Journal of Communications Vol. 11, No. 5, May 2016

Transcription

1 Jorna of Commncatons Vo., No. 5, May 6 Dynamc Rate Aocaton for Mtpath Rotng nder Path Stabty and Prortzed Traffc Sesson Constrants for Cogntve Rado Ad Hoc Networks wth Sefsh Secondary Users Long Zhang, Chnhong Ba, Fan Zho, Hatao X, and We Hang 3 Schoo of Informaton and Eectrca Engneerng, Hebe Unversty of Engneerng, Handan 5638, Chna Schoo of Compter and Commncaton Engneerng, Unversty of Scence and Technoogy Bejng, Bejng 83, Chna 3 Energy Effcency Evaaton Center, Chna Eectrc Power Research Insttte, Bejng 9, Chna Ema: zhangong@hebe.ed.cn; {chhba, fnzho}@otook.com; xhatao@stb.ed.cn; hangwe@epr.sgcc.com.cn Abstract In ths paper, we nvestgate the dynamc rate aocaton for mtpath rotng wth the node-dsjont paths n cogntve rado ad hoc networks composed of sefsh secondary sers (SUs) whch are nwng to forward the packets. We mode the path stabty factor of each path nder a certan traffc sesson by dstrbtng avaabe channe n vew of the carrer freqency of the avaabe pnk channe on each path. Then the traffc sensng factor s aso devsed to characterze the prortes of dfferent traffc sessons aong the node-dsjont paths. In addton, we propose the dfferenta game mode of dynamc rate aocaton for mt-path rotng by takng nto accont both the mobty and random natre of the sefsh SUs. A set of the non-cooperatve optma sotons to the dynamc rate aocaton mode s frther obtaned by dervng an optmzaton probem. Moreover, the dstrbted optma rate pdate re s deveoped to dynamcay regate the rate of traffc sesson over a certan path. Nmerca rests confrm that the effectveness of the proposed dynamc rate aocaton mode. Index Terms Cogntve rado, ad hoc networks, rate aocaton, path stabty, prortzed traffc sesson I. INTRODUCTION Cogntve rado (CR) or dynamc spectrm access n [] has newy emerged as a promsng soton to mprove the spectrm tzaton by aowng ncensed secondary sers (SUs) to access de censed spectrm. In a CR network, SUs can perodcay sense the censed spectrm and opportnstcay access the spectrm hoes or spectrm opportntes (SOPs) noccped by prmary Manscrpt receved November 6, 5; revsed May, 6. Ths work was spported by the Natona Natra Scence Fondaton of Chna (6447, 6546, 6343), the Natra Scence Fondaton of Hebe Provnce of Chna (F3439), the Scentfc Research Fondaton of the Hgher Edcaton Instttons of Hebe Provnce of Chna (QN348), the Research Program of Scence and Technoogy of Hebe Provnce of Chna (5444D-), and the Research Project of Hgh-eve Taents n Unversty of Hebe Provnce (GCC46) Correspondng athor ema: zhangong@hebe.ed.cn. do:.7/jcm sers (PUs). Most of the exstng research efforts n CR networks many focs on the sses of the physca and MAC ayers nder an nfrastrctre based snge-hop scenaro, ncdng cooperatve spectrm sensng, spectrm access and sharng technqes []-[7]. In addton, SUs can frther form a mt-hop ad hoc network wthot the spport of nfrastrctre. In a cogntve rado ad hoc network (CRANET) [8], SUs can ony access the SOPs by seekng to nderay, overay, or nterweave ther sgnas wth those of exstng PUs wthot sgnfcanty mpactng ther commncatons. In comparson wth the ower ayer sotons expored by the exstng research efforts as reported before, recent work ndcates that there are many new chaenges towards the rotng probem pon the network ayer n mt-hop CRANETs, sch as the copng between rotng and spectrm-awareness, path stabty and spectrm avaabty, and rote mantenance, etc [9]-[]. In ths context, rotng desgn n CRANETs dffers sffcenty from the ad hoc networks, and mst be carefy consdered to dea wth the creaton and the mantenance of mt-hop paths among SUs by seectng both the next hop SUs and the spectrm avaabty on each nk of the path de to the presence of PUs. To factate the dstrbton of the traffc oad and aso to avod the rote fare, there s a need to expot mtpath rotng from the perspectve of path dversty to mprove the mt-hop transmsson effcency and stabty n CR networks, notaby []-[6]. Aso, rate aocaton for mt-path rotng n CR networks s a key technoogy to ensre the effectve packet devery n dfferent paths at the same tme as we as the traffc oad baance. Frthermore, effectve rate aocaton w aso hep to mnmze end-to-end deay and contro message overhead. Varos rate aocaton methods have been deveoped for CR networks from dfferent perspectves, ncdng end-to-end deay optmzaton [7], QoSconstraned b-objectve optmzaton [8], jont rate and power aocaton [9], etc. 6 Jorna of Commncatons 46

