Towards Intelligent Edge Sensing for ISCC Network: Joint Multi-Tier DNN Partitioning and Beamforming Design

Peng Liu, Zesong Fei, Xinyi Wang · 2025

Read Towards Intelligent Edge Sensing for ISCC Network: Joint Multi-Tier DNN Partitioning and Beamforming Design by Peng Liu, Zesong Fei, Xinyi Wang (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-2504-21409.txt.

Introduction

Peng Liu and colleagues study smart edge sensing in this two thousand twenty-five paper. Phones and sensors can now sense their surroundings and send data to base stations for heavy computing. Sending all raw data costs bandwidth and time. The authors propose a three-tier setup to share the work. Small devices, edge servers, and cloud servers each run part of one neural network.

Only mid-level features travel up the chain, which cuts the load on the links. The paper also covers how to split the network and aim the radio beams at once. Their design targets sensing tasks that need fast answers. It should interest engineers who build networks where sensing, radio links, and computing must work as one system.

Catalog record

CREATORPeng Liu, Zesong Fei, Xinyi Wang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE67 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-21409

Browse the full library · About the Tower of Babel

Full text

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

Towards Intelligent Edge Sensing for ISCC Network: Joint Multi-Tier DNN Partitioning and Beamforming Design Thanks: Peng Liu, Zesong Fei, and Xinyi Wang are with the School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, China (e-mail: [email protected], [email protected], [email protected]). Thanks: Xiaoyang Li is with Shenzhen Research Institute of Big Data, The Chinese University of Hong Kong (Shenzhen), Shenzhen 518172, China (e-mail: [email protected]). Thanks: Weijie Yuan is with Department of Electronic and Electrical Engineering, Southern University of Science and Technology, Shenzhen 518055, China (e-mail: [email protected]). Thanks: Yuanhao Li and Cheng Hu are with the Radar Technology Research Institute, School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, China(e-mail: [email protected], [email protected]). Thanks: Dusit Niyato is with the College of Computing and Data Science, Nanyang Technological University, Singapore (e-mail: [email protected]). Peng Liu, Zesong Fei, Xinyi Wang Abstract The combination of Integrated Sensing and Communication (ISAC) and Mobile Edge Computing (MEC) enables devices to simultaneously sense the environment and offload data to the base stations (BS) for intelligent processing, thereby reducing local computational burdens. However, transmitting raw sensing data from ISAC devices to the BS often incurs substantial fronthaul overhead and latency. This paper investigates a three-tier collaborative inference framework enabled by Integrated Sensing, Communication, and Computing (ISCC), where cloud servers, MEC servers, and ISAC devices cooperatively execute different segments of a pre-trained deep neural network (DNN) for intelligent sensing. By offloading intermediate DNN features, the proposed framework can significantly reduce fronthaul transmission load. Furthermore, multiple-input multiple-output (MIMO) technology is employed to enhance both sensing quality and offloading efficiency. To minimize the overall sensing task inference latency across all ISAC devices, we jointly optimize the DNN partitioning strategy, ISAC beamforming, and computational resource allocation at the MEC servers and devices, subject to sensing beampattern constraints. We also propose an efficient two-layer optimization algorithm. In the inner layer, we derive closed-form solutions for computational resource allocation using the Karush-Kuhn-Tucker conditions. Moreover, we design the ISAC beamforming vectors via an iterative method based on the majorization–minimization and weighted minimum mean square error techniques. In the outer layer, we develop a cross-entropy-based probabilistic learning algorithm to determine an optimal DNN partitioning strategy. Simulation results demonstrate that the proposed framework substantially outperforms existing two-tier schemes in inference latency. Keywords: Integrated sensing and communication, mobile edge computing, DNN partitioning, multiple-input multiple-output, cross-entropy. I Introduction Burgeoning applications, including autonomous driving, unmanned aerial vehicles (UAVs), extended reality, and industrial automation, demand not only ultra-reliable and low-latency communication, but also real-time and high-resolution environmental sensing. These dual demands challenge conventional wireless systems, where the separation of sensing and communication often results in inefficient spectrum usage and redundant hardware deployment. Integrated Sensing and Communication (ISAC) has emerged as a promising paradigm to overcome these limitations