A Novel DSA-Driven MAC Protocol for Cognitive Radio Networks

Transcription

1 Wreless Sensor Networ, 29, 2, 6-2 do:.4236/wsn Publshed Onlne July 29 ( A Novel DSA-Drven MAC Protocol for Cogntve Rado Networs Hua SONG, Xaola LIN School of Informaton Scence and Technology, Sun Yat-sen Unversty, Guangzhou, Chna Emal: songhua@mal2.sysu.edu.cn, lnxl@mal.sysu.edu.cn Receved February 9, 29; revsed Aprl 29, 29; accepted May 6, 29 Abstract Wth the deployment of more wreless applcatons, spectrum scarcty becomes an ssue n many countres. Recent reports show that the reason for ths spectrum shortage s the underutlzaton of some spectrum resources. Fortunately, the emergence of open spectrum and dynamc spectrum assess (DSA) technology n cogntve rado networs releves ths problem. In ths paper, we propose a novel DSA-drven cogntve MAC protocol to acheve hghly effcent spectrum usage and QoS provsonng. In the proposed protocol, secondary users are dvded nto several non-overlappng groups, and all leftover channels are allocated among groups tang the groups bandwdth requrements nto consderaton. Moreover, the allocaton of vacant channels can be adjusted dynamcally when members jon/leave groups or prmary users return/leave the current networ. Smulatons show that the proposed MAC protocol greatly mproves the qualty of servce for secondary users and maxmzes the utlzaton rato of spectrum resources. Keywords: Cogntve Rado, DSA-Drven MAC Protocol, QoS Provsonng, Dynamc Spectrum Access. Introducton The deployment of wreless servces and devces has been ncreasng rapdly n recent years, but current usable spectrum has almost been allocated to varous spectrum-based servces, whch greatly bloes the development of wreless communcaton. However, extensve reports ndcate that the reason for ths spectrum shortage s not the scarcty of the rado spectrum, but the low utlzaton (only 6%) of the lcensed rado spectrum n most of the tme []. Ths underutlzaton of spectrum resources has prompted the emergence of cogntve rado. In 23, Federal Communcatons Commsson (FCC) suggested a new concept/polcy for dynamcally allocatng the spectrum [2]. Thus, a promsng mplementaton technque called cogntve rado s proposed to allevate the scarcty of spectrum bandwdth. Based on cogntve rado, open spectrum and dynamc spectrum access (DSA) technolo- * Ths wor was supported n part by NSFC under Projects , U735, and 985 II fund under Project 373. ges have shown great nterest recently [3]. In ths technology, prmary users (lcensed users) have hgh prorty to use ther spectrum; secondary users (unlcensed users) are allowed to opportunstcally access the spectrum only when the spectrum s not used by prmary users. Although the research communty has proposed several cogntve MAC protocols to address varous ssues n cogntve networ [4-8], commonly, they pay more attenton to save the number of transcevers, and mprove throughput of the whole system or decrease sesson delays. However, all these protocols do not lay emphasze on qualty of servce for secondary users and hgh usage of leftover spectrum wth dynamcally adjustng allocaton. For nstance, n [5], a decentralzed protocol, called hardware-constraned cogntve MAC protocol (HC- MAC), for managng and coordnatng spectrum access s proposed. Under HC-MAC, a par of secondary users can use several channels to communcate smultaneously after they have sensed the vacant channels, but f the leftover spectrum allocated to the par of secondary users s more than they can utlze, ths part of surplus spectrum s wasted and cannot be used by other users whose

2 H. SONG ET AL. 3 bandwdth requrements have not been satsfed. To effectvely provde QoS for secondary users and acheve hghly effcent spectrum usage, n ths paper, we propose a novel DSA-drven MAC protocol whch s of sgnfcant mportance n ad hoc cogntve networ to guarantee the QoS requrements of secondary users. Dfferent from the exstng cogntve MAC protocols, the man advantages of the proposed MAC protocol nclude: ) Maxmzng the utlzaton rato of spectrum resources; 2) Usng bondng/aggregaton and dynamcal channel allocaton technques to guarantee the QoS requrements of secondary users; 3) Ensurng the farness of channel allocaton for groups. The rest of the paper s organzed as follows. Related wor s dscussed n Secton 2. The prelmnares and system model s ntroduced n Secton 3. The proposed MAC protocol s presented n Secton 4. Performance of the proposed MAC protocol s evaluated by experments n Secton 5. Fnally, concluson s drawn n Secton Related Wor Over the past several years there have been ncreasng nterests n cogntve rado. In addton, wreless MAC protocol has a prncpal part n spectrum reuse and effcency management. Therefore, varous cogntve MAC protocols have been proposed for more flexble