit architecture. But GPUs are only one approach to acceleration – the more traditional FPGA option remains very important, though the nature of that particular beast is changing significantly, 30-plus years from its introduction.
Xilinx, the FPGA leader, has expanded its platform into something broader and AI-enhanced, which it dubs ACAP (adaptive compute acceleration platform), and it has just announced an ACAP-based product targeted at telcos, as well as a vRAN contract with Samsung. And Intel, which owns the second-largest FPGA maker, Altera, is combining its solutions with other programmable options such as enhanced ASIC to surround its core Xeon CPUs with accelerators and chase the overriding goal of all major silicon vendors in the 5G space – an end-to-end portfolio that can address the multi-faceted requirements of this challenging but highly valuable segment.