Wi-Fi sensing: fusion of non-cooperative and device-based RF sensors for short-range localization

Through the years, target localization has captured the attention of both academic and industrial worlds, thanks to the huge amount of applications which require the knowledge of the position information. Several works can be found on this topic, where the target localization has been addressed in different ways, depending on the type of target, the specific application and the surrounding scenario. The main goal of this thesis is the definition of innovative methodologies able to solve the problem of the localization of human targets and small objects in local area environments in any operating conditions. In addition to the achievement of important improvements in positioning accuracy, we are also interested in performing the localization for the entire observation time where the target stays in the area of interest. To achieve this result, in this work we decided to propose the joint use of different positioning techniques, based on their fusion in a unified system. The advantage of this fusion lies in the possibility of compensating for the intrinsic limitations of each proposed methodology, especially when complementary techniques are employed. Two different sensors are considered in this work. Both exploit the Wi-Fi transmissions, based on the IEEE 802.11 Standard, therefore also the same receiver can be employed to receive measurements and information about the target present in the area of interest from multiple sensors, without increasing the complexity of the receiving system. Specifically, the first candidate to be used is the Passive Bistatic Radar (PBR) that exploits the Access Point (AP) as illuminator of opportunity. Due to the possibility to obtain the human target position without the necessity for the target to carry a device, this technique can be inserted into the group of the “Device-free localization” methodologies. It makes the WiFi-based passive radar attractive for local area surveillance and monitoring applications, especially where the targets cannot be assumed to be cooperative, as in typical security applications. With reference to the second sensor, the Passive Source Location (PSL) is another possible strategy to estimate the target position. In contrast to the PBR, this is a device-based technique that uses the device transmissions to perform the localization of the specific target. Considering the characteristics of the proposed strategies, it is evident that they present complementary aspects. We can take advantage from this complementarity in several ways. Firstly, due to the Time Division Multiple Access (TDMA) approach used in the Wi-Fi Standard, devices and AP cannot transmit simultaneously, so we can compensate for the lack of signals from one sensor with the measures estimated by the other one. Secondly, we can use the device-based strategy when the target is stationary, and the Passive Radar cannot estimate its position because of the cancellation stage performed during the processing. On the other hand, the Passive Radar is necessary when the target does not carry an active mobile device, or it does not want to be localized (surveillance and monitoring activities). Finally, we can discriminate even very closely spaced target (if both carry an active mobile device) thanks to the possibility to read the MAC Address written into the packets of their devices. The first part of this thesis is dedicated to the characterization of the single sensors, based on the description of the measurement extraction and the evaluation of the related positioning techniques. With respect to the measurement extraction, the PBR provides the target position through the combination of different sets of measures as range/Doppler/Angle of Arrival (AoA). For the PSL, the Time Difference of Arrival (TDoA) and the AoA can be exploited for the same purpose. Since the properties of the PBR have been extensively defined by our research group in the past, in this work more attention has been dedicated to the PSL description. In particular, proper techniques for measurement estimation are reviewed and innovative techniques for TDoA estimation of the PSL sensor are proposed, which provide improved performance with respect to existing techniques. The accuracies achieved with different positioning techniques exploiting several combinations of the estimated measurements are then evaluated. The results show that in short range applications it is desirable to use only AoA measurements, if possible. After the characterization of the sensors, the localization performance of the two techniques are analyzed and compared. This analysis has shown both the effectiveness of the two sensors in target localization and their inherent limitations. In particular, we have studied the relationship between data traffic conditions and performance, and we have seen that it is strictly linked to the number of data available for the estimation of the measures of interest. In addition, the complementarity of the two methodologies has been demonstrated through the evaluation on experimental data acquired in appropriate measurement campaigns, in different network traffic conditions. In this phase, a tracking stage has not been performed. In order to improve the localization performance and carry out the desired sensor fusion, the second part of the thesis has been dedicated to the definition of innovative techniques for target tracking which exploit the characteristics of the employed sensors. Specifically, a new Sensor Fusion tracking filter is proposed. It uses a modified version of the Interacting Multiple Model (IMM) approach, where a Modified Innovation (MI) is introduced, together with Data Fusion techniques. In particular, in this strategy the information related to the presence or absence of the PBR estimates is used to help the choice between the employed filters, in order to improve the localization performance of human targets in the typical “stop & go” motion scenario. The performance of the proposed strategy has been evaluated on both simulated and experimental data. The performance has shown that the IMM-MI outperforms the other strategies, since it provides the best performance in terms of positioning accuracy, target motion recognition capability and percentage of acquisition time covered by this strategy.