2 Jorna of Commncatons Vo., No. 5, May 6 As an aternatve framework for modeng, gametheoretc approach has ganed more attenton as an economc too to stdy the rate aocaton probem n wreess networks [], []. In [], a non-cooperatve game-theoretc framework for dstrbted asynchronos power and rate aocaton n ad hoc networks was proposed. The rate aocaton wthn ths framework deas wth ts convergence to network-wde acceptabe eqbrm states nder stochastc commncaton channes. In [], a probem of jont channe and rate assgnment was formated as a statc non-cooperatve game wth chargng scheme, n whch the dfferent ndvda demand constrants n ad hoc networks are taken nto accont. The exstng rate aocaton modes [], [] sng the statc game theoretc approach cannot take nto accont the nderyng constrant that CR networks n practce were sbject to the dynamc tmevaryng network topoogy wheren the assgnment of rate shod be contnos n tme. Another body of work formates the dynamc rate aocaton probem for mtpath rotng n CR networks as dfferenta game mode [], [3]. In [], an effcency-awareness rate assgnment probem of mt-path rotng was devsed as a non-cooperatve dfferenta game mode wth the hop nmber and bandwdth constrants, and the eqbrm of the game s aso obtaned. In [3], a mt-person noncooperatve dfferenta game mode for stabe-aware traffc oad sharng over mt-path rotng was presented, and the feedback Nash eqbrm of the game s derved. However, athogh [] and [3] sed dfferenta game theoretc approach to stdy the dynamc rate aocaton probem, they dd not consder some practca factors sch as path stabty and prortzed traffc sesson for mt-path rotng n dynamc rate aocaton mode. Or work n ths paper many focses on the nderay CRANET scenaro de to the hgh spectrm tzaton and the smpcty of mpementaton. Both mobty and sometmes random natre of SUs rest n the dynamcs of the CRANET topoogy wth respect to tme dependency. In genera, gven the dynamc tme-varyng network topoogy, t s not reastc to keep the traffc rate nchangeaby for mt-path rotng. Hence, t s crca to dynamcay aocate the rate for mt-path rotng accordng to the dynamc natre of tme dependency. Meanwhe, n order to avod the rerote operatons de to the break of the end-to-end paths, an avaabe and stabe path shod be seected as one of the node-dsjont mt-path rotes, amng to acheve the effectve assgnment of the traffc oad aong the node-dsjont mt-path rotng. Takng nto consderaton the mpact of the dfferentated types of traffc sessons on the demand of transmsson qaty, ths observaton motvates s to characterze the prortes of the dfferent traffc sessons to dea wth the rate aocaton probem. Wth ths respect, the dynamc rate aocaton framework for the node-dsjont mt-path rotng s formated bearng n mnd the constrants of path stabty and prortzed traffc sesson. In ths work, we stdy the dynamc rate aocaton for mtpath rotng wth node-dsjont paths by takng nto accont the constrants of the path stabty and the prortzed traffc sesson for CRANET. We defne the path stabty factor of each path nder a certan traffc sesson n mtpath rotng scenaro. Specfcay, the path stabty factor s cacated by dstrbtng the avaabe channe accordng to the carrer freqency of the avaabe pnk channe on each path wth dfferent dstance. To characterze the prortes of dfferent traffc sessons aong the node-dsjont paths, we aso devse the traffc sensng factor to evaate the mpact of dfferentated