= ) number of (4) Where Ψ stands for decision statistics.

Transcription

1 207 IEEE 7th Internatonal Advance Computng Conference on-unform Quantzed Data Fuson Rule Allevatng Control Channel Overhead for Cooperatve Spectrum Sensng n Cogntve Rado etworks Arpta Chakraborty, and Jyot Sekhar Baneree Electroncs & Communcaton Engneerng Department Bengal Insttute of Technology, Kolkata, Inda Emal: chakraborty_arpta2006@yahoo.com Abr Chattopadhyay Electroncs & Communcaton Engneerng Department Unversty of Engneerng & anagement, Kolkata, Inda Emal: abr_chattopadhyay@yahoo.co.n Abstract In ths correspondence, the authors propose a nonunform quantzed data fuson (-QDF) rule allevatng control channel overhead for energy detecton based cooperatve spectrum sensng scheme n cogntve rado systems. Though soften hard or quantzed data fuson (QDF) technque carres few-bt overhead from each user but t prescrbes an mproved soluton between detecton performance and complexty. Agan hgher-bt QDF provdes greater detecton probablty than lower-bt QDF, due to the loss of more nformaton n lower-bt QDF. In ths paper, we derve a non-unform quantzed data fuson (-QDF) rule that smultaneously enhances the detecton probablty for a gven false alarm probablty & hgher bt QDF wth mnmum control channel overhead. We have conducted an extensve smulaton study where the performance of varable-bt -QDF technque s compared wth dfferent unform.e. 3, 4, 5 bt QDF technques wth respect to dfferent parameters to valdate our proposed scheme. Keywords Cogntve rado; fuson rules, cooperatve spectrum sensng;quantzed fuson rule. I. ITRODUCTIO ow the most promsng key enablng technology to defeat spectrum scarcty and spectrum usage neffcency s Cogntve rado (CR) [6] - [8]. Usng the Dynamc Spectrum Access (DSA) polcy, CR can opportunstcally explot the un-utlzed spectrum holes.e. Whte Spaces [3] wthout creatng any notable nterference to the Prmary Users (PU) or lcensed users. oreover, from the report [] t s clear that the spectrum often un-utlzed at varous places at varous tmes. Hence, spectrum sensng process s the most crucal factor n Cogntve Rado etwork (CR) to ensure whether the lcensed users are present or not. Three popular spectrum sensng [7], [20] technques are matched flter method, energy detecton method, and cyclostatonary detecton method. In ths paper we ve chosen energy detecton for spectrum sensng process as t does not demand any pror knowledge regardng the lcensed user sgnals and t s less complex than the other technques. The prmary detecton [0] probablty n spectrum sensng s manly affected by three factors.e. multpath fadng, shadowng and the recever uncertanty problem. Cooperatve spectrum sensng (CSS) [6], [8], [9] s very effectve technque to counter shadowng, multpath fadng, & reduce the recever uncertanty problem [2], [6] whle CSS also acheves spatal dversty n mult-user CR networks. The presence or absences of the lcensed users are decded dependng on the nformaton receved from varous CR users n cooperatve spectrum sensng scheme []-[3]. Intally, hard combnaton technque s dscussed n [4] & [5], where every CR user communcates ts decson of one-bt to the fuson center (FC). On the contrary, n soft combnaton technque, nstead of makng one-bt local decson, the entre results obtaned from local sensng are transmtted to the fuson center by dfferent CR users. Usng approprate combnng rules, these data are combned at FC and decson s made [22], Soft combnaton demands hgher control channel bandwdth but provdes best detecton result where hard combnaton demands ust one-bt control channel overhead on each CR user but provdes degraded prmary user detecton performance for omttng the nformaton. In [2], two-bt quantzed data fuson scheme s descrbed whch provdes mproved prmary detecton than one-bt hard decson. Dfferent researchers then proposed more number of bt quantzed data fuson technque.e. 3-bt, 4-bt even 5-bt [22]-[23] scheme are also proposed to mprove detecton performance where control channel overhead s also ncreasng. In ths paper, we concentrate on the bt reducton of the control channel overhead for multuser CR networks. Our maor contrbutons may be dscussed as follows. Frstly, we have proposed a new non-unform quantzed data fuson rule whch consder varable bt overhead for every CR user. Secondly, we also derve a data rate compresson rato wth effcency of the