Medum ccess Control for Mult-Channel Parallel Transmsson n Cogntve Rado Networs Tao Shu, Shuguang Cu, and Marwan Krunz Department of Electrcal and Computer Engneerng Unversty of rzona Tucson, Z 85721 {tshu, cu, runz}@ece.arzona.edu bstract mult-channel parallel transmsson protocol s proposed for the medum access control n cogntve rado networs CRNs). Ths protocol contans two ey elements: mult-channel assgnment and mult-channel contenton. For an ncomng flow-based connecton request, the mnmum number of parallel channels are assgned to satsfy the rate and nterference mas constrants. For the contenton of the assgned channels, our protocol provdes an extenson of the sngle-channel RTS- CTS-DT-CK handshang of the IEEE 802.11 scheme. The proposed MC coherently ntegrates optmzaton results nto a practcal mplementaton. Through numercal examples, we verfy that our protocol provdes lower connecton blocng probablty and hgher system throughput for CRNs than ts sngle-channel counterpart. I. INTRODUCTION Over the past few years, we have wtnessed rapd prolferaton of wreless networs n varous envronments home, offce, publc hotspot, and so on). Ths trend wll ultmately lead to ubqutous networng, where not only computers but ordnary electronc devces wll be connected n a wreless way. The large populaton of connected devces mposes a great demand on spectrum resource, whch s now crowded wth most frequency segments beng statcally and exclusvely allocated to specfc types of wreless devces. On the other hand, a recent report from the FCC Spectrum Polcy Tas Force ndcated that at any gven tme and place, less than 10 percent of the allocated spectrum s beng utlzed [1]. To solve ths dlemma and mprove spectrum utlzaton, cogntve rados CRs), have been proposed to mplement opportunstc and dynamc spectrum access [2]. In ths paper, we assume the avalablty of cogntve rado technologes at the physcal layer, and focus our attenton on desgnng a medum access control MC) protocol for multple one-hop pont-to-pont communcatons n a cogntve rado networ CRN). Partcularly, we focus on extendng the contenton-based carrer sensng multple access wth collson avodance CSM/C) scheme defned n the IEEE 802.11 standard due to ts maturty and wde deployment n practcal networs. basc requrement for CRNs s that the communcaton between CRs should not lead to unacceptable nterference to legacy prmary rados PRs) that share the same channel. The spectrum-sharng nature of CRNs mposes tremendous challenges for MC desgn. Frstly, mult-channel parallel transmssons over a ln are necessary for CRs; to support a prescrbed data rate under a gven nterference mas, a CR needs to use multple parallel channels to reduce ts nterference level by spreadng power over multple bands. The mult-channel assgnment and contenton dffer from the conventonal sngle-channel assgnment and contenton used n current protocols. Secondly, when mult-channel parallel transmssons are adopted, we need to decde the rate and power allocaton across these parallel channels n such a way that certan objectves such as total nterference, rate, or power) are optmzed. Consderng a dedcated control channel for CRNs and N shared data channels between prmary rado networs PRNs) and CRNs, ths paper descrbes and evaluates a novel RTS-CTS-DT-CK based 802.11-le MC protocol, whch s desgned to support mult-channel parallel transmssons n CRNs. We nvestgate the mnmumpower/rate allocaton across parallel channels under constrants on rate and CR-to-PR nterference. Numercal examples are used to verfy the performance of the proposed protocol n terms of connecton blocng probablty and system throughput. The rest of the paper s organzed as follows. The system model and problem statement are gven n Secton II. Interference-optmal power/rate allocaton over parallel channels s presented n Secton III. Our contenton-based protocol s descrbed n Secton IV. Numercal results are presented n Secton V and conclusons are gven n Secton VI. II. MODEL DESCRIPTION ND PROLEM STTEMENT. System Model We consder a hybrd system consstng of N types of legacy PRNs and one CRN. The N + 1 