types of traffc sessons on the rate aocaton probem. By anayzng the characterzed factors, the payoff fncton of a certan path from the pstream sorce SU s frther presented based on two cost fnctons ncdng the path stabty cost and the traffc sesson cost. Owng to both the mobty and sometmes random natre of the sefsh SUs, we propose the dfferenta game mode of dynamc rate aocaton for mt-path rotng. In ths mode, the transmsson rate of traffc sesson over a certan path from the pstream sorce SU s dynamcay regated accordng to the dynamc natre of tme dependency. Moreover, we devse an optmzaton probem to obtan a set of the noncooperatve optma sotons to the dynamc rate aocaton mode. The dstrbted optma rate pdate re s aso empoyed to dynamcay regate the rate of traffc sesson over a certan path. The remander of ths paper s organzed as foows. In Secton II, we ntrodce the premnares ncdng network mode, path stabty mode, and prortzed traffc sesson mode. The dynamc rate aocaton mode s formated n Secton III. Then the non-cooperatve optma soton s descrbed n Secton IV. Secton V presents the nmerca rests. Fnay, we concde the whoe paper n Secton VI. A. Network Mode II. PRELIMINARIES We consder a dstrbted nderay CRANET scenaro, where PUs send ther packets to the prmary base staton throgh the cear prmary networks wth a fnte set of ces. We empoy the ndependent and dentcay dstrbted aternatng ON-OFF process to mode the occpaton tme ength of PUs n pnk channes. Specfcay, the OFF state ndcates the de state where the noccped pnk channes or the SOPs can be freey occped by SUs. Let t and T denote the startng tme and the termna tme of the dynamc rate aocaton for mt-path rotng n ths scenaro, respectvey. By the means of coaboratve spectrm sensng, SUs can ony everage the OFF state to access the SOPs over the athorzed pnk channes to frther form a dstrbted mt-hop nderay CRANET wth a set of statonary SUs wthn tme nterva be the set of, 6 Jorna of Commncatons 463

3 Jorna of Commncatons Vo., No. 5, May 6 traffc sessons. We assme that there s a traffc sesson from a snge pstream sorce SU to a downstream destnaton SU aong n nodedsjont paths. Let,, n be the set of the nodedsjont paths. Note that a path P, nder the traffc sesson s a sbset of SUs. B. Path Stabty Mode We assme that there exst k nks wth dfferent engths on path P, over m avaabe pnk P channes. Let be the -th nk of path P, for,,,k. The ength of the -th nk of path P s assmed to be P d. We then rank the ength of k nks n descendng order and denote by sorted nk, for,,, k P L the -th, where s the abe n descendng order. We remark that the greater vae of w rest n the shorter ength of the correspondng nk. P Anaogosy, we denote by f the carrer freqency of the avaabe pnk channe on path P, for,,,m. The m avaabe pnk channes w aso be ranked n ascendng order based on the vaes of the carrer freqency. The ascendng order s denoted by, for,,,m. Notce that the greater vae of yeds the greater vae of the carrer freqency. Therefore, the stabty copng factor of the -th sorted nk defned as P L on path P, can be formay, f k, f k where s a constant, for. Note that Eq. () P ndcates that the -th sorted nk L can fnd ts most approprate channes n vew of the ength of nks and the carrer freqency of pnk channes. Obvosy, k s smaer than k. We assme that there have A sorted nks wth stabty copng factor k and k A sorted nks wth stabty copng factor k. For the convenence of cacaton, we assme that the vae of n k A () sorted nks are represented by G, G,, Gk A, respectvey. Hence, we defne the path stabty factor of path P, as, whch s the sm of the stabty copng factor of the -th sorted nk path stabty factor of path P L on path P. Specfcay, the P can be wrtten as k A A k () k G C. Prortzed Traffc Sesson Mode Appeang to the Enhanced Dstrbted Channe Access (EDCA) mechansm for the spport of appcatons wth QoS reqrements n IEEE 8.e standard [4], we devse the traffc sensng factor nder the