Cognitive radio (CR) is being considered as a vital technology to provide solution to spectrum scarcity in next generation network, by efficiently utilizing the vacant spectrum of the licensed users. Cooperative spectrum sensing in cognitive radio network has a promising performance compared to the individual sensing. However, the existence of the malicious users’ attack highly degrades the performance of the cognitive radio networks by sending falsified data also known as spectrum sensing data falsification (SSDF) to the fusion center. In this paper, we propose a double adaptive thresholding technique in order to differentiate legitimate users from doubtful and malicious users. Prior to the double adaptive approach, the maximal ratio combining (MRC) scheme is utilized to assign weight to each user such that the legitimate users experience higher weights than the malicious users. Double adaptive threshold is applied to give a fair chance to the doubtful users to ensure their credibility. A doubtful user that fails the double adaptive threshold test is declared as a malicious user. The results of the legitimate users are combined at the fusion center by utilizing Dempster-Shafer (DS) evidence theory. Effectiveness of the proposed scheme is proved through simulations by comparing with the existing schemes.

1. Introduction

Wireless network technologies are the most promising technologies in the twentieth century. Today, we already have over a dozen wireless technologies in use: Wi Fi, Bluetooth, Zig Bee, NFC, LTE, earlier 3G standards, satellite services, etc. Due to the proliferation of these wireless networks and the increase in the number of users the spectrum scarcity problem is raised. On the other hand, various reports have shown that the spectrum is inefficiently utilized such that the spectrum is underutilized at a fixed frequency and at a random geographical area [1]. Federal Communication Commission (FCC) states that temporal and geographical variations in the utilization of the assigned spectrum vary from 15% to 85% [2]. One promising solution to this problem is proposed by Joseph Mitola, i.e., “Cognitive Radio (CR)” [3, 4].

CR is a vital technology to improve spectrum utilization. A major challenge in CR is spectrum sensing that identifies the presence of the licensed/primary user (LU) in the network and whenever the LU is detected, secondary user (SU) needs to vacate the channel [5].

Sensing reliability of a single SU degraded by fading and hidden terminal problems. This problem is overcome by the use of cooperative spectrum sensing (CSS), which involves exchange of local sensing decision between multiple SUs. SUs send the sensing result to fusion center by utilizing either the hard decision or soft decision rules.

There are two types of CSS, one is centralized CSS and the other one is distributed CSS. In the centralized CSS, all SUs sense the environment and send their information about the presence of LU to the data fusion center (DFC) and the DFC gives the final decision about the presence of LU. In the distributed CSS in which there is no central node, every SU senses the radio environment and different SUs share their information and make their own decision with distributed manner [1].

Various detection techniques are utilized in the literature to detect the presence of LU. Various detection techniques can be categorized as energy detector based sensing technique, waveform-based sensing technique, cyclostationarity based sensing technique, and matched-filtering technique [6]. Among the techniques, energy detector gives an effective spectrum sensing performance with low complexity. Energy detector is a noncoherent detector which detects the presence of the LU signal by measuring its energy and comparing it with a predetermined threshold. Furthermore, this technique does not require any prior information about LU and it is easy to implement and to be extended for the other spectrum sensing.

Meanwhile, CR networks (CRNs) are highly vulnerable to security threats. Security for wireless networks is an important part which ensures secure operation of the underlying network infrastructure [7]. Various attacks are studied in the literature, which highly degrades the performance of CRN. The most common attacks in CRN are primary user emulation attack (PUEA) and spectrum sensing data falsification (SSDF) attack [8]. In the PUEA, a malicious user behaves like an incumbent transmitter so as to enforce SUs to vacate the spectrum band. In the SSDF attack, the malicious users send false information about the presence or absence of LU to the fusion center. The SSDF attacks severely degrade the spectrum sensing reliability and spectrum utilization.

For secure CSS from the SSDF attacks, various schemes have been proposed. In [9], a scheme was proposed to prevent the SSDF attacks by calculating and updating the credit value of the SUs; malicious users are excluded to avoid the SSDF attacks in cooperative spectrum sensing. In [10], a cooperative scheme based on adaptive threshold is proposed which utilizes matched filter detector as a second stage detector in a confused region between signal and noise, and in the clear region between signal and noise, energy efficient energy detector is used as a first stage detector. Main problem of the conventional energy detector is its low detection performance at low signal-to-noise-ratio (SNR) region new approach has been proposed in [11] to solve the problem. In [11] a bilevel thresh holding approach for energy detection is proposed. In [12], an improved soft fusion-based algorithm was proposed. The authors made improvements in the traditional soft fusion algorithm by establishing the reputation mechanism according to the SU’s past service qualities. The different SUs’ reputation degrees are utilized for allocation of weights to the SUs in the fusion, and using this scheme the effect of the malicious users can be reduced. In [1315], the authors explored Dempster-Shafer (D-S) evidence theory in CSS. It includes four consecutive procedures, which are (1) basic probability assignment (BPA) with this approach, (2) holistic credibility calculation, (3) option and amelioration for BPA, and (4) evidence combination via the D-S rule, respectively [15]. In [16], a new method was proposed, which estimated the attack strength and applies it in the k−out−N rule to obtain the optimum value of k that minimizes the Bayes risk.