networs overlap wth each other geographcally. PRs n the same PRN operate over the same frequency band, but the dfferent PRNs are allocated dfferent non-overlappng) bands. lthough n realty a PRN may occupy more than one band, such a networ can be equvalently treated as multple vrtual) PRNs worng over dfferent bands. For the th PRN, we denote ts carrer frequency and bandwdth by f and W, respectvely. The motvaton for usng CRN s to enhance the spectrum utlzaton by allowng CRs to share the same spectrum, W 1... W N, wth the PRNs. Here we consder flowbased traffc for ndvdual CRs. When a CR ntates a flow of rate R 0, t chooses m channels, 1 m N. ccordngly, the data flow s converted nto m parallel sub-flows, whch are transmtted smultaneously over the m selected channels wth approprately assgned powers and rates. Wthout loss of
generalty, let the selected set of channels be 1, 2,..., m) and let the transmsson power of CR at channel be denoted as P ), = 1,..., m. To ensure a feasble channel sharng, we have the followng four constrants: 1) Maxmum number of parallel channels: CR can smultaneously use up to M channels from channels 1 to N. Due to the lmtaton of cost, typcally m M N. 2) CR-to-PR nterference spectrum mas: The average transmsson power { of} CR for channel must be constraned by E P ) P ) mas. The vector P mas = P 1) mas,..., P N) mas ) s referred to as the CR-to-PR nterference spectrum mas. Ths mas s needed to ensure that transmssons from a CR wll not cause unacceptable nterference to the co-channel PRs. The determnaton of P mas s certanly an mportant ssue worth nvestgaton, but s out of the scope of ths paper. Here, we smply assume that P mas s gven. 3) Sum-transmsson-power constrant: The total average transmsson power over the m selected channels should { be lmted by a maxmum value P max,.e., m } E P ) P max. Ths constrant s mposed, for example, by the CR s battery. 4) ggregate flow-rate constrant: Let the data rate provded by channel be R.e., the rate of the th subflow). The rate allocaton must satsfy m R = R 0. Here we only consder the case when R 1,..., R m are constants. Ths corresponds to the scenaro where the channel selecton and rate allocaton are decded at the begnnng of the flow, e.g., at the hand-shang phase, and are mantaned throughout the duraton of the flow. To facltate channel sharng, we assume that the th CR node perodcally senses channels and estmates the nstantaneous nterference to the CR at ndvdual channels as a vector I = I 1),..., IN) ). Ths nterference vector s used to drve the channel selecton, rate allocaton, and power control algorthms, as descrbed later. In addton, to coordnate channel access between CRs, we assume that an out-of-band spectrum segment, W N+1, can be assgned to the CRN as a dedcated control channel. CR s capable of ether transmttng or recevng control pacets on ths channel at any tme. It s worth notng that n a pure cogntve rado envronment, a common n-band dedcated control channel may not exst due to the spectrum heterogenety caused by the coexstence of heterogeneous PRNs. The allocaton of an out-of-band control channel greatly smplfes the coordnaton among CRs, but may somehow counter the motvaton of mprovng the spectrum utlzaton through channel sharng. However, because of the short length of control pacets, the bandwdth of the control channel s typcally neglgble compared wth that of the data channels,.e., W N+1 N W. Thus, we argue that the allocaton of a dedcated control channel does not sgnfcantly nfluence the overall spectrum utlzaton of the system.. Problem Statement Wthout loss of generalty, we focus on an arbtrary sourcedestnaton CR par whose flow-rate demand s R 0. The medum access process conssts of two ey elements: channel assgnment and channel contenton. Channel assgnment s executed at the destnaton node, whle channel contenton s conducted n an coordnated way between the source and destnaton CRs. 1) Channel ssgnment: The tas of channel assgnment s to optmally decde the channel selecton and the assocated power/rate allocaton for the selected channels subject to the nterference