EDCA mechansm to qantfy the prortes of the dfferent traffc sessons. Or objectve s to characterze the prortes of dfferent traffc sessons aong the nodedsjont paths bearng n mnd the dfferentated types of traffc sessons. From the pont of vew of the prorty, we cassfy the traffc sesson types for mt-path rotng nto for access categores (ACs) [4]. Let q be an nteger ndex of the assocated traffc type to an AC, for q,4. As shown n Tabe I, for ACs correspond to for types of traffc sessons, ncdng Voce Traffc (VO), Vdeo Traffc (VI), Best Effort Traffc (BE), and Backgrond Traffc (BK). We denote by F q the traffc sensng factor for traffc sesson nder traffc type q. Accordng to the EDCA mechansm [4], the traffc sensng factor gven as where F q, for can be F q Random q asottme AIFSN q (3) Random q s a psedo-random nteger drawn,cw q from a nform dstrbton over the nterva, asottme s a constant tme nmber n mcroseconds based on dfferent physca ayer technqes n the EDCA mechansm, and AIFSN q s the arbtraton nterframe space nmber of the EDCA mechansm, whch s frther formated n Tabe I. It shod be noted that mn, max CW q CW q CW q s an nteger wthn the range of vaes of the contenton wndow mts mn CW max q. CW q and TABLE I. TRAFFIC SESSION TYPES AND CONTENTION PARAMETERS Traffc Sesson Prorty CWmn(q) CWmax(q) AIFSN(q) Type VO Pr() (acwmn+)/4- (acwmn+)/- VI Pr() (acwmn+)/- acwmn BE Pr(3) acwmn acwmax 3 BK Pr(4) acwmn acwmax 7 III. DYNAMIC RATE ALLOCATION MODEL Practcay, drng the mt-path transmsson wth the traffc sesson from sorce SU to destnaton SU, SUs n charge of forwardng the packets appear to the sefshness. Ths s a natra dea de to the fact that the SUs are nwng to forward the packets on behaf of other SUs snce these operatons w 6 Jorna of Commncatons 464

4 Jorna of Commncatons Vo., No. 5, May 6 ead to the consmpton of ther own resorces, sch as energy and avaabe bandwdth. Becase of the sefsh SUs n forwardng actons, the pstream sorce SU as the rate controer of the traffc oad nder the traffc sesson shod pay for each downstream SUs aong path P to accommodate the consmpton of resorces. Based on both the mobty and sometmes random natre of the sefsh SUs, dfferenta game s adopted to constrct the dynamc rate aocaton mode for mt-path rotng by takng nto accont the tmecontnos rate change of the traffc sesson. The objectve of the pstream sorce SU s to mnmze the cost or the so caed payoff fncton of path P nder the constrant of the path stabty and prortzed traffc sesson. Let r s denote the nstant transmsson rate of traffc sesson over path P, from the pstream sorce SU at tme s t, T. Accordng to Shannon s capacty theorem, the bandwdth of path P can be approxmatey cacated as r s og S N, where S N s the sgna-to-nose rato over path P. Under ths descrpton, the copng constrant between the traffc sensng factor F q and the bandwdth of path as P,, denoted by, s frther represented og r s SN F q To formate the payoff fncton of path (4) P that the pstream sorce SU shod pay, we then proceed to propose the path stabty cost fncton and the traffc sesson cost fncton. Let hs be the crrent prce that pstream sorce SU s wng to pay at tme s t T,. Reca that the sm of the stabty copng factor of the -th sorted nk L on path P s characterzed by the path stabty factor Hence, the path stabty cost of path tme s t T, denoted by PSC,, P, at s, s modeed as PSC s h s (5) To qantfy the mpact of prortzed traffc sesson on the dynamc rate aocaton, we devse the traffc sesson cost of path s t T, denoted by TSC P, at tme s, whch can be cacated as, TSC s r s (6) h s where over path s a prce coeffcent of traffc sesson P. Therefore, based on the cost fnctons as dscssed above, the tota cost of path tme s t T s gven by, h s F qog S N P, at r s h s Appeang to dfferenta game theory by [5], can be regarded as the state varabe, and (7) hs r s can be consdered as the strategy. The strategy means the choce of acton or behavor by payer n