RF-LSCM: Pushing Radiance Fields to Multi-Domain Localized Statistical Channel Modeling for Cellular Network Optimization
Bingsheng Peng, Shutao Zhang, Xi Zheng, Ye Xue, Xinyu Qin · 2025
Read RF-LSCM: Pushing Radiance Fields to Multi-Domain Localized Statistical Channel Modeling for Cellular Network Optimization by Bingsheng Peng, Shutao Zhang, Xi Zheng, Ye Xue, Xinyu Qin (2025) 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-2509-13686.txt.
Introduction
Bingsheng Peng and five co-authors upgrade wireless channel modeling in a 2025 wireless networks paper. Optimizing a cellular network needs accurate local predictions of how signals spread. Those predictions are summarized by an angular power spectrum. The standard modeling framework infers that spectrum from received-power measurements. But it handles only one cell, one grid area, and one carrier frequency at a time.
Their framework, RF-LSCM, represents signal loss and multipath inside a radiance field. It borrows machinery from computer graphics. A frequency model built on physics supports predictions across frequencies. Point-cloud environment data extends coverage to many cells and grids. A low-rank tensor method with a new angular modeling algorithm cuts memory and training time. Real-world tests cut coverage prediction error by up to about thirty percent.
Catalog record
| CREATOR | Bingsheng Peng, Shutao Zhang, Xi Zheng, Ye Xue, Xinyu Qin |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 59 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-13686
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
RF-LSCM: Pushing Radiance Fields to Multi-Domain Localized Statistical Channel Modeling for Cellular Network Optimization Thanks: Manuscript created June, 2025; Bingsheng Peng, Xinyu Qin are with the School of Science and Engineering, The Chinese University of Hong Kong, Shenzhen 518172, China, and also with Shenzhen Research Institute of Big Data, Shenzhen 518172, China (e-mail: [email protected]; [email protected]). Shutao Zhang and Xi Zheng are with the Networking and User Experience Laboratory, Huawei Technologies, Shenzhen 518129, China (e-mail: [email protected]; [email protected]). Ye Xue is with Shenzhen Research Institute of Big Data, Shenzhen 518172, China, and also with the School of Data Science, The Chinese University of Hong Kong, Shenzhen 518172, China (e-mail: [email protected]). Tsung-Hui Chang is with the School of Artificial intelligence, The Chinese University of Hong Kong, Shenzhen 518172, China, and also with Shenzhen Research Institute of Big Data, Shenzhen 518172, China. (e-mail: [email protected]). Bingsheng Peng, Shutao Zhang, Xi Zheng, Ye Xue, Xinyu Qin Abstract Accurate localized wireless channel modeling is a cornerstone of cellular network optimization, enabling reliable prediction of network performance during parameter tuning. Localized statistical channel modeling (LSCM) is the state-of-the-art channel modeling framework tailored for cellular network optimization. However, traditional LSCM methods, which infer the channel’s Angular Power Spectrum (APS) from Reference Signal Received Power (RSRP) measurements, suffer from critical limitations: they are typically confined to single-cell, single-grid and single-carrier frequency analysis and fail to capture complex cross-domain interactions. To overcome these challenges, we propose RF-LSCM, a novel framework that models the channel APS by jointly representing large-scale signal attenuation and multipath components within a radiance field. RF-LSCM introduces a multi-domain LSCM formulation with a physics-informed frequency-dependent Attenuation Model (FDAM) to facilitate the cross frequency generalization as well as a point-cloud-aided environment enhanced method to enable multi-cell and multi-grid channel modeling. Furthermore, to address the computational inefficiency of typical neural radiance fields, RF-LSCM leverages a low-rank tensor representation, complemented by a novel Hierarchical Tensor Angular Modeling (HiTAM) algorithm. This efficient design significantly reduces GPU memory requirements and training time while preserving fine-grained accuracy. Extensive experiments on real-world multi-cell datasets demonstrate that RF-LSCM significantly outperforms state-of-the-art methods, achieving up to a 30% reduction in mean absolute error (MAE) for coverage prediction and a 22% MAE improvement by effectively fusing multi-frequency data. Fig. 1: System Workflow of RF-LSCM. The framework models the cellular environment as a 3D voxel grid represented by a low-rank tensor and integrates a novel FDAM within a radiance field while using point clouds to regularize the radiance field parameters. This allows the system to render the channel APS, enabling accurate prediction of base station coverage performance for diverse antenna configurations. I Introduction The rapid advancement of the fifth-generation (5G) mobile communication technology and the ongoing exploration of the sixth-generation (6G) systems have made cellular network optimization