Spectrum Co-existence of IEEE b and a Networks Using Reactive and Proactive Etiquette Policies

Transcription

1 Moble Netw Appl (2006) 11: DOI /s z Spectrum Co-exstence of IEEE b and a Networks Usng Reactve and Proactve Etquette Polces Xangpeng Jng Dpankar Raychaudhur Publshed onlne: 4 May 2006 C Scence + Busness Meda, LLC 2006 Abstract Ths paper presents an nvestgaton of spectrum co-exstence between IEEE b and a networks n the same shared frequency band usng cogntve rado technques wth dfferent levels of complexty. Smple reactve nterference avodance algorthms as well as proactve spectrum coordnaton polces based on etquette protocols are proposed and compared n terms of achevable spectrum effcency n a shared W-F/W-Max scenaro. In reactve nterference avodance methods, rado nodes coordnate spectrum usage wthout exchange of explct control nformaton ths s done by adaptvely adjustng transmt PHY parameters such as frequency, power and tme occupancy based on local observatons of the rado band. Because local observatons provde nformaton only about transmtters, they may not be suffcent for resolvng spectrum contenton n scenaros wth hdden recevers. Proactve coordnaton technques solve the hdden-recever problem by utlzng a common spectrum coordnaton channel (CSCC) for exchange of transmtter and recever parameters. Rado nodes can cooperatvely select key PHY-layer varables such as frequency and power by broadcastng messages n the CSCC channel and then followng specfed spectrum et- Research supported by NSF grant # CNS and # Paper submtted for publcaton after presentaton n part at IEEE DySpan 2005, Nov. 8 11, 2005, Baltmore, MD. X. Jng ( ) D. Raychaudhur Wreless Informaton Network Laboratory (WINLAB), Rutgers Unversty, 671 Route 1 South, North Brunswck, NJ, 08902, USA e-mal: xjng@wnlab.rutgers.edu Tel.: ext 653 Fax: D. Raychaudhur e-mal: ray@wnlab.rutgers.edu quette polces. An ns2 smulaton model s developed to evaluate both reactve and proactve etquette polces n scenaros wth co-exstng IEEE b and a networks. The densty of rado nodes n the coverage regon, and ther degree of spatal clusterng are key parameters n the system evaluaton. Detaled smulaton studes were carred out for a varety of scenaros ncludng both sngle and multple b hotspots per a cell wth and wthout spatal clusterng. Our results show that smple reactve algorthms can mprove system throughput when suffcent free space (n frequency, power or tme) s avalable for PHY adaptaton. In more congested scenaros wth spatally clustered nodes and hdden recevers, the proposed CSCC etquette can sgnfcantly mprove overall system performance over reactve schemes. Keywords Cogntve rado. Spectrum etquette protocol. CSCC. Co-exstence. Dynamc spectrum access 1. Introducton In ths paper, we nvestgate the feasblty of spectrum coexstence between IEEE b (W-F) and a (W- Max) [1] networks usng both reactve nterference avodance methods and the CSCC (common spectrum coordnaton channel) etquette protocol. CSCC has been proposed as an explct spectrum etquette protocol whch uses a common edge-of-band control channel for coordnaton between transcevers usng dfferent rado technologes. In an earler paper [2], t was shown that a smple CSCC mplementaton can be used to sgnfcantly reduce nterference between b and Bluetooth devces operatng n close proxmty. Ths motvated us to next consder the mportant emergng

