Millimeter Accuracy Indoor Localization System Using an Attention Convolution Model

This paper presents a novel deep learning model for achieving millimeter accuracy for indoor localization. The model comprises a multi-head self-attention model and a convolutional neural network (CNN), allowing for robust feature extraction from the captured wireless signals. To further enhance the localization accuracy, we also introduced a data augmentation method to increase the size and diversity of the dataset by creating synthetic variants. The performance of the proposed model is tested on an open dataset containing measured channel state information (CSI) signals from a massive multiple-input multiple-output (MIMO) system. The validation accuracies for all three cases are more than seven times higher than the state of the art. The model has also been further evaluated in the COST INTERACT CA20120 Machine Learning Challenge. The performance of our model is competitive with the measured position and significantly outperforms other teams. The proposed model and the associated approaches contribute to the development of practical millimeter-level indoor localization systems using deep learning architectures.