proposed scheme, whch, however, performs better than the conventonal quantzed data fuson technque. The paper has been organzed as follows. The system models along wth problem formulaton are dscussed n secton II. Secton III provdes our proposed non-unform quantzed data fuson rule for CSS n CR. Secton IV provdes dfferent results from the smulatons and some dscussons. Fnally, the concluson s descrbed n secton V. II. SYTE ODEL In ths letter a centralzed cooperatve spectrum sensng network model has been consdered that conssts of a number of Cogntve Rado (CR) [4], [5], [9] users or Secondary User (SU) and an Access pont or Fuson Center (FC). We have employed Parallel Fuson odel - [2] the most popular and /7 $ IEEE 2 20 DOI 0.09/IACC

2 domnatng approach of cooperaton among CR users for spectrum sensng, (see fg.) as t emphaszes the sensng process comely. All the synchronzed CR users to FC, ndvdually sense the presence or absence of Prmary User (PU) and ther local sensng data are forwarded to the FC va common control channels. FC takes the central cooperatve decson regardng the presence or absence of the PU by combnng the sensng data from dfferent cooperatng SUs. Fg. Parallel Fuson Scheme of CSS n Cogntve rado System = ) number of SUs n a geographcal area and among them only number of SUs can take part n cooperaton due to avalablty of free b channels. We use a varable ch to denote the channel avalablty: b, ch b B of th SU s avalable ch = () 0, ch b B of th SU s un - avalable Here B s the set of equal-bandwdth channels avalable n that geographcal area for both PUs and SUs. Agan ( V) may be represented as: = ch b (2) = V Suppose each SU utlzes K (such that K = { k =,2,..., K} ) number of samples from the sgnal receved for the purpose of energy detecton [2] durng the process of spectrum sensng. Hence sensng of spectrum can be bascally reduced to an dentfcaton problem and modeled as bnary hypothess test consstng of two hypotheses H 0 Let there are V (such that V {,2,..., V} (Absence of PU) and H (Presence of PU). The sgnals under hypothess are of the followng form: H0 : Y(k) = η (k) (3) H : Y(k) = h.s(k) + η (k) Where Y(k) s the sgnal receved by SUs, S(k) are samples of the sgnal transmtted by (PU), η (k ) s the nose of the recever for the th CR user, that s consdered to be an..d. random process wth zero mean, unt varance and ndependent of the prmary sgnal under H. To study the mpact of the detecton ablty of th SU, Detecton probablty ( Pd ) and false alarm probablty ( Pf ) are defned as follows: = Pr { Ψ H } Pd { Ψ H } Pf = Pr = 0 Where Ψ stands for decson statstcs. A. Varous Fuson Rules Cooperatve Spectrum Sensng (CSS) s carred out n threesteps: sensng locally, reportng of data, and fnally fuson of data whch s bascally a method of blendng local sensng data for hypothess testng. There are three methods of combnng the sensed results forwarded to the FC as follows: Hard decson fuson, Soft data fuson, and Quantzed data fuson. A. Hard Decson Fuson In ths process, each SU ndvdually senses the spectrum of nterest and dentfes the absence or presence of the PU and forwards ts decson of one bt to the data fuson center. FC then apples ether OR or AD or Votng rule on the bnary decsons and takes the ultmate global decson on the exstence of PU. The AD rule concludes that PU s present f and only f all the SUs have perceved a sgnal. The cooperatve decson employng the AD rule may be expressed as below: AD Rule H : Ψ = H 0 : otherwse (5) The OR rule concludes the presence of a sgnal when any one of the users perceve the exstence of a sgnal. Hence, the cooperatve decson employng the OR rule may be expressed as below: O R Rule H : Ψ H 0 : otherwse (6) Votng rule s the thrd rule that dctates on the presence of a sgnal when at least of the users have perceved the exstence of a sgnal wth. The cooperatve decson s expressed as below: Votng Rule H : Ψ H 0 : otherwse (7) The sgnfcant mert of ths scheme s that t requres lmted band wdth as data overhead of control channel s only one bt per SU. On other hand, the serous lmtaton of ths scheme s the detecton ablty s not commendable. A.2 Soft Data Fuson In ths case, SUs forward ther local sensng result ( E ) entrely to the FC nstead of takng any local decson and the (4) 22 2

