Social Spider Algorithm-based Spectrum Allocation Optimization for Cognitive Radio Networks

Transcription

1 Socal Sder Algorh-based Secru Allocaon Ozaon for Cognve Rado Neworks Bnh Thanh Dang Faculy of Inforaon Technology Indusral Unversy of HCMC Ho Ch Mnh Cy Vena Mnh Cong Vo Faculy of Couer Scence Unversy of Inforaon Technology Faculy of Inforaon Technology Indusral Unversy of HCMC Ho Ch Mnh Cy Vena Tung Khac Truong Faculy of Inforaon Technology Indusral Unversy of HCMC Ho Ch Mnh Cy Vena Absrac In recen years alyng nellgen algorhs n solvng hard robles has becoe a favore oc n nex generaon neworkng research Alhough classcal ehods have acheved any oran oucoes hs new rend roses sgnfcan resuls Ths aer addresses he secru allocaon roble n Cognve Rado Neworks (CRN) n whch we roosed a new soluon based on Socal Sder Algorh () o search for he oal allocaon schee The nuercal resuls rove he sueror of our aroach n coarson o oher ehods Keywords: Secru allocaon; ozaon; cognve rado neworks; swar-based algorh INTRODUCTION In recen years wreless secru allocaon ozaon has becoe a favore oc n nex generaon neworkng research [] Tradonally he secru s anaged by he sae and regulaons n whch he ajor aroach s consrucng a fxed allocaon syse As an aaren resul he an drawbacks of hs acc are low ulzaon of he bandwdh wase of sarse or unassgned channels whle ohers are so crowded and lead o undesred nerference level n he curren councaon channel Those dsadvanages ay no be a serous roble n he as bu wh he rad growh n he quany of he wreless devces connecng o he nework whch s forng he Inerne of Thngs (IoT) he suaon has changed draacally where unozed ulzaon of he secru resources are no acceable any ore Cognve rado [2][3][4] s coonly consdered o be he soluon for he case: rosed ore nellgen secru allocaon schees n a ore dynac and fruful syle As e goes by any roosals o solve he secru allocaon roble have been resened [5] In [6] he auhors nvesgaed he nfluence of ul-cell ul-oeraor nerference on he wreless resources n case any oeraors co-exs and consue a shared secru reosory Moreover hey suggesed a fraework desgned o rove he bandwdh ulzaon whle rovdng a reasonable QoS degree by ensurng only a slgh nerference level aong oeraors Anoher aroach s usng nellgen algorhs o search for he ou allocaon schee as leened n [7] [8] and [9] The dencal dea aeared n hese works are usng swarbased algorhs o carry ou a global search for he bes answer sasfyng a re-defned hreshold In hs aer we roose a cenralzed oal secru allocaon (COSA) schee ha s ossble o sasfy QoS requreens whch were descrbed n [0] a a beer degree han revous soluons Meanwhle we also resen a odfed verson of Socal Sder Algorh () [] o benef s ably n allocaon arx decson ozaon The res of hs aer consss of below ars Frsly relaed works are descrbed n ar 2 Then Cognve Rado Neworks Secru Allocaon Proble (CRNSAP) s enoned n ar 3 Aferwards ar 4 resens our roosed -based secru allocaon schee where ajor odfcaons n he algorh o suor cognve rado neworks are clarfed Our sulaon rearaon rocess and nuercal resuls are dscussed n ar 5 Evenually conclusons and fuure consderaons are rovded n he las secon 3879

