基於名譽與頻譜感知資料融合以減輕認知無線電網路偽造攻擊之機制設計

Abstract

Cognitive Radio Network appears to be one of the suitable technologies to realize shared accesses of licensed spectrum by secondary users for increasing spectrum efficiency. One main advantage of Cognitive Radio Network is the possibility to implement a distributed cooperative spectrum sensing mechanism. Although it has been proven that the distributed sensing paradigm has many advantages respect to a centralized approach, it leaves rooms for new security threats such as Spectrum Sensing Data Falsification attack, where malicious users inject fake spectrum information to disrupt spectrum sharing. This can lead to missing transmission opportunities and cause interferences to the licensed users. In this thesis we consider a Cognitive Radio Network with three entities: Primary Users (PUs), Secondary Users (SUs) and Fusion Center (FC). PU is the licensed user that has transmission priority over the spectrum, SUs sense the spectrum and opportunistically transmit through channels unused by the PU. FC collects the sensing results from SUs, takes decision about spectrum occupancy and coordinates the spectrum access of SUs. In this CRN, there is the need to mitigate the falsification attacks by malicious secondary users (MSUs) so that FC decisions on channel occupancy are resilient to fake reports. There are two prominent research problems: I How to mitigate negative impacts by MSUs on FC decisions through non-MSUs’ sensing and assessment? II How to achieve Non-Repudiation of non-Malicious users so that non-Malicious users are not banned from the network? In this research thesis, we look at the two problems from a prospective that is missing in most of the previous approaches. First, we consider the whole scenario as a multi-agent system in which two different types of users pursue different objectives. From this point of view, we design an Effective Reputation Mechanism for Security, called ERMeS. This data fusion mechanism is based on the concept of reputation and exploits local sensing information. The first innovation of this research work is the exploitation of devices’ cognitive capability to build reputations indexes among clustered neighboring second users. Reputation i) is built on the comparison between user’s and neighbor’s sensed spectrum occupancy information and ii) is updated according to neighbors’ assessments and past reputation history. For each user we then propose a general framework, based on oneself and neighbors sensing and reputation data, to model the reputations updating. The second innovation is the design of ERMeS, a fusion mechanism to combine the distributed reputations assessed by users and obtain a unique reputation value for each neighbor device. More precisely, from all the reputations a user got from its neighbors, we obtain a value that represent its global reputation in the network. The FC then exploits these values as the weights for users’ spectrum reports during the final decision. The third innovation is the proof of a Nash equilibrium of the system in presence of only non-malicious users and perfect correlation among spectrum sensed data. We analyze the interaction among non-malicious users as a Nash Game and find requirements on the non-malicious users reputation updating policy to induce a Nash equilibrium with desirable properties. Under these assumption, we also provide a bound on the maximum percentage of malicious users that the mechanism can mitigate. The forth innovation is the analysis of imperfect sensing and malicious users impact on mechanism performance. In particular we address how uncorrelated sensing and malicious users can affect the reputation among non-malicious users and how this can impact the mechanism error rate. The fifth innovation is the design and analysis of a strategy for non-malicious users resilient to noise and malicious users presence that is still able to satisfy the condition of the Nash equilibrium in perfect case. For our scenario, we consider two types of secondary users: Honest (HSU) and Malicious (MSU). We assume that HSUs aim to increase their transmission opportunity and do not lie on spectrum reports, but we do admit that they can change their reputation indexes. Opposite, MSUs aim to disrupt the sensing mechanism and lie systematically on spectrum data and reputation, but they do not care about transmission resources. Under this scenario, in case of noisy sensing numerical simulation on the mechanism behavior has been performed to find the following: F1. A noise tolerant, malicious user resilient strategy for non-malicious users based on a probabilistic analysis on noise level and channel numbers. F2. Maximum level of noise tolerable by the noise tolerant strategy for non-malicious user. More precisely, we use both probabilistic analysis to find the best value for the error tolerance threshold and the verify it by simulating the mechanism. Then we focus on the noise level and make simulation to understand how this impacts the mitigation capability of this strategy. We considered a scenario with 10 channels and 12 users and simulate 100 sensing rounds 100 times. Our simulations show that in this case if the sensing mismatch probability is lower than 0.1, our mechanism can effectively reduce the FC decision error rate also in the presence of 5 malicious users.