The HSDRAN architecture (Yu, Xue, Bennis, Chen, and Han, IEEE Transactions on Vehicular Technology, 2018) showed that a two-layer hierarchical software-defined radio access network — a global controller in the core plus local controllers co-located with macro base stations — can solve the joint user-association and downlink resource-allocation problem by a delegated dual-subgradient method with provable optimality and per-controller storage and communication guarantees, but only in a static, single-operator setting. The authors explicitly flagged that user mobility and channel fluctuation may prevent convergence of the iterative scheme, that the iterate-averaging primal feasibility transformation is unsuitable under frequent dynamics, and that deeper control hierarchies and multi-operator resource competition with dynamic slicing remain unaddressed. No prior work carries the HSDRAN delegated distributed-optimization framework to dynamic settings with convergence guarantees, provides an online feasibility transformation, or models multi-operator competition within that framework. The problem of carrying hierarchical delegated distributed optimization in software-defined RANs from a static proof of concept to dynamic, multi-operator operation remains open, with the online-convergence core as its precise center.

Background

Software-defined networking (SDN) separates a network's control logic (the control plane) from the forwarding equipment (the data plane), concentrating decisions in a programmable software controller. Applied to cellular radio access networks (RANs), this idea — pioneered by SoftRAN (Gudipati, Perry, Li, and Katti, 2013), which proposed a single centralized controller making transmission decisions for all base stations — promises flexible management of heterogeneous networks (HetNets), where dense deployments of low-power small base stations (SBSs) are underlaid by macro base stations (MBSs), creating strong interference coupling and heavy management overhead. A fully centralized RAN control plane, however, suffers large backhaul latency, heavy computation and communication load, and poor responsiveness, because every user- and cell-level decision must round-trip to one controller in the network core.

HSDRAN (Hierarchical Software-Defined RAN; Yu, Xue, Bennis, Chen, and Han, 2018) addressed this with a two-layer control plane: a global controller (GC) in the core plus local controllers (LCs) co-located with each MBS, each LC managing its MBS and the SBSs associated with it. To demonstrate that coordinated optimization can run on this hierarchy, the paper studies the canonical problem of joint user association (choosing which base station serves each user) and downlink resource allocation, maximizing the weighted proportional-fair utility ∑UwUlog⁡rU\sum_U w_U \log r_U over user rates rUr_U — the standard fairness objective in wireless scheduling — subject to shared bandwidth limits, per-base-station backhaul capacity, a network-wide on-grid power budget, and almost-blank-subframe (ABS) ratios αM\alpha_M, where ABS is an interference-mitigation technique in which a macro base station silences a fraction αM\alpha_M of its subframes so that nearby small-cell users can be served with less interference. The resulting program is convex and satisfies Slater's condition, so strong duality holds and it is solved by a dual subgradient method (DSM): Lagrange multipliers (dual variables) are attached to the constraints and updated by projected subgradient steps, and the update rules are decomposed so that every primal and dual variable is stored and updated at exactly one controller — a delegation scheme following radio locality (for example, a user's variables are held at the LC of its nearest MBS, while the single network-power multiplier sits at the GC). Because subgradient iterates are generally primal-infeasible, the paper adds a primal feasibility transformation (PFT): time-averaging the iterates and applying per-base-station linear scaling produces a feasible network configuration at every iteration. Analysis and simulations show that this delegated scheme reaches the optimum while drastically reducing per-controller storage and inter-controller communication compared with a centralized controller; a companion paper (Chen, Han, Chang, Xue, Zhang, and Bennis, WCNC 2017) studied downlink power adaptation on the same hierarchical architecture.

Three limitations are stated openly in the paper and constitute the open question addressed in this statement. First, the framework is static: RAN dynamics from user mobility and channel fluctuation may prevent the iterative scheme from converging, and the averaging-based PFT is unsuitable when the network changes faster than averaging can adapt, so the authors defer advanced online optimization and PFT methods to future work. Second, the two-layer hierarchy may not suffice for more complex communication models (device-to-device links, vehicular communications, or wireless sensor overlays hosted by powerful SBSs), which are said to need additional control layers together with new distributed optimization and computation-offloading mechanisms. Third, the model assumes a single network operator, whereas real RANs are dynamically shared among competing operators through network slicing — the partitioning of a shared infrastructure into isolated virtual networks, as envisioned by the RadioVisor slicing plane (Gudipati, Li, and Katti, 2014) — which the authors defer as a separate research direction. In the intervening years, industry has standardized a two-tier hierarchical RAN control architecture in O-RAN (with non-real-time and near-real-time RAN intelligent controllers, i.e., software controller platforms operating at different timescales; see Polese et al., 2023), and the multi-operator slicing literature has grown rapidly, but with heuristic and learning-based coordination rather than the provably convergent delegated distributed optimization that HSDRAN introduced.