2 RELATED WORKS In hs secon we wll focus on secru allocaon roveens based on ozaon algorhs We choose o descrbe here one of he os faous ones secru allocaon based on Parcle Swar Ozaon () [2] whch was nroduced n [9] Ths ar of he aer wll also descrbe he base algorh for our roosed schee Socal Sder Algorh [] A for Cognve Rado Neworks The algorh whch was resened by Kennedy and Eberhar n [2] s an evoluonary algorh based on he relcaon of he socal behavor of a brd s oulaon Ths oulaon s called a swar whle he ndvduals n he oulaon are naed arcles A arcle reresens a robable soluon for he ozaon roble Parcle a eraon s deerned by s oson n D- x x x x D denson search 2 f x evaluaed by he fness funcon The arcles are In each arcle ravels owards he bes oson The obly and he velocy are affeced by wo asecs: he ersonal bes oson of each arcle so-called bes 2 D and he global bes oson of he enre oulaon naed gbes 2 g g g gd The velocy of he arcle s reresened as v v v v 2 D The velocy and he oson of arcle are brough u-o-dae deendng on he below forulas: v v c r x c r x () d d d d 2 2 gd gd x x v (2) d d d where s he nera wegh; c and c 2 are learnng raes; r and r 2 are wo unforly dsrbued arbrary nubers roduced ndeendenly n 0 For he urose of [9] s o ake he os of he hroughu of he enre cognve rado nework he fness funcon was oulned as follows f x n N n C s x r SI I PI P n n l n l ohers Sewse rocess of he -based secru allocaon algorh wren n [9] s as follows: Se : Generae channel avalably arx L channel bandwdh arx B channel sae nforaon arx G nerference consran arx C and rans ower arx P Se 2: Esablshen of he araeers requred by he algorh: he quany of he arcles learnng facors c c 2 reeon es nera wegh he orgnal eeraure T 0 he coolng rae and he velocy range V V The nex asks are randoly assgnen of he osons and veloces of he arcles n search sace ensurng ha he osons of he arcles ee he execaons of he consran arx C calculang and fndng he global bes oson g Se 3: Calculae he acceance ossbly value roulee oron and he Q a curren eeraure Execue roulee aroach o ck he global bes oson g fro Se 4: Modfy he osons and veloces akng sure ha he oson of he arcles leases he resrcons of arces L and C If he velocy f v n V ake v n v n V ake V v n V Se 5: Reckon he fness values of all arcles ake bes u o dae and decrease he eeraure and Se 6: If he eraon es reaches eraon es so he loo and ouu he global bes oson he fness of g If no reurn o se 3 B Socal Sder Algorh () g as well as [] s a eaheursc global ozaon algorh encouraged by he behavor of socal sder In he soluon of an ozaon roble s characerzed by he oson of an arfcal sder on he hyer-denson sder web The sders ove on he web and hey share he oson nforaon va vbraons Each sder could ove ndeendenly as long as does no leave he web Whenever a sder oves o a new locaon roduces a vbraon whch s roagaed hrough he web Deend on he receve vbraons fro oher ones a sder wll be guded o he oal oson Deals of wll be descrbed n he followng subsecons 3880

3 a) Sder The arfcal sders are he ajor funconng enes of Every sder says a a secfc oson on he hyer-denson sder web and he fness value of hs oson s lnked o he sder Addonally he sder has a eory sace used o sore s condon along wh ozaon araeers ncludng s resen oson resen fness value subsequen vbraon a earler eraon nacve level recedng oveens and a denson ask These characerscs hel he sder o look for he deal soluon b) Vbraon Vbraon s a val conce n I s one of he key feaures ha searae fro dfferen eaheursc algorhs uses wo roeres o descrbe a vbraon ha s he source oson and he source nensy of he vbraon The followng equaon defnes how o coue he source vbraon nensy Is ( ) log( ) f () s C where f() s s he fness rae of he sder s and C s a consan n order ha he nu fness values are larger han C The equaon below calculaes he aenuaon when he vbraon s ransed fro sder s o sder s ' : D( s s ') I( s s ') I( s) ex ra where r (0 ) a (4) s a user-conrolled oerand Ths araeer anulaes he aenuaon rae of he vbraon nensy over dsance Meanwhle s he sandard devaon of all sder osons along each denson c) Search Paern rocesses a oulaon of sders hrough a sequence of ozaon hases Parcularly each eraon of could be