On the other hand, above schemes do not consider the doubtful region where it is not clear whether a certain user is legitimate or malicious. While [10] considered the scenario with the doubtful region, matched filter detection technique is used. It is worthy of note that the matched filter detection has a complexity issue where it needs prior knowledge about LU.

In our proposed scheme, we utilize D-S evidence theory with double adaptive threshold method to differentiate among legitimate, doubtful, and malicious users. In the proposed scheme, weights are assigned to each user by utilizing maximal ratio combining (MRC). The weights are assigned to individual SUs, once weights are assigned and masses, i.e., basic probability assignment (BPA), are updated. The legitimate users have the highest weights and the malicious users have the lowest weights. The doubtful users have the weights between those of the legitimate and the malicious users. To ensure its credibility as a legitimate user a fair chance is given through the proposed adaptive algorithm. If it proves its credibility, it is declared as a legitimate user. Otherwise it is declared as a malicious user and withdrawn from the final decision. Through the simulation results, we verify that our proposed scheme is effective and efficient compared to the existing schemes.

The remaining of the paper is organized as follow: system model description is given in Section 2. Section 3 gives a detailed description of the proposed double adaptive threshold scheme and the proposed algorithm at the fusion center. In Section 4, we evaluate the performance of the proposed scheme and compare with the existing schemes. Finally, the paper is concluded in Section 5.

2. System Model

We consider a cognitive radio network that consists of secondary users (SUs), malicious users (MU), and a common receiver, which plays a role as a fusion center as shown in Figure 1. All SUs in the network perform spectrum sensing and transmit sensing result in the fusion center determining if a licensed user (LU) is present or absent in the network. All the SUs including both the legitimate and malicious users participate in cooperation to determine the status of the licensed user in the network.

We assume that each SU independently performs local sensing by using energy detector. The local sensing is a binary hypotheses testing under the absence H0 or the presence H1 of the LU in the network, which is given bywhere z(n) denotes the additive white Gaussian noise (AWGN) and s(n) denotes the transmitted signal from the LU.

Since we consider the energy detection technique for the collection of information on the existence of the LU in the network, the test statistics is equivalent to an estimation of the received signal which is given by each SU aswhere , in which is the sensing duration and W is the bandwidth, and denotes the j-th sample of the received signal. According to the central limit theorem (CLT), when the value of N is large enough, e.g., , the combined signal can be well approximated as a Gaussian random variable under hypotheses H0 and H1, with means and and variances and , which are given by [17] where is the signal to noise ratio (SNR) the LU at the SUs.

In D-S evidence theory, the frame of discriminant A can be defined as , where describes whether of the hypotheses is true or not. Based on parameters of means and variances, the BPA mH0(i) and mH1(i) are determined as a cumulative distribution function, respectively, by using hypotheses of the absence and the presence as follows [18]:where and denote masses of the BPA values for the absence or presence of the LU in the network.

3. Proposed Scheme

In this section, we provide detailed description of the proposed scheme. In CSS, SUs utilize an energy detection technique to sense the existence of the LU in the network. After performing the spectrum sensing, the SUs measure their mass values by using the BPA. Once the mass values of the SUs are measured, the SUs can be classified into three different categories. If the mass value of an SU is less than the lowest threshold (thr1), the SU is identified as a MU and discarded from the final decision. If the mass value of an SU is greater than the highest threshold (thr2), the SU is identified as a legitimate SU. Finally, the mass value of a SU lies between the lowest threshold (thr1) and highest threshold (thr2); the SU is categorized as a doubtful user. In order to provide a fair opportunity to doubtful users and ensure their credibility, we apply the proposed algorithm to those users. If the credibility is ensured, the user is considered as a legitimate SU; otherwise, it is declared as an MU and discarded from the final decision at the fusion center. The proposed double adaptive threshold is graphically described in Figure 2.

In Figure 2, it is shown a single threshold, fixed double threshold, and the proposed adaptive threshold. The single threshold does not take the doubtful users into consideration. It categorized the SU as either legitimate user or an MU, which degrades the performance of the system. In fixed double threshold, the lowest threshold and highest threshold are fixed. In the proposed scheme, we consider legitimate, doubtful, and MU in a double adaptive thresholding scenario and provide a fair chance to doubtful user to prove their credibility.

The mass value of each SU is measured in (4). After the mass values of the SUs are determined, the next step is to measure the weighting factor for each SU to update the masse value. The weightage of each user is determined bywhere SNR(i) is the signal to noise ratio of the i-th SU, and Total SNR is the sum of all the SUs’ SNRs.

In the proposed algorithm, if the mass value of an SU is less than the lowest threshold (thr1) then the weight assignment to the SU is zero and considered as an MU. If the mass value of an SU is greater than the lowest threshold (thr1) but lower than the highest threshold (thr2), its mass value is updated and compare with the proposed adaptive threshold. If its mass value is still lower than the highest threshold (thr2), it is categorized as an MU and discarded from the final decision at the fusion center. Finally, the updated mass value is sent to fusion center for final decision.

