AI applications for RAN – tried, but not tested
AI has been touted as the panacea to operators’ network complexity woes. By deploying AI applications for the radio access network (RAN), it’s hoped that operators will be able to automate and optimize base station processes. Many operators are already investing in the technology: spending on AI-driven network management software is forecast to grow to more than $1.9 billion 2021, while annual outlay on AI network operations solutions is expected to amount to $7.4 billion by 2025. Yet, while artificial intelligence (AI) and machine learning (ML) have been talked up in telecoms, there’s been far less noise around validating the performance of these approaches.
Next generation networks = next level of complexity
Next generation networks will be tasked with enabling a broad spectrum of 5G use cases and a huge number of diverse end-points. Increasing deployment of network functions virtualization and software-defined networking will strain legacy networks and throw up integration challenges. Finally, next-generation telecoms infrastructure must also support technologies like millimetre wave, carrier aggregation, massive MIMO, and beamforming – with each optimized to deliver the best possible network performance and highest quality end-user experience.
Just as we’re seeing the emergence of 5G use cases and deployments, so too are we seeing AI solutions discussed, marketed and adopted to ensure networks continue to perform – for both end-users and operators’ revenues. Ericsson, for example, recently announced it had joined the O-RAN Alliance, where it will pursue research on the open interworking between RAN and network orchestration and automation, with a focus on AI-enabled automation and optimization. Nokia, meanwhile, launched its Cognitive Collaboration Hubs just in time for MWC19, which will help support its efforts to improve 5G network planning (including massive MIMO and beamforming configuration) using ML.
Both of these technologies – massive MIMO and beamforming – are crucial in delivering dynamic user-based coverage for 5G radio access networks. Beamforming is the ability to generate and shape multiple beams using a much larger antenna array by manipulating the signals and directing energy to an end-user’s specific service area. This reduces interference and optimizes the use of RF spectrum.
As the industry migrates to 5G, the number of antennas – and therefore beams – involved in network architectures will multiply, creating further complexities in scheduling and configuration. Failure to optimize beamforming can affect the quality of end-user experience, so it’s little surprise that a growing number of operators are looking to implement AI solutions at base-station level, to ensure maximum throughput.
Validating the future of 5G networks
However, while it may seem that everyone’s talking about AI, it appears very few are discussing AI validation. Adopting AI for the RAN – with all the promised benefits we’re hearing from vendors – will be a considerable investment for operators. AI deployed in the RAN is still in its relative infancy, with little in the way of data-supported results into its performance. As such, how can operators be sure that the ‘AI-optimized’ beamforming is actually delivering the best results? How can they be certain that they’re getting the return on their AI spend?
VIAVI Solutions is already helping operators validate over-the-air performance. Our CellAdvisor 5G features beam analyzer functionality, which assesses individual beam IDs, power level, and corresponding signal-to-noise ratios. It also offers a 5G route map for coverage verification, mapping in real time the physical cell identity and beam strength, as well as providing coverage data for post-processing.
Our years of R&D and leadership in this field means we know exactly what the optimum beamforming performance looks like. This means we also know exactly when, why and to what extent operators are not achieving maximum network performance and throughput. VIAVI Solutions can validate the RAN to ensure that AI solutions are delivering what vendors are promising, referencing the performance of RAN elements against established benchmarks.
This process can be done in the lab, using our TM500 network testing solution. The TM500 can validate the performance of AI repeatedly, which is particularly important following software upgrades to the RAN, for bug fixing and maintenance. The TM500 can be used in this scenario to ensure that KPIs defined for AI performance continue to be accurate and remain unchanged.
The TM500 also provides scalability, supporting a huge number of UEs in a virtualized environment. The alternative (testing beamforming in the field with real UEs, the approach taken by many operators) is complex, especially when testing real-world scenarios like connected cars in a smart city.
AI deployments for the RAN may be in their early stages, but VIAVI Solutions’ expertise and product development means we’re ahead of the game. By validating the performance of AI-enabled beamforming, we’re able to support operators as they transform their networks and ready their businesses for the 5G future.