2 540 Moble Netw Appl (2006) 11: scenaro n whch both wde-area and short-range rado technologes could co-exst n the same unlcensed band wth a small amount of coordnaton, ether explct or mplct. It s generally accepted that current unlcensed band etquettes (such as lsten-before-talk) are not applcable to the wde-area/short-range hybrd scenaro under consderaton due to hdden-recever problems and the need to support stream servces such as VoIP or vdeo. As a result, we beleve that t s approprate to consder new cogntve rado [3] technques whch allow dynamc sharng of spectral resources between multple rado devces n the same band. Cogntve rado methods can be categorzed n terms of ther protocol and hardware complexty, coverng a wde range of optons from reactve nterference avodance to explct protocol-based coordnaton, or even network-based collaboraton [4]. Reactve cogntve rado technques are based on channel sensng and dstrbuted adaptaton of transmt parameters such as frequency, power, bt-rate and tme occupancy. Reactve adjustment of PHY parameters s based only on local observatons, whch may sometmes be nsuffcent such as n scenaros where there are hdden recevers. The hdden-recever problem occurs when a recever s located n between two potental transmtters whch cannot sense each other s presence and hence may cause unntended nterference at the recever. Ths problem wll be dscussed further n Secton 4. The CSCC protocol coordnates rado nodes n a proactve way, where a common spectrum coordnaton channel at the edge of avalable spectrum bands s allocated for announcement of rado parameters such as frequency, power, modulaton, duraton, nterference margn, servce type, etc. Each node s equpped wth a low bt-rate, narrow-band control rado (or software-defned rado) for lstenng to announcements and broadcastng ts own parameters at the CSCC channel. Rado nodes recevng CSCC control nformaton can then ntate approprate spectrum sharng polces, such as FCFS (Frst-Come-Frst-Served), prorty or dynamc prcng aucton, to resolve conflcts n spectrum demand and share the resource more effcently by adaptng PHY parameters such as frequency or power. The hddenrecever problem mentoned above can also be solved because the range of CSCC control can be desgned to exceed that of regular servce data, and recevers can also explctly announce ther presence to further optmze spectrum use. The specfc problem studed n ths paper s that of evaluatng both reactve and proactve etquette polces for coexstence between W-F and W-Max [1] networks sharng the 2.4 GHz ISM band. Both smple scenaros wth one a cell and one b hotspot and more realstc scenaros wth multple hotspots are smulated usng an ns2 [5] system model. Varatons of node geographc dstrbuton (clustered vs. unform) are studed. The densty of rado nodes n the coverage regon and ther degree of spatal clusterng are key parameters n the system evaluaton. Clusterng regmes where CSCC can sgnfcantly mprove the network throughput by solvng the hdden-recever problem are dentfed. The rest of paper wll be organzed as follows: Secton 2 presents cogntve rado background; then the proposed reactve and proactve etquette polces wll be ntroduced respectvely n detal n Sectons 3 and 4; the co-exstng network framework s presented n Secton 5; smulaton parameters and results wth dscussons are demonstrated n Secton 6; we conclude wth future work n Secton Cogntve rado background A number of approaches have been proposed for mproved spectrum sharng over the past decade. Notable methods beng dscussed n the techncal and regulatory communtes nclude property rghts regmes [6, 7], spectrum clearnghouse [8], unlcensed bands wth smple spectrum etquette [9], open access [10 12] and cogntve rado [13 15] under consderaton here. The dstnctons between unlcensed spectrum regmes, open access and cogntve rado approaches are relatvely subtle as they are all based on the concept of technology neutral bands to be used by a varety of servces usng rado transcevers that meet certan crtera. For example, cogntve rado may be vewed as a specal case of open access or unlcensed regmes n whch rado transcevers are requred to meet a relatvely hgh standard of nterference avodance va physcal and/or network layer adaptaton. The cogntve rado prncples currently under consderaton by the FCC [15] and the research communty (such as DARPA XG Program [16]), span a farly wde range of possble functonaltes both at physcal and network layers, as outlned n Fg. 1, whch shows the protocol complexty and rado hardware complexty regmes for a number of possble coordnatonschemes. The agle wdeband rado scheme [17, 18] shown at the lower rght sde of Fg. 1 s the most prevalent concept for cogntve rado n whch transmtters scan the channel and autonomously choose ther frequency band and modulaton waveform to meet nterference mnmzaton crtera wthout any protocol-level coordnaton wth neghborng rado nodes. We observe here that although the agle rado has the least protocol complexty, t requres rapd waveform and modulaton adaptaton whch may have a hgh level of hardware complexty. Wthout explct coordnaton, t suffers from serous lmtatons due to near-far problems and hdden-recever problem due to the fact that nterference s a recever property whle spectrum scannng alone only provdes nformaton about transmtters. Ths can be overcome by a small amount of explct protocol level coordnaton n