and effcent use of spectrum resources [4-8]. Hamdaou and Shn [4] propose the OS-MAC protocol. Ths protocol dvdes the secondary users nto several groups, at each Opportunstc Spectrum Perod, the channel used by a group can be adjusted dynamcally accordng to the channels' states of the whole system. The OS-MAC protocol also dscusses the non-cooperatve mode between prmary users and secondary users, but t does not gve a feasble soluton (we wll present our soluton on ths problem n subsequent paper). Furthermore, as only one channel can be used n a group at anytme, the spectrum resources cannot be used to the maxmum wth tang secondary users QoS nto consderaton. Ja et al. [5] present the HC-MAC protocol. Ths protocol uses -stage loo-ahead method to sense unused channels wth hgh effcency and taes hardware-constrants nto consderaton (ncludng sensng constrant and transmsson constrant). Besdes the flaw mentoned n Secton, there s a problem called sensng exposed termnal problem n ths protocol. That s, f a secondary par senses unused channels whle ther neghbors who ddn't receve the C-RTS/C-CTS pacets are performng ther operatons freely, ths par of users can not sense the vacant channels accurately. Su and Zhang [6] propose the cross-layer based opportunstc MAC protocols. In the protocols two sensng polces, the random sensng polcy and the negotaton-based sensng polcy, are presented. Le [5], the protocols also use bondng/aggregaton technque to transmt data through several channels. In essence, the man contrbuton of ther wor s to reveal the tradeoff between throughput and delay, whch provdes the gudelnes to support the dfferent QoS requrements over cogntve rado based wreless networs. But, at anytme, only a par of secondary users can use vacant channels, thus the leftover channels cannot be used effcently. Thoppan et al. [7] propose a CSMA-Based MAC protocol. In the protocol, each node mantans a lst of favorable channels for each of ts neghbors based on the prevous hstory of communcaton on each of the channels, and a secondary par chooses the most favorable channel for communcaton. As t does not consder the channels utlty of the whole system, t also cannot use the rado spectrum resources effcently. Smlarly, the methods n [9-2] all do not address the ssues of the QoS of secondary users and the spectrum resources utlty of the entre system. 3. Prelmnares and System Model We frst present an ntroducton to the channel bondng/aggregaton technque and the man framewor of the system. 3.. Channel Boundng/Aggregaton Technque From the defnton of channel bondng/aggregaton technque n [3] we can see that channel bondng s used for contguous channels, but channel aggregaton s for dscrete ones. So, f several contguous channels can be used, channel bondng s the approprate technque. Otherwse, we can adopt channel aggregaton technque. In MAC perspectve, channel bondng ncurs no addtonal overhead as all control messages are transmtted only once, and an access pont (AP) wth channel bondng also has much greater control and more freedom on resource allocaton and transmt power. In contrast, wth channel aggregaton, the overhead ncreases consderably wth the number of channels used, and for an effectve channel aggregaton soluton, features such as sophstcated schedulng, load balancng and channel management are needed. Fgure (a) and Fgure (b) compare the aggregate throughput and effcency of the two technques respectvely. From Fgure, we can see that channel aggregaton ncurs much more overhead than channel bondng and these two technques are desgned for medum-hgh loads. Also, ths fgure presents that channel bondng can offer much better channel utlzaton wth less overhead. In [5], hardware constrans of bondng/aggregaton are ponted out: spectrum used by a secondary user has max-

3 4 H. SONG ET AL. Aggregate throughput (Mbps) aggregated channels 3 aggregated channels 2 bonded channels 3 bonded channels Overall load (Mbps) (a) Aggregate throughput aganst overload. Effcency aggregated channels 3 aggregated channels.2 2 bonded channels 3 bonded channels Overall load (Mbps) (b) Effcency aganst overload. Fgure. Channel bondng vs. channel aggregaton. mum bandwdth lmt and maxmum fragmentaton number lmt. For convenence, we wll not tae these constrans nto consderaton presently, but they are ncluded n our future research plan Secondary Users Devces The whole avalable rado spectrum n a CR networ s dvded nto a number of non-overlappng data channels (DCs) and a common control channel (CC), for transmttng data nformaton and control nformaton respectvely. Further, we assume the CR networ s a wreless ad hoc networ formed by a great many secondary users, each of whch s equpped wth two wreless transcevers. One s called data transcever used to sense leftover