Attack Detection Scheme Against Cooperative Spectrum Sensing Data Falsification on Common Control Channel in Cognitive Radio Networks

Abstract

Cognitive radio is an intelligent technology designed to help secondary users (SUs) increase their opportunity to access unused spectrum channels while avoiding interference with the primary users. In cognitive radio networks (CRNs), to find the available channels, SUs execute cooperative spectrum sensing and exchange channels-related control information, namely an available channels list (ACL), on a common control channel (CCC) before determining which channels they may transmit. However, some SUs, defined as attackers, could create a security issue by sharing false ACL information with other SUs to increase their own utilization of the available channels, which significantly decreases the performance of CRNs. In this paper, we propose an efficient detection scheme for CCC security to identify any attacker among the cooperating SUs. In the proposed scheme, all SUs share their ACL information on the CCC, with an associated reputation, which is updated according to its own behavior in each cooperation round, to cooperatively identify attackers. An attacker will be excluded from cooperating group with the result that its updated reputation value exceeds a certain threshold. Simulation results show how to further improve the performance of the proposed scheme by choosing optimized thresholds. In addition, we also illustrate that the proposed scheme can achieve considerable performance improvement compared with a attack detection technique COOPON for secure ACL information exchange.

Appendix 1

A proper value for the initial penalty parameter $\gamma ^{(1)}$ plays an important role in implementing the IPF method [28]. Because of the unconstrained minimization of $\phi (\vec {\lambda }, \gamma ^{(i)})$ is to be carried out for a decreasing sequence of $\gamma ^{(i)}$, so it might appear that we can avoid an excessive number of minimizations of the function $\phi$ by choosing a very small value for $\gamma$. From a computation perspective, it is easier to minimize the unconstrained function $\phi (\vec {\lambda }, \gamma ^{(i)})$ if $\gamma ^{(i)}$ is large (according to the DS method). However, as [28] explains, the minimum $\phi ^{(i)}$ will be farther away from the desired minimum if $\gamma ^{(i)}$ is large. Thus, a trade-off initial value has to be chosen for the penalty parameter $\gamma ^{(1)}$. Based on the theoretical analysis given in [28], a value of $\gamma ^{(1)}$ that makes the value of $\phi (\vec {\lambda }^{(1)}, \gamma ^{(1)})$ approximately equal to 1.1 to 2.0 times the value of $f(\vec {\lambda }^{(1)})$ has been found to be quite satisfactory in achieving fast convergence in the procedure. Therefore, with a feasible initial value of $\vec {\lambda }^{(1)}$, $\gamma ^{(1)}$ can be taken as,

$$\begin{aligned} \gamma ^{(1)} \simeq (0.1\ {\text {to}}\ 1.0)\frac{f\left(\vec {\lambda }^{(1)}\right)}{\frac{1}{\lambda _1^{(1)}}+\frac{1}{\lambda _2^{(1)}}} \end{aligned}$$

In this paper, we chose the value 0.5 at random. Once the initial value of $\gamma ^{(1)}$ is chosen, the subsequent value of $\gamma ^{(i+1)}$ can be taken according to $\gamma ^{(i+1)}=c_1\gamma ^{(i)}$. Here we set $c_1 = 0.5$.