3 Moble Netw Appl (2006) 11: Unlcensed band + proactve protocols Internet Server-based Spectrum Etquette Ad-hoc, Mult-hop Collaboraton Cogntve Rado schemes Protocol Complexty (degree of coordnaton) Unlcensed Band wth DCA (e.g x) Rado-level Spectrum Etquette Protocol Internet Spectrum Leasng Statc Assgnment Reactve Rate/Power Control UWB, Spread Spectrum Agle Wdeband Rados Open Access + smart rados Fg. 1 Hardware Complexty Hardware and protocol complexty chart for potental cogntve rado approaches whch control nformaton s exchanged between transmtters and recevers. Another smple technque s reactve control of transmt frequency/rate/power/tme, n whch rado nodes do not have any explct coordnaton wth neghbors but seek equlbrum resource allocaton usng reactve algorthms to control frequency [19 21],bt-rate,power [22] and tme occupancy, analogous to the way the TCP protocol reactvely adjusts source bt-rate over the Internet when congeston occurs. Fgure 2 shows a scenaro where reactve schemes are deployed. Two transmt pars AB and CD may use dfferent wreless technologes, but they are flexble n controllng ther operatng frequences (channels), ther transmsson rates, ther transmt powers and ther transmt tme. Usng smple reactve schemes, rado nodes can explore and fll the gaps n resource dmensons of frequency, space/power or tme by scannng each channel and sensng the nterference power. A&B s spectrum band C&D s spectrum band Range wth power control Fg. 2 A Range wthout power control Reactve schemes of frequency or power aglty D B C Range wth power control Range wthout power control For example, when C and D communcate, they wll sense that the frequency band taken by A and B has a hgh nterference power and other bands have a low nterference power, so C and D wll dynamcally select a clearer frequency band to avod nterference between two systems. In cases when there s no avalable degree of freedom n frequency, rado nodes can explore the dmenson of space by reactve transmt power control (RTPC) to ncrease spatal reuse. Both AB and CD can calculate the mnmum transmt powers possble for ther communcatons to reduce ther nterference areas to other systems. As llustrated n Fg. 2, both AB and CD can transmt n parallel n the same frequency band by reducng ther transmt powers. Another smple technque s reactve control of transmt tme by changng transmt probabltes based on nterference condtons. When the nterference power s hgh, transmt probablty s reduced to avod more congested stuatons n usng the spectrum. Otherwse transmt probablty can be ncreased when nterference power s low and channel condtons are good. Snce reactve adaptatons are based on local observatons, they may be expected to suffer from hdden-recever problems. Wth a slghtly hgher level of protocol complexty, proactve cogntve rado technques can mprove coordnaton between rado nodes by spectrum etquette protocols, usng ether a Common Spectrum Coordnaton Channel (CSCC) at the edge of the shared frequency band or Internet-based spectrum servces [4]. The CSCC concept s to standardze a common control protocol between dfferent rado systems for spectrum coordnaton purposes. A smple way s to equp a Common Control Rado (CCR) wth each node, whch s a low bt-rate, narrow-band rado, such as a prototype IEEE

4 542 Moble Netw Appl (2006) 11: Fg. 3 CSCC etquette protocol Frequency Rcscc Band #N Band #N-1 Network A Ad-hoc Network B Ad-hoc Network A Rcscc Band #2 Band #1 Network B Network C Ad-hoc Network C CSCC Tme Rcscc b 1 Mbps rado (coverng a range of about 600 m). Note that ths approach requres some protocol coordnaton ablty ncludng the use of a common physcal layer for coordnaton, but may not requre full-fledged agle rado capabltes wth programmable waveforms. A small amount of spectrum (called the Common Spectrum Coordnaton Channel [2]) at the edge of the shared spectrum bands can be allocated for the CCR, as llustrated n Fg. 3 (Frequency vs. Tme) where the shared spectrum s splt nto Band#1, Band#2,..., Band#N for data communcaton and the CSCC band at the lower edge for control purposes. Rado nodes can lsten to announcements and broadcast ther own parameters n the CSCC channel. Based on shared control nformaton on the CSCC, approprate spectrum sharng polces can be ntated, such as FCFS (Frst-Come-Frst-Served), prorty or dynamc prcng aucton [23], to resolve conflcts n spectrum demand and share the resource more effcently by adaptng PHY parameters such as frequency or power. For example, n Fg. 3, each ad hoc network A, B and C can set up approprate operatng channels or transmt powers to avod nterference. The hdden-recever problem mentoned above can also be solved because the range of CSCC control can be desgned to exceed that of regular servce data, and rado recevers can also explctly announce ther presence to overcome the hdden recever problem dscussed earler. 3. Reactve cogntve rado technques In ths secton, three reactve cogntve rado technques are ntroduced: Dynamc Frequency Selecton (DFS), Reactve Transmt Power Control (RTPC) and Tme Aglty (TA), whch reactvely adapt n dmensons of frequency, power and transmt tme respectvely Dynamc frequency selecton (DFS) In the DFS scheme, rado nodes perodcally scan the spectrum band and measure nterference power level n each avalable channel. When rado nodes have data to transfer, Fg. 4 An example of the dynamc frequency selecton algorthm Interference Power Level Perodcal Channel Scannng Tme 4 6 #1 #2 #3 #4 #5 #6 #7 #8 #9 #10 # Frequency (802.11b Channel)

5 Moble Netw Appl (2006) 11: they choose the channel wth the least nterference power. The concept s llustrated n Fg. 4, n whch each node keeps a record of the nterference power level of each channel and selects a sequence of channel #6, #9, #9,..., #4,etc. for communcaton. The updatng nterval can be determned by the statstcs of the traffc, e.g., randomly chosen n the order of a short data sesson ( 100 ms for about 50 packets wth sze of 512 Bytes at 2 Mbps bt-rate). Note that too frequent channel swtchng may cause packet loss due to lnk-level nterruptons. On the other hand, nfrequent swtchng may result n a slow response to channel condton changes. To prevent unnecessary channel swtchng, a new channel s used only f nterference power of a clearer channel s at least 10% less than current nterference level Reactve transmt power control (RTPC