DCs and exchange data through these channels, but ths transcever s not able to operate n sensng and transmttng mode concurrently. The other transcever s called control transcever used to send or receve messages from CC. Note that, the transcevers are half-duplex, thus they are not able to send and receve messages smultaneously. We also assume that every secondary user can use aggregated spectrum and has full spectrum sensng ablty Control Mechansm Based on dstrbuted envronment, we dvde secondary users nto several non-overlappng groups, and each gro- up has a leader (the leader can be adjusted dynamcally) who s responsble for group members' management, group channel's management and group channels' applcaton. Moreover, there s a manager n the whole wreless ad hoc networ who s elected among leaders and s used to manage all the leftover channels and allocate these channels to groups farly. The manager communcates wth leaders through CC, and the control messages n a group are exchanged through DCs. So the common control channel (CC) s lght loaded and wll not be the bottlenec of the networ. As the manager and leaders can be changed dynamcally, each of them s mpossble to become a sngle falure pont. Thus, the system s provded good scalablty and extensblty. 4. Our Proposed DSA-Drven MAC Protocol In ths secton, we present the desgn of our proposed DSA-drven MAC protocol for cogntve rado networs. 4.. Defntons and Notatons In order to present the proposed MAC protocol more clearly, we would now le to ntroduce some defntons and notatons here. ) Tables: These tables are used for the management of the whole networ. GroupTable: Ths table s mantaned by the manager. It contans the group number, the leader and members n the group. MemberTable: Ths table s mantaned by the leader. It contans the lst of group members. Note that each member has a MemberTable. 2) Control frames: These frames are used for the control of protocol's realzaton. The frst fve frames are sent through CC, the LeaveReq frame s sent through DC and the last one can be sent through CC or DC. Invte: It contans the manager's d. JonReq: It contans the d of a secondary user whom the request user wants to communcate wth. JonACK: It contans the leader's d of chosen group. Sense: It contans the group number and the sensng range of channels. Allocate: It contans the group number and a lst of channels. LeaveReq: A member uses ths frame to apply for leavng the current group. Notfy: It contans ManagerChanged and Leader- Changed felds. 3) Tmers: These tmers are used for the mantenance of the whole system. ManagerHeartBeatTmer: Ths tmer s started at the manager when t broadcasts the Invte frame on CC. In essence, ManagerHeartBeatTmer has two functons, the frst one s to let secondary users

4 H. SONG ET AL. 5 now who s the manager currently and the other one s to mae leaders assure the manager s stll alve. ManagerFalureTmer: Ths tmer s started at a leader when t receves an Invte frame on CC. If ths tmer expres before the leader receves another Invte frame, t nows that the manager s faled. LeaderHeartBeatTmer: Ths tmer s started at a leader when t broadcasts the LeaderHeartBeat frame on DC n a group. As soon as members receve the LeaderHeartBeat frames they now that the leader s alve. LeaderFalureTmer: Ths tmer s started at a member when t receves a LeaderHeartBeat frame from DC. If ths tmer expres before the member receves another LeaderHeartBeat frame, t nows that the leader s faled Overvew There are two nds of control frames n our proposed MAC protocol: nter-group control frame and ntra-group control frame. Inter-group control frames are transmtted between the leaders and manager through CC, whle ntra-group control frames are exchanged between the leader and members through DCs n a group. If several leaders want to send messages to the manager, they must negotate wth each other va the contenton-based algorthms, such as IEEE 82. Dstrbuted Coordnaton Functon (DCF) [4] and p-persstent Carrer Sense Multple Access (CSMA) protocols [5]. In order to guarantee relable transmsson, we use acnowledge mechansm n control messages exchange. Fgure 2 shows the prncple of our proposed MAC protocol. It dvdes tme nto Perods and each of whch conssts of three consecutve phases: Sensng Phase, Allocatng Phase and Transmttng Phase. At any tme, only one par of members n a group can exchange nformaton. In order to let all members n the same group use ch- Data Channels unused by prmary users Common Control Channel DC DCn CC Sensng Phase Secondary users start sensng vacant channels Manager coordnates all groups sensng vacant channels Perod Allocatng Phase Transmttng Phase 2 m Each leader dvdes ths phase Each group nto several slots accordng to swtches to the number of