3 central decson s taken at FC from these results employng sutable combnng rules such as square law combnng (SLC), maxmal rato combnng (RC) and selecton combnng (SC). The Square Law Combnng (SLC) s very smple lnear soft combnng scheme where the sensed energy of every SU s forwarded to the FC where they wll be summed up altogether. Then by comparng ths summaton wth a pre-defned threshold the presence or absence of the PU gets determned and the decson statstc s gven as below: E SLC = E (8) Here E sgnfes the statstc of the th CR user. In axmum Rato Combnng (RC) scheme the energy obtaned by each SU s transmtted to the FC and s consdered wth a normalzed weght and then summed up. The weght s set as per the receved SR of the dfferent SU. The statstcal decson of ths method s gven as below: E RC = W E (9) Selecton Combnng (SC) fnds the FC to select the branch wth hghest SR γsc = max ( γ, γ 2, γ 3,..., γ ) Soft combnaton scheme proves to have better detecton performance over hard combnaton, but t demands more bandwdth for the control channel [9]. It also produces more burdens on the control channel n comparson to the hard combnaton scheme. A.3 Quantzed Data Fuson (QDF) In ths scheme a tradeoff has been drawn between the control channel overhead and the detecton performance. As an alternatve to one-bt hard combnng, n whch there s only a sngle pre-defned threshold for dvdng the entre range of the local sensng energy nto two sectons, an mproved detecton performance may be realzed f the number of thresholds are ncreased to get more regons of observed energy. In hard combnaton all the SUs above the threshold are nomnated wth the equal weght rrespectve of the vable sgnfcant changes n ther recorded energes. Hence to acheve enhanced detecton ablty the entre range of the recorded energy s dstrbuted nto more regons wth varable weghts, lke larger weghts are assgned to the hgher-energy regons and smaller weghts to the lower-energy regons. ext the exstence of the concerned sgnal gets determned at the FC employng the equaton below: = Q DF Rule H : W T (0) H 0 : otherwse Where (such that = { =,2,..., } ) s the number of regons n the whole energy range. Each energy regon, ts correspondng energy and weghts may be denoted by λ, E and W respectvely. The number of users fallng n the th energy level ( λ ) may be assumed to be.the threshold parameter T s equvalent to the weght of the hgher-energy regon and may be optmzed as per requrement. A three bt QDF scheme wth seven thresholds ( λ, λ2, λ3,..., λ7 ) ndcatng eght dfferent energy regons along wth ther weghts ( W 0, W, W 2,..., W7 ) are shown n the fg.2 below: One must note the followng relaton among these weghtng factors: W < W < W,..., < W () ( ) 0 Fg.2 prncple of 3-bt softened hard or quantzed data Fuson technque 2 Cooperatve Spectrum Sensng wth QDF rule has mproved the detecton performance to a great degree at the cost of data overhead on the control channel. It demands more band wdth than hard combnng scheme but less than soft combnng schemes. III. PROPOSED O-UIFOR QUATIZED DATA FUSIO RULE (-QDF) Bandwdth s the most prced commodty n today s world and ts profuse requrement s a maor constrant of any communcaton system. So far QDF technque has shown ts excellence puttng an adustment between detecton ablty and data overhead. Stll ths technque burdens the control channel wth mmense number of data bts requred to represent each energy regon. Hence to mprove the data overhead caused by QDF technque and stll mantanng detecton performance, we have proposed a new on-unform QDF (-QDF) technque n ths paper. Smlar to QDF scheme, here also the entre range of recorded energy s dstrbuted nto number of regons wth varable weghts (hgher-energy regons wth larger weghts and lower-energy regons wth smaller weghts), but the code

4 length ( L ) representng each energy level ( ) same for all the code words ( C ) λ s no more. The proposed -QDF scheme ntutvely reduces more number of bts from code words representng hgher energy levels wth larger weghts whle code words representng lower energy levels wth smaller weghts are almost unaltered. Thus the varable bt length code allevates the exstng data overhead problem of control channel wthout hamperng the detecton performance. The decson statstc by -QDF scheme s gven as: Q DF Rule H : W T (2) H = 0 : otherwse A. athematcal Formulaton of -QDF Let us assume that the sensed energy range s dvded nto = =,2,..., ) regons and the (such that { } number of