History based forward and feedback mechanism in cooperative spectrum sensing including malicious users in cognitive radio network

In cognitive radio communication, spectrum sensing plays a vital role in sensing the existence of the primary user (PU). The sensing performance is badly affected by fading and shadowing in case of single secondary user(SU). To overcome this issue, cooperative spectrum sensing (CSS) is proposed. Although the reliability of the system is improved with cooperation but existence of malicious user (MU) in the CSS deteriorates the performance. In this work, we consider the Kullback-Leibler (KL) divergence method for minimizing spectrum sensing data falsification (SSDF) attack. In the proposed CSS scheme, each SU reports the fusion center(FC) about the availability of PU and also keeps the same evidence in its local database. Based on the KL divergence value, if the FC acknowledges the user as normal, then the user will send unified energy information to the FC based on its current and previous sensed results. This method keeps the probability of detection high and energy optimum, thus providing an improvement in performance of the system. Simulation results show that the proposed KL divergence method has performed better than the existing equal gain combination (EGC), maximum gain combination (MGC) and simple KL divergence schemes in the presence of MUs.


Introduction
The demand for radio spectrum is on the rise and is considered asa serious issue. Cognitive Radio Network recently emerged as the prime method for the efficient utilization of flexible radio spectrum [1]. The key responsibility of the Cognitive Radio Networks is sensing spectrum of the primary users(PUs), dynamic spectrum access, spectrum management and efficient spectrum utilization. In spectrum sensing, secondary users (SUs) monitor the activities of the PUs to find out spectrum holes for the transmission of SUs without any interference to the PUs [2][3][4]. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 this average statistics information to the FC. The decision to forward either average free or occupied energy statistics is based on the current sensing observation of the channel.
The proposed method is tested against the existence of always Yes, always No, opposite and random opposite MUs. It is exhibited that this change in the sensing and reporting procedure results in more accurate and sophisticated detection of the PU as compared to the traditional KL divergence, EGC and MGC schemes as in [12,17] with optimum energy consumption by each SU.
The rest of the paper is organized as follows. In Section 2, the system model is explained. Section 3 addresses how FC utilizes the individual energy reports of all SUs to generate global decision of the PU detection and sends back the individual KL divergence values to each SU. Experimental results are presented in Section 4. Section 5 concludes the paper.

Data model
All SUs in the centralized CSS as in Fig 1 report FC about the existence of PUs with local spectrum sensing information. FC combines the received sensing notifications from all SUs with his own sensing results and generates a global decision about the free and the occupied status of the PU spectrum.
Based on the spectrum sensing information by each SU in a particular band decision between H 1 and H 0 isas follows: Where H 0 and H 1 are the hypothesis about the absence and presence of the PU. y j (l) is the received signal from the j th SU, n j (l) is the additive white gaussian noise at the l th time slot for the j th SU, h j is the channel gain value between the j th SU and PU and s(l) is the signal transmitted from the PU. According to the hypothesis H 1 and H 0 the received signal energy of the channel by the j th SU user at the i th sensing interval is: where M is representation of the number of samples in the i th sensing interval. The number of samples M is to be considered large enough such that the energy reported by each SU resembles a gaussian random variable under both H 0 and H 1 hypothesis [46].
Here η j is the SNR value between the j th SU and the PU. ðm 0 ; s 2 0 Þ, ðm 1 ; s 2 1 Þ are the mean and variance values of the energy under H 0 and H 1 hypothesis.
As the KL divergence value between the two probability distribution functions(PDFs) a(x) and b(x) both normally distributed is calculated as follows [47].
Similarly, the KL divergence for functions a(x) with mean and variance ðm a ; s 2 a Þ and function b(x) with mean and variance values ðm b ; s 2 b Þ is further calculated as: The result in Eqn.(5) clearly shows that for functions a(x) and b(x) with similar PDF occurrence, has "0"KL divergence value.

Proposed algorithm for the detection of MUs based on Kullback-Leibler divergence
In the proposed work, the total number of MUs considered is less than the total number of cooperating SUs. All SUs report FC about the existence of PUs with local spectrum sensing information and also stores this data locally.FC combines the individual reports and generate a global decision of the PU spectrum. FC also creates a feedback report for each SU about its individual detection performance as in Fig 3 by measuring the KL divergence score for each SU.Before SU reports any sensing information, it compares the detection results feedback received from the FC with a target value. Based on the feedback from the FC, if the detection results are achieved on behalf of a user,then this particular user will further participate in the sensing process by combining current sensing results with its local history to report a more solid PU status to FC. SUs not declaredas normal will forward their current sensing energy of the PU channel to the FC, while theconfirmednormal user will sense the channel and forward meanenergy of the reports already made under H 1 and H 0 hypothesis.

Pseudo code of the proposed method is as below
• For i = 1 to sensing limit Update the KL distance score K j (i) for the j th SU as K j ðiÞ ¼ X N i K j ði À 1Þ þ DK L;j ðiÞ and send feedback report of K j (i) to the j th SU.

End of loop
The combine KL divergence is determined as ΔK T (i) = ∑ j W j × ΔK L,j (i). Where W j is the weighting factor assigned to the j th SUs decision.
Update mean μ j1 and variance s 2 j1 for next iteration.