users n the the allocated group, and only a par of users channels can communcate n a slot Manager allocates channels for each group Managng group members jonng and leavng Reelectng the leaders and manager f needed Fgure 2. The prncple of our proposed MAC protocol. t annels farly, the Transmttng Phase s dvded nto several tme slots accordng to the number of users. Furthermore, a few of tme slots are reserved for ntra-group control frames exchange. In Sensng Phase, manager coordnates all groups sensng vacant channels (t's an effcent method to chec all unused channels n a short tme). Then, based on the feedbac of sensng results and bandwdth requrements from all groups, the manager calculates out a best allocaton scheme and allocates the avalable channels to each group n Allocatng Phase. When groups gan ther new allocated channels, they swtch to these channels mmedately and begn to transmt nformaton n Transmttng Phase Detals of the Proposed MAC Protocol In the proposed MAC protocol, each secondary user n the networ wll be n one of the followng phases at any gven tme. ) Intalzaton Phase: If a secondary user s not nvolved n any group and sees channels to transmt nformaton, t wll lsten to CC. a) If the secondary user receves an Invte frame from CC, t sends a JonReq control frame to apply for jonng a group. After recevng the JonReq frame, the manager loos up ts GroupTable to decde whch group ths user should jon n accordng to the user's request. If the manager fnds an approprate group, t updates ts GroupTable wth a JonACK control frame to send bac. When the secondary user receves the JonACK frame, t communcates wth the leader of the chosen group through CC and tunes ts data transcever to the accordng DCs. However, the manager may not fnd a sutable group for the secondary user to jon n, then t checs whether t's possble to establsh a new group for ths user. If so, the manager creates a new group and maes ths secondary user a leader; otherwse a REJ control frame s responded. b) If the ManagerFalureTmer expres before the secondary user receves an Invte frame, the user nows that t s the only secondary user n the networ. Then t establshes a new group, maes tself a leader and manager, creates a MemberTable and a GroupTable, and broadcasts an Invte frame on CC. 2) Sensng Phase: all secondary users cooperate to sense the channels unused by prmary users. a) The manager allocates sensng channels among all groups accordng to GroupTable, and sends a Sense frame to each leader va CC. b) Each group senses vacant channels whose range s ndcated by the Sense frame. c) Leaders report sensng results to the manager through CC.

5 6 H. SONG ET AL. 3) Allocatng Phase: When the manager gets the vacant channel nformaton from all groups, t allocates these channels among groups. a) Each leader sends bandwdth requrement of ts group to manager va CC. b) The manager determnes a channel allocaton scheme accordng to the bandwdth requrement of each group, and sends ths scheme to leaders. c) Each group swtches to the allocated channels. 4) Transmttng Phase: As soon as groups swtch to the allocated channels, they can use bondng/aggregaton technque to transmt data. Analyze the type of the frames Receve Sense Reply frames from leaders Compute vacant channels for each group Send Allocate frame to leaders LeaveReq Modfy the GroupTable Send ACK frame Member Manager allocate vacant channels Allocate Tell members to swtch to the allocated channels Leave the group Send Invte frame Perform operaton The coordnator s manager? Yes Assgn a new leader and a new manager Send GroupTable to the new manager Send Notfy frame to members and leaders Where are the frames from? Manager Analyze the type of the frames Sense Coordnate all members to sense vacant channels Report sensng results and bandwdth requrement to the manager (a) Create a new group and mae the new comer a leader (b) Receve frames No Manager organze sensng process Assgn a new leader of the group Send Notfy frame to members and the manager New comer JonGrp Approved by the manager? Yes Send group channels nformaton Yes Analyze the type of the frames No Possble to establsh a new group? Compute sensng channes for each group Send Sense frame to leaders No JonReq Send REJ frame No No The coordnator s manager? Yes Loo for a group for the new comer Fnd an approprate group? Yes Update GroupTable and send JonACK frame Fgure 3. The operaton flow chart of a coordnator. a) As leaders dvde the whole Transmttng Phase nto several tme slots, n each group, the members use channels n turn. b) At the end of Transmttng Phase, the manager and leaders exchange group nformaton through CC n order to update GroupTable and MemberTable respectvely. Also, n each group the leader broadcasts GroupTable to all