bts requred to represent { } th =,2,..., energy level ( λ ) of QDF scheme be L = n, where n 2 = log. We also assume that the occurrence of all the energy levels ( λ ) s equ-probable and so the probablty of occurrence of each code s equal and s represented by p( λ ) = p( C ) p. Hence for number of cooperatng = SUs, the data overhead (DO) on the control channel engendered by QDF scheme becomes: DO QDF = L avg = n. (3), QDF = Where L avg = p( C ).L = n (4), QDF = In case of -QDF scheme the lower energy regon whch s pvotal for ascertanng the presence of PU, s quantzed wth smaller step sze and the hgher energy regon s quantzed wth larger step sze. Consequently L s no more fxed for dfferent code words representng dfferent energy levels ( λ ). ostly the lower energy levels wth lower weghts are represented by n number of bts and as the weghts of the energy levels are ncreasng, L are decreasng gradually (lke n, {n-}, {n- 2},., ). Thus for number of cooperatng SUs, the data overhead on the control channel generated by -QDF scheme becomes: DO = L n". QDF avg, QDF, (5) = Where avg p( C). = n" and n" < n, QDF = L L (6) = = Dvdng equaton (5) by equaton (3) we get the followng relaton n equaton (7) whch reflects the excellence of proposed -QDF scheme. DO DO QDF QDF = L n". n" = = < n. n DO < DO (7) QDF QDF L avg avg, QDF, QDF Let us assume that each code may be wrtten n -QDF scheme as C, and hence ts prevous & next code may be wrtten as C, & C + respectvely. Suppose L s the numbers of bts are used for a partcular code C, R s the number of 0 s from SB untl gettng any for a partcular code C & S s the number of s from SB untl gettng any 0 for a partcular code C. ow let us consder a unform bt sequence ( C ) havng bt length L = n. Therefore, the total numbers of codes are = 2 n n + n. Agan, n = or n = s the mddle bt of Odd-n (or Even-n) number of bt. In the followng, we dscuss our proposed -QDF algorthm; 24 23

5 In the above mentoned algorthm the authors have explaned the proposed -QDF scheme holstcally. Ths s a generalzed algorthm for both even and odd n number of bt length. In ths correspondence we have shown the 3 (n=odd) and 4 (n=even) number of bt combnaton by QDF scheme and ts correspondng bt combnaton by -QDF scheme (shown n Table I) obtaned by the algorthm. It s observed that startng from n bt length (where n= 3 and 4 n the Table I) gradually the bt lengths are decreasng unformly. For 3 and 4- bt -QDF schemes we fnd the average bt lengths are.75 and 2.68 respectvely followng equaton (6). Lkewse Fg.4 descrbes the ROC plot of detecton probablty versus Sgnal-to-ose Rato (SR) wth varyng Probablty of False Alarm and for all the cases detecton performance of -QDF scheme s comparable wth exstng QDF scheme. TABLE I. REPRESETATIO OF COPRESSED BIT STREA PROPOSED BY -QDF SCHEE Fg. 3 Pd versus Pf of the tradtonal Unform and Proposed on-unform Quantzed Fuson Rule usng dfferent bt-length. IV. RESULTS AD DISCUSSIO In ths segment, we have nvestgated how well the proposed -QDF scheme performs and evaluated t wth respect to Probablty of Detecton (Pd) as shown the followng Fg.3 and Fg.4. We have smulated the code n ATLAB envronment to have the comparatve study of the detecton performance of the proposed scheme over the exstng schemes. We found -QDF scheme produces less control channel overhead n comparson to QDF scheme, but detecton probablty s not compromsed. Fg.3 llustrates the ROC plot of detecton probablty versus Probablty of False Alarm employng dfferent bt length and for all the cases detecton performance of -QDF scheme s not sacrfced. Fg.4 Pd versus SR (n db) of the tradtonal Unform and Proposed on- Unform Quantzed Fuson Rule usng dfferent Probablty of False Alarm. V. COCLUDIG REARKS In ths correspondence we proposed a non-unform quantzed data fuson (-QDF) rule whch can reduce control channel overhead and reach a qualty tradeoff between complexty & detecton performance. Our am s to enhance the detecton probablty wth the help of more number of quantzed data fuson technque.e.3, 4 or 5-bt but effectvely by usng less number of overhead bts. In ths letter, we have tred to compare the performance of proposed -QDF scheme wth correspondng QDF scheme. Smulaton results sgnfy that the 25 24