searaed no he below ses FITNESS EVALUATION On he sar of an eraon he fness values of he osons of all sders s reassessed These fness values wll be used n he vbraon generaon and broadcas rocedure VIBRATION GENERATION Frsly a new vbraon s creaed for each sder Subsequenly ha vbraon s sread o all oher sders n he web wh dsance consderaon: he vbraon wll be aenuaed hough he sace Laer he larges aenuaed vbraon nensy s carefully chosen deends on he receved vbraons and coare wh he revous one As a resul he larger nensy vbraon s sored as he arge vbraon If he sder decdes o aler s sored vbraon he nacve level s ncreased by one oherwse s se o zero value Ths level urose s o hel he algorh avod local oa MASK CHANGING Afer he followng vbraons of every sders are coued hs hase wll udae her osons In hs sage a denson ask s used o gude he oveen [] Each sder holds a bnary vecor ask whose lengh s he denson of he ozaon roble In each eraon a sder has a n robably N c o odfy s ask for clarfcaon N n s he sder s nacve nuber Whenever a ask s decded o be changed each b n he ask has a robably o ge a value and a robably o be se a 0 value RANDOM WALK A hs hase each sder carres ou a rando walk o rove her osons Afer he walk he new soluon should be fxed o ensure ha no sder oves ou of he web A ore dealed descron of he rando walk consderaons could be found n [] COGNITIVE RADIO NETWORKS SPECTRUM ALLOCATION PROBLEM (CRNSAP) In a wreless nework a user s an objec ha uses a channel (a ece of he rado secru) o send and receve nforaon They are sl no wo yes: rary users and secondary users The basc rule s rary users always have hgher rory over all secondary users n her regsered frequency bands In anoher way secondary users ay ulze hese channels only when hey are no beng eloyed by rary users Moreover secondary users us gve u hese channels on any occason he rary users wan he In hs aer we wll oulne he roble as descrbed n [0] Suose ha every user (boh rary or secondary) uses an on-dreconal anenna and could anulae he ranssson ower and consequenly s nerference range We call ( ) d n he nerference range of user n whose s beng assgned wh channel and have he ye s where 388

4 could be eher or s whch sands for a rary user or a secondary user resecvely Frsly all rary users decde her favore channels and he corresondng nerference ranges (by rulng he ranssson owers) Then he secondary users could fx he uer-bound ranssson owers and he nerference ranges n order ha here should no be any nerferences wh rary users Because of he hardware laon he nerference range should be consraned gven by d d ( n ) d for user n and channel n Assue ha he nework consss of N secondary users and M orhogonal channels We could buld he channel n n 0 based on he NM osons and he nerference ranges of all rary and secondary users In hs arx l has he eanng ha avalably arx L l l channel s ready for secondary user n o use Oherwse l wll be se o zero n Addonally we also defne he channel reward arx B b n N M n Each eleen b sybolzes he reward when a secondary user n chooses o use channel Furherore we llusrae he nerference aong he secondary users by he nerference consran arx C c c n k n k 0 NNM n where c eans n k ha user n wll nerfere wh user k f boh of he eloy c s se he zero value channel Oherwse n k Evenually he channel assgnen arx A an an 0 s used o secfy whch NM channels are ered o be used by he secondary users where a les ha channel s allocaed o secondary user n n and an 0 eans ha channel s no assgned o secondary user n Nex we wll dscuss abou syse consrans If all secondary user channel assgnens whch are carred by arx A only allocaed wh channels whch do no conflc wh any oher users we