members. c) If a member wants to leave a group, the followng operatons wll be done: If the member s not a leader or manager, t sends a LeaveReq frame to the leader through DC, then the leader modfes the GroupTable wth an ACK frame s repled. If the member s a leader, t assgns a new leader from rest members of ths group and broadcasts a Notfy frame on DCs n the group; t also sends ths frame to the manager va CC. In case ths leader s the last one n the group, t sets leader's d NULL n the LeaderChanged feld of Notfy frame. If the member s a manager, frst, t selects a new leader from rest members n ts group and a new manager from all leaders; second, t sends GroupTable to the new manager; last, t broadcasts a Notfy frame on DCs n the group, as well as posts ths frame to leaders va CC. d) If a secondary user wants to jon a group, t wll do the operaton as shown n Intalzaton Phase. In the followng, we call the manager or leader coordnator. Fgure 3 delneate the operaton flow of a coordnator accordng to the protocol's detals mentoned above. For concseness, we only gve the foremost control operatons Channel Allocaton Mechansm We suppose a rado spectrum system consstng of M groups whch apply for n vacant channels. For each group, the allocaton optmzaton problem can be descrbed as: max st.. c c b j j n where c (,, M) c n M b mn( r,( r B / r)) j j n th ( B / n/ M n) () s the number of leftover channels allocated to the group, b ( j,, c ) denotes the bandwdth of the j th channel whch s allocated to the th group (In ths paper, we assume that all channel s bandwdths are not the same and they are unformly j

6 H. SONG ET AL. 7 dstrbuted), r (,, M) s the bandwdth requrement of the th group, and B (,, n) s the bandwdth of the th vacant channels. The last constrant ndcates that the bandwdth allocated to each group cannot exceed a certan upper bound whch s defned as n M n mn( r,( r B. The / r)) ( / / B n M n) augend of the upper bound represents each group s deserved bandwdth accordng to ts requrement and the addend s an adjustng factor whch guarantees the channel allocaton s accurate and even. The object of problem () s to maxmze the utlty of vacant channels. For ths nd of problems, branch and bound algorthm [6, 7] s an effectve method to search the optmal solutons. In ths paper, we propose a branch and bound algorthm, whch s more sutable for the channel allocaton optmzaton problem n cogntve rado networs than the general one. For each group, the detal steps are as follows: ) Intalzaton: The feasble soluton can be presented as P { x,, x t }, Where t ( t n) s the number of avalable vacant channels for the current group, and x {,} (,, t). If the value of x s, t means that the th channel s allocated to the current group. We defne P {,,, }. In ths algorthm, the channels are sorted by B (,, t) whch s the band- th wdth of the channel, that s Bj B, j, j, I,; because the channel wth hgher bandwdth has larger mpact to the optmal soluton. Let S { P} to ndcate the ntal feasble set and ntalze. The ntal lower and upper bound of the optmal soluton can be represented as t B (3) 2) Branchng: Based on the characterstcs of the channel allocaton optmzaton problem, the branchng method puts each P S n duplcate to set S and set the value of x. () P { P x, x P} (2) P { P x, x P} 3) Boundng: The lower bound of the optmal soluton n a set can acheve by a greedy method. In ths method, the channel wth hgh bandwdth has the prorty to be chosen n order to acheve the optmal capacty. Thus, the lower and upper bound of the optmal value n P s (2) (4) (5) ( P) xb (6) ( P) x B B t (7) The lower and upper bounds of the optmal soluton n the entre feasble regon D are max ( P) (8) P S max ( P) (9) P S 4) Prunng: For P S, t should be pruned f one of the followng condtons s satsfed: a) The constrants are volated. b) The upper bound of ths set s smaller than the maxmum lower bound, whch s ( P). In ths step, many subsets are pruned to accelerate the searchng speed. 5) Convergence: In ths step, examne whether t s the t th teraton. a) If t, and go to the branchng step. b) If t, the optmal solutons of all the unpruned subsets are confrmed. In the convergence step of the t th teraton, end the teraton and enumerate the optmal solutons of all unpruned subsets n P t to search the global optmal soluton. 6) Return: After gettng the optmal soluton for current group, the algorthm returns to calculate for the next group. Thus, the channels allocated to the current group wll be excluded n the next round, and the number of avalable channels t wll also be modfed Electon Mechansm If the coordnator s crashed, members must elect a new one. In our proposed MAC protocol, we use a smple and hgh effcent algorthm to realze the electon mechansm. When any member notces the coordnator s not functonng, t sends an Electon frame on CC/DCs (manager electon frame on CC and leader electon frame on DCs) to apply for becomng the new coordnator. If several members detect the coordnator's malfuncton concurrently, they have to compete for the CC/DCs through the IEEE 82. random access scheme. Recall that for a successful transmsson, an ACK frame wll be sent bac to the sender. Under our proposed MAC protocol, as all members are lstenng to CC/DCs at all tmes, each member s able to receve ths Electon frame, and f anyone reples, the ACK frame wll be heard by all members. The member who s the frst to successfully

