Rotatable RIS-Assisted Edge Computing: Orientation, Task Offloading, and Resource Optimization

Bin Li, Dongdong Yang, Lei Liu · 2025

Read Rotatable RIS-Assisted Edge Computing: Orientation, Task Offloading, and Resource Optimization by Bin Li, Dongdong Yang, Lei Liu (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-2503-16879.txt.

Introduction

Bin Li, Dongdong Yang, and Lei Liu published a paper in 2025 on smart reflective panels that can rotate to help phones offload computing. Edge servers near users can take over heavy tasks, saving battery, but weak wireless links waste that promise. A wall of tiny adjustable reflectors can steer signals, and letting the whole panel rotate adds a further gain.

The authors treat phones moving across time slots and aim to cut total energy use. Their plan jointly sets panel angle, reflector phases, server resources, sending power, and what each phone offloads. A learning algorithm from the actor-critic family handles the moving parts, while simpler rules fill in the rest. Against a fixed panel, rotation cut energy use by up to about forty-seven percent.

Catalog record

CREATORBin Li, Dongdong Yang, Lei Liu
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-16879

Browse the full library · About the Tower of Babel

Full text

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

Rotatable RIS-Assisted Edge Computing: Orientation, Task Offloading, and Resource Optimization Bin Li, Dongdong Yang, Lei Liu Abstract The rotatable reconfigurable intelligent surface (RIS) can enhance mobile edge computing (MEC) performance by optimizing its orientation to improve the gain of received and transmitted signals. This correspondence investigates a rotatable RIS-assisted MEC system, aimed at minimizing energy consumption for multiple moving user equipment (UEs) through the joint design of RIS orientation, discrete phase shift, computation resource allocation, transmitting power and task offloading strategies. Considering the mobility of UEs, this problem is formulated as a sequential decision-making across multiple time slots. To address this challenge, a soft actor-critic (SAC)-based algorithm is proposed to optimize RIS orientation, phase shift and task offloading strategies, while computation resource allocation and transmitting power are determined based on the actions. Numerical results demonstrate that the proposed scheme exhibits superior convergence and performance compared to benchmarks. Additionally, the rotatable RIS scheme reduces total energy consumption by up to 47.3% compared to the fixed RIS, enhancing MEC system performance. I Introduction Reconfigurable intelligent surface (RIS) has gained recognition as a key technology in 6G for its ability to improve channel quality, along with its passivity, reconfigurability and cost-effectiveness [ 1 ] . By integrating RIS with mobile edge computing (MEC), the advantages of RIS can be fully leveraged to significantly enhance offloading links, providing substantial benefits in MEC systems. For example, [ 2 ] introduced an optimization strategy in RIS-assisted MEC non-orthogonal multiple access networks, using \epsilon -constraint approach and a successive convex approximation-based algorithm to minimize the offloading delay and maximize transmission rate. To further enhance energy and computation efficiency, [ 3 ] proposed a novel algorithm based on deep learning. In addition, considering the task offloading security, [ 4 ] explored the benefits of RIS in MEC-enabled networks, utilizing a deep deterministic policy gradient-based algorithm to maximize the weighted sum secrecy computation efficiency among devices. Given that the antenna’s receiving and transmitting gain is affected by the angle of wave incidence and transmission [ 5 ] , the RIS orientation optimization becomes a crucial aspect worth investigating. Nevertheless, existing studies have predominantly focused on performance analysis and optimization in fixed or single-UE scenarios, which may not adequately address the complexities of scenarios involving multiple moving UEs in MEC systems. For instance, [ 6 ] evaluated the impact of RIS position and orientation in a single mobile UE environment, highlighting that the orientation significantly influences the received power. Similarly, [ 7 ] demonstrated that optimizing RIS orientation can achieve higher average throughput in low earth orbit satellite communication systems. To the best of our knowledge, the potential of rotatable RIS in MEC systems has not been thoroughly investigated in existing studies. Notably, for rotatable RIS-assisted MEC systems, the phase shift is highly coupled with orientation and offloading strategies, and typically discrete under practical constraints, making it challenging for traditional algorithms to identify optimal solutions [ 8 ] . Nevertheless, deep reinforcement learning (DRL) offers innovative approaches to addressing such optimizing problems in time-varying scenarios. For example, [ 9 ] proposed a soft actor-critic (SAC)-based joint beamforming and base station (BS)-RIS-UE association design to maximize the sum-rate in millimeter wave communication systems. Similarly, in the RIS-assisted MEC context, [ 10 ] developed a twin delayed deep deterministic policy gradient-based algorithm for