AMD SoC Fuels Open RAN Designs for Meta’s Evenstar Effort
By leveraging AMD’s RFSoC device, radio units for Meta’s Evenstar program are well-positioned to speed the adoption of Open RAN infrastructure.
Signifying a critical step toward adding momentum to the Evenstar program and the Open RAN movement, today AMD (formerly Xilinx) announced that its Xilinx Zynq UltraScale+ RFSoC is being designed into multiple Evenstar RUs (radio units).
In this article, we put the Evenstar program and Open RAN into context, examine AMD’s RFSoC as an Open RAN radio system, and share insights from our interview with Gilles Garcia, AMD’s Senior Director Business Lead of AMD’s Data Center and Communications Group.
The Evenstar Program and Open RAN
Ever since Meta, the parent company of Facebook, launched its Evenstar program in 2020, the effort has been gaining momentum and has become a leading force pushing the idea of Open RAN-based infrastructure, challenging the traditional model of wireless telecom system providers and its monolithic solutions.
Led by Meta Connectivity, the Evenstar program is a joint initiative between operators (service providers} and technology vendors. Meta’s goal for the Evenstar program is to craft adaptable, efficient, and “metaverse-ready” RAN reference designs for 4G and 5G networks in the Open RAN ecosystem. Additionally, Evenstar aligns with 3GPP and O-RAN specifications.
In single-vendor RANs, hardware and software are part of one system. Open RANs decouple these elements and use open standards-based interfaces. Image (modified) used courtesy of AMD
For a traditional RAN, all the elements of the network, including the RRU (remote radio unit) hardware, the DU (distributed unit) hardware, and CU (centralized unit) software are sold as a set package from a networking vendor like Nokia or Ericsson.
In contrast, the approach of Open RAN and the Evenstar program is to decouple these elements and provide alternative solutions supporting standard interfaces. The Open RAN model calls for a multi-vendor, disaggregated, interoperable RAN based on open interfaces. While we’re years away from a completely open model, as the diagram above shows, eventually the Open RAN approach could eliminate the need for local base station units in cell towers.
However, a valid counter-argument to the Open RAN approach is that traditional wireless network system vendors offer one-stop shopping for service providers. “The problem with one-stop shopping is that operators are not sure of what they’re buying, or they’re often paying for more equipment than they need," says Garcia. "With Open RAN, operators can find hardware and software vendors that let them customize gear suited to the way they want to monetize their networks.”
AMD’s Adaptive RFSoC as a Single-chip Radio Subsystem
According to AMD, the Evenstar RUs with Xilinx Zynq RFSoC architecture offer flexibility to meet the requirements of 4G/5G, mmWave, and sub-6 GHz networking based on the same hardware. This is expected to make it easy to use the platform to address diverse radio configurations and emerging standards so that radio suppliers can react fast to market opportunities.
In reality, the AMD RFSoC isn’t just part of these RU designs, it is actually the whole radio. The device isn’t a classic FPGA, but rather an adaptive SoC. The device claims to provide SoC and processing functionality combined with a lot of ASIC (application-specific IC) cell block hardware while leaving some programmable logic so that designers can differentiate themselves.
Importantly, the RFSoC features a generous block of analog functions on-chip, including RF ADCs, RF DACs, down converter functions, and more. Altogether, the RFSoC provides a hardened radio system capable of implementing a single-chip 8T8R FDD/TDD (frequency division duplex/time division duplex) radio.
The RFSoC is not an FPGA. It is an Adaptive SoC combining processing, analog and RF function blocks, and some programmable logic. Image used courtesy of AMD
Each analog device block on the RFSoC enables part of the radio functionality. The RF DACs and RF ADCs provide 7.125 GHz direct-RF bandwidth. The DPD (digital pre-distortion) block supports traditional as well as new (400 MHz) GaN (gallium nitride) power amplifiers. The CFT (crest factor reduction) block provides up to 400 MHz of instantaneous bandwidth (IBW). Multi-carrier and multiband support is enabled by the DUC (digital up-converter) and DDC (digital down conversion blocks). Lastly, the signal processing unit on the RFSoC provides re-sampling and equalizer functionality for the radio.
RFSoC Facilitates Diverse Carrier Needs
The broad functionality and flexibility of the RFSoC-based radio could translate into some powerful capabilities for service operators. Aside from support for 400 MHz (8T8R), the device claims to also support up to 1,600 MHz IBW mmWave IF transceivers. The SoC’s embedded processing offers greater compute-per-antenna configurations or alternatively, more antennas per radio. By providing more bandwidth per radio, systems can have fewer radio units, which translates to fewer power amplifiers and the associated lower system costs.
The RFSoC enables aggregation of up to 8 carriers per antenna on an 8T8R FDD radio. Image (modified) used courtesy of AMD
Meanwhile, the RFSoC could be well-suited for situations where multiple operators share the same RAN. The device enables multi-band operation with carrier aggregation of up to 8 carriers-per-antenna. Moreover, 4G LTE and 5G can run on the same radio (multi-band, multi-mode). More data pipes through the same radio, once again, means lower system costs.
For his part, Garcia says he has not yet seen an RFSoC design used by more than two operators. “In Europe especially, we are seeing two to three operators working together. But very often they have only two that are working very closely,” he says. That said, the capability is there for several carriers to share the same radios.
Solution for Accelerating Open RAN Adoption
All in all, the marriage of AMD’s RFSoC technology with Meta’s Evenstar program seems to add juice to the movement toward Open RAN architectures in wireless networking. The shift won’t happen overnight, but the ability to do multiple radio designs using the same hardware could have huge implications for lowering cost and complexity. This can only help smooth the way toward the so-called metaverse-ready networks of the future.