mas and transmsson power constrants. s mentoned earler, a flow from a source CR s converted nto m parallel sub-flows, whch are transmtted smultaneously over the m selected channels accordng to the assgned rates and powers. To ensure an effcent utlzaton of each channel, the optmal channel assgnment should allocate each flow the mnmum number of channels that satsfy both the rate requrement and nterference constrants. We present the optmal channel assgnment algorthm n Secton III. 2) Channel Contenton: The actual occupyng of the assgned channels s accomplshed by the channel contenton functonalty. In our protocol, the channel contenton mechansm s desgned to ensure non-overlappng local channel occupancy between CRs,.e., a channel whch s occuped by a CR cannot be allocated to other CRs n ts onehop communcaton range. Ths mechansm excludes CR-to- CR nterference although t stll suffers from the co-channel PR-to-CR nterference, thus largely smplfyng the CR-to- PR nterference control process. The ncluson of CR-to-CR nterference by allowng multple CRs to share overlappng channels locally s a more general scenaro for CRNs. However, t demands dstrbuted teratve power adjustment of ndvdual CRs, leadng to potental convergence ssues that can be extremely dffcult to address. Related wor on such a problem has been dscussed n the context of nterference channels, whch s a well-nown open problem [6]. In ths paper, we focus on the smplfed case mentoned above. The detaled contenton protocol s presented n Secton IV. III. OPTIML CHNNEL SSIGNMENT For an ncomng CR access request, we dvde the channel assgnment nto two routnes that are executed teratvely. Frst, for a gven channel assgnment, we compute the optmal power/rate allocaton that leads to the mnmum CR transmsson power under the nterference mas. Second, we change the channel combnaton and repeat the frst procedure. Ths algorthm explores dfferent channel combnatons to see the optmal one that has the mnmum number of channels and requres the mnmum transmsson power whle satsfyng the CR-to-PR nterference mas. Once the optmal channel combnaton s determned, the assocated power/rate allocaton s also determned by routne one.. Optmal Power/Rate llocaton for a Gven Channel Combnaton We consder a one-hop source-destnaton CR par, ) that requres a flow rate of R 0. For an arbtrary channel combnaton, label these channels by 1,..., m. ssumng
WGN and treatng nterference as nose, the maxmum errorfree rate of the th channel s gven by the Shannon capacty formula [7]: R = W ln 1 + P ) g) I ) ), = 1,..., m 1) where P ) s the transmsson power of CR on channel, s the dstance-dependent sgnal-power attenuaton over channel, I ) s the nstantaneous nterference-plus-wgn power at CR on channel a random varable n our model), and R s the rate measured n nats/second. To provde a constant rate of R on channel, the transmsson power must satsfy P ) = I) e r 1) 2) def where r = R /W. ccordngly, the expected transmt power s gven by P ) = er 1) Ī) where Ī) = I ) 0 fi) )di) wth fi) ) beng the p.d.f. of the random varable I ). The optmal power/rate control problem can be formulated as optmzng the rate allocaton across channels 1 to m such that the total average transmt power s mnmzed. Mathematcally, ths optmzaton s expressed as mnmze {r1,...,r m} such that m er 1) Ī) g ) m r W = R 0 0 r ln 1 + P ) mas g) Ī ) ) 3) = 1,..., m. 4) The upper bound on r s due to the CR-to-PR nterference mas, whch s gven as e r 1) Ī) P ) mas. n observaton of 4) ndcates that t s a strctly convex optmzaton problem wth upper and lower bounds on ndvdual varables. The optmal soluton r o 1, r o 2,..., r o m) to 4) can be derved by frst removng the upper bounds and then sequentally determnng the varables that exceed ther upper bounds. Specfcally, ths sequental algorthm s descrbed as follows: 1) We frst solve the optmzaton problem where the upper bound on r s not mposed,.e., mnmze {r1,...,r m} such that m er 1) Ī) g ) m r W = R 0 r 0, = 1,..., m. 5) Obvously, 5) s a convex problem and ts Lagrangan s gven by L r, κ, ɛ) = 6) where κ and ɛ = ɛ 1,..., ɛ m ) are the