Proximity and occupancy detections using Bluetooth Low Energy (BLE) beacons are two common sensing techniques for many emerging IoT applications today. This thesis addresses several generic challenges in using BLE beacons for these sensing techniques. First, it is difficult to achieve high-resolution proximity detection in indoor environments when BLE beacons are densely deployed. An adaptive scanning method is proposed to achieve accurate detection. Conversely, proximity detection becomes unreliable when there is insufficient observation owing to malfunctioning beacons. A compressive sensing framework along with an efficient evolutionary computation method is introduced to re­cover the observation. In certain scenarios, where users do not need to carry their smart devices, it is prefe...[ Read more ]

Proximity and occupancy detections using Bluetooth Low Energy (BLE) beacons are two common sensing techniques for many emerging IoT applications today. This thesis addresses several generic challenges in using BLE beacons for these sensing techniques. First, it is difficult to achieve high-resolution proximity detection in indoor environments when BLE beacons are densely deployed. An adaptive scanning method is proposed to achieve accurate detection. Conversely, proximity detection becomes unreliable when there is insufficient observation owing to malfunctioning beacons. A compressive sensing framework along with an efficient evolutionary computation method is introduced to re­cover the observation. In certain scenarios, where users do not need to carry their smart devices, it is preferable to have device-free occupancy detections in certain scenarios. A novel deep learning method using a denoising-contractive autoencoder is therefore devel­oped to filter out the noise while capturing the received signal strength (RSS) variations useful to detect the presence of an occupant. In some extremes, Internet connection might not always available to deliver the remote IoT applications with BLE beacons. An ad­vanced online-to-offline deep Q-learning method is proposed to train the beacon overlays on the mesh network to forward data for remote sensing and monitoring purposes. In summary, in this thesis, we successfully identify and overcome several key limitations of using BLE beacons for emerging IoT sensing applications. The proposed methods and analysis may also serve as effective references for related IoT applications with BLE.