dfferenta game. We emphasze that the path P refers to the payer of the game. For convenence, we vew the startng tme of the game as t throghot the paper. From Eq. (7), the am of path P at tme s, T s to mnmze the tota cost fncton, whch can be formay expressed as Mnmze : og s s T F q S N h s (8) h s e ds where s a constant dscont factor, for. Accordng to dfferenta game theory [5], the state varabe hs n Eq. (8) s assmed to satsfy the dfferenta eqaton as foows h s h s r s h h r (9) Therefore, Eqs. (8) and (9) consttte the dfferenta game mode of dynamc rate aocaton for mt-path rotng. IV. NON-COOPERATIVE OPTIMAL SOLUTION In ths secton, we appy the dfferenta game theory [5] to derve the non-cooperatve optma rate of traffc sesson over path P nder or dynamc rate aocaton mode n Eqs. (8) and (9). Specfcay, we devse an optmzaton probem to obtan a set of the noncooperatve optma sotons to the dynamc rate aocaton mode as foows Mnmze : Sbject to : T F qog S N s hs e ds hs hs r s h h Let t h t h t h,,,,,, n, s r h s () be the set of the non-cooperatve optma sotons to the optmzaton 6 Jorna of Commncatons 465

5 Jorna of Commncatons Vo., No. 5, May 6 probem n Eq. (). Moreover, we assme that there V t, h, exsts contnosy dfferentabe fncton satsfyng the foowng parta dfferenta eqatons V t, h t mnmze : F q S N og s r hs e h s V t, h ht j t, h r t h j \ t () V T, h () Theorem : Under the condton of t V t, h A t b t B t e, the set of the noncooperatve optma sotons to the optmzaton probem n Eq. () s formated as foows og t h t h F q S N A tht (3),, where A t and B t satsfy foowng eqatons og F q S N A t A t A t 4 B t og Aj t A t (4) F q S N B t j \ j Proof: Performng the mnmzaton operaton of Eq. (), we obtan og, F q S N V t h h t t, h h t e Sbstttng th, (), we have (5) n Eq. (5) nto Eqs. () and t V t, h A t b t B t e (6) The non-cooperatve optma soton to the optmzaton probem n Eq. () s formated as Note that og F q S N (7) t, h A t h t A t and B t are two extra ntrodced axary varabes to better represent V t, h, where A t and B t can be gven n Eq. (4). Hence, ths competes the proof. For notatona smpcty, we set F q og S N =, and we aso omt the tme t n the axary varabes foowng paper. A t and B t throghot the Theorem : Let c denote a constant nmber. A n the non-cooperatve optma soton to the optmzaton probem n Eq. () can be frther formated as A tc H H e H e tch (8) (9) 4 j ( ) 4 j H () f and ony f the foowng condton s satsfed A H e A H ( tc) H Proof: Consderng the symmetrc form of A and () n Eq. (4), we can mmedatey denote Eq. (4) by Rccat eqaton da dt 4 j j B A A () Wthot oss of generaty, the form of Eq. () can be rearranged by the foowng dfferenta eqaton X A, t dt Y A, t da (3) X A t A A and 4 j j Y A, t. where, In vew of, Y A, t X A t A t, t s not dffct to recognze that Eq. () s not exact. Hence, we ntrodce U A, t whch can make a non-zero ntegratng factor the eqaton an exact form by mtpyng t on both sdes of Eq. (3). Here, for easy obtan where, U A, t 4, we can A A (4). So Eq. (3) mtped by 4 j U A t s exact, and then we obtan j,,,,, U A t X A t dt U A t Y A t da du A t Hence, we have U A, t A A A U A, t t (5) (6) 6 Jorna of Commncatons 466

6 Jorna of Commncatons Vo., No. 5, May 6 Let H 4. If and ony f A A, by ntegratng Eq. (6) wth respect to A, we have A H U A, t n H A H U A, t t b t We can easy obtan bt sovng Eq. (8) as foows b t (7) t. Let U A, t c. Upon A n H t c H A H so we have Eq. (8). Hence, ths competes the proof. (8) Agorthm : Dstrbted Optma Rate Update Re : Identfy the set,, n ; : Intaze F q for nder traffc type q, and for P ; 3: Intaze r of over P ; 4: t ; 5: oop 6: for each do 7:, H ( ) 4 ; 4 j 8: Compte A t gven by Eq. (8); 9: r t t, h wthn tme nterva,t ; : end for : f q q or t T then : Termnate the oop; 3: end f 4: end oop In smmary, the non-cooperatve optma rate of traffc sesson over path P n or dynamc rate aocaton mode s strcty derved as foows th, og F q S N t c