7 8 H. SONG ET AL. delver an Electon frame s automatcally apponted as the new coordnator. Therefore, upon recevng an ACK frame notfyng a successful recepton, ths member consders tself the new leader. Also, any other members who hear the ACK frame, and hence would now that someone else s apponted to be the new leader and need not send ts own acnowledgement. In Fgure 4, an example of electon algorthm s gven. The group conssts of eght processes, numbered from to 7. Member 7 s the coordnator whch has just crashed. Member 4 s the frst one to notce ths, so t sends an Electon frame on CC/DCs. After hearng the Electon frame from Member 4, Member 6 affrms Member 7 s malfuncton and reples wth an ACK frame. Upon recevng ths ACK frame, Member 4 nows that t has been permtted to be the new coordnator. Furthermore, because the rest members n the same group all can hear the Electon and ACK frames, they are also aware of Member 7 s crash, as well as the new coordnator's generaton. Note that the electon mechansm s very effcent, t explots the already exstng ACK mechansm and does not requre any extra message exchange, thus ncurrng no bandwdth overhead. 5. Smulaton and Performance Evaluaton In ths secton, we present the smulaton results for the performance evaluaton of the protocol. The smulatons are performed usng the networ smulator ns-2 [8]. The used parameters are presented n Table. In the smulaton, the proposed MAC protocol s compared wth the followng exstng MACs: OS-MAC and CO-MAC. ) OS-MAC: OS-MAC [4] s an opportunstc spectrum MAC. It dvdes tme nto perods each of whch s called OSP (Opportunstc Spectrum Perod) and conssts of three consecutve phases: Select, Delegate, and Update. Select Phase: Each SUG (Secondary User Group) selects a best DC, and uses t for communcaton untl the end of the current OSP. Delegate Phase: On each DC, a DSU s apponted among members to represent the group durng the Update Phase. 2 Electon (a) 2 ACK Fgure 4. Electon algorthm. (b) Table. Parameters used n the smulatons. Parameter Number of secondary user group 5, 8 Number of DCs 3 Bandwdth of each DC (not the same) Requred bandwdth of each group Value ~.5Mbps 4.5Mbps Transmt power.w Transport protocol UDP Update Phase: All DSUs swtch to CC to update each other about ther channel condtons whle all non-dsus contnue communcatng on ther DCs. 2) CO-MAC: Le OS-MAC, CO-MAC [6], an opportunstc MAC protocol, uses two transcevers, a dedcated CC, and N DCs. The tme s dvded nto a number of perodcal tme slots and each slot s dvded nto two phases, namely, Reportng Phase and Negotatng Phase. Reportng phase can be further dvded nto n mn-slots, each of them correspondng to one of the n lcensed channels. Reportng Phase: Each secondary user s equpped wth one SDR transcever, and by usng ths transcever only one of n lcensed channels can be sensed. Thus the secondary user s unable to accurately now the states of all the channels by tself. However, the goal of the Reportng Phase s to empower the secondary users to have a large pcture of all the channels states through ther cooperaton. In partcular, each secondary user senses a channel n correspondng mn-slot, f the channel s dle, the user sends a beacon durng ths mn-slot over the control channel. Otherwse, no beacon s posted. Negotatng Phase: the secondary users use the control transcevers to negotate about the data channels among the secondary users by exchangng request-to-send (RTS) and clear-to-send (CTS) pacets over the control channel. Meanwhle, the only secondary user whch s the wnner n contendng for the data channels durng the last tme slot uses the SDR transcever to transmt data pacets over all the unused lcensed channels n the current tme slot. For concseness, we defne the unused channels percentage n the entre system as pn n/ N, where N s the number of channels lcensed to prmary users. 5.. Throughput Analyss In ths secton, we analyze the throughput wth dfferent numbers of unused channels under three protocols: the proposed MAC, OS-MAC and CO-MAC. The throughput allows us to evaluate the protocols performance, that s, the hgher throughput, the hgher performance.