Once the weights of each SU are measured using (5), then the mass values of SUs are updated aswhere is the updated masses of hypotheses of absence of LU as reported by i-th secondary user and is the updated masses of hypotheses of presence of LU as reported by i-th SU.

The performance of spectrum sensing in CR is enhanced by keeping the highest value of probability of detection () and lowest value of the probability of false alarm (). According to IEEE 802.22 (WRAN), to prevent any interference between the LU and SU the probability of detection () needs to be as high as possible. To prevent underutilization of spectrum, probability of false alarm () needs to be kept as low as possible. Thus, the threshold should be selected such that we receive optimum value of   and  . Thus, by picking up different threshold values to obtain the lowest possible values and highest , we obtain an optimal threshold.   and    are calculates by (7) and (8), respectively: where thr(k) is the range of thresholds.

The upper and lower limit of the double adaptive threshold is selected based on the requirement of the decision. The optimal threshold for masses is also calculated using the same formula except thr(k) is replaced by a range of masses for minimal and maximal .

Then, the fusion center is applied in Algorithm 1.

Input: energy of the sample (mass value  ), lowest threshold (thr1), highest threshold
If energy of the sample (mass value) < lowest threshold (thr1)
Set the corresponding weights to zero
Else if lowest threshold (thr1) < energy of the sample (mass value) < highest threshold (thr2)
Update masses value using (6)
Don’t update mass values
End if
Output: Update mass value .

Once the mass values of the SUs for both hypotheses and are updated by the proposed algorithm by utilizing (6), the updated mass values of the SUs are sent to fusion center for final decision.

According to the D-S evidence theory, the combination of updated masses at the fusion center can be given aswhere and the operator is the sequential combination of the mass values.

The final decision is determined based on the following simple strategy:

The overall flowchart of the proposed scheme is given in Figure 3.

4. Numerical Evaluation

In this section, we discuss the simulation results of the proposed scheme and compare its performance with the existing schemes. In the simulation environment, we placed five SUs randomly and measured their local energy by utilizing energy detector technique. The probability of appearance of the LU is 0.5, and the bandwidth is 6 MHz, and sensing period is 50 μsec. The simulation environment is developed by utilizing MATLAB as an implementation tool. The parameters for the simulations are summarized in Table 1.

In Figure 4, the performance comparison of the proposed scheme to other existing schemes is shown, with and without existence of MU in the network. In this scenario, we consider 20% malicious users for “Always Yes”. It can be clearly observed that without malicious user in the network, for     of about 0.15,   of the proposed scheme is 0.9 when the highest    from all the other schemes is almost 0.85. Similarly, with the presence of MU in the network,     of the proposed scheme drops but it is still higher than other existing schemes. At    of 0.1,     of the proposed scheme is almost 0.68 whereas the highest    of the other existing scheme is 0.64.

Figure 5 shows the performance comparison of the proposed scheme with other existing schemes with and without having malicious user in the system. The malicious users attack considered is “Random” attack scenario. It can be observed from Figure 5 that without any malicious user in the network, for of about 0.1, of the proposed scheme is 0.86 when the highest from all the other schemes is almost 0.79. is increased in case of a “Random” malicious user due to random behavior of the malicious user. When the malicious user is added to the network, of the proposed scheme is still higher than the other existing schemes. At of 0.1, offered by the proposed scheme is almost 0.67 which is still higher than of the other existing schemes.

Figure 6 shows the comparison of the proposed scheme with other existing schemes, with or without an “Always No” malicious user attack in the system. It is clear from Figure 6 that without any malicious user in the network, for of about 0.2, of the proposed scheme is almost 0.92 when the highest from all the other schemes is almost 0.9. When the malicious user is added to the network, the performance of the proposed is better than the other existing schemes. Specifically, when the value of is 0.2, of the proposed scheme is almost 0.8 and still better than the other existing scheme.

5. Conclusion

The spectrum sensing data falsification attacks falsifies the sensing results, which highly degrades the performance of cooperative spectrum sensing. In this paper, we proposed a double adaptive approach in cognitive radio networks to deal with legitimate, doubtful, and malicious users in the networks. Maximal ratio combining scheme is utilized for weighting of the secondary users and the proposed double adaptive thresholding approach categorized legitimate, doubtful, and malicious users. A fair opportunity is provided to doubtful user to ensure its credibility. At the fusion center Dempster-Shafer evidence theory is utilized for combining legitimate secondary users’ and making the final decision. The performance of the proposed scheme is tested in the presence of various types of malicious users’ attacks and compared the results with the existing schemes. The results showed that the proposed scheme outperforms the existing schemes in case of “Always yes”, “Always No”, and Random attacks.

Data Availability

The data used to support the findings of this study are included within the article.

Conflicts of Interest

The authors declare no conflicts of interest.


This work was supported in part by the MSIT (Ministry of Science and ICT), Korea, under the ITRC (Information Technology Research Center) support program (IITP-2018-0-01426) supervised by the IITP (Institute for Information and Communication Technology Promotion) and in part by the National Research Foundation (NRF) funded by the Korea government (MISP) (no. NRF-2017R1A2B1004474).