H H e j H t ch ht e (9) Accordngy, based on Eq. (9), we devse Agorthm to acheve the optma rate pdate re n a dstrbted manner. V. NUMERICAL RESULTS Consder a dstrbted CRANET scenaro depcted n Fg., nvovng 5 SUs ocated randomy n the range of m m sqare area. We assme that there s a traffc sesson from a snge pstream sorce SU to a downstream destnaton SU aong 4 n node-dsjont paths, denoted by P, P 3 and P 4. The dstance P, d of each nk s cear n ths scenaro. We assme that the nmber of the avaabe pnk channe on n 4 node-dsjont paths s set to m 7. The choce of the channe for nks are gven n Fg., where P dstance of the -th nk of path,, and means that the P s ranked the - th and the vae of the carrer freqency of seected channe s ranked the -th. The stabty factor of n 4 node-dsjont paths can be obtaned by means of combnaton between Fg. and Eq. (). In Eq. (), we set the constant n Eq. () to.. In the dfferenta game mode of dynamc rate aocaton for mt-path rotng, we assme that the prce coeffcent for n 4 node-dsjont paths has the same vae, whch s defned as. The sgna-to-nose rato S N s set to 7dB, and the constant c n Eq. (8) s assmed to be -. The contenton wndow mts of the EDCA mechansm n IEEE 8.e standard are gven by acwmn 3 and acwmax =3. Range(m) Fg.. The smaton scenaro P() Soce SU Destnaton SU Range(m),4,3,,,, 3 P(),,5 4,6,4 P() P(),,6 5,3, P() 3,3,7 3,,5 6,7,3,,3,4,6,3, 4 3,,3,4,5,3,4,, 4 Fg.. Lnks and the correspondng channes 4 4 7,5, 6 Jorna of Commncatons 467

7 Jorna of Commncatons Vo., No. 5, May 6 Frsty, we show the non-cooperatve optma rate comparson of n 4 node-dsjont paths nder dfferent dscont factor. We assme that the BK as the type of data traffc s empoyed by the traffc sesson from a snge pstream sorce SU to a downstream destnaton SU aong n 4 node-dsjont paths. Ths, we choose the arbtraton nterframe space nmber T F, AIFSN 7. As for the traffc sensng factor we choose the psedo-random nteger Random 49 and the constant asottme s. Fg. 3 compares the non-cooperatve optma rate of n 4 node-dsjont paths s nder dfferent dscont verss game tme t,4 factor. From Fg. 3, we can see that the noncooperatve optma rate w ncrease when the dscont factor ncreases. Ths can be expaned by the fact that s the dscont factor whch s a rato of expected earnngs n the ftre and present vae. The hgher dscont factor w rest n the hgher rate for data transmsson. Meanwhe, Fg. 3 aso shows noncooperatve optma rate changes of for paths nder dfferent stabty factor wth the growth of game tme t,4 s. It s apparent that the non-cooperatve optma rate of for paths nder the same type of traffc sesson depends on the stabty factor accordng to Eq. (3). It can be observed the non-cooperatve optma rate of P 3 and P s obvosy arger than that of P, P 4. Ths can be expaned by the fact that the hgher stabty factor can redce the rato of the break of the end-to-end paths. 5 5,=.7 P(),=.7,=.7,= Tme(s) 5 5,=.3 P(),=.3,=.3,= Tme(s) 5 5,=.5 P(),=.5,=.5,= Tme(s) 5 5,=. P(),=.,=.,=. 3 4 Tme(s) Fg. 3. The non-cooperatve optma rate of for node-dsjont paths nder dfferent dscont factor Fg. 4 compares the payoff of n 4 node-dsjont paths verss game tme t,4 s nder dfferent types of traffc sessons. From Fg. 4, t s apparent that the ncrease of the path stabty factor of path rest n an enhancement of the payoff of path P w P. In genera, hgh path stabty factor needs hgh average payoff. Ths can be expaned by the fact that hgher stabty factor w rest n more payoff that path P needs to pay accordng to Eq. (7). Moreover, t s cear that the payoff of path P wth traffc sessons BE and BK cod be ess than that of VO and VI. The reason for ths s that the prorty of traffc sesson VO and VI s hgher than that of BE and BK. Then we fnd that the ower traffc sensng factor F q w ead to the rsng of the factor. Pay off of Pay off of BK,=.7 BE,=.7 VI, =.7 VO,= Tme(s) BK,=.7 BE,=.7 VI, =.7 VO,= Tme(s) Pay off of P() Pay off of BK,=.7 BE,=.7 VI, =.7 