Problem Statement

The HSDRAN framework (Yu et al., 2018) proved that a two-layer hierarchical SD-RAN control plane — a global controller (GC) in the core and local controllers (LCs) co-located with macro base stations — can execute, by delegated dual-subgradient iteration, a distributed solution of the joint user-association and downlink resource-allocation problem with per-controller storage and communication guarantees, but only in a static, single-operator setting. The problem is to extend this delegated distributed-optimization paradigm, with the same style of guarantees, to the environments in which RANs actually operate. Its central, precisely stated component is:

(1) Online/dynamic operation. Under user mobility and channel fluctuation, the source paper notes that the dual-subgradient iterations may not converge, and that the iterate-averaging primal feasibility transformation is unsuitable under frequent network dynamics. Design and analyze an online variant of the delegated distributed algorithm — including an online feasibility transformation in place of iterate averaging — that provably converges to, or tracks, the optimal network objective under an explicitly stated model of RAN dynamics that captures the regime flagged in the source: user mobility and channel fluctuation frequent enough that the static dual-subgradient iteration may fail to converge and iterate averaging becomes unsuitable, so that the dynamics cannot be reduced to finitely many change epochs separated by stationary intervals over which the static scheme and averaging remain valid. The variant must preserve the locality-based delegation of variables and the per-controller storage/communication overhead bounds that motivate the architecture.

The same research program, as named in the source, further comprises two architectural extensions whose relation to (1) is that they ask the same question — what distributed-optimization mechanism runs on a hierarchical SD-RAN control plane, and with what guarantees — in broader settings:

(2) Multi-layer control planes. More complex communication models (for example, device-to-device, vehicular, or wireless-sensor overlays hosted by powerful small base stations) are stated to require more than two control layers for scalability and load balancing; develop the associated distributed optimization and computation-offloading mechanisms for such deeper hierarchies, with overhead accounting in the style of the source's storage and communication analysis.

(3) Multi-operator operation. The source assumes a single operator and defers 'modeling and coordinating multi-operator competition': each operator would run its own HSDRAN control plane inside a network slice, with a higher-hierarchy slice manager dynamically allocating radio resources among operators. Formulate this multi-operator resource-competition problem and provide a coordination mechanism whose performance (equilibrium behavior or optimality gap) can be analyzed.

The common scientific objective is to carry hierarchical delegated distributed optimization in software-defined RANs from a static single-operator proof of concept to dynamic, heterogeneous, multi-operator operation, without surrendering the convergence, feasibility, and per-controller scalability guarantees that distinguish the approach from heuristic or purely learning-based control. The central open question of this statement is component (1): an online algorithm with analyzed convergence or tracking behavior under the flagged dynamics regime, together with an online feasibility transformation, preserving locality-based delegation and the per-controller overhead bounds. Components (2) and (3) are further source-named extensions that pose the same question in broader settings; they are described here as context and directions of the same research program, not as additional conditions on component (1).