Else
Update mean μ j0 and variance s 2 j0 for next iteration.
End IF

• End sensing limit
Local decision and history maintenance by SU In this step pre-sensing check is done by each SU, before forwarding, local sensing information to the FC based on its KL distance feedback information received from the FC.
Where ∑ i K j (i − 1) is the KL distance value received by the j th SU and M 1j (i), M 0j (i) are the mean sample values of all sensing energies reported by the j th SUs under H 1 and H 0 hypothesis based on the history results.
If it is the first time, sensing is done by the j th SU or if the KL divergence satisfaction score is not achieved by a particular SU then, according to Eqn.(6) the sense energy Z j (i) = E j (i) is reported by the SU to the FC and stores this energy locally for future implication.
Similarly, if detection results for the j th SU is met by achieving the KL divergence satisfaction score, then the user is declared as normal. The normal user will search local history and calculate the mean of all high reporting energies as M 1j (i) and of low energies as M 0j (i) and will no more send energy E j (i) to the FC as: The normal SUs further forward these mean energy samples to the FC during the current and in the following sensing intervals according to the observation of the channel to forward decision M 1j or M 0j to the FC.

KL divergence at the FC
Based on the energies reported by the j th SU and the previous mean and variance values, new values of the mean and variances in the i th sensing interval is calculated for all SUs at the FC as follows: z 1 and z 2 are constants with z 1 ¼ kÀ 1 k and z 2 ¼ 1 k . Here k is the effecting level of the received energy to corresponding mean and variance of SUs PDF.
The KL divergence value for the j th SU is determined as: Where K j1 (i) is the KL divergence under the presence hypothesis for the j th SU and K j0 (i) is the KL divergence for the j th SU under absence hypothesis. Difference in the probability distribution function ΔK L,j (i) for the j th SU under H 1 and H 0 hypothesis is calculated as: The total KL divergence value K j (i) of the j th user is further updated as below: This updated value of K j (i) is sent by the FC to the j th SU in order to utilize this information prior to any further reports.

Global decision at the FC
Based on the KL divergence values of all SUs, the global decision G B (i) is made at the FC as follows: where W j is the weighting value assigned to the j th SU for data fusion combination. The lower

Updating mean and variance for the next iteration
A perfect values of (μ j1 , μ j0 ) and ðs 2 j0 ; s 2 j1 Þ for calculating KL divergence is not possible due to unavailability of exact information about the PU. Therefore, universal decision G B (i) value calculated previously is further taken as an estimate of the PU signal for calculating and updating mean and variance values, which is used in the next sensing interval for KL divergence value calculation.
Therefore, based on the universal decision results generated by the FC updated values of mean and variances are calculated. If the global decision G B (i) = 1, mean and variance μ j1 and s 2 j1 are updated as: Similarly, if G B (i) = 0, then mean and variance μ j0 , s 2 j0 are updated for all SUs as: Cooperative spectrum sensing including malicious users in cognitive radio network where d is window size related to the history of the sensing performance for estimated mean and variance.
A flowchart diagram representing the detail operation of the proposed scheme is shown in   A similar comparison is shown in Figs 8-10 by drawing the probability of error against probability of detection for the proposed, KL [12], MGC and EGC schemes. The graphical results showed improved detection results for the proposed scheme against traditional KL, MGC and EGC schemes at all number of cooperative and malicious SUs. By the inspection of these results, it is noticeable that the error in terms of detection of the licensed user for the proposed scheme decreases more quickly as compared with previous fusion schemes and has less vulnerability to the introduction of increasing MUs.

Simulation results and discussions
Probability of error results for each individual SU is drawn against the SNRs varying from -20 dB to -10 dB in Figs 11-13. The graphical results showed that with the increasing average SNR values, the proposed method results showed sophisticated improvements and is able to reduce the channel sensing error quickly in comparison with all other fusion schemes. Similarly, it can be seen that for a given average SNR value, the probability of error decrease even further by varying the total number of cooperative SUs from 4 to 10 in Cooperative spectrum sensing including malicious users in cognitive radio network cases when 20, 25 and 30 cooperative SUs participate in CSS. The energy transmitted by all SUs increases when the number of cooperative SUs is increased from 20 to 30for a given total number of MUs. The MUs are selected for energy comparison with 25% always Yes, 25% always No, 25% opposite and 25% random opposite MUs. These energy plots display that the proposed scheme results in overall savings of the transmitting energy for the proposed scheme under all 20, 25 and 30 total cooperative SUs. The simulation results show effectiveness of the proposed scheme in getting higher detection results of the PU, which results in lower error with optimize transmission energy.

Conclusion
As MUs misinform other SUs about the PU spectrum, it is therefore mandatory to withdraw information provided by MUs in the CSS environment. KL divergence is a tool used for the detection of MU based on the PDF dissimilarity of a normal user and MUs in CSS. The proposed scheme is using the KL divergence with a modified pre-sensing check for each SU before forwarding the observed spectrum information to the FC. The SUs with reputation score in the form of KL divergence feedback by the FC attained, will report FC about the PU status with sensed energy based on the current and past results from its local database. Simulation Cooperative spectrum sensing including malicious users in cognitive radio network results demonstrate the effectiveness of the proposed scheme in terms of sophisticated detection while exercising comparatively less total transmission energy.
This study has limited analysis of the different fusion schemes in the presence of always Yes, always No, opposite and random opposite MUs for sensing merely one PU spectrum. The proposed scheme could be further enhanced for sensing more than one PU spectrum by introducing primary user emulation attack category of MU, resembling the behavior of the PU to misguide other SUs. Similarly, the behavior of the proposed technique can be checked with other schemes by assigning lower and higher SNR values to MUs in comparison with normal SUs to confirm if this method is able to identify and separate MUs under low and higher SNRs.
The KL divergence scheme adopted in this work for identifying MUs is based on the energy distribution of the individual SU report. In future, the KL divergence measurement will be estimated based on sensed information of an individual SU with the average of the sensing information provided by all other SUs. A reputation of the KL divergence for individual SUs can be used in the subsequent sensing intervals to avoid sensing information from the confirmed MUs.  Cooperative spectrum sensing including malicious users in cognitive radio network Writing -review & editing: NG IMQ.