Building Adaptive Full Stack AI Pipelines for Federated MLOps in Beyond 5G (B5G) Networks Open Radio Access Network (O-RAN) Frameworks

This blog is directed at data science professionals and AI modelers, as well as those in the telecommunications industry with a research obligation on developing federated AI Pipelines for future B5G O-RAN frameworks.

By 2025 the world is expected to exceed 37 billion devices, with two-thirds of those devices comprising “things” (IoT devices). This equates to approximately 4.3 devices for every man, woman, child, cat, and dog living on this planet. Just focusing on the bandwidth problem (side-barring the energy, emissions, and skillset problems for later), the world’s current state-of-the-art 5G cellular communication systems can only admit approximately 50,000 (narrowband) IoT devices per cell. 

With user demand for bandwidth capacity outpacing current 5G supply and current cellular base-stations physically incapable of meeting the High-Performance Computing (HPC) requirements of computationally intensive cooperative algorithms (which perform signal processing, encoding, and decoding), a radically new Radio Access Network (RAN) approach is needed.

This new virtualized and Artificial Intelligence (AI) enabled Open-Source Radio Access Network (O-RAN) architecture will revolutionize emerging Beyond-5G (B5G) networks and future 6G mobile communication networks by eliminating global connectivity limitations for bandwidth, latency, and performance. Powered by intelligent O-RAN applications with embedded Machine Learning (ML), O-RAN is expected to transform IoT, fully connecting billions of people and machines.

B5G’s emerging Open Radio Access Network (O-RAN) is a disaggregated virtual/physical network design, premised on collaborative open-AI development. Unlike traditional RAN designs, O-RAN data will be processed close to the mobile user, at the Multi-access Edge Computing (MEC) device. O-RAN will necessitate a heavy dependence on automation using Machine Learning (ML) and Deep Learning (DL) techniques. This lofty skillset induces risk since O-RAN’s heavy dependence on AI/ML pipeline techniques presents an immediate skills gap issue that Telecom, Edge, Cloud, and Mobility companies need to first address.

To meet the Quality of Service (QoS) required for B5G use cases, workloads need to be placed as close to the user as possible. For B5G, workloads will be executed at either a public or private MEC device. This will help to ensure Key Performance Indicator (KPI) targets, requiring B5G to service 1 million IoT devices per 0.4mi2/1km2 with 1-millisecond latency and 20Gb/s peak data rates are met.

To directly address the burgeoning user demand placed on cellular base station computational capacity—predicted to increase 3.5X by 2024, largely due to advancements in video-streaming capabilities—O-RAN will require Full Stack AI pipelines driven by automated, secure, and operationalized ML techniques (ie., MLDevSecOps). However, O-RAN’s stringent KPIs will bear a new breed of AI Pipeline, since distributed, Massively Parallel Processing (MPP) techniques—innate to HPC, will be necessary to support Federated-HPC ML/inference at the MEC device.

Still, software-driven AI pipelines can only get us so far … and with hardware ultimately the key limiting factor for wireless capacity, this must eventually be addressed. While advances in hardware are occurring, such as massive MIMO (Multiple Input Multiple Output) wireless antenna breakthroughs, and beamforming technologies, more innovation is needed to sustain the ITU’s (International Telecommunication Union) forecasted 120% IoT device growth rate by 2030. At this time, practical use cases are forecasted for autonomous vehicles and swarm systems, intelligent automation, aerial and satellite networks, volumetric media streaming, as well as, multi-sensory (haptics), and immersive extended reality systems.

To further put this IoT bandwidth challenge in perspective, while Automotive has been active in 5G development since 2017, as evidenced by 5G’s V2X (Vehicle-to-Everything) cellular enhancements; Manufacturing is only just now getting started. By 2026, Manufacturing is expected to be the largest and fastest-growing B5G market with stringent KPIs that include 8-9s of reliability, sub-millisecond latency requirements, and positioning accuracy of ~7.87in/20cm. The telecommunications industry is undergoing a rapid technological revolution. With commercial 5G implementations barely off the ground, telecom stakeholders are already working on greenfield projects for the next release of B5G network, powered by O-RAN implementations for key industries, like automotive and manufacturing. Driven by IoT device growth forecasts over the next decade, visionary researchers are simultaneously working on B5G and O-RAN AI pipelines in parallel with next-gen 6G wireless communications technologies. Many researchers, like myself are looking to Quantum to both satisfy and future-proof the world’s dynamic telecommunication needs.

References:

1. Alain Mourad, Rui Yang, Per Hjalmar Lehne, & Antonio De La Oliva. (2020). A Baseline Roadmap for Advanced Wireless Research Beyond 5G. Electronics, 9(2), 351. https://doi.org/10.3390/electronics9020351
2. Akhtar, M. W., Hassan, S. A., Ghaffar, R., Jung, H., Garg, S., & Hossain, M. S. (2020). The Shift to 6G Communications: Vision and Requirements. Human-Centric Computing and Information Sciences, 10(1). https://doi.org/10.1186/s13673-020-00258-2
3. Attiah, M. L., Isa, A. A. M., Zakaria, Z., Abdulhameed, M. K., Mohsen, M. K., & Ali, I. (2020). A Survey of mmWave User Association Mechanisms and Spectrum Sharing Approaches: An Overview, Open Issues and Challenges, Future Research Trends. Wireless Networks (10220038), 26(4), 2487–2514. https://doi.org/10.1007/s11276-019-01976-x
4. Brik, B., Boutiba, K., & Ksentini, A. (2022). Deep Learning for B5G Open Radio Access Network: Evolution, Survey, Case Studies, and Challenges. IEEE Open Journal of the Communications Society, Communications Society, IEEE Open Journal of the, IEEE Open J. Commun. Soc, 3, 228–250. https://doi.org/10.1109/OJCOMS.2022.3146618
5. Kim, M., Venturelli, D., & Jamieson, K. (2019). Leveraging Quantum Annealing for Large MIMO Processing in Centralized Radio Access Networks. Proceedings of the ACM Special Interest Group on Data Communication, 241–255. https://doi.org/10.1145/3341302.3342072
6. Murali, P., Linke, N. M., Martonosi, M., Abhari, A. J., Nguyen, N. H., & Alderete, C. H. (2019). Full-Stack, Real-System Quantum Computer Studies: Architectural Comparisons and Design Insights. 2019 ACM/IEEE 46th Annual International Symposium on Computer Architecture (ISCA), Computer Architecture (ISCA), 2019 ACM/IEEE 46th Annual International Symposium On, 527–540.
7. Rodrigo, S., Abadal, S., Alarcon, E., & Almudever, C. G. (2020). Will Quantum Computers Scale Without Inter-Chip Comms? A Structured Design Exploration to the Monolithic vs Distributed Architectures Quest. 2020 XXXV Conference on Design of Circuits and Integrated Systems (DCIS), Design of Circuits and Integrated Systems (DCIS), 2020 XXXV Conference On, 1–6. https://doi.org/10.1109/DCIS51330.2020.9268630
8. Rufino Henrique, P. S., & Prasad, R. (2021). 6G Networks for Next Generation of Digital TV Beyond 2030. Wireless Personal Communications: An International Journal, 121(2), 1363–1378. https://doi.org/10.1007/s11277-021-09070-2
9. Saini, S., Khosla, P., Kaur, M., & Singh, G. (2020). Quantum Driven Machine Learning. International Journal of Theoretical Physics, 59(12), 4013–4024. https://doi.org/10.1007/s10773-020-04656-1