A Bluetooth Indoor Positioning System Based on Deep Learning with RSSI and AoA

Traditional received signal strength indicator (RSSI)-based and angle of arrival (AoA)-based positioning methods are highly susceptible to multipath effects, signal attenuation, and noise interference in complex indoor environments, which significantly degrade positioning accuracy. To mitigate the impact of the above deterioration, we propose a deep learning-based Bluetooth indoor positioning system, which employs a Kalman filter (KF) to reduce the angular error in AoA measurements and utilizes a median filter (MF) and moving average filter (MAF) to mitigate fluctuations in RSSI-based distance measurements. In the deep learning network architecture, we propose a convolutional neural network (CNN)-multi-head attention (MHA) model. During the training process, the backpropagation (BP) algorithm is used to compute the gradient of the loss function and update the parameters of the entire network, gradually optimizing the model's performance. Experimental results demonstrate that our proposed indoor positioning method achieves an average error of 0.29 m, which represents a significant improvement compared to traditional RSSI and AoA methods. Additionally, it displays superior positioning accuracy when contrasted with numerous emerging indoor positioning methodologies.