A benchmarking framework for PON-based fronthaul network design
Egemen Erbayat 1 , Gustavo B. Figueiredo 2 , Shih-Chun Lin 3 , Motoharu Matsuura 4 , Hiroshi Hasegawa 5 , Suresh Subramaniam 1 · 2026
Read A benchmarking framework for PON-based fronthaul network design by Egemen Erbayat 1 , Gustavo B. Figueiredo 2 , Shih-Chun Lin 3 , Motoharu Matsuura 4 , Hiroshi Hasegawa 5 , Suresh Subramaniam 1 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-14480.txt.
Introduction
Egemen Erbayat and five co-authors set shared ground rules for network design in this 2026 paper. Mobile networks are moving toward 5G and 6G architectures. Passive optical networks offer a critical solution for cost-effective fronthaul transport, the links that carry data from antennas inward. Many optimization strategies compete in this space.
The literature lacks standardized evaluation models, so objective comparison of those strategies is hard. The team proposes a unified benchmarking framework. It standardizes cost catalogs and deployment scenarios for PON-based fronthaul design. Readers get a common yardstick, so future design claims can be checked against the same facts.
Catalog record
| CREATOR | Egemen Erbayat 1 , Gustavo B. Figueiredo 2 , Shih-Chun Lin 3 , Motoharu Matsuura 4 , Hiroshi Hasegawa 5 , Suresh Subramaniam 1 |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 76 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-14480
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A benchmarking framework for PON-based fronthaul network design Egemen Erbayat 1 , Gustavo B. Figueiredo 2 , Shih-Chun Lin 3 , Motoharu Matsuura 4 , Hiroshi Hasegawa 5 , Suresh Subramaniam 1 Abstract As mobile networks transition toward 5G and 6G RAN architectures, Passive Optical Networks (PONs) offer a critical solution for cost-effective fronthaul transport. However, the lack of standardized evaluation models in current literature makes an objective comparison of diverse optimization strategies difficult. This paper addresses this gap by proposing a unified benchmarking framework that standardizes cost catalogs and deployment scenarios. We formulate the network design problem using Integer Linear Programming (ILP) to establish optimality bounds and evaluate three scalable heuristic strategies: a Genetic Algorithm, K-Means Clustering (KMC+), and a graph-based Randomized Successive Splitter Assignment (RSSA+) algorithm. Simulation results show that a time-limited ILP remains a strong reference point, even when optimality is not reached. Despite being rarely used in prior fronthaul planning studies, it consistently yields solutions superior to those produced by standard heuristic methods. Among scalable approaches, RSSA+ reliably attains near-ILP performance while ensuring feasibility across all evaluated scenarios, which underscores the importance of advanced, constraint-aware algorithmic designs over simpler heuristics. The complete benchmarking framework and datasets are publicly shared in [ 22 ] . 1 { erbayat, suresh } @gwu.edu , 2 [email protected] , 3 [email protected] , 4 [email protected] , 5 [email protected] I Introduction The unprecedented demand for ultra-high-speed, low-latency, and reliable connectivity is reshaping mobile network architecture as the industry transitions from 5G to the vision of 6G [ 12 , 27 ] . Driven by applications such as immersive extended reality (XR), autonomous mobility, massive Internet of Things (IoT), and the tactile Internet, future wireless systems must support orders-of-magnitude growth in data volume, device density, and service diversity [ 34 ] . To meet these performance goals, Radio Access Networks (RANs) are becoming increasingly disaggregated, with baseband functions centralized and cell deployments densified. While Centralized or Cloud-RAN (C-RAN) architectures provide enhanced spectral efficiency and lower operational cost, they place heavy demands on the fronthaul segment linking remote radio heads (RRHs) with centralized baseband units (BBUs) [ 7 , 10 , 11 ] . As fronthaul links are required to simultaneously provide multi-gigabit throughput, sub-millisecond latency, and synchronization precision, transport acts as a critical bottleneck for scalable 5G and 6G deployments [ 55 , 1 , 39 ] . Passive Optical Networks (PONs) provide a high-capacity, low-latency, and cost-effective solution for delivering fiber-based transport from the central office to numerous cell towers [ 31 ] . Since PONs facilitate shared fiber infrastructure and avoid the need for a dedicated fiber per tower, they are uniquely suitable for scenarios where multiple cell sites need to be served economically [ 51 ] . Modern standards such as 10-Gigabit-capable Symmetric PON (XGS-PON), Next-Generation PON 2 (NG-PON2), and 50-Gigabit-capable PON (50G-PON) offer bandwidths up to 10–50 Gbps per wavelength and support time and/or wavelength division multiplexing [ 38 , 40 , 43 , 54 ] . Consequently, they