VO,= Tme(s) BK,=.7 BE,=.7 VI, =.7 VO,= Tme(s) Fg. 4. The payoff of n 4 node-dsjont paths nder dfferent types of traffc sessons Fg. 5 compares the non-cooperatve optma rate of path t,4 s nder dfferent P verss game tme types of traffc sessons. From Fg. 5, we can see that the non-cooperatve optma rate of traffc sesson VO and VI s sgnfcanty hgher than that of BE and BK. Ths can be expaned by the fact that the prorty of VO and VI s hgher than that of BE and BK BK,=.3 BE,=.3 VI, =.3 VO,= Tme(s) Fg. 5. The non-cooperatve optma rate comparson of dfferent types of traffc sessons P nder 6 Jorna of Commncatons 468

8 Jorna of Commncatons Vo., No. 5, May 6 VI. CONCLUSION In ths paper, we have deveoped the dfferenta game mode of dynamc rate aocaton for mt-path rotng n CRANETs by takng nto accont the mobty and random natre of the sefsh SUs. We devse the payoff fncton of a certan path from the pstream sorce SU based on two cost fnctons ncdng the path stabty cost and the traffc sesson cost. By devsng an optmzaton probem, we aso derve the non-cooperatve optma rate of traffc sesson over a certan path nder or dynamc rate aocaton mode. Moreover, we propose a dstrbted optma rate pdate re wth the prpose of dynamcay regatng the rate of traffc sesson over a certan path. REFERENCES [] S. Haykn, Cogntve rado: bran-empowered wreess commncatons, IEEE Jorna on Seected Areas n Commncatons, vo. 3, no., pp. -, Feb. 5. [] D. Xe, E. Ekc, and M. C. Vran, Cooperatve spectrm sensng n cogntve rado networks sng mtdmensona correatons, IEEE Transactons on Wreess Commncatons, vo. 3, no. 4, pp , Mar. 4. [3] V. Rakovc, D. Denkovsk, V. Atanasovsk, P. Mahonen, and L. Gavrovska, Capacty-aware cooperatve spectrm sensng based on nose power estmaton, IEEE Transactons on Commncatons, vo. 63, no. 7, pp , J. 5. [4] D. L, Y. X, X. Wang, and M. Gzan, Coatona game theoretc approach for secondary spectrm access n cooperatve cogntve rado networks, IEEE Transactons on Wreess Commncatons, vo., no. 3, pp , Mar.. [5] H. Km and K. G. Shn, Effcent dscovery of spectrm opportntes wth MAC-ayer sensng n cogntve rado networks, IEEE Transactons on Mobe Comptng, vo. 7, no. 5, pp , May. 8. [6] N. Zhang, H. Lang, and N. Cheng, Dynamc spectrm access n mt-channe cogntve rado networks, IEEE Jorna on Seected Areas n Commncatons, vo. 3, no., pp , Nov. 4. [7] C. Y and J. Ca, Two-stage spectrm sharng wth combnatora acton and Stackeberg game n recabased cogntve rado networks, IEEE Transactons on Commncatons, vo. 6, no., pp , Oct. 4. [8] I. F. Akydz, W. Y. Lee, and K. R. Chowdhry, CRAHNs: cogntve rado ad hoc networks, Ad Hoc Networks, vo. 7, no. 5, pp , J. 9. [9] M. Cesana, F. Como, and E. Ekc, Rotng n cogntve rado networks: Chaenges and sotons, Ad Hoc Networks, vo. 9, no. 3, pp. 8-48, May.. [] S. C. Ln and K. C. Chen, Spectrm-map-empowered opportnstc rotng for cogntve rado ad hoc networks, IEEE Transactons on Vehcar Technoogy, vo. 63, no. 6, pp , J. 4. [] X. Jn, R. Zhang, J. Sn, and Y. Zhang, TIGHT: A geographc rotng protoco for cogntve rado mobe ad hoc networks, IEEE Transactons on Wreess Commncatons, vo. 3, no. 8, pp , Ag. 4. [] P. Y. Chen, S. M. Cheng, W. C. Ao, and P. Y. Chen, Mt-Path rotng wth end-to-end statstca QoS provsonng n nderay cogntve rado networks, n Proc. IEEE INFOCOM, Shangha, Chna, Apr., pp. 7-. [3] A. Bhattacharya, S. C. Ghosh, and B. P. Snha, Mt-Path rotng n cogntve rado networks for mtmeda commncaton sng sampe dvson mtpexng, n Proc. IEEE GLOBECOM, Anahem, USA, Dec., pp [4] L. G, S. Zo, and X. Zhong, Dstrbted best-rote seecton for mtpath rotng n cogntve rado ad hoc networks, Eectroncs Letters, vo. 48, no. 5, pp , Dec.. [5] T. Y. Jang, Y. S. Chen, and C. L. Hs, An effcent mt-path rotng scheme for cogntve Ad Hoc networks, n Proc. 7th IEEE Internatona Conference on Ub-Meda Comptng, Uaanbaatar, Mongoa, J. 4, pp [6] B. Khanna and R. Ram, Interference-aware mtpath rotng n a cogntve rado ad hoc network, n