Lagrange multplers, κ s an arbtrary real number, and ɛ s are nonnegatve real numbers. Wthout loss of generalty, we ran the channels such that g1) W1 Ī 1) g m) W m Ī m) closed form: and g2) W2 Ī 2).... The optmal soluton to 5) can be obtaned n κ = exp r o = max R 0 K1 W ) { ln κg) W Ī ) K 1 Ī) W, 0 } ) W K1 W 7) 8) where K 1, 1 K 1 m, s determned n a way such that fk 1 ) > 1 and fk 1 + 1) 1, wth the functon fn) defned as: fn) = gn) W n Ī n) R 0 n exp n W ) Ī) W ) W n W 9) for 1 n m. The unqueness of such a K 1 can be proved n a smlar way as n [4]. Denote the channel set n the desgn space by V, whch ntally ncludes {1,..., m}. 2) Due to the monotoncally-ncreasng property of the objectve functon over r s [5], f any of the r o n 7) volates the upper bound on r n 4), then the correspondng bounded optmal soluton r o the upper bound tself,.e., r o = ln must ) be. 1 + P ) mas g) Ī ) Denote the set of channels whose unbounded optmal solutons exceed ther upper bounds by U, and we set V = V U. 3) Wth the nowledge of r o for U, the objectve functon n 4) s modfed to the followng form mnmze {r V} V e r 1) Ī). 10) ccordngly, we set R 0 = R 0 U ro W, and the rate constrant n 4) s updated as r j W j. 11) V r W = R 0 j U The modfed optmzaton problem, whch conssts of the objectve functon 10), the rate constrant 11), and r 0, V, s a degenerated verson of 4). The un-bounded optmal soluton to ths new optmzaton problem has the same form as n 7) and 8). 4) Steps 2) and 3) are repeated untl all un-bounded solutons r o of the degenerated problem are wthn ther bounds. The mnmzed sum transmsson power of CR can be e r 1) Ī) +κr 0 r W ) ɛ r, derved by substtutng the optmal rate allocaton r1, o..., rm) o nto the objectve functon of 4). Ths value s compared aganst the transmsson power upper bound P max to decde
whether the gven channel combnaton s feasble. Dfferent channel combnatons wll be tested usng the above algorthm and the optmal channel assgnment s found among those feasble combnatons, as descrbed n the subsequent secton. In addton, to mantan the assgned rates for ndvdual channels through the flow duraton, CR conducts closedloop power control based on the feedbac of the recever-sde nstantaneous nterference level. The optmal power control polcy n channel s derved by substtutng r o nto 2), whch ) leads to the desred average power P.. Optmal Channel ssgnment The optmal channel selecton s a feasble channel combnaton that contans the mnmum number of channels among all feasble combnatons nvolvng wth no more than M channels. In case that more than one feasble combnatons contan the same mnmum number of channels, the one requrng the mnmum sum transmsson power wll be selected. The search process can be effcently mplemented by sequentally explorng the channel combnatons, startng from those contanng only one channel, wth the goal of fndng the mnmum-sze optmal combnaton. If there s no feasble channel combnaton n current round, the algorthm proceeds to those combnatons contanng two channels, and so on. The search contnues untl an optmal combnaton s found, or all combnatons contanng no more than M channels have been tested. Therefore, n the worst case, a total number of M C N channel combnatons need to be tested by the algorthm, where Cx y x! = y!x y)! for ntegers x, y, y x. IV. PROTOCOL DESCRIPTION ased on the optmal channel assgnment algorthm presented n Secton III, we now descrbe the proposed MC protocol for CRNs. Ths protocol s an extenson of the snglechannel RTS-CTS-DT-CK handshang scheme used n the 802.11 standard. For brevty, we only focus on the parallelchannel transmsson aspect n our descrpton. To facltate mult-channel contenton, each CR needs to mantan a local free-channel table FCT). Ths table contans the set of channels that are un-occuped by other CRs wthn the node s one-hop communcaton range. Intally, the FCT contans all N data channels and s contnuously updated accordng to the channel access dynamcs. The proposed protocol s specfed as follows. 