A Machine Learning Model for Indoor Positioning of Bluetooth Receivers

Context

This proposal is presented in the context of a tracking system for forklifts (and other indoor factory floor machinery) based on Bluetooth beacons. The beacons are placed at fixed locations in a factory floor, and periodically send packets with their ID.

The forklift is a moving vehicle, equipped with an antenna array which receives the beacons and can determine its stationary position on the warehouse floor by knowing the relative Angles-of-Arrival (AoA) of each incoming beacon. For example, if the factory has one Bluetooth beacon in each corner of a square floor, the forklift can compute its position after receiving all 4 beacons. By knowing the Angle-of-Arrival of each transmission, the position of the forklift can be determined.

Objectives

This work aims to study machine learning models to estimate the Angle-of-Arrival of transmissions, based on data related to the phase of the signals that is captured by the antenna array of a receiver board.

Proposed Work Plan

• Estimation of angle of arrival from phase differences

• Generate synthetic datasets (with different levels of noise) for use in a regression algorithm which can estimate an Angle-of-Arrival from a set of phase differences

• Design an adequate regression algorithm (decision tree, SVM, NN, etc)

• Tune and evaluate the classification accuracy of the model

• Estimation of final receiver position from a set of angles (i.e., outputs of the previous classifier)

• Generate datasets where a set of angles corresponds to a given ground-truth position

• Design an adequate regression algorithm (decision tree, SVM, NN, etc)

• Tune and evaluate the classification accuracy of the model

• Experiment with integration of the inference model with an existing position and tracking simulator and/or an existing receiver board with 8 antennas, and evaluate the achievable positioning accuracy

Details

• Status: Open!
• Student: None (yet).