by enabling joint design and operation of sensing and communication within a unified framework [ 1 ] . By sharing spectral and hardware resources, ISAC enhances spectral efficiency while reducing system cost. To fully unlock the potential of ISAC in 6G networks, extensive research has been devoted to key enabling technologies, particularly in waveform design [ 2 , 3 ] , beamforming optimization [ 4 , 5 , 6 ] , multiple access [ 7 , 8 ] , and resource allocation strategies [ 9 , 10 ] . Despite its numerous advantages, the sensing process in ISAC systems typically generates a large volume of raw data to be processed for target recognition and classification. These applications typically require intensive computation and substantial processing power, which often exceeds the capabilities of devices with limited energy and computational resources. To address this challenge, mobile edge computing (MEC) has been introduced as a complementary technology that enables computation offloading to nearby edge servers offering higher processing capacity and low latency for intelligent processing [ 11 ] . The integration of MEC with ISAC not only enables low-latency processing of sensing data but also facilitates the realization of edge intelligence [ 12 , 13 , 14 ] . This convergence has led to the emerging paradigm of Integrated Sensing, Communication, and Computing (ISCC) [ 15 ] , which aims to support the growing demand for real-time, data-driven, and computation-intensive intelligent applications in 6G networks. Recent research has explored the potential of ISCC from the perspectives of system architecture design and performance optimization [ 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 ] . In [ 16 ] , the authors proposed an ISCC framework in which terminals simultaneously perform sensing and computation offloading over shared spectral resources. By jointly optimizing beamforming vectors and computational resource allocation, the energy consumption of terminals was significantly reduced. Building on this, the authors in [ 17 , 18 ] developed a three-tier ISCC architecture comprising a cloud server, multiple MEC servers, and several devices, where ISAC beamforming matrices and offloading strategies were jointly optimized to minimize the average execution latency of sensing tasks. In [ 19 ] , a short-packet transmission-based ISCC paradigm was introduced, where ISAC devices transmit sensing data to edge servers, with the packet size and computational resources jointly optimized to effectively reduce device energy consumption. Li et al. proposed a novel ISCC framework integrated with over-the-air computation (AirComp), which enables the functional computation of signals based on the analog wave superposition property of multiple access channels [ 20 , 21 ] . Furthermore, the authors in [ 22 ] investigated an AirComp-enabled ISCC system tailored for wireless sensor networks, where the beampatterns of the sensors are jointly designed to minimize AirComp errors. Different from the aforementioned works where terminal devices are equipped with ISAC capabilities, in [ 23 ] , a Non-Orthogonal Multiple Access (NOMA)-aided multi-tier ISCC architecture was proposed, where the base station (BS) performs sensing while concurrently supporting multiple users with edge computing, assisted by a cloud server. Furthermore, the authors in [ 24 ] investigated the deployment of multiple cloud servers and an ISAC BS, and proposed a two-tier offloading strategy to minimize device energy consumption. Although prior studies have investigated the offloading of sensing tasks to edge or cloud servers, they typically treated the sensing task as a monolithic entity. This requires transmitting large volumes of raw sensing data, which will lead to significant communication delays and raise privacy concerns. To address these challenges, deep neural network (DNN) partitioning has emerged as a promising solution [ 25 ] . Since a typical DNN comprises multiple layers with millions of parameters, executing the entire model incurs high computational cost. Fortunately, the intermediate features generated by DNN layers are often significantly smaller than the raw input data, enabling feature-level offloading to effectively reduce transmission overhead [ 26 ] . By leveraging the hierarchical structure of DNNs, partitioning allows for flexible distribution of inferring workloads across devices, edge servers, and the cloud, thereby reducing latency, enhancing privacy, and improving overall sensing efficiency. Recently, several studies have investigated the integration of DNN partitioning into MEC systems. In [ 27 ] , the authors proposed a channel-adaptive partitioned inference strategy, where a U-shaped network is collaboratively executed on the MEC server and the devices, with the partition point dynamically adjusted according to wireless channel conditions. Furthermore, the authors in

Read the complete text