are well-suited to carry fronthaul traffic for low-layer functional splits such as 3rd Generation Partnership Project (3GPP) Option 7.2 [ 3 ] . A wide range of research studies have examined the utilization of PONs for fronthaul within 4G, 5G, and anticipated 6G networks. While some works focus on physical-layer feasibility by showing analog and digital transmission of radio signals across PON topologies [ 14 , 60 ] , others introduce optimization models for planning and dimensioning cost-effective and low-latency PON deployments [ 8 , 58 , 57 , 53 , 16 , 2 , 47 , 52 , 6 , 45 , 25 , 26 , 23 , 18 , 20 , 21 , 17 , 19 , 9 ] . These studies use a variety of functional splits and network models to provide a broad perspective. This diversity enriches the field, but it creates context-specific results which are not always suited for direct comparison. The evolution toward open and disaggregated architectures, such as those promoted by the Open RAN (O-RAN) Alliance [ 55 ] 1 1 1 To align with O-RAN specifications, we adopt the terms Distributed Unit (DU) and Radio Unit (RU) throughout this paper. These functionally correspond to the BBU and RRH, respectively, as commonly referenced in traditional D-RAN and C-RAN literature. , further intensifies the need for objective assessment of transport solutions. This capability is crucial to ensure interoperability, scalability, and cost-effectiveness. However, we recognize that specific cost parameters and technical constraints are inherently site-dependent and fluctuate based on the chosen technology and geographical location. This variability creates a significant challenge for comparative research, as differences in reported results often stem from mismatched input assumptions rather than algorithmic performance. Consequently, a key limitation within the current literature is the absence of standardized benchmarking scenarios and unified evaluation frameworks; without these, performance claims remain context-dependent and cannot be fairly compared. In this paper, we address this gap by proposing the first standardized benchmarking framework for PON-based fronthaul design and enable reproducible comparisons that was previously impossible. Rather than relying on isolated case studies, our framework utilizes a standardized cost catalog derived from broad industry averages to mitigate the impact of parameter variability. By defining fixed functional split scenarios, normalized techno-economic models, and standard deployment templates, we establish a controlled and reproducible environment for evaluation. We further validate this framework by quantitatively benchmarking an exact Integer Linear Programming (ILP) formulation against three scalable heuristic strategies. These include a Genetic Algorithm (GA), an improved K-means clustering method (KMC+), and a graph-based Improved Randomized Successive Splitter Assignment (RSSA+). The latter two build on [ 21 ] and are extended in this study to address practical limitations, with KMC+ introducing a principled mechanism for determining the number of clusters and RSSA+ incorporating cost heterogeneity into the assignment process. Taken together, these methods are evaluated within a unified benchmarking framework. Through this benchmarking, we enable transparent comparisons of PON fronthaul architectures for 5G and beyond and ensure that performance differences remain interpretable across diverse optimization methods. Beyond comparative evaluation, our results yield several concrete design insights. First, a time-limited ILP, an exact formulation solved under a fixed runtime budget without requiring optimality, serves as a strong and informative baseline. Its suboptimal solutions consistently outperform heuristics and provide meaningful lower bounds. Second, heuristic performance depends strongly on explicit constraint handling and cost awareness. Simple methods such as GA degrade under strict physical constraints, while structured approaches such as RSSA+ and KMC+ closely track ILP performance at scale, with RSSA+ showing greater flexibility under heterogeneity due to its constructive nature. Third, the results expose clear trade-offs between DU availability, site-related costs, and infrastructure sharing that fixed clustering methods cannot capture. Together, these findings reinforce the need for constraint-aware and cost-sensitive optimization and demonstrate the value of standardized benchmarking for fronthaul design in future mobile networks. II Background II-A Modern RAN Architectures and Fronthaul Requirements Fronthaul networks play a critical role in modern RAN deployments by enabling centralized processing, improved spectral efficiency, and more flexible resource allocation across radio sites. High-capacity, low-latency fronthaul links allow baseband resources to be pooled, reducing hardware duplication, lowering energy consumption, and simplifying network management. These benefits have been a