Résumé : Various positioning techniques have been developed to localize Internet of Things (IoT) devices accurately. Because IoT communications are often narrowband, efficient localization can be achieved by deducing the device position from the estimated signal Angle of Arrival (AOA) at multiple arrays of antennas. It has recently been shown that significant accuracy gains can further be obtained by iterating between the AOA estimation and multi-lateration steps. However, the existing method relies on the knowledge of the transmitted signal (Data-Aided (DA) estimation) which makes it impractical for narrowband communications where the preamble is short. Non-Data-Aided (NDA) estimation is recommended to improve the positioning accuracy for low capacity IoT sensors. This paper proposes an NDA iterative (NDA-It) algorithm using AOA measurements to determine the position of an IoT sensor. Simulation results show that the proposed algorithm significantly outperforms the DA-It in a Bluetooth Low Energy (BLE) context because it can use a much higher number of samples (snapshots); however, it needs more iterations to converge. The computational complexity analysis proves the competitiveness of the proposed NDA-It. The performance of the algorithms is further investigated in multipath and Non-Line-of-Sight (NLOS) propagation environments. Finally, an experimental setup is built to validate the performance of the proposed algorithms.