Delay-Aware Task Offloading for Heterogeneous VLC-RF-based Vehicular Fog Computing

Nan An 1 , Hongyi He 1 , Fang Yang 1 2 , Chang Liu 3 4 , Jian Song 5 , Zhu Han 6 , and Binbin Zhu 7 · 2026

Read Delay-Aware Task Offloading for Heterogeneous VLC-RF-based Vehicular Fog Computing by Nan An 1 , Hongyi He 1 , Fang Yang 1 2 , Chang Liu 3 4 , Jian Song 5 , Zhu Han 6 , and Binbin Zhu 7 (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-2601-10978.txt.

Introduction

Nan An and co-authors design delay-aware task offloading for vehicular fog computing. Cars can hand computing tasks to nearby fog nodes to cut delay. That serves the delay-sensitive services of next-generation transport. Classic schemes rely on radio frequency links. Radio struggles in dense traffic, where channels crowd and fade.

The authors build a heterogeneous setup that mixes visible light communication with radio. Light links offer fresh bandwidth where radio saturates. Offloading choices must weigh delay against the quirks of each link. The work targets smart roads where many vehicles compute at once and every millisecond of delay can matter for safety services.

Catalog record

CREATORNan An 1 , Hongyi He 1 , Fang Yang 1 2 , Chang Liu 3 4 , Jian Song 5 , Zhu Han 6 , and Binbin Zhu 7
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-10978