could conclude ha arx A s n a nonconflcng condon Ths can be defned by equaon (6) below a a c n k N M n k n k Besdes because of hardware ls every rado nerface n a CRN syse should have a laon C on he u nuber of channels alloed [0] Ths can be saed as M an C n N (7) Ulaely he urose of solvng he CRNSAP s zng he reward obaned fro an assgnen A Ths could be characerzed by he uly funcon U( A ) As enoned n [0] we defne he uly as: ) Max-Su-Reward (MSR): U ( A) a b N M (8) MSR n n n 2) Max-Mn-Reward (MMR): U ( A) n a b (9) MMR n n nn 3) Max-Prooronal-Far (MPF): N M N 6 MPF ( ) n n 0 n U A a b M (0) In hese objecve funcons MSR and MMR ry o boos he reward of he enre syse and ha of he os dsadvanaged user resecvely Meanwhle MPF s desgned o ensure he farness n channel assgnen The soluon o he roble s an assgnen arx A Those assgnens whch sasfy boh consrans defned n equaons (6) and (7) ogeher for he reasonable soluon se Generally CRNSAP could be exressed as subjec o U( A) A a a c n k n k M n n k N M a C n N where U( A ) could be U ( A ) U ( A ) U ( A ) MSR MMR MPF PROPOSED -BASED SPECTRUM ALLOCATION ALGORITHM FOR CRN We have ade any odfcaons o he orgnal algorh o suor he secru allocaon roble n CRN A dealed flowchar of our verson of for he CRNSAP s rovded n Fg The below sub-secons wll clarfy wha we alered o he orgnal roosal A Soluon Reresenaon A soluon for he CRNSAP s a channel assgnen arx called A n whch he coluns sand for channels and rows 3882

5 reresen secondary users (SU) For exale he arx below: 0 A shows ha SU s graned channels and 3 SU2 s assgned channels 2 and 3 whle SU3 ges channel In our algorh an neredae channel assgnen arx used o be called X Noe ha he fnal X afer all alernaons and evoluonary s exacly he A arx The shae of X n he rocessng rocedure could be a bnary vecor or a arx START Daa Se Inalzaon Sder Poulaon (X) Creaon Conver X o Bnary For CRN Consrans Processng Assgn Paraeers Fness Calculaon B Real-o-Bnary Nuber Converson Funcon In our roosed algorh he sgod funcon s ulzed o conver real nubers o bnary ones Ths funcon was also used n [3] [4] and [5] Our real-o-bnary nuber converson funcon lays s ar a wo laces n our sulaon code: afer naon of he assgnen arx and afer he rando walk The osons are n real values hen and hey should be convered o bnary values by usng he equaon (3) below: X s 0 f rand () S( Ps ( )) ( ) f rand () S( Ps ( )) (2) where S() s he sgod funcon for ransforng he velocy o he robably as he followng equaon: S( Ps ( )) (3) Ps ( ) e Fg2 wll descrbe he grah drawn fro he ouus of he sgod funcon No Yes Sore Var = Vbes Choose Bes Vbraon fro he Web Vbes > Var Rando Walk Deerne new oson for every sders (X) Conver X o Bnary For CRN Consrans Processng Song Creron Yes Ouu he Soluon No Use Var fro he revous eraon C Consrans Resoluon END A secfc channel assgnen arx wll be checked wh hese suffs o be vald: Fgure : Our -based CRN Secru Allocaon Algorh Channel avalably arx L Inerference consran arx C Maxu nuber of channels alloed C Rules fox handlng a channel assgnen arx X are as follows Frsly we check X s eleens wh arx L If L()=0 3883

6 hen X() s se he zero value for n range o he oulaon sze and ranges fro o he nuber of densons Secondly he resulng X wll be ached wh he nerference consran C Sgod Funcon Ouu P = [2 4] In addon our sulaon uses he araeers descrbed n he able below Noe ha he oulaon sze nuber of channels nuber of rary users and secondary users nuber of densons of he ozaon roble u eraon u ranssson ower of secondary users nu ranssson ower of secondary users are he sae for he wo algorhs Oher varable secfc o each algorh s used as defaul values whch are descrbed n [7] and [9] Table : Sulaon