Hybrid fronthaul planning is becoming a key factor in scaling CF-mMIMO in O-RAN. We analyze how two-level optimization reduces costs and maintains capacity.
The problem does not manifest immediately — until the moment when the network density (ultra-dense networks, UDN) begins to hit fronthaul limitations. In the CF-mMIMO architecture, numerous distributed access points (AP) must synchronously exchange data with computational nodes (DU). This creates high requirements for throughput and latency. The traditional approach of fully utilizing optics provides stability but does not scale well in terms of cost. Alternatives like mmWave and FSO add flexibility but introduce environmental and reliability constraints. As a result, the task boils down not to choosing a technology, but to combining them.
The authors propose a two-level model of hybrid fronthaul planning, where the topology is formed first, followed by the optimization of total cost of ownership (TCO). The first level is implemented through the NOFAC (Near-Optimal Fronthaul Association and Configuration) algorithm. It groups APs using K-means, then balances the groups through split/merge rules and minimizes intra-group distances. For radio stripes (RS), a combination of TSP and nearest neighbor is used, while for the hierarchical scheme (HS) — Minimum Spanning Tree. The second level is formalized as an integer linear programming (ILP) problem, where the type of fronthaul (fiber, mmWave, FSO) is selected considering constraints on capacity, QoS, and cost.
A key engineering aspect is working with functional split. The FS8 option requires about 2.95 Gbps per AP, while FS7.2x reduces the load to ~1.73 Gbps by offloading some PHY functions to the edge. This directly impacts technology choice: higher throughput increases pressure on fronthaul and makes wireless options less efficient. The model also accounts for overhead control plane through a coefficient, bringing the calculations closer to real O-RAN scenarios. A worst-case approach is used — all APs are planned for maximum load, which increases reliability but raises costs.
Insights from the model reveal stable patterns. Fiber dominates in decentralized configurations due to stable capacity (10 Gbps per link) and predictability. mmWave proves effective with moderate centralization, where distances are limited and beamforming with high SNR can be utilized. FSO acts as a filling technology — it covers “gaps” in coverage but suffers from atmospheric losses and depends on environmental conditions. Meanwhile, the hierarchical scheme (HS) reduces the risk of a single point of failure compared to RS, where the sequential connection of APs makes the chain vulnerable.
A separate compromise is the topology structure. RS minimizes the number of connections and simplifies routing but worsens fault tolerance. HS adds redundancy through a tree structure, increasing resilience but potentially raising management complexity. In both cases, grouping APs and selecting a leading node (leading AP) are critical, as they determine the connection points to the DU and, consequently, the load on the fronthaul.
For the industry, this appears as a pragmatic shift from a “single technology” to a hybrid architecture. Hybrid fronthaul planning allows for reducing TCO by adapting to specific topology and load. FS7.2x becomes the preferred option, as it balances network and edge computing requirements. Practical application involves planning with respect to network geometry: distances, AP density, and the level of centralization should directly influence the choice between fiber, mmWave, and FSO. This approach reduces infrastructure redundancy and provides controllable scalability.
Information source
arXiv is the largest open preprint repository (since 1991, under the auspices of Cornell), where researchers quickly post working versions of papers; the materials are publicly accessible but do not undergo full peer review, so results should be considered preliminary and, where possible, checked against updated versions or peer‑reviewed journals. arxiv.org