computation rate maximization. Motivated by the DRL methods, a policy-based SAC algorithm is adopted to tackle the problem of minimizing energy consumption in the rotatable RIS-assisted MEC system with multiple moving UEs, where RIS orientation, discrete phase shift, computation resource allocation, transmitting power, and task offloading strategies are jointly involved. In particular, the SAC algorithm employs a stochastic strategy with entropy regulation as its main feature, enabling it to learn near-optimal policies with minimal environmental information. II System Model and Problem Formulation As illustrated in Fig. 1 , we consider a rotatable RIS-assisted MEC system, consisting of K single-antenna moving UEs, a rotatable RIS with N RIS elements, and a single-antenna BS integrated with an MEC server. The set of UEs and RIS elements are denoted by \mathcal{K}=\{1,...,k,...,K\} and \mathcal{N}=\{1,...,n,...,N\} , respectively. Since the direct links between the BS and UEs are prone to be blocked, the RIS is strategically deployed to maintain line-of-sight (LoS) paths between them. Additionally, the RIS can dynamically adjust its orientation according to the relative positions of the BS and UEs. Fig. 1 : System model and the corresponding angle variations during RIS rotation. II-A Communication Model In this correspondence, we assume that all UEs operate within a task cycle of duration T , equally divided into Q intervals, each with length \tau=T/Q , where the set of time slots is represented as \mathcal{Q}=\{1,\ldots,q,\ldots,Q\} . During each time slot q , the channels from UE k to the RIS and from the RIS to BS are denoted by {\bm{h}_{k,{\rm{R}}}}[q] and \bm{v}_{\rm{R},\rm{B}}[q] , respectively. These overall channels are determined by the individual channels between UE k and each RIS element h_{k,\rm{R}}^{n}[q] , as well as the channels between each RIS element and the BS v_{\rm{R},\rm{B}}^{n}[q] , both modeled as Rician distribution [ 5 ] : \begin{aligned} h_{k,\rm{R}}^{n}[q]&=\sqrt{\frac{\rho_{0}}{{d_{k,\rm{R}}^{n}[q]}^{\alpha_{1}}}}\left(\sqrt{\frac{K_{1}}{1+K_{1}}}\bar{h}_{k,\rm{R}}^{n}[q]+\sqrt{\frac{1}{1+K_{1}}}\tilde{h}_{k,\rm{R}}^{n}[q]\right),\end{aligned} (1) \begin{aligned} v_{\rm{R},\rm{B}}^{n}[q]&=\sqrt{\frac{\rho_{0}}{{d_{\rm{R},\rm{B}}^{n}[q]}^{\alpha_{2}}}}\left(\sqrt{\frac{K_{2}}{1+K_{2}}}\bar{v}_{\rm{R},\rm{B}}^{n}[q]+\sqrt{\frac{1}{1+K_{2}}}\tilde{v}_{\rm{R},\rm{B}}^{n}[q]\right),\end{aligned} (2) where {\rho_{0}} denotes the path-loss factor at a reference distance of 1 meter; d_{k,\rm{R}}^{n}[q] ( d_{\rm{R},\rm{B}}^{n}[q] ) signifies the distance between UE k (BS) and the n^{\rm th} RIS element; \alpha_{1} and \alpha_{2} represent the path loss exponents; K_{1} and K_{2} are the Rician factors. The LoS components are given by {{\bar{h}}_{k,\rm{R}}^{n}}[q]=\exp\left(-j\frac{2\pi}{\lambda}d_{k,\rm{R}}^{n}[q]\right) and {{\bar{v}}_{\rm{R},\rm{B}}^{n}}[q]=\exp\left(-j\frac{2\pi}{\lambda}d_{\rm{R},\rm{B}}^{n}[q]\right) , while the non-LoS components {{\tilde{h}}_{k,\rm{R}}^{n}}[q] and {{\tilde{v}}_{\rm{R},\rm{B}}^{n}}[q] follow a circularly symmetric complex Gaussian distribution with zero mean and unit variance. Since the LoS paths between UEs and the BS are blocked, the channel from UE k to the BS h_{k,\rm{B}} is modeled as Rayleigh fading channel. The angle between UE k and the RIS plane, as well as between the RIS plane and BS are denoted by {\theta_{k}[q]} and {\theta_{\rm B}[q]} , respectively, which can be calculated as follows {\theta_{k(\rm{B})}[q]}={\theta_{k(\rm{B})}^{0}[q]-\delta[q]}, (3) where \delta[q] denotes the rotation angle, and \theta_{k(\rm{B})}^{0}[q]=\left\langle n_{0},n_{k(\rm{B})}[q]\right\rangle represents the angle between UE k (BS) and the initial orientation RIS, with n_{k(\rm{B})}[q] indicating the direction between UE k (BS) and the RIS, and n_{0} denoting the initial RIS orientation. To ensure successful task offloading to the BS, \theta_{\rm{B}}[q] must be maintained within the range \theta_{\rm{B}}[q]\in[0,\pi] . Thus, the constraint on the rotation angle \delta[q] is given by \delta[q]\in[\theta_{\rm{B}}^{0}-\pi,\theta_{\rm{B}}^{0}]. (4) Considering the rotation of the RIS, some UEs may fall outside the reflection half-space as it rotates. Therefore, i_{k} is introduced to indicate whether UE k remains within the half-space, defined by {i_{k}}[q]=\left\{{\begin{array}[]{*{20}{c}}{1,}&{{\theta_{k}}[q]\in[0,\pi]},\\ {0,}&{{\rm{otherwise.}}\quad\;}\end{array}}\right. (5) The location of UE k at time slot q is defined as {l_{k}}[q]=[{x_{k}}[q],{y_{k}}[q],0] , and updated following the Guass-Markov random model [ 11 ] . The reflection coefficient matrix of the RIS in time slot q is represented by the diagonal matrix \mathbf{\Phi}[q]={\rm{diag}}({e^{j{\phi_{1}}[q]}},\ldots,{e^{j{\phi_{n}}[q]}},\ldots,{e^{j{\phi_{N}}[q]}}) , where {e^{j{\phi_{n}}[q]}} denotes the phase shift of the n^{\rm{th}} RIS element in time slot q . It is worth noting that discrete phase shift must be considered due to practical hardware limitations. Therefore, the b -bit discrete phase shift set is characterized as \mathcal{F}=\{{e^{j{\phi_{n}}[q]}}|{\phi_{n}}[q]\in\{0,{2^{1-b}}\pi,...,(2-{2^{1-b}})\pi\}\}. (6) Since the BS and UEs are situated in the far-field region of the RIS,

Read the complete text