araeers Paraeer Varable nae Value Nuber of ndvduals o_sze 20 Nuber of denssons d MN Maxu eraon _er 000 Fgure 2: Sgod funcon used n our roosed algorh Rae of vbraon aenuaon when roagang over he web r a In arcular he odfcaons n a channel assgnen arx X wll be conrolled by he value of an eleen C(j) n he arx C If C(j)= hen here wll be a collson when boh X() and X(j) are se he value s In hs conex X() or X(j) wll be se he value 0 based on a rando robably Aferwards X wll be checked wh C Each row n he X arx wll be exaned and reassgn he value f he requreen enoned n equaon (7) s volaed PERFORMANCE EVALUATIONS We conduc all he sulaons on he sae hardware syse and use only on syse araeers se for far coarson beween our roosed algorh and he referenced soluon An Inel(R) Xeon(R) E cores (2 hreads) CPU wh 8GB RAM couer syse s used Each algorh s sulaed 30 es and he values deonsraed n he dagras are he average ones derved fro hese ess In each run our -based algorh uses he loo of 000 eraons o search for he bes soluon For equy he -based sulaon uses he sae nuber of loo As enoned before we have rary and secondary users n a CRN syse We decde o creae K rary users and assgn channels o he Noe ha here are M channels MK and soe rary users could share he sae channel Aferwards he arces L B C are generaed based on he seudo codes gven by [0] An nal soluon X wll also be roduced We hen conver X o a bnary arx o fulfl he requreens of he CRNSAP Inerference range (roecon area) DPR 2 Probably of changng he ask c 07 The robably for each b of he vecor o be assgned he value f he ask changed 0 To secure he farness we use he sae daa se whch s generaed ror o he sulaons as he sandard roble for boh and -based secru allocaon schees Addonally here are oher suffs o be roduced before he ess ncludng he arces L B C and secfcaons for rary and secondary users We used he seudo codes gven n [0] and [6] o creae hese daa and all codes are rograed n MATLAB For CRNSAP s a zaon ssue whle s roosed for nzaon we conduc soe odfcaons o he fness funcon as n [7] Raher han zng U( A ) we decde o nze he uly funcon U '( A) 000 U( A) n he eraons of Fnally we ouu U( A ) 000 U '( A ) as he concluded soluon for bes each sulaon bes 3884

7 Max Su Reward value and sauraes a C =4 whle -based one could no ncrease he desred value snce C =6 Subsequenly he erforance of our algorh s very balanced whls based one shows ha does no deals wh hs suaon very well A reason for hs crcusance s he song creron Whle he ouu could be beer he song creron revens he algorhs o dg deeer no he search sace Noneheless our algorh connually show ha s a beer choce n any conexs C Fgure 3: MSR coarson beween and Max Mn Reward In hs aer we sudy he ac on he aoun of ered allocaed channels o he users by ncreasng he value of C fro o 20 (se=) A clear concluson could be drawn fro he bencharks s he saller C s he saller he nuber of channels could be gven o he users and as a resul he saller he fness values could be acheved Our sulaon resuls are deonsraed n below dagras Each grah vsualzes he erforance of -based and based algorhs for secru allocaon roble n CRN n a dfferen requreen Fg 3 shows her caables n a syse focus on zaon of he oal of reward values of channel assgnens Our roosed algorh always asses he resuls aaned by -based schee The dfferences vary fro 292% o 004% for each C value Fg4 and Fg5 are where our roosed ehod ouerfors he -based schee Fg4 clarfes he erforance of he wo aroaches when ryng o ze he nu reward values when we ncrease he C value fro o 20 Whle -based echnque has unsable ouus -based algorh always has sgnfcanly hgher MMR Moreover our soluon also roves ha could acheve he bes soluon faser