1) Whenever a CR s dle,.e., nether transmttng nor recevng data, t lstens to the control channel and updates ts FCT as descrbed n STEP 3. When a CR ntends to establsh a flow-based connecton, t transmts a requestto-send RTS) pacet to the destnaton CR over the control channel usng the largest transmsson power and the physcal carrer sensng scheme. Specfcally: a) ll channels n the FCT of the source CR are ndcated n ths RTS pacet. The duraton of the flow DOF), whch s gven by dvdng the flow length n the unts of bts) by the desred flow rate, s also gven n the RTS pacet. Ths nformaton provdes other CRs wth an estmate of the endng tme of the underlyng flow. b) If the FCT at the source CR s empty, the source CR defers ts RTS transmsson, bacs off, and retransmts the RTS pacet at the end of bacoff. Durng the bacoff perod, the source CR contnues to lsten to the control channel and eeps updatng ts FCT as descrbed later n STEP 3. c) Other nodes recevng the RTS on the control channel defer ther control-pacet transmsson untl the appearance of the ECTS pacet descrbed later n STEP 3) or the tmeout of a predefned perod. The tmeout value should be set reasonably larger than the typcal nterval between the RTS and the ECTS to avod nterruptng normal hand-shang. 2) Upon successful recepton of the RTS pacet but before transmttng the clear-to-send CTS) pacet, the destnaton CR conducts the logc and operaton between ts FCT and the source CR s FCT. Those channels that appear n both FCTs are tagged as avalable channels for the data communcaton. ased on ths set of avalable channels, the followng actons are taen: a) The channel assgnment algorthm descrbed n Secton III s executed to test varous combnatons of the avalable channels. b) If a partcular combnaton of m channels s selected, the denttes of these channels and the assocated power/rate allocaton nformaton are ndcated n the CTS pacet. In addton, the destnaton CR also copes the DOF nformaton from the RTS pacet to the CTS pacet. If the set of avalable channels s empty or f no feasble channel combnaton s found by the channel assgnment algorthm, an empty flag wll be ndcated n the CTS. 3) When the source CR receves the CTS pacet, t transmts an Echo-CTS ECTS) pacet over the control channel, ncludng n ths pacet the channel assgnment and DOF nformaton provded n the CTS. fter that, the source and destnaton CRs begn the parallel dataflow communcaton on the assgned channels usng the assgned powers and rates. ll other CRs that overhear the CTS and ECTS pacets wll remove the assgned channels from ther local FCTs for a DOF amount of tme. When the DOF-equvalent tme expres, these channels wll be appended bac nto ther local FCTs. If an empty flag s ndcated n the CTS and ECTS, the source CR enters bacoff and retres afterwards. 4) Durng the flow transmsson, both the source and the destnaton CRs eep lstenng to the control channel for CTS and ECTS pacets from other CRs, and updatng ther FCTs accordngly as descrbed n STEP 3. In addton, to mantan the assgned rates for ndvdual subflows through the flow duraton, the source CR conducts closed-loop power control based on the feedbac of the recever-sde nstantaneous nterference level. The rule for ths power control has been gven by 2).
V. PERFORMNCE EVLUTION To evaluate the effectveness of the proposed MC protocol, we conducted numercal experments usng MTL and also smulated a hybrd system that conssts of 2 PRNs and 1 CRN. Nodes n these networs are unformly dstrbuted over a 100- meter-radus crcular area. The frst PRN operates n the 900 MHz band, occupyng fve non-overlappng 1-MHz channels that are labelled as channels 1 to 5 n the smulaton. The numbers of PRs n each channel are 100, 200, 300, 400, and 500, respectvely. The second PRN operates n the 2.4 GHz band, also occupyng fve non-overlappng 1-MHz channels that are labelled as channels 6 to 10 n the smulaton. The numbers of PRs n each channel of PRN 2 are 100, 200, 300, 400, and 500, respectvely. The sgnal strength s attenuated by d 4 wth d the propagaton dstance [7]. We dvde the tme nto slots, each of length 10 ms. t any gven slot, each PR n the frst and the second PRNs attempts to transmt wth a probablty of 0.1 and 0.4, respectvely. The