Malicious User Detection in Spectrum Sensing for WRAN Using Different Outliers Detection Techniques

Transcription

1 Malcous User Detecton n Spectrum Sensng for WRAN Usng Dfferent Outlers Detecton Technques Mansh B Dave #, Mtesh B Nakran #2 Assstant Professor, C. U. Shah College of Engg. & Tech., Wadhwan cty , Surendranagar Guarat Inda 2 Assstant Professor, C. U. Shah College of Engg. & Tech., Wadhwan cty , Surendranagar Guarat Inda Abstract In cogntve rado t s of prme mportance that the presence of Prmary Users (PU) s detected correctly at each of the tme. In order to do so the help from all present Secondary Users (SU) s taken and such a taken s known as co-operatve spectrum sensng. Ideally t s assumed that all the secondary users gve the correct result to the control center. But there are certan condtons under whch the secondary users delberately forward wrong result to the control center so as to degrade the performance of the cogntve network. In ths paper we study the dfferent technques for detectng the malcous users or outlers. We take nto consderaton practcal envronmental condton such that the receved sgnal of the secondary users s made to undergo fadng and nose s also ntroduced n the sgnal. We further go on to examne each of the outler detector technques and fnd out the most sutable at varous nstants. Keywords- Cogntve Rado, Co-operatve Spectrum Sensng, WRAN, Cyclo-statonary detector, Malcous Users, Outlers, Medan Absolute Devaton (), Medcouple, Skewness. I. INTRODUCTION Cogntve Rado s a new technology whch s stll n ts early stages. It s technology wheren the lcensed bands are used by the unlcensed users thereby better utlzng the underutlzed bands []. In cogntve rado, a frequency band s checked for the presence of prmary users (PU) and f empty the unlcensed users or the secondary users (SU) are allowed to operate n the band. The detecton poses a serous performance ssue as each of the PU s use dfferent modulaton and transmtted power. There are number of technques for detectng the presence of PU n the frequency band whch are elaborated n [2]. In order to mprove the performance of the system results regardng the presence of PU are not lmted to a sngle sensng SU but rather at dfferent sensng SUs whch s known as co-operatve spectrum sensng. In ths method the sensed data from dfferent ponts s brought to a control centre wheren the fnal decson s made. A sgnfcant performance gan s acheved when usng ths method [3-4] In deal case the nformaton from varous SU s s made avalable to the control centre and fnal decson s made by averagng all the nformaton. But n practcal condtons there are many such nstances when a SU delberately reports false nformaton to the control centre [5]. Such users are known as malcous users. As a result of ths false report the performance of the system s degraded substantally. In statstcs generally data whch are far away from the mean dstrbuton are classfed as outlers. In the case of malcous users problems they can be classfed as outler because they wll gve nformaton whch wll vary n large amount from the mean dstrbuton of the accumulated data. Thus detectng these malcous users can be consdered as smply detectng outlers n a gven dstrbuton. There number of methods for detectng outlers whch are elaborated n [6-7]. The paper s organzed n 4 sectons. Followng ths ntroducton secton a number of outler detecton methods are dscussed n secton II, whch are used to elaborate the performance n terms of detectng the malcous users. System model used to fnd out the results s dscussed n secton III. The performance of dfferent outler detecton methods are presented n secton IV. Ths s followed by concluson n secton IV. The paper proposes to fnd the outler detecton method whch gves better performance n comparson to others under dfferent number of malcous users. II. OUTLIER DETECTION TECHNIQUES A node can be malcous due to devce fault or for selfsh reasons. In case of selfsh reasons t mght happen that a SU reports that a PU s present and force the control centre to such a decson and use the channel to transmt ts own sgnal. Such types of malcous users are known as Always Yes users. On the other hand n order to cause nterference to the PU some SU s mght force the control centre to thnk that the channel s empty. Such users can be classfed as Always No users [7-8]. Another subcategory of malcous users can be malcous at specfc nstants of observaton and can be ether Yes or No. In case of Always Yes users the reported energy level wll always be much greater than the threshold level decded by the control centre and wll devate by a large amount from ISSN: Page 326