han he oher one The ouu of -based algorh quckly reaches he eak C Fgure 4: MMR coarson beween and The sae henoenon reaears n Fg5 The rendlnes n Fg5 also show a huge ga beween our algorh and he based one The searaons beween he wo algorhs ranges fro 5073% o 722% a C =2 The dfference s so large ha we have o research o fnd ou why hs could haen The draac caacy of our algorh could be exlaned by he naure of he aroaches has a novel soluon where every sder conrbues o he decson of where o ove o n he nex se of a secfc one A sder suors oher ones and s wegh on a secfc oveen reles on s aenuaed vbraon a he one ha should decde wha o do Anher fac should be ake no consderaon s ha our roosal reaches he sauraon on a a hgher C value han ha of he soluon for CRNSAP Ths suaon could be exlaned by he dfferen rores of he objecve funcons More dealed nuercal resuls are rovded n he Table II 3885

8 Nu Run = 30 Table II Perforance Analyss Resuls Max Mean Sd Max Mean Sd C = C = C = C = E (a) Max Su Reward Nu Run = 30 Max Mean Sd Max Mean Sd C = E C = E C = C = E (b) Max Mn Reward Nu Run = 30 Max Mean Sd Max Mean Sd C = C = C = C = (c) Max Prooronal Far The sub-ables of Table II once agan reresen he donaon of our algorh over he referenced one I s frly beer on and ean values whle he sandard devaon (Sd) s sgnfcanly lower han -based aroach CONCLUSIONS In hs aer we have nroduced a rosng aroach n solvng he CRNSAP Is aazng erforance roves ha here s always soe way o enhance he syse erforance esecally when we ulze a revoluonary swar-based Max Prooronal Far C Fgure 5: MPF coarson beween and algorh o search for bes soluon for hard robles Our roosed algorh shows soe unsable resuls n MSR objecve funcon There also ore ways o oze he erforance of he algorh such as arallelze he coung syse or cobne wh soluons fro oher algorhs Tha wll be where we focus n rovng our roosal n he nex works REFERENCES [] J Marnho and E Monero Cognve rado: Survey on councaon roocols secru decson ssues and fuure research drecons Wrel Neworks vol 8 no [2] I F Akyldz W-Y Lee M C Vuran and S Mohany NeX generaon/dynac secru access/cognve rado wreless neworks: A survey Cou Neworks vol 50 no [3] J Mola and G Q Magure Cognve rado: akng sofware rados ore ersonal IEEE Pers Coun vol 6 no [4] Bebe Wang and K J R Lu Advances n Cognve Rado Neworks: A survey IEEE J Sel To Sgnal Process vol 5 no

9 [5] E Z Tragos S Zeadally A G Fragkadaks and V A Srs Secru Assgnen n Cognve Rado Neworks A Corehensve Survey IEEE Coun Surv Tuorals vol 5 no [6] G Alnwa K Arshad and K Moessner Dynac Secru Allocaon Algorh wh Inerference Manageen n Co-Exsng Neworks IEEE Coun Le vol 5 no [7] A Y S La and V O K L Checal Reacon Ozaon for Cognve Rado Secru Allocaon 200 IEEE Glob Telecoun Conf GLOBECOM [8] T Sddque and A Aza Secru Ozaon n Cognve Rado Neworks usng Genec Algorhs Bleknge Insue of Technology 200 [9] Z Je and L Tejun Secru Allocaon n Cognve Rado wh Parcle Swar Ozaon Algorh [0] C Peng H Zheng and B Y Zhao Ulzaon and farness n secru assgnen for oorunsc secru access Mob Neworks Al vol no [] J J Q Yu and V O K L A socal sder algorh for global ozaon Al Sof Cou J vol [2] J Kennedy and R Eberhar Parcle Swar Ozaon Encycloeda of Machne Learnngachne learnng Srnger US [3] J Kennedy and R C Eberhar A dscree bnary verson of he arcle swar algorh 997 IEEE In Conf Sys Man Cybern Cou Cybern Sul vol [4] M A Khanesar M Teshnehlab M A Shoorehdel and E E Faculy A novel bnary arcle swar ozaon 2007 Mederr Conf Conrol Auo vol no [5] S Mrjall S M Mrjall and X S Yang Bnary ba algorh Neural Cou Al [6] A Y S La V O K L J J Q Yu and S Meber Power-Conrolled Cognve Rado Secru Allocaon wh Checal Reacon Ozaon vol 2 no