Evaluation of Machine Learning Algorithms for NB-IoT Module Energy Consumption Estimation Based on Radio Channel Quality

Abstract

In this study, we propose a method to estimate energy consumption in battery-powered Narrowband Internet of Things (NB-IoT) devices using the statistical data available from the NB-IoT modem, thereby circumventing the need for additional circuitry to measure battery voltage or current consumption. A custom edge node with an NB-IoT module and onboard current measurement circuit was developed and utilized to generate a labeled dataset. Each data point, generated upon UDP packet transmission, includes metadata such as radio channel quality parameters, temporal parameters (TX and RX time), transmission and reception power, and coverage extension mode. Feature selection through variance and correlation analysis revealed that coverage extension mode and temporal parameters significantly correlate to the energy consumption. Using these features, we tested 11 machine learning models for energy consumption estimation, assessing their performance and memory footprint, both of which are critical factors for resource-constrained embedded systems. Our best models achieved up to 93.8% of fit with measured values, with memory footprints below 100 KB, some as low as 3 KB. This approach offers a practical solution for the energy consumption estimation in NB-IoT devices without hardware modifications, thereby enabling energy-aware device management.

Keywords

• NB-IoT
• energy consumption estimation
• LPWAN
• machine learning