[论文解读] Toward Integrated Sensing, Communications, and Edge Intelligence Networks

Mattia Merluzzi, Miltiadis C. Filippou+2 more|arXiv (Cornell University)|Mar 24, 2026

Age of Information Optimization被引用 0

一句话总结

论文提出一个面向计算感知的联合优化框架，用于在传感、通信和边缘AI推理之间共享资源的三功能网络，显示了集成资源分配相较于独立策略的优点。

ABSTRACT

Wireless systems are expanding their purposes, from merely connecting humans and things to connecting intelligence and opportunistically sensing of the environment through radio-frequency signals. In this paper, we introduce the concept of triple-functional networks in which the same infrastructure and resources are shared for integrated sensing, communications, and (edge) Artificial Intelligence (AI) inference. This concept opens up several opportunities, such as devising non-orthogonal resource deployment and power consumption to concurrently update multiple services, but also challenges related to resource management and signaling cross-talk, among others. The core idea of this work is that computation-related aspects, including computing resources and AI models availability, should be explicitly considered when taking resource allocation decisions, to address the conflicting goals of the services coexistence. After showing the natural coupling between theoretical performance bounds of the three services, we formulate a service coexistence optimization problem that is solved optimally, and showcase the advantages against a disjoint allocation strategy.

研究动机与目标

• Motivate the triple-functional network concept that shares infrastructure for ISAC, communications, and edge AI inference.
• Formulate a computation-aware resource sharing problem that jointly optimizes communications and inference under sensing and communication needs.
• Show that coupling computation resources (AI models) with radio resource allocation improves coexistence performance over disjoint allocation strategies.

提出的方法

• Model a multi-service wireless system with a base station serving UL EI data uploads, DL ISAC data, and monostatic sensing under a single FD MIMO BS.
• Define KPIs for EI inference delay and goal effectiveness, and ISAC metrics including CRB-based sensing accuracy and DL rate.
• Formulate a mixed integer non-linear program that jointly optimizes DL transmit powers and data batch size under CRB and rate constraints, solved via convex optimization for fixed batch sizes using CVXPY.
• Derive the CRB for angle estimation as a function of DL spectral resources and model parameters, linking sensing performance to DL time allocation (rho_dl) and communication delay.
• Propose a solution approach that fixes n_b (batch size) and solves a convex subproblem; evaluate compute-aware versus compute-unaware baselines.

Figure 1 : The considered multi-service wireless system and frame structure.

实验结果

研究问题

• RQ1 How can resources be jointly allocated to support ISAC, EI, and sensing without orthogonal partitioning of time and power?
• RQ2 What is the impact of AI model choice and data representation (compression) on the joint performance of DL rate, inference delay, and sensing accuracy?
• RQ3 Can a compute-aware optimization yield better trade-offs between DL transmit power and edge inference quality compared to disjoint allocation strategies?
• RQ4 How does the available computation (AI model complexity) constrain or enable frame-level resource allocation for tri-functional networks?

主要发现

• A computation-aware resource sharing scheme improves the trade-off between ISAC costs and EI performance compared to disjoint allocation.
• Increasing DL transmit power generally improves goal effectiveness for EI by allowing more resources for inference but is balanced by CRB and DL rate constraints.
• Using more capable AI models (up to a point) can enhance EI quality, enabling better trade-offs with DL power, while very heavy models may negate gains due to computation delay.
• The framework shows a natural coupling among communication, computation, and sensing, where the chosen inference model determines the time budget for DL transmission and sensing accuracy.
• Simulation results indicate that for 8.18 GFLOPs, higher goal effectiveness is achievable with significantly lower DL power than with heavier models, under the proposed compute-aware optimization.
• Relaxing sensing requirements (CRB threshold) changes the balance between DL power and EI goal effectiveness, illustrating cross-layer interdependencies.

(a)