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Delay-Aware Task Offloading for Heterogeneous VLC-RF-based Vehicular Fog Computing Nan An 1 , Hongyi He 1 , Fang Yang 1 2 , Chang Liu 3 4 , Jian Song 5 , Zhu Han 6 , and Binbin Zhu 7 Abstract Vehicular fog computing (VFC) is a promising paradigm for reducing the computation burden of vehicles, thus supporting delay-sensitive services in next-generation transportation networks. However, traditional VFC schemes rely on radio frequency (RF) communications, which limits their adaptability for dense vehicular environments. In this paper, a heterogeneous visible light communication (VLC)-RF architecture is designed for VFC systems to facilitate efficient task offloading. Specifically, computing tasks are dynamically partitioned and offloaded to idle vehicles via both VLC and RF links, thereby fully exploiting the interference resilience of VLC and the coverage advantage of RF. To minimize the average task processing delay (TPD), an optimization problem of task offloading and computing resource allocation is formulated, and then solved by the developed residual-based majorization-minimization (RBMM) algorithm. Simulation results confirm that the heterogeneous VLC-RF architecture with the proposed algorithm achieves a 15% average TPD reduction compared to VFC systems relying solely on VLC or RF. I Introduction The growth of delay-sensitive applications in transportation networks, such as autonomous driving, has intensified the demand for low-delay processing of computation-intensive tasks, which often exceeds the ability of onboard computing resources. Although cloud computing provides abundant resources, its inherent delay renders it unsuitable for delay-critical applications [ 3 ] . Besides, while edge computing is geographically close [ 11 ] , it is hindered by bandwidth limitations and resource competition when serving multiple vehicles simultaneously [ 14 ] . To bridge this gap, vehicular fog computing (VFC) has emerged as a promising paradigm for next-generation networks [ 17 ] , where vehicular computing tasks are offloaded to nearby vehicles with idle computing resources, effectively reducing the task processing delay (TPD). Since TPD encompasses computing and transmission delays, both task offloading and transmission have become key issues in VFC systems. For task offloading, bipartite task partitioning divides computing tasks into two components, which are then processed by onboard central processing unit (CPU), edge servers, or dedicated service vehicles [ 7 ] . Nevertheless, such a scheme only leveraged the computing resources from a limited number of vehicles, failing to exploit substantial idle resources in the network. In the case of multi-partition offloading, tasks were pre-split and offloaded to multiple idle vehicles based on fuzzy logic [ 2 ] and overlapping coalition game [ 17 ] , yet static task partitioning ignored dynamic variations in achievable data rates and available vehicle computing resources, leaving significant optimization potential. Furthermore, dynamic multi-partitioning of tasks and computing resources trading for task execution were enabled via multi-agent reinforcement learning [ 14 ] , but its simplified assumption of interference-free parallel transmissions undermines its applicability in realistic multi-vehicle scenarios. For task transmission, existing VFC implementations predominantly depend on radio frequency (RF) communications for broad coverage, suffering from spectral congestion and competition-induced failures, especially in dense vehicular networks [ 9 ] . In contrast, via directional line-of-sight (LoS) links between headlamps/taillights and photodiodes, visible light communication (VLC) inherently mitigates inter-vehicle interference, which was utilized for task offloading in edge computing [ 16 ] . Under this circumstance, the heterogeneous VLC-RF architecture aims to combine the interference resilience of VLC and the extended coverage of RF [ 10 ] , which is expected to reduce the transmission delay and ensure the timely completion of computing tasks. The primary contributions of this paper are summarized as follows. Firstly, to the best of our knowledge, it is the first time to integrate heterogeneous VLC-RF communications into VFC, with a comprehensive system model being constructed, where the dynamic multi-partitioning of tasks is enabled to fully utilize the idle computing resources. Secondly, the average TPD minimization problem is formulated, and the corresponding residual-based majorization-minimization (RBMM) algorithm is designed. Thirdly, numerical results demonstrate a 15% average TPD reduction of the heterogeneous VFC system with the proposed algorithm compared to VFC systems relying solely on VLC or RF. In this paper, the system model of heterogeneous VFC is described in Section II . The formulation of the average TPD minimization problem is detailed in Section III , and solved by the RBMM algorithm in Section IV . Numerical simulations are shown in Section V , and finally Section VI draws the conclusion. II System Model Fig. 1: The model of the heterogeneous VLC-RF-based VFC system. In Figure 1 , a heterogeneous VLC-RF-based VFC system is considered in transportation network, which comprises N vehicles with width D_{1} and length D_{2} driving on a multi-lane road. The spatial coordinate of vehicle n is denoted as (x_{n},y_{n}) . While a vehicle is driving, it handles arriving computation-intensive tasks, where the input size and corresponding computation intensity of vehicle n are indicated by \rho_{n} and \kappa_{n} , respectively. Moreover, the local computing (LC) resource (measured in CPU cycles per second) of vehicle n is indicated by C_{n} . To ensure timely task execution, computing tasks are dynamically partitioned between local execution and offloading to multiple idle vehicles via both RF and VLC. In this context, vehicles with task arrivals are designated as task vehicles (TVs) and idle vehicles as service vehicles (SVs), with the set of TVs and SVs represented by \mathcal{U}=\{1,2,\dots,U\} and \mathcal{V}=\{1,2,\dots,V\} , respectively. For convenience, a virtual SV V+1 is created, where the task offloaded to it is locally computed by the corresponding TV. Consequently, the augmented set of SVs is expressed as \mathcal{V}_{0}=\mathcal{V}\cup\{V+1\} . The task partition matrix is defined as \bm{M}\in\mathbb{R}^{(V+1)\times U} , where \mathbb{R} indicates the field of real numbers. Its element m_{v,u} denotes the task proportion offloaded from TV u to SV v , and m_{V+1,u} corresponds to the locally processed proportion at TV u . Besides, the computing resources allocation matrix is defined as \bm{F}\in\mathbb{R}^{V\times U} , with its element f_{v,u} specifying the proportion of computing resources allocated by SV v to process the task of TV u . Therefore, the computing delay for SV v to process the task of TV u is calculated by Z_{v,u}=m_{v,u}\rho_{u}\kappa_{u}/(f_{v,u}C_{v}) . To reduce TPD, each TV simultaneously offloads tasks to multiple SVs via both RF and VLC, leveraging both communication systems to maintain sustainable offloading during vehicle mobility. To allow concurrent transmission among multiple vehicles, orthogonal frequency-division multiple access is employed in both VLC and RF subsystems, with subchannels denoted as \mathcal{K}=\{1,2,\dots,K\} and \mathcal{L}=\{1,2,\dots,L\} , respectively. Furthermore, subchannel assignment is indicated by binary variables a_{v,u,k} and b_{v,u,l} . Specifically, a_{v,u,k}=1 designates that the subchannel k in the VLC subsystem, while b_{v,u,l}=1 describes the subchannel l in the RF subsystem of TV u is allocated to SV v , and 0 otherwise. For RF subsystem, the channel gain on subchannel l between TV u and SV v is denoted by g_{v,u,l} as modeled in [ 8 ] , and the transmission power of vehicle TV u on subchannel l is signified by q_{u,l} , with the collective power matrix as \bm{Q} . Subsequently, the achievable data rate on subchannel l of the RF link between TV u and SV v is represented as S_{v,u,l}=\Gamma_{\mathrm{R}}\log_{2}\left(1+\frac{|g_{v,u,l}|^{2}q_{u,l}}{\sum_{i\in\mathcal{U}\backslash u}|g_{v,i,l}|^{2}q_{i,l}+\mu_{\mathrm{R}}\Gamma_{\mathrm{R}}}\right), (1) where \Gamma_{\mathrm{R}} and \mu_{\mathrm{R}} are the subchannel

Read the complete text