6 proposed -QDF rule performs better than correspondng QDF scheme wth respect to dfferent parameters. VI. ACKOWLEDGEET The authors deeply acknowledge the support from the department of ECE, Bengal Insttute of Technology (A unt of Techno Inda Group) and the department of ECE, Unversty of Engneerng & anagement, and Kolkata, (A unt of IE UE Group). VII. REFERECES [] FCC, ET Docket o otce of Proposed Rule akng and Order, Dec [2] F. Akyldz, B. F. Lo, and R. Balakrshnan, "Cooperatve spectrum sensng n cogntve rado networks: A survey," Physcal Communcaton (Elsever) Journal, vol. 4, no., pp , arch. 20. [3] S. Haykn, Cogntve rado: bran-empowered wreless communcatons, IEEE J. Select. Areas Commun., vol. 23, no. 2, pp , Feb [4] J.S. Baneree, A.Chakraborty, and A.Chattopadhyay, Relay node selecton usng analytcal herarchy process (AHP) for secondary transmsson n mult-user cooperatve cogntve rado systems. n Proc. ETAEERE 206 (Communcated), LEE-Sprnger, Dec [5] J.S. Baneree, A.Chakraborty, and A.Chattopadhyay, Fuzzy based relay selecton for secondary transmsson n cooperatve cogntve rado networks. n Proc. OPTROIX 206 (Press), Sprnger, Inda, Aug [6] J.S. Baneree and A.Chakraborty, Fundamentals of software defned rado and cooperatve spectrum sensng: a step ahead of cogntve rado networks, In. Kaabouch, & W. Hu (Eds.) Handbook of Research on Software-Defned and Cogntve Rado Technologes for Dynamc Spectrum anagement, Informaton Scence Reference, Hershey, Pennsylvana, USA, pp , 205. [7] J.S. Baneree and A.Chakraborty, odelng of software defned rado archtecture & cogntve rado, the next generaton dynamc and smart spectrum access technology, In.H. Rehman & Y. Faheem (Ed.), Cogntve Rado Sensor etworks: Applcatons, Archtectures, and Challenges, Informaton Scence Reference, Hershey, Pennsylvana, USA, pp.27-58, 204. [8] J.S. Baneree, A.Chakraborty, and K.Karmakar, Archtecture of cogntve rado networks, In. eghanathan & Y.B.Reddy (Ed.), Cogntve Rado Technology Applcatons for Wreless and oble Ad Hoc etworks, Informaton Scence Reference, Hershey, Pennsylvana, USA, pp.25-52, 203. [9] J.S. Baneree and K.Karmakar, A Comparatve Study on Cogntve Rado Implementaton Issues. Internatonal Journal of Computer Applcatons, vol.45, no.5, pp. 44-5, ay.202. [0] A.Chakraborty and J.S. Baneree, An advance Q learnng (AQL) approach for path plannng and obstacle avodance of a moble robot. Internatonal Journal of Intellgent echatroncs and Robotcs, vol.3, no., pp.53 73, 203. [] W. Zhang, R. allk, and K. Letaef, Cooperatve spectrum sensng optmzaton n cogntve rado networks, n Proc. IEEE Int. Conf.Commun., pp , [2] G. Ganesan and Y. (G.) L, Cooperatve spectrum sensng n cogntve rado part I: two user networks, IEEE Trans. Wreless Commun., vol. 6, no. 6, pp , June [3] G. Ganesan and Y. (G.) L, Cooperatve spectrum sensng n cogntve rado part II: multuser networks, IEEE Trans. Wreless Commun., vol. 6, no. 6, pp , June [4] S.. shra, A. Saha, and R. W. Brodersen, Cooperatve sensng among cogntve rados, n Proc. IEEE Int. Conf. on Commun., vol. 4, pp , June [5] A. Ghasem and E. S. Sousa, Collaboratve spectrum sensng for opportunstc access n fadng envronments, n Proc. IEEE Int. Symp. on ew Fronters n Dynamc Spectrum Access etworks, pp.3 36, ov [6] D. Cabrc, S.. shra, and R. W. Brodersen, Implementaton ssues n spectrum sensng for cogntve rados, n Proc. Aslomar Conf. on Sgnals, Systems, and Computers, vol., pp , ov [7] Yucek,T. & Arslan, H., A Survey of Spectrum Sensng Algorthms for Cogntve Rado Applcaton, IEEE Communcatons Surveys & Tutorals, (), Frst Quarter [8] Edward Peh, Yng-Chang Lang, Optmzaton for Cooperatve Sensng n Cogntve Rado etworks, WCC -5, pp , arch [9] Jayakrshnan Unnkrshnan and Venugopal V. Veeravall, Cooperatve Spectrum Sensng and Detecton for Cogntve Rado, IEEE GLOBCO, pp , ov [20] T. Jang and D. Qu, On mnmum sensng error wth spectrum sensng usng countng rule n cogntve rado networks, n Proc. 4th Annual Int. Conf.Wreless Internet (WICO 08), Brussels, Belgum, pp. 9, [2] F. F. Dgham,. -S. Aloun, and. K. Smon, On the energy detecton of unknown sgnals over fadng channels, n Proc. IEEE Int. Conf. on Commun., vol. 5, pp , ay [22] J. a and Y. L, Soft combnaton and detecton for cooperatve spectrum sensng n cogntve rado networks, n Proc. IEEE Global Telecomm. Conf., pp , [23] D.Tegug, B. Scheers, and V. Le r. "Data fuson schemes for cooperatve spectrum sensng n cogntve rado networks." n Proc. IEEE Int. Conf. Communcatons and Informaton Systems Conference (CC), 202 ltary. IEEE,