Cross-Layer Energy Management in Embedded AIoT Sensor Networks: A Structured Review, Full-Cycle Energy Model, and Break-Even Analysis

Abstract

Energy management in battery-powered embedded Artificial Intelligence of Things (AIoT) nodes requires coordinated control of sensing, computation, and wireless communication. This structured review combines a documented search of Scopus, Web of Science Core Collection, and IEEE Xplore with screening of 548 unique records and full-text source-to-construct coding of 56 database-derived sources. The synthesis shows that acquisition, Tiny Machine Learning (TinyML) inference, communication, protocol-state overhead, adaptive sensing, and energy harvesting are commonly optimized separately, while field-validated joint control remains limited. The review develops a full-cycle energy model and cross-layer framework linking sensing, processing, payload representation, radio policy, and energy availability. Dimensionless break-even analysis shows that the admissible local-processing budget is bounded by fixed communication overhead and non-machine-learning edge costs. A targeted Long-Term Evolution for Machines (LTE-M) experiment comprising 40 complete transactions showed that payload length alone was insufficient to characterize communication energy; a state-aware model incorporating measured signaling duration achieved R2=0.980 and predicted a temporally separated second session with 1.66% relative root mean square error (RMSE). Precision agriculture and precision apiculture illustrate modality- and network-dependent trade-offs. The findings support full-cycle benchmarking and context-aware orchestration rather than isolated optimization of inference or communication.