Radio Frequency Pattern Matching - Subscriber Location in 5G Networks

Received Signal Strength measures have been collected at the Base Station antenna array of a wireless network operating at 28 GHz mmWaves, and virtually deployed using Open Street Maps and Matlab®. These radio frequency patterns imprinted by a geolocated subscriber transmitting along the campus, have been used to automatically discover the characteristics of the area of interest by using k-means clustering into the proposed unsupervised method. This technique has been integrated into supervised ML methods based on K-Nearest Neighbors, in order to provide an accurate estimation of the subscriber position by performing the match between the received RF patterns and the stored fingerprints. New results exhibit an improved accuracy over previous works based on supervised ML methods. Furthermore, the impact of the operation frequency band over positioning has been evaluated within the range of 3.7–30 GHz. Results show that accuracy degrades at lower frequencies and some mitigation methods are discussed.