Offloading of Indoor Positioning System Using Cloudlet Framework

Abstract

This study evaluates the performance of Indoor Positioning System (IPS) by combining offloading and fingerprinting techniques using Convolutional Neural Network (CNN) implementing the cloudlet framework. The aim of the research are to measure the prediction time and battery consumption at various prediction execution location of the CNN model. The execution locations are mobile device (MD), cloudlet, cloud, and using automatic prediction execution location selection. The Received Signal Strength Indication (RSSI) data from 8 Bluetooth Low Energy (BLE) beacons are collected using an Android application and will be used to train the CNN model. The trained CNN model is then used as radio map to predict the coordinates of the MD. Evaluation is conducted by measuring the battery consumption and prediction speed over 30 minutes while continuously running predictions at four execution locations. The prediction speed is measured from the start of the prediction until the end of the prediction. The study results show that the prediction approach using the cloudlet is more efficient on battery consumption compared to other execution locations. Additionally, the proposed automatic selection method of selecting prediction execution location demonstrates faster execution speed compared to performing the prediction s at a single location. These findings are important for understanding the balance between prediction speed and battery consumption efficiency.