Home // UBICOMM 2023, The Seventeenth International Conference on Mobile Ubiquitous Computing, Systems, Services and Technologies // View article

Deep Learning based Indoor Positioning Approach Using WiFi CSI/RSSI Fingerprints Technique

Authors:
Marco Mühl
Wiem Fekih Hassen

Keywords: CSI, RSSI, private dataset, Raspberry Pi, SVR, LSTM, CNN.

Abstract:
Indoor Positioning Systems (IPSs) play a vital role in various applications, ranging from asset tracking to location-based services. With different approaches being explored in the last years, Wi-Fi-based IPSs utilizing Channel State Information (CSI) and Received Signal Strength Indicator (RSSI) have gained increased attention. This research aims to develop a Wi-Fi based indoor positioning system using CSI and RSSI measurements, specifically focusing on datasets collected at the University of Passau, since datasets used in related work are private. Additionally, after the acquired data is subjected to preprocessing and data cleaning techniques, the study explores the potential of Machine Learning (ML) techniques, including Support Vector Regression (SVR), Long Short-Term Memory (LSTM), and Convolutional Neural Networks (CNN), to enhance positioning accuracy. These models are trained and evaluated using appropriate performance metrics, including Mean Squared Error (MSE) and distance error. The experimental results, focusing on the prediction of vertical and horizontal coordinates within the laboratory room, demonstrate the effectiveness of the proposed system. For unseen RSSI data, the best distance error based on MSE achieved was 29.5 cm using SVR, while for unseen CSI Amplitude data, the lowest distance error based on MSE was 37.9 cm with a CNN approach. A comparison is conducted within the different methods. All tested models consistently achieve a distance error based on MSE of under 50 cm, proving the high quality of the collected dataset. Future research directions and areas for improvement are also suggested.

Pages: 22 to 27

Copyright: Copyright (c) IARIA, 2023

Publication date: September 25, 2023

Published in: conference

ISSN: 2308-4278

ISBN: 978-1-68558-106-0

Location: Porto, Portugal

Dates: from September 25, 2023 to September 29, 2023