Non-cooperative emitter classification and localization with vector sensing and machine learning in indoor environments

Advancements in wireless technology have led to an increased demand in the enhancement of wireless security, especially in indoor environments as GPS and cellular services degrade in performance. Recent developments in wireless security for indoor environments have focused mainly on developing radio frequency fingerprinting approaches through machine learning for device classification or localization. The work performed and discussed herein describes a developed system that can simultaneously perform device classification and localization in indoor environments using designed vector sensing antenna and artificial intelligence concepts. The devices evaluated are considered to be non-cooperative emitters that convey wideband code division multiple access (WCDMA) information found in universal mobile telecommunication systems (UMTS). However, the designed approaches can be extended to other protocols such as Global System for Mobile communications (GSM), Long-Term Evolution (LTE), and Code Division Multiple Access (CDMA). Device classification is performed in line-of-sight (LoS) scenarios with a developed vector sensor based on statistical features are extracted from the received power spectra and evaluated by two machine learning models, i.e. support vector machine (SVM) and weighted-K-nearest neighbor (WKNN). The final analysis experimentally validates the localization of the UMTS devices in an indoor environment by means of a comparison between dimensionally reduced features extracted from a short-time Fourier transform matrix along with three-dimensional received signal strength features, all acquired by the designed vector sensor antenna. Extension to other wireless protocols is assessed by evaluation of narrowband GSM signals for localization whilst being compared to the localization performance of the wideband UMTS non-cooperative emitters via WKNN.