A Multiobjective Neural Network for Scenario-Aware Priority Adaptation in LoRaWAN

View at publisher Summarise with VIRAT
Authors: Vijay G. Dubey, Mandavi Dubey, Nen-Fu Huang
Cited by: 0
References: 38

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

LoRaWAN relies on a pure ALOHA uplink mechanism that lacks native support for distinguishing urgent or high-importance messages. This limitation becomes critical in dense or dynamic deployments, where congestion, interference, and bursty traffic degrade the reliability of time-sensitive IoT data. To address this issue, this paper introduces the Scenario-Aware Priority Deep Classifier (SAP-DC), an artificial neural network (ANN)-based model that incorporates operational context into priority prediction. SAP-DC classifies messages into priority levels {Q1,Q2,Q3,Q4} using a unified cross-layer feature set that combines physical-layer measurements, link-quality indicators, and contextual attributes derived from node density and traffic load. The proposed architecture employs dense layers, residual connections, LeakyReLU activations, and batch normalization to ensure stable learning under heterogeneous LoRaWAN conditions. SAP-DC is evaluated across four operating environments, namely Base, Density-Aware, Traffic-Aware, and Unified, while maintaining strong generalization without requiring architectural modification. Experiments based on LoRaFREE simulation traces show that SAP-DC achieves high classification accuracy, ranging from 0.89 to 0.98, with a compact model size of 0.45 MB. Compared with classical machine learning baselines, SAP-DC demonstrates competitive or superior performance while requiring substantially lower memory. These results validate the effectiveness of learning-based priority prediction and support efficient scheduling and adaptive MAC-layer control in future LoRaWAN deployments.

Fields of Study & Subjects

IoT Networks and Protocols Energy Efficient Wireless Sensor Networks IoT and Edge/Fog Computing

Publication details

Journal / source
IEEE Sensors Journal
Volume
26
Issue
9
Pages
14178–14193
Publication date
26 March 2026
Language
EN