Joint communication-computation-routing optimization for energy-aware, deadline-constrained smart agriculture

Abstract

The integration of Internet of Things (IoT) sensing with edge–cloud processing is fundamental to realizing scalable Smart Agriculture. However, jointly orchestrating communication, computation, and routing across resource-constrained rural deployments remains an open challenge. This paper presents a Joint Communication–Computation–Routing (JCCR) framework that unifies dual-mode radio access, hierarchical edge-cloud task placement, multi-commodity backhaul routing, and per-pipeline deadline constraints based on a modeled deterministic service-time approximation into a single mixed-integer non-linear program. The model explicitly captures Low-Power Wide-Area (LPWA) and Wireless Local Area Network (WLAN) interfaces, multi-stage agricultural data processing pipelines with flow conservation, and a realistic energy footprint combining cubic Central Processing Unit (CPU) frequency scaling with transmission costs. The latency model captures deterministic transmission and processing service-time bounds, while explicit path-dependent propagation and queueing delays are not modeled. To solve the resulting NP-hard problem, we develop a Joint Communication–Computation–Routing Optimization Framework (JCCR)- Block Coordinate Descent (BCD) algorithm with McCormick-envelope convex relaxation, which decomposes the problem into three tractable subproblems for radio admission, task placement and computation allocation, and backhaul routing. The proposed algorithm yields a non-increasing objective sequence for the relaxed problem and converges to a finite objective value. Extensive simulations on a realistic 2×2km2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2 \times 2 \text { km}^2$$\end{document} farm deployment demonstrate that the proposed framework achieves approximately 60 Mbps of timely processed throughput under moderate load and 90 Mbps under dense device deployment. Compared with decoupled, cloud-only, edge-only, and heuristic baselines, the proposed framework achieves superior throughput scalability, lower energy consumption, and a more favorable energy–modeled-latency trade-off. In particular, the proposed method maintains an admission rate of approximately 75%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$75\%$$\end{document} under heavy congestion while keeping the modeled service-time latency within the prescribed deadline constraint and converges within 15–20 iterations, confirming its practicality for scalable, energy-aware precision agriculture.