transmsson power of each PR s 1 W when t s on. We smulated 10 pars of one-hop source-destnaton CRs. To smplfy our smulaton, we assume that the dstance between each par s equal, and we tae the path loss to be - 30 d. We assume that all CRs are wthn the transmsson range of each other, so that any control pacet sent from a CR can be heard by all other CRs. The nstantaneous nterference sensed by a CR n a certan channel s the sum of the nterference from all actve co-channel PRs. The flow generaton at each source CR follows a Posson process wth parameter λ flows/second. Each flow has an exponentally dstrbuted duraton wth mean 1/µ second. The flow from the th source CR requres a constant data rate of 0.5 MegNats/second. We assume that a CR can use up to two data channels smultaneously. We set the CR-to-PR nterference spectrum mas to P 1) mas =... = P 10) mas = 20 mw, and the transmsson-power upper bound to P max = 20 mw. We compare the performance of the mult-channel parallel transmsson MC protocol wth the mult-channel RCS MC protocol proposed n [3]. In contrast to our proposed mult-channel parallel transmsson strategy, the mult-channel RCS protocol only selects the best avalable channel for data transmsson.e., a node uses only one channel at a tme). lthough t s not orgnally desgned for a CRN, we adapt t to the CRN applcaton by modfyng the channel selecton condton: f the average transmsson power assocated wth the best avalable channel satsfes the CR-to-PR nterference mas, then t wll be selected; otherwse no channel wll be assgned and the ncomng access request wll bac off. The performance crtera to compare nclude the connecton blocng rate plotted n Fgure 2) and the system throughput plotted n Fgure 3). The connecton blocng rate s defned as the rato between the number of requests that end n bacoff to the total number of connecton requests. The system throughput s defned as the average volume of CR traffc transmtted by the system n one second. The smulaton results verfy that sgnfcant reducton n the connecton blocng probablty and consderable ncrease n throughput are acheved by the proposed MC protocol. Connecton blocng probablty percentage) Networ throughput Mnats/second) 40 35 30 25 20 15 10 5 mult-channel RCS mult-channel parallel 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Fg. 1. 14 13 12 11 10 9 8 7 6 5 4 3 2 1 traffc rate λ/µ) Connecton blocng rate vs. traffc load. 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Fg. 2. mult-channel RCS mult-channel parallel traffc rate λ/µ) System throughput vs. traffc load. VI. CONCLUSIONS mult-channel parallel transmsson protocol was proposed for medum access control n cogntve rado networs. Ths protocol contans two ey elements: mult-channel assgnment and mult-channel contenton. The proposed MC coherently ntegrates the optmzaton results n mult-channel assgnment nto a practcal mplementaton of the multchannel contenton. Compared wth a reference protocol, the proposed one provdes better spectrum utlzaton n terms of smaller connecton blocng probablty and larger system throughput. REFERENCES [1] Federal Communcatons Commsson, Spectrum Polcy Tas Force, Rep. ET Docet no. 02-135, Nov. 2002. [2] S. Hayn, Cogntve rado: ran-empowered wreless communcatons, IEEE Journal on Selected reas n Communcatons, vol. 23, no. 2, pp. 201-220, Feb. 2005. [3] N. Jan, S. R. Das, and. Naspur, multchannel CSM MC protocol wth recever-based channel selecton for multhop wreless networs, Proc. IEEE ICCCN, 2001. [4] J. Xao, S. Cu, Z. Q. Luo, and. J. Goldsmth, Power schedulng of unversal decentralzed estmaton n sensor networs, IEEE Transactons on Sgnal Processng, vol. 54, no. 2, pp. 413-422, Feb. 2006. [5] S. Cu, J. Xao,. J. Goldsmth, Z. Q. Luo, and H. V. Poor, Estmaton dversty wth multple heterogeneous sensors, to appear at Proc. IEEE ICC, June 2006. [6] W. Yu, W. Rhee, S. oyd, and J. Coff, Iteratve water-fllng for Gaussan vector multple access channels, IEEE Transactons on Informaton Theory, vol. 50, no. 1, pp. 145-151, Jan. 2004. [7] T. S. Rappaport, Wreless Communcaton Prncples and Practce, Prentce Hall, 1996.