On the properties of device-free multi-point CSI localization and its obfuscation

The use of Channel State Information (CSI) as a means of sensing the environment through Wi-Fi communications, and in particular to locate the position of unaware people, was proven feasible several years ago and now it is moving from feasibility studies to high precision applications, thus posing a serious threat to people's privacy in workplaces, at home, and maybe even outdoors. The work we present in this paper explores how the use of multiple localization receivers can enhance the precision and robustness of device-free CSI-based localization with a method based on a state-of-the-art Convolutional Neural Network. Furthermore, we explore the effect of the inter-antenna distance on localization, both with multiple receivers and with a single MIMO receiver. Next we discuss how a randomized pre-filtering at the transmitter can hide the information that the CSI carries on the location of one person indoor. We formalize the pre-filtering as a per-frame, per-subcarrier amplitude multiplication based on a Markovian stochastic process, and we discuss different signal clipping and smoothing methods highlighting the existence of a trade-off between communication performance and obfuscation efficiency. The methodology can in any case guarantee almost unhampered communications with very good localization obfuscation. Results are presented discussing two different ways of exploiting the multi-receiver or multi-antenna redundancy and how, in any case, properly randomized pre-distortion at the transmitter can prevent localization even if the attack is carried out with multiple localization devices (receivers controlled by the attacker) and not only with a multi-antenna (MIMO) receiver.