8 H. SONG ET AL. 9 Fgure 5 shows the throughput comparson for our proposed MAC protocol wth the other two MAC protocols under each of the two networ scenaros: (Fgure 5(a)) the number of secondary user groups wth M=5 and (Fgure 5(b)) the number of secondary user groups wth M=8. Frst note that wth the ncrease of unused channels' percentage ( p n ), the system throughput of these three protocols all ncreases. Also, observe that the throughput of the proposed MAC protocol ncreases sharply whle CO-MAC attans a bound rapdly and OS-MAC ncreases lttle after reachng a fxed value. Ths demonstrates that our proposed protocol can suffcently use the vacant channels wth the ncrement of p n. In OS-MAC, as only one channel can be used by a group and all channel s bandwdths are unformly dstrbuted between Mbps and.5mbps, when each group monopolzes a channel (.e., ths group does not share the channel wth other groups), the overall throughput won't ncrease much wth the ncrement of p n but wth the number of secondary user groups (M). For example, n Fgure 5 (a), the system throughput of OS-MAC n pn 4% s a ltter hgher than whch n p n 3%, because wth the ncreasng of p n, although each group cannot gan another channel, t has more opportunty to swtch to a Transmsson Throughput (Mbps) proposed MAC OS-MAC CO-MAC (a) The number of secondary user groups M = 5. Transmsson Throughput (Mbps) Proposed MAC OS-MAC CO-MAC (b) The number of secondary user groups M = 8. Fgure 5. The throughput wth dfferent number of unused channels. 9 8 Proposed MAC 7 OS-MAC CO-MAC (a) The number of secondary user groups M = 5. Unused Channels Usage (%) Unused Channels Usage (%) 9 8 Proposed osed MAC 7 OS-MAC CO-MAC (b) The number of secondary user groups M = 8. Fgure 6. The unused channels usage wth dfferent number of unused channels. best one, Moreover, n Fgure 5(b), the throughput of OS-MAC s much hgher than the counterpart n Fgure 5(a), because more vacant channels are utlzed by groups as M s ncreasng. In CO-MAC, snce only a par of secondary users uses multple vacant channels at anytme and each par's requred bandwdth s the same n our best one, Moreover, n Fgure 5(b), the throughput of OS-MAC s much hgher than the counterpart n Fgure 5(a), because more vacant channels are utlzed by groups experment, as soon as the requred bandwdth s satsfed, the entre throughput wll not change. Another pont that requres attenton s that, n Fgure 5(a), the curve represented for the system throughput of our proposed protocol s closed to a fxed value when pn 8%. Ths s because of the throughput of the entre system mantans a certan value and more vacant channels are not needed when all groups bandwdth requrements are satsfed Rato of Unused Channels Utlzaton Analyss Fgure 6 depcts the varaton of unused channels utlzaton rate ( ) of the three MAC protocol wth the change of unused channels quantty (p n ) under each of

9 2 H. SONG ET AL. the two networ scenaros: (Fgure 6(a)) the number of secondary user groups wth M=5 and (Fgure 6(b)) the number of secondary user groups wth M=8. There are three observatons to mae. Frst, the unused channels usage ( ) of our proposed MAC protocol fluctuates between.9 and., whle the other two protocols unused channels' usage ( ) decreases wth the ncrease of p n. Second, we fnd that, n the proposed protocol, unless the secondary users bandwdth requrements are satsfed, they utlze the vacant channels to ther fullest extent and conduct hgh channel usage. However, the unused channels usage ( ) n OS-MAC drops rapdly wth p n due to the utlzaton of fxed number of vacant channels except for the ncrement of M. Thrd, n Fgure 6 (a), the curve represented for under our proposed MAC protocol declnes qucly after pn 8% caused by the satsfacton of all groups bandwdth requrements. However, unle CO-MAC, n the proposed protocol, wth the ncrement of p n, part of unused channels wll not be occuped by some secondary users whose bandwdth requrements are already satsfed, thus these channels are able to be allocated to other users who need them. Ths s an mportant feature. Throughput (Mbps) group group2 group3 group4 group5 3% 6% 9% (a) The number of secondary user groups M = 5. Throughput (Mbps) group group2 group3 group4 group5 group6 group7 group8 3% 6% 9% (b) The number of secondary user groups M = 8. Fgure 7. Group throughput wth dfferent p n n our proposed protocol QoS and Farness Analyss In ths subsecton, we evaluate QoS and farness of the proposed MAC protocol. Fgure 7 presents each group s throughput aganst the dfferent number of unused channels ( pn 3%, pn 6% and pn 9% ) under each of the two networ scenaros: (Fgure 7(a)) the number of secondary user groups wth M = 5 and (Fgure 7(b)) the number of secondary user groups wth M = 8. Commonly, the throughput comparson of all groups wth certan p n reflects the farness of the whole system, and the comparson of certan group s throughput wth dfferent p n shows the QoS support of the networ. In Fgure 7, we observe that, wth certan p n, the throughputs of all groups are much closer. Especally when p n = 9%, the groups throughputs are almost the same as those n Fgure 7(a). We also note that the throughput of certan group ncreases monotoncally wth the ncrement of p n. For nstance, n Fgure 7(b), the throughput of Group rases from.365mbps to 3.67Mbps as p n ncreases from 3% to 6%. Based on the smulaton results, we can mae the followng conclusons. Frst, our proposed MAC protocol s shown to be more effcent than the other two protocols not only from throughput aspect, but also from unused channels utlzaton aspect. Second, our proposed MAC protocol greatly mproves the qualty of servce for secondary users and guarantees the farness of channels allocaton for groups. 6. Conclusons In ths paper, we have proposed a novel DSA-drven MAC protocol for cogntve rado networs, In whch, secondary users are dvded nto several non-overlappng groups, and each group uses bondng/aggregaton technque to transmt data. All leftover channels are allocated among groups tang the groups bandwdth requrements nto consderaton. Moreover, the allocaton of vacant channels can be adjusted dynamcally when members jon/leave groups or prmary users return/leave the current networ. The smulatons ndcate that our proposed MAC protocol greatly mproves the qualty of servce for secondary users and maxmzes the utlzaton rato of spectrum resources. 7. References [] M. Mchenry, Spectrum whte space measurements, New Amerca Foundaton Broadband Forum, June 23.