Proc. IEEE MILCOM, Batmore, USA, Nov., pp [7] L. G, X. Zhong, and S. Zo, Traffc assgnment agorthm for mt-path rotng n cogntve rado ad hoc networks, n Proc. IEEE WCNC, Shangha, Chna, Apr. 3, pp [8] L. Akter and B. Natarajan, Dstrbted approach for power and rate aocaton to secondary sers n cogntve rado networks, IEEE Transactons on Vehcar Technoogy, vo. 6, no. 4, pp , May.. [9] Y. W, L. Qan, and L. Meng, Sensng based jont rate and power aocatons for cogntve rado systems, IEEE Wreess Commncatons Letters, vo., no., pp. 3-6, Apr.. [] S. Kcera, S. Assa, K. Yamamoto, and S. Yoshda, Asynchronos dstrbted power and rate contro n ad hoc networks: A game-theoretc approach, IEEE Transactons on Wreess Commncatons, vo. 7, no. 7, pp , J. 8. [] S. Q, H. L, J. L, Y. Ma, and F. Wang, Jont channe and traffc assgnment n ad hoc networks sng game theory wth chargng scheme, n Proc. IEEE PIMRC, Sydney, Astraa, Sept., pp [] L. Ln, A. Wang, X. Zho, and X. Mao, Noncooperatve dfferenta game based effcency-aware traffc assgnment for mtpath rotng n CRAHN, Wreess Persona Commncatons, vo. 6, no., pp , Jan.. [3] L. Ln and J. Tan, Stabe-aware traffc assgnment for mt-path rotng n cogntve rado ad hoc networks, n Proc. IEEE CyberC, Bejng, Chna, Oct., pp [4] Part : Wreess LAN Medm Access Contro (MAC) and Physca Layer (PHY) Specfcatons, IEEE Standard, Jn Jorna of Commncatons 469

9 Jorna of Commncatons Vo., No. 5, May 6 [5] D. W. K. Yeng and L. A. Petrosyan, Cooperatve Stochastc Dfferenta Games, New York: Sprnger, 5. Long Zhang receved the B.S. degree n commncaton engneerng from Chna Unversty of Geoscences, Whan, Chna, n Jne 6 and the Ph.D. degree n commncaton and nformaton systems from Unversty of Scence and Technoogy Bejng (USTB), Bejng, Chna, n Janary. Snce Febrary, he has been wth the Schoo of Informaton and Eectrca Engneerng from Hebe Unversty of Engneerng, Handan, Chna. Hs crrent research nterests ncde opportnstc Internet of Vehces, cogntve rado ad hoc networks, satete and space commncatons, deay and dsrpton toerant networks, resorce aocaton n wreess networks, network modeng and optmzaton, and cross-ayer desgn. Chnhong Ba receved the B.S degree n eectroncs and nformaton engneerng from Hebe Unversty of Engneerng, Handan, Chna n 4. She s crrenty workng toward the M.S. degree n compter technoogy at the Department of Commncaton Engneerng, Hebe Unversty of Engneerng, Handan, Chna. Her crrent research nterests ncde cogntve rado, game theory, and wreess resorce optmzaton. Fan Zho receved the B.S degree n commncatons engneerng from Hebe Unversty of Engneerng, Handan, Chna n 4. He s crrenty workng toward the M.S. degree n compter technoogy at the Department of Commncaton Engneerng, Hebe Unversty of Engneerng, Handan, Chna. Hs crrent research nterests ncde cogntve rado, rotng agorthm, and cross-ayer desgn. Hatao X receved hs B.S. degree n Department of Eectroncs Commncaton & Software Engneer from Sn Yat-Sen Unversty, Chna, n 7, and hs M.S. degree n Facty of Engneerng from Unversty of Brsto n 9. He obtaned the Ph.D. degree n Department of Commncaton Engneerng, Schoo of Compter & Commncaton Engneerng, from Unversty of Scence and Technoogy Bejng. Now he s a ectrer n Unversty of Scence and Technoogy Bejng. Hs research nterests ncde cogntve rado networks, sef-organzed networks, mathematca mode, game theory and next-generaton networks. We Hang receved the B.S. degree n atomaton from the Schoo of Eectronc and Informaton Engneerng, Unversty of Scence and Technoogy Laonng, Laonng, Chna, n Jne 4, and the M.S. degree n contro scence and engneerng from the Schoo of Atomaton and Eectrca Engneerng, Unversty of Scence and Technoogy Bejng (USTB), Bejng, Chna, n Janary. Crrenty, he s an Engneer of Energy Effcency Evaaton Center of Eectrc Power Research Insttte, Bejng, Chna. Hs crrent research nterests ncde energy effcency evaaton, power demand management, and commncaton technoogy. 6 Jorna of Commncatons 47