a central focus of research [ 1 ] . In 5G networks, the optimization of wireless cellular networks is increasingly challenging due to the scaling-up of the number of parameters to be optimized, including those associated with antennas, handover processes, beamforming techniques, and carrier configurations [ 2 ] . Traditional network optimization methods, which depend on engineering expertise and multiple rounds of drive tests [ 3 , 4 ] , are often labor-intensive and unable to meet the performance requirement of a large-scale cellular network. Recently, the concept of digital twins [ 5 ] has motivated the use of data-driven approaches to perform offline network optimization, avoiding risks caused by online parameter tuning. This approach allows for the modeling and simulation of wireless networks without the need of multi-round manual intervention, thereby enhancing efficiency and reducing cost in the network optimization process. Channel modeling is one of the most fundamental components for network optimization since an accurate channel model provides reliable predictions of network performance during the parameter optimization process. Existing 5G channel models, including GBSMs [ 6 , 7 ] and ray-tracing based models [ 8 , 9 , 10 ] , are inadequate for the network optimization task. Specifially, GBSMs are designed for typical communication scenarios and cannot provide localized modeling for targeted scenarios, while ray tracing cannot perform well without accurate map information [ 11 , 12 , 13 ] . Simplified empirical models such as COST231-Hata [ 14 ] are too simple to capture the complexity and randomness of 5G networks. These limitations result in unsatisfactory optimization configurations that underperform in real-world scenarios. To address these limitations, data and physic based localized statistical channel modeling (LSCM) [ 15 ] has been proposed as a tailored solution for cellular network optimization. LSCM characterizes the multi-path structure of localized propagation environments by estimating the fine-grained angular power spectrum (APS) of wireless channels based on multi-beam Reference Signal Received Power (RSRP) measurements. Notably, LSCM has been successfully implemented in the SRCON platform [ 16 , 17 ] , a real-world telecom optimization tool used for 5G network. SRCON integrates statistical mechanisms and builds large-scale simulation models of real-world networks, including multi-path channels, user distribution, and base station responses. This approach has enabled significant achievements in 5G network optimization, such as leading the European P3 assessments in Spain, Germany, the Netherlands, and Switzerland [ 18 ] . Fig. 2: The global environmental impact on channel propagation. On the left, the illustration shows that different grids may exhibit similar channel characteristics, with angle differences less than a small value \Delta\omega , due to environmental reflections. On the right, it demonstrates that channels from different cells share the same characteristics when radio waves traverse the same space and intersect. However, the original LSCM approach primarily focuses on modeling the APS between a base station (BS) and a single geometric grid, only using RSRP data from that grid while ignoring the geometric relationships across different grids. Although subsequent works [ 19 , 20 ] have attempted to extend LSCM to multi-grid scenarios, these methods still rely on oversimplified statistical priors between adjacent grids, thus fail to model the intricate interactions between radio wave propagation and the physical environment, as illustrated in Fig. 2 . Compounding by this issue, their applicability remains strictly confined to single-cell and single-carrier frequency scenarios. This critical limitation hinders their ability to leverage the intrinsic dependencies within multi-domain data (multi-carrier frequency, multi-cell, and multi-grid data), which in turn restricts their effectiveness for large-scale network optimization tasks. Consequently, there is a clear and growing necessity for an advanced localized channel modeling approach, one that can faithfully characterize the influence of the environment on radio wave propagation and truly address the limitations of existing methods. Recent developments in deep learning have markedly enhanced the simulation of radio-wave propagation, offering great promise for more accurate channel modeling. A recent state-of-the-art method, named neural radio-frequency radiance fields (NeRF 2 ) [ 21 ] , employs a neural radiance field (RF) to model the radio signal of indoor scenarios by tracing rays from all possible directions. However, NeRF 2 primarily focuses on radio map construction, not on the fine-grained APS modeling required for LSCM. APS modeling from RSRP is intrinsically an ill-imposed estimation problem and is more challenging to solve than the tasks considered in [ 21 ] . Moreover, while this neural radiance field method has shown strong potential, its direct application to outdoor cellular