Wireless Positioning Using Deep Learning with Data Augmentation Technique

Recently, wireless positioning has been widely studied in various fields. Traditional models, including random regression forests and k-nearest neighbors, are difficult to guarantee positioning performance in an environment where obstacles and path loss may have a great impact. Deep learning (DL) emerges as a powerful method for positioning problem, which can deal with complicated environments. In this paper, we propose a DL-based scheme for wireless positioning. It uses the received signal strength and coordinates of the receiving nodes as the features to determine the source position. Moreover, in order to deal with the insufficient sample size of collected data, a data augmentation technique is proposed. It can well generate new samples based on the features in the original database. Simulation results demonstrate that proposed DL-based scheme with data augmentation technique can achieve promising positioning performance.