10 H. SONG ET AL. 2 [2] FCC, Et docet no , November publc/attachmatch/fcc A.pdf [3] I. F. Ayldz, W. Y. Lee, M. C. Vuran, and S. Mohanty, Next generaton/dynamc spectrum access/cogntve rado wreless networs: A survey, Computer Networs, 26. [4] K. G. Shn and B. Hamdaou, OS-MAC: An effcent MAC protocol for spectrum-agle wreless networs, Transactons on Moble Computng, 27. [5] J. Ja, Q. Zhang, and X. Shen, HC-MAC: A hardware-constraned cogntve MAC for effcent spectrum management, IEEE Journal on Selected Areas n Communcatons, Vol. 26, No., pp. 6 7, January 28. [6] H. Su and X. Zhang, Cross-layer based opportunstc MAC protocols for QoS provsonngs over cogntve rado wreless networs, IEEE Journal on Selected Areas n Communcatons, Vol. 26, No., pp. 8 29, January 28. [7] T. Mans, V. S., and P. Rav, CSMA-based MAC protocol for cogntve rado networs, World of Wreless, Moble and Multmeda Networs, IEEE Internatonal Symposum, pp. 8, June 27. [8] A. Mshra, A mult-channel MAC for opportunstc spectrum sharng n cogntve networs, Proceedngs of MILCOM 26, October 26. [9] L. Ma, X. Han, and C. C. Shen, Dynamc open spectrum sharng MAC protocol for wreless ad hoc networs, n IEEE Internatonal Symposum on New Fronters n Dynamc Spectrum Access Networs, pp , 25. [] J. So and N. Vadya, Mult-channel MAC for ad-hoc networs: Handlng mult-channel hdden termnals usng a sngle transcever, n ACM Proceedngs of MOBIHOC, 24, pp [] J. Zhao, H. Zheng, and G. H. Yang, Dstrbuted coordnaton n dynamc spectrum allocaton networs, n IEEE Internatonal Symposum on New Fronters n Dynamc Spectrum Access Networs, pp , 25. [2] N. Ne and C. Comancu, Adaptve channel allocaton spectrum etquette for cogntve rado networs, n IEEE Internatonal Symposum on New Fronters n Dynamc Spectrum Access Networs, pp , 25. [3] C. Cordero, K. Challapal, D. Brru, and S. Shanar, IEEE 82.22: An ntroducton to the frst wreless standard based on cogntve rados, Journal of Communcatons, Vol., No., pp , Aprl 26. [4] IEEE Standard , Wreless LAN medum access control (MAC) and physcal layer (PHY) specfcatons, November 999. [5] L. Klenroc and F. Tobag, Pacet swtchng n rado channels: Part I - carrer sense multple-access modes and ther throughput-delay characterstcs, IEEE Transactons on Communcatons, No. 23, pp. 4 46, December 975. [6] P. Aparan and H. D. Tuan, Robust control va concave mnmzaton local and global algorthms, IEEE Transactons on Automatc Control, Vol. 45, No. 2, pp , February 2. [7] W. Wang, T. Peng, and W. Wang, Optmal power control under nterference temperature constrants n cogntve rado networ, Wreless Communcatons and Networng Conference, WCNC 27 IEEE, pp. 6 2, March 5, 27. [8] The networ smulator (ns-2).