2 levels of rest users. In case of Always No users the reported energy wll be much lesser than that of the threshold. Accordng to the defntons of outlers from [9] An outler s an observaton whch devates so much from the other observatons as to arouse suspcons that t was generated by dfferent mechansms. So the levels reported by malcous users can be classfed as outlers. There are number of methods to detect the outlers some of whch are elaborated further n ths secton and are used to remove the malcous users from beng consdered for the fnal decson. ) Mean Dfference (): Varance are good measure for checkng the varablty n a set of data but t gves superor results n scope to the data whch are almost or exactly normally dstrbuted. Mean dfference doesn t have such lmtaton [0]. The mean dfference s gven as E ( X X 2 where X and X2 beng data at two dfferent nstant. For the case of random sample of sze N, the mean dfference can be calculated as the arthmetc mean of the absolute dfference of all values as N N x x 2 N 2) Medan Absolute Devaton (): s smlar to standard devaton n terms of estmatng the spread of data n a gven dstrbuton but has a breakdown pont of 50% lke medan. It gves the medan of absolute devaton of data values from the medan of the sample under consderaton. s gven as [] medan x medan( x ) 3) Estmator (): It has been found that gves better result when the sample dstrbuton s symmetrcal. Thus ths property can be of sgnfcant dsadvantage n case of skewed dstrbuton. So for skewed dstrbutons, better estmators are proposed n [2]. Frst of them s the estmator whch s gven as.92* Medan Medan x x s smlar to n manner that both use combnaton of medans. In terms of dfference, takes nto account the devaton from the central value of the dstrbuton whereas estmator consders the dstances between any two observatons. 4) Estmator (): It s smlar to the estmator and has the same propertes whle the only excepton that t smooth s out any dscontnutes n the result obtaned from estmator. The estmator s gven as Q 2.2* st Quartle x x : n 5) Medcouple (): In case of malcous users false notfcaton the reported level dstrbuton becomes skewed. In order to measure ths skewness we can apply the medcouple [3]. The medcouple () s defned as ( x ms) ( ms x ) ( S) Medan xmsx x x Q h ( ) IQR; Q h ( ) IQR l 3 where x and x are sampled ndependently from the dstrbuton S, ms s the medan of the dstrbuton and IQR s the nter-quartle range. It can be sad from the defnton that the value of medcouple wll le between - and +. For a rght skewed dstrbuton t wll have a postve value and for left skewed the value wll be negatve. For the present case we take nto account the exponental model of the medcouple whch s gven as h.5* exp( 3.5* ) h l r.5*exp(4* ) Note that n each of the schemes the x and x are reported energy levels of dfferent SU s under consderaton. III. SYSTEM MODEL We consder a group of N secondary or cogntve users. Practcal envronment are taken nto consderaton by takng nto account the effect of nose and fadng. All the SU s measure the energy of the channel at ther respectve stes and forward ther measured energy level to the control or the decson centre va control channels. It has been assumed that control channels are perfect. At the control centre the dfferent measurement data are collected and data fuson methods are used to obtan the fnal decson. Snce man am of cogntve rado s to provde mnmal nterference to the PU s we have consdered only Always No users n ths paper. We have assumed that the malcous users produce extreme false values. Contnung from our prevous work SU s measure the channel energy wth the help of cyclo-statonary detector work [4]. The result s sent to the control centre and averagng scheme s used to obtan the detecton probablty. Based on the threshold obtaned from the dfferent outler detecton methods malcous users are detected and they are gnored when makng the fnal decson at the control center. r ISSN: Page 327

3 IV. SIMULATION RESULTS We consder a system havng 50 Secondary Users (N=50). The average receved SNR at each of the SU locaton s -0 db. In addton Raylegh fadng s ntroduced between PU transmtter and SU s. The energy levels are obtaned wth the help of the cyclo-statonary detector [4]. Now to the energy level of all the SU s each of the outlers detecton methods are appled and a threshold s decded for each of the technques. The threshold obtaned s compared wth the energy level of each of the SU s. If the reported level s less than threshold then t s classfed as outler or malcous user. Dfferent tme nstants are consdered such that the fadng coeffcent for each of the N users vares and threshold s calculated for each of the samplng nstant. Fgure dsplays the dfferent values of threshold for 50 dfferent tme nstants wth number of malcous users equal to 0. The teratve procedure s contnued tll a stoppng condton s reached. The condton can be maxmum number of teraton or dfference between two thresholds values be less than a predefned tolerance Observaton Instants Fg. values for dfferent outlers methods for 50 observatons wth 0 malcous users. We consdered 50 dfferent observaton nstants and counted the number of malcous users detected by a partcular outler method each tme. In fgure 2 plot of the number of malcous users versus the number of tmes the outler technques have detected that number of malcous users. Fg. 2 No of tmes correct detecton of malcous users for dfferent outlers technques (actual outlers=0). Fgure 3 dsplays the threshold levels at dfferent nstants wth number of malcous users equal to 5. For the threshold levels obtaned the count for 50 observatons s plotted n fgure 4. We have consdered dfferent number of malcous users and found out how many tmes dfferent technques correctly dentfy the malcous users. Fgure 5 dsplays the case when the actual number of malcous users s equal to 7 whereas the stuaton of actual number of malcous users s depcted by fgure 6. Table shows the correct detecton rates for varous outler detecton technques. It presents good comparson between dfferent methods as to whch performs better n terms of malcous user detecton. It can be nferred from the table that the most consstent and accurate of the fve technques s the Medan Absolute Devaton or the. Ths s followed by and estmator as there s barely and dfference between the two. Next n the lst s the Mean Dfference () as for less number of tmes t has detected correctly as compared to followed by medcouple Observaton Instants Fg. 3 values for dfferent outlers methods for 50 observatons wth 5 malcous users. ISSN: Page 328

