Editor’s observe: I’m within the behavior of bookmarking on LinkedIn, books, magazines, motion pictures, newspapers, and data, issues I believe are insightful and attention-grabbing. What I’m not within the behavior of doing is ever revisiting these insightful, attention-grabbing bits of commentary and doing something with them that may profit anybody apart from myself. This weekly column is an effort to appropriate that.
What’s there to say in regards to the double-edged sword that’s the radio entry community (RAN) that hasn’t already been mentioned? Operators need to near-constantly undergo the capital-intensive strategy of densifying, modernizing, and upgrading the RAN to broaden protection and ship extra capability. After which there’s the ever-increasing ramp in operational complexity and price that goes together with delivering service and in any other case maintaining the factor working. As such, operators maintain their heads on a swivel for alternatives to take capex and opex out of the RAN whereas determining the right way to monetize the factor past simply promoting connectivity.
Since circa 2018, Open RAN — by way of the operator-led O-RAN Alliance — has been working to standardize interfaces between central, distributed, and radio models, whereas additionally defining and commercializing the RAN Clever Controller (RIC) so varied rApps and xApps can be utilized to, once more, tame complexity and price whereas including capabilities that may, in flip, be monetized. Then it was all about Cloud RAN, a vendor-led initiative to show the radio community right into a horizontal cloud platform that can be utilized to, once more, tame complexity and price whereas including capabilities that may, in flip, be monetized. Now, in step with the bigger international know-how cycle, it’s all about AI RAN the place synthetic intelligence is embedded all through the RAN to be able to, once more, tame complexity and price whereas including capabilities that may, in flip, be monetized.
Be aware the sample. Additionally observe that the RAN remains to be operators’ greatest value middle, and that operators’ revenues, usually, stay flat.
Dell’Oro Group Vice President Stefan Pongratz, in a LinkedIn publish at present, teased new analysis on AI RAN. The important thing takeaways are:
- “AI RAN is occurring.”
- “The know-how will attain vital scale within the second half of 5G.”
- And “AI RAN just isn’t anticipated to develop the general RAN pie.”
Quick-term, it’s “extra about effectivity features than new income streams. There may be sturdy consensus that AI RAN can enhance the person expertise, improve efficiency, scale back energy consumption, and play a essential position within the broader automation journey. Unsurprisingly, nonetheless, there’s better skepticism about AI’s functionality to reverse the flat income trajectory that has outlined operators all through the 4G and 5G cycles.” Oof.
The view from the AI-RAN Alliance
Through the current Telco AI Discussion board (accessible on demand right here), I had the chance to meet up with AI-RAN Alliance Chairman Alex Choi, additionally a analysis fellow with SoftBank. At Cellular World Congress Barcelona 2024, the AI-RAN Alliance unveiled its foundational imaginative and prescient for AI for RAN, AI and RAN, and AI on RAN. These respectively communicate to “advancing RAN capabilities by way of AI to enhance spectral effectivity, integrating AI and RAN processes to make the most of infrastructure extra successfully and generate new AI-driven income alternatives, [and] deploying AI providers on the community edge by way of RAN to extend operational effectivity and provide new providers to cellular customers.”
As an apart, Pongratz mentioned incumbent RAN distributors Ericsson, Huawei, Nokia, Samsung, and ZTE, “are well-positioned within the preliminary AI-RAN part, pushed primarily by AI-for-RAN upgrades leveraging the present {hardware}.”
Again to our dialog with Choi. “We’ve made super progress since our debut.” He mentioned membership has handed 90 firms from 18 international locations, and the alliance has demonstrated proofs-of-concept breakthroughs in areas like AI-native air interface design, deep studying for uplink channel estimation, mobility-aware energy saving, superior spectrum sensing, and dynamic AI mannequin partitioning. The takeaway, Choi mentioned, is “all these demos usually are not simply hypothetical.” They validate the group’s mission and sign motion “from idea to early actual world experimentation.”
To scale this out in the true world, operators will want extra RAN compute sources. Which means extra highly effective CPUs and GPUs to speed up AI workloads whether or not that’s in service of AI for RAN, AI and RAN, and/or AI on RAN. Bottomline, Choi mentioned, the entire thing is in flux and there received’t be a one-size-fits-all strategy to accelerating AI workloads.
He tracked the normal strategy to RAN processing as anchored in digital sign processors contained in an ASIC-based structure. However, “We’re seeing clear indications it’s evolving to a hybrid computing platform that means combining CPUs and GPUs.” CPUs, as an illustration, are well-suited for Layer 2 and Layer 3 management aircraft duties. Whenever you get into compute-intensive Layer 1 duties like sign processing or huge MIMO beamforming, there’s a case for GPU-enabled parallel processing. “It delivers clear value/efficiency benefits,” Choi mentioned.
“Realistically,” Choi mentioned, “we are going to very probably see hybrid CPU and GPU platforms rising because the dominant structure for AI RAN within the coming years.” In some deployment situations, notably dense city environments with excessive and variable capability demand, “It’s more and more probably that accelerators like GPUs might be wanted.”
Choi and I additionally mentioned AI and RAN — suppose GPU-as-a-Service, what meaning for community engineering groups, and so forth. Take a look at the session. However extending past AIOps for the RAN the place AI delivers operational efficiencies to operators, the case for AI RAN probably turns into essential within the 6G period when the contra-forces of huge channel bandwidths and lack of spectrum will create a forcing operate making AI RAN essential.
6G, AI and MRSS
Choi described multi-radio entry spectrum sharing (MRSS) because the mechanism to manipulate shared spectrum throughout 5G, 6G, Wi-Fi, satellite-based, and different radio entry mediums. “As we transfer towards 6G, we’re coming into an period the place spectrum sharing, huge bandwidth, and ultra-dense deployments make…typical, guide community administration utterly not possible.”
AI fashions can predict spectrum occupancy and demand patterns based mostly on historic and real-time knowledge to allow proactive spectrum allocation earlier than congestion or interference happens. Reinforcement studying brokers can repeatedly calibrate the optimum approach to allocate spectrum throughout completely different entry applied sciences. And AI brokers, working throughout radio and edge websites, can autonomously steadiness site visitors masses.
“With out synthetic intelligence know-how, the complexity and pace required for the environment friendly MRSS operation in 6G could be virtually unmanageable, making AI-native RAN structure in 6G an important basis for the long run,” Choi mentioned.
Now let’s return to the connection between Open RAN, Cloud RAN, and AI RAN. This got here up in a webinar I hosted final week with some good of us from LitePoint, Spirent Communications, and VIAVI Options. We had been speaking in regards to the state of mess around Open RAN however to be able to make {that a} complete dialog it essentially wanted to (and did) embody dialogue of Open RAN, Cloud RAN and AI RAN. You may try the complete session right here.
Suffice to say, my line of pondering is that these are complementary approaches to RAN evolution which are maybe being characterised as mutually unique as a result of that’s expedient to how established RAN distributors promote and ship tools. However the incumbent distributors engaged in O-RAN Alliance standardization efforts to guard their pursuits and steer the ecosystem. In doing so, they opened up some components of their RAN stacks for integration with third-party {hardware} and software program. Count on extra of that within the AI RAN period as a result of nobody firm — properly, perhaps precisely one firm — can ship every part wanted for operators to do all three flavors of AI RAN.
To place that extra succinctly, Open RAN is the interoperable basis. Cloud RAN is the structure. And AI RAN is the ambition.