References

  1. Ruozhou Yu, Guoliang Xue, Mehdi Bennis, Xianfu Chen, Zhu Han. "HSDRAN: Hierarchical Software-Defined Radio Access Network for Distributed Optimization." IEEE Transactions on Vehicular Technology 67(9):8623–8636, 2018. DOI: 10.1109/TVT.2017.2691735. https://doi.org/10.1109/TVT.2017.2691735
  2. Xianfu Chen, Zhu Han, Zheng Chang, Guoliang Xue, Honggang Zhang, Mehdi Bennis. "Adapting Downlink Power in Fronthaul-Constrained Hierarchical Software-Defined RANs." In: Proc. IEEE Wireless Communications and Networking Conference (WCNC) 2017, pp. 1–6. DOI: 10.1109/WCNC.2017.7925773. https://doi.org/10.1109/WCNC.2017.7925773
  3. Aditya Gudipati, Daniel Perry, Li Erran Li, Sachin Katti. "SoftRAN: Software Defined Radio Access Network." In: Proc. Second ACM SIGCOMM Workshop on Hot Topics in Software Defined Networking (HotSDN) 2013, pp. 25–30. DOI: 10.1145/2491185.2491207. https://doi.org/10.1145/2491185.2491207
  4. Aditya Gudipati, Li Erran Li, Sachin Katti. "RadioVisor: A Slicing Plane for Radio Access Networks." In: Proc. Third ACM SIGCOMM Workshop on Hot Topics in Software Defined Networking (HotSDN) 2014, pp. 237–238. DOI: 10.1145/2620728.2620782. https://doi.org/10.1145/2620728.2620782
  5. Wenchao Xia, Tony Q. S. Quek, Jun Zhang, Shi Jin, Hongbo Zhu. "Programmable Hierarchical C-RAN: From Task Scheduling to Resource Allocation." IEEE Transactions on Wireless Communications 18(3):2003–2016, 2019. DOI: 10.1109/TWC.2019.2901684. https://doi.org/10.1109/TWC.2019.2901684
  6. Payam Rahimi, Chrysostomos Chrysostomou, Haris Pervaiz, Vasos Vassiliou, Qiang Ni. "Joint Radio Resource Allocation and Beamforming Optimization for Industrial Internet of Things in Software-Defined Networking-Based Virtual Fog-Radio Access Network 5G-and-Beyond Wireless Environments." IEEE Transactions on Industrial Informatics 18(6):4198–4209, 2022. DOI: 10.1109/TII.2021.3126813. https://doi.org/10.1109/TII.2021.3126813
  7. Chia-Cheng Hu. "Profit-Based Algorithm of Joint Real-Time Task Scheduling and Resource Allocation in C-RANs." IEEE Internet of Things Journal 8(2):941–950, 2021. DOI: 10.1109/JIOT.2020.3008173. https://doi.org/10.1109/JIOT.2020.3008173
  8. An Du, Jie Jia, Schahram Dustdar, Jian Chen, Xingwei Wang. "Online Service Placement, Task Scheduling, and Resource Allocation in Hierarchical Collaborative MEC Systems." IEEE Transactions on Services Computing 18(2):983–997, 2025. DOI: 10.1109/TSC.2025.3536307. https://doi.org/10.1109/TSC.2025.3536307
  9. Sisamouth Hongvanthong. "Novel Four-Layered Software Defined 5G Architecture for AI-based Load Balancing and QoS Provisioning." In: Proc. 5th International Conference on Computer and Communication Systems (ICCCS) 2020, pp. 859–863. DOI: 10.1109/ICCCS49078.2020.9118463. https://doi.org/10.1109/ICCCS49078.2020.9118463
  10. Joao F. Santos, Wei Liu, Xianjun Jiao, Natal V. Neto, Sofie Pollin, Johann M. Marquez-Barja, Ingrid Moerman, Luiz A. DaSilva. "Breaking Down Network Slicing: Hierarchical Orchestration of End-to-End Networks." IEEE Communications Magazine 58(10):16–22, 2020. DOI: 10.1109/MCOM.001.2000406. https://doi.org/10.1109/MCOM.001.2000406
  11. Junling Li, Weisen Shi, Peng Yang, Qiang Ye, Xuemin Sherman Shen, Xu Li, Jaya Rao. "A Hierarchical Soft RAN Slicing Framework for Differentiated Service Provisioning." IEEE Wireless Communications 27(6):90–97, 2020. DOI: 10.1109/MWC.001.2000010. https://doi.org/10.1109/MWC.001.2000010
  12. Hao Zhou, Medhat Elsayed, Melike Erol-Kantarci. "RAN Resource Slicing in 5G Using Multi-Agent Correlated Q-Learning." In: Proc. IEEE 32nd Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC) 2021, pp. 1179–1184. DOI: 10.1109/PIMRC50174.2021.9569358. https://doi.org/10.1109/PIMRC50174.2021.9569358
  13. Arman Sanaei, Massoud Reza Hashemi. "Adaptive, Profit-Aware RAN Slicing for Multi-Operator Networks via Spatio-Temporal Prediction and Energy-Aware SAC." IEEE Transactions on Network and Service Management 23:4849–4865, 2026. DOI: 10.1109/TNSM.2026.3694799. https://doi.org/10.1109/TNSM.2026.3694799
  14. Michele Polese, Leonardo Bonati, Salvatore D'Oro, Stefano Basagni, Tommaso Melodia. "Understanding O-RAN: Architecture, Interfaces, Algorithms, Security, and Research Challenges." IEEE Communications Surveys & Tutorials 25(2):1376–1411, 2023. DOI: 10.1109/COMST.2023.3239220. https://doi.org/10.1109/COMST.2023.3239220