4 Fgure 7 gves the plot of detecton probablty versus the threshold (ROC) for malcous users equal to 5. The black dash dot (WM) plots the detecton probablty whle consderng the malcous users. All the detecton schemes gve better result than n the presence of malcous users. The medcouple gves the hghest probablty because half of the users have been consdered malcous. The most consstent scheme gves better result than WM but no as hgh as. In fgure 8 number of malcous users consdered s 0. Snce n ths case the worst performng scheme s Mean Dfference, the detecton s hgh as almost half of users have been consdered malcous. Fg. 4 No of tmes correct detecton of malcous users for dfferent outlers technques (actual outlers=5) Pd WM Fg. 5 No of tmes correct detecton of malcous users for dfferent outlers technques (actual outlers=7) Fg. 7 Pd vs Th (ROC) for malcous users = Pd WM Fg. 6 No of tmes correct detecton of malcous users for dfferent outlers technques (actual outlers=2). Actual No. of Malcous Users TABLE I CORRECT DETECTION BY VARIOUS SCHEMES Number of tmes correct detecton of malcous users (out of 50) () () () () () Fg. 8 Pd vs Th (ROC) for malcous users =0 V. CONCLUSION The smulaton results clearly ndcate that the ROC curve cannot be the only plot upon whch the performance of an outler detecton scheme can be udged. The reason beng that n both cases the maxmum probablty of detecton s for those schemes n whch the number of malcous users detected s much greater than actual value. Thus for checkng the performance of the dfferent outler technques ISSN: Page 329

5 the number of tmes correctly detected malcous users must also be consdered. From the results we can conclude that the scheme s good at detectng the exact number of malcous users and also has a good detecton rate REFERENCES [] FCC, "Spectrum Polcy Task Force Report," vol. ET Docket No , November [2] T. Yucek and H. Arslan, "A survey of spectrum sensng algorthms for cogntve rado applcatons," Communcatons Surveys Tutorals, IEEE, vol., no., pp. 6-30, quarter [3] S. Atapattu, C. Tellambura, and Ha Jang, "Energy Detecton Based Cooperatve Spectrum Sensng n Cogntve Rado Networks," Wreless Communcatons, IEEE Transactons on, vol. 0, no. 4, pp , aprl 20. [4] A. Ghasem and E.S. Sousa, "Collaboratve spectrum sensng for opportunstc access n fadng envronments,", nov. 2005, pp [5] S. M. Mshra, A. Saha and R. W. Brodersen, Cooperatve sensng among cogntve rados," Internatonal Conference on Communcaton ICC'06", vol. 4, pp , June [6] R. Chen, J.-M. Park, and J. Reed, Defense aganst PU emulaton attacks n CR networks," IEEE J. Sel. Areas Commun., vol. 26, pp. 2537, Jan [7] P. Kalgneed, M. Khabbazan and V. K. Bhargava, "Secure Cooperatve Sensng Technques for Cogntve Rado System" IEEE Internatonal Conference on Communcatons, Beng, 9-23 May2008, pp [8] Kalgneed, P., M. Khabbazan, and V. K. Bhargava, "Malcous User Detecton n a Cogntve Rado Cooperatve Sensng System", Wreless Communcatons, IEEE Transactons on, vol. 9, no. 8, pp , aug., 200. [9] HAWKINS,D.M. Identfcaton of Outlers. Chapman and Hall, London and New York,980. [0] Ytzhak, Shlomo, Gn's Mean dfference: a superor measure of varablty for non-normal dstrbutons, Metron - Internatonal Journal of Statstcs, LXI, ssue 2, p , [] Seo Songwon, "A Revew and Comparson of Methods for Detectng Outlers n Unvarate Data Sets", Master's Thess, Unversty of Pttsburgh, [2] PJ Rousseeuw, C Croux, "Alternatves to the medan absolute devaton", Journal of the Amercan Statstcal Assocaton, pp DOI:0.080/ [3] Vandervere, E. & Huber, M. An adusted boxplot for skewed dstrbutons. COMPSTAT2004 Symposum: proceedngs n computatonal statstcs(pp ), [4] M. B. Dave, "Optmal number of users n Co-operatve spectrum sensng n WRAN usng Cyclo-Statonary Detector", Internatonal Journal of Engneerng Trends and Technology(IJETT), V4(7), pp July 203. ISSN: arxv: v[cs.ni]. ISSN: Page 330