Localization In The Presence of Spatially Obfuscated Sensor Reports Supriyo Chakraborty and Mani B. Srivastava UCLA {supriyo,mbs}@ee.ucla.edu Chatschik Bisdikian and Raghu K. Ganti IBM T.J. Watson Research Center {bisdik,rganti}@us.ibm.com Abstract—Detection and localization of events is an impor- tant activity of military coalition networks characterized by distributed sensing and information exchange. However, the variability in mutual trust between members results in the formulation of sharing policies which often require reports to be deliberately obfuscated while maintaining the plausibility of their content. In this paper, using localization reports as evidences in support of (or against) hypotheses about event locations, we develop the foundations of an evidential reasoning- based approach that uses subjective logic for information fusion and inferencing for localization in the presence of incomplete and conflicting knowledge. To do so, we exploit our recent extensions of subjective logic that accommodate the spatial relationships that naturally exists between location reports from the different members. After highlighting our spatial extensions, we apply them in building an inferencing algorithm for a specific example scenario of primary user localization in a cognitive radio network. The reports are provided by various secondary users in the network. Through extensive simulations, we analyze the performance and the effect of various design parameters, showing a 90% accuracy in localization. Finally, we provide comparison results with other localization techniques via simulations. I. I NTRODUCTION Detection and localization of events (e.g., sniper activity, armored vehicles) is an important activity for many military operations. It relies on distributed sensing and rapid dissem- ination of information to the operation participants. Aside from the inherent challenges in disseminating information over unreliable military networks, coalition settings, where the variability in mutual trust, in conjunction with heterogeneity in data handling and sharing policies between participants may result in information obfuscation, makes the dissemination of sufficiently useful information even more challenging. We define obfuscation as a process which leads to degrada- tion in information quality. It results from deliberate transfor- mations applied with the aim of information hiding (at least to a certain degree) while maintaining sufficient information utility [4]. For example, for an event associated with param- eters < type, location coordinates >, a sharing policy may dictate to accurately reveal the type of the event but selectively hide the location coordinates by providing only an obfuscated region. On one hand the multiplicity of sources provides sensing diversity, while on the other the increased uncertainty and possibly conflicting reports from the sources adversely affect the decision accuracy. Thus, we need techniques to explicitly quantify uncertainty and reason about the obfuscated data. Specifically, our solutions approach is based on evidential reasoning about the location reports, i.e., the evidences, of an event location from sources of different trust levels that may also obfuscate their reports. In our case, this reasoning takes place against a backdrop of location hypotheses for the event and the outcome is an inference about the event location of high belief. Evidential reasoning is anchored on the Dempster-Shafer (DS) theory of evidence [14] which is a logic-based technique that deals with uncertainties in knowledge and allows reasoning in the presence of incomplete and conflicting pieces of evidence, as the case may be when obfuscation occurs. Evidential reasoning and DS theory has often been used for performing analysis (i.e., reason) of events observed during network operation for reasons such as anomaly detection [3] and network fault detection [17] or in intrusion detection [7] and DDoS [16]. In this paper we investigate the use of evidential reasoning in coping with the possibility of obfuscated location reports from sources in coalition networks. In doing so, we first adopt a generalization of the DS theory, that of subjective logic (SL) [9][10] that explicitly considers uncertainty that arises from the available evidences in addition to the belief and plausibility in DS. It provides a framework for subjective quantification of the belief, disbelief and uncertainty associated with a report when objective probabilities are hard to compute (due to inadequate information) [6]. However SL (and evi- dential reasoning) only deals with concrete (i.e., categorical) hypotheses, such as the state of a node being “good” or “bad” [17], a fault being of type “A”, “B” or “C” [1] etc. SL does not provision for the case where there are the spatial and, hence, measurable relationships between the hypotheses – a case which naturally arises when dealing with location reports and localization inferences. In response to this limitation, we have recently developed saSL, a spatial-aware extension to SL to allow consideration of “distances” between location hypotheses, comprising of spatial regions, and “influences” of one location reports to such hypotheses. In this paper, we highlight saSL and use it in developing an efficient window- based search algorithm, which finds the region of maximum belief as the event region. To evaluate our algorithm, we consider an example of a cognitive radio network (CRN) which utilize radios that can change their transmission and reception parameters au- tonomically. This allows for improved utilization of licensed spectrum by allowing unlicensed users (the “secondary”) to access the licensed spectrum while avoiding potential conflicts with the licensed (the “primary” users) and other secondary users. Localization will typically involve location reports from secondary users about the location of a primary user. In