Precise WiFi Indoor Positioning using Deep Learning Algorithms

TL;DR

This paper presents a hybrid WiFi indoor positioning system leveraging deep learning algorithms, combining RSSI and AoA data, and demonstrates high accuracy with less than 250 mm error in real environments.

Contribution

Introduces a novel hybrid localization approach using deep learning with RSSI and AoA, and a fitting function in MATLAB for improved path loss estimation.

Findings

Hybrid CNN model achieves less than 250 mm average error in large classrooms.

Deep learning algorithms effectively reduce noise and interference impacts.

Performance varies across different indoor environments.

Abstract

This study demonstrates a WiFi indoor positioning system using Deep Learning algorithms. A new method using fitting function in MATLAB will be utilized to compute the path loss coefficient and log-normal fading variance. To reduce the error, a new hybrid localization approach utilizing Received Signal Strength Indicator (RSSI) and Angle of Arrival (AoA) has been created. Three Deep Learning algorithms would be utilized to decrease the adverse influence of the noise and interference. This paper compares the performance of two models in three different indoor environments. The average error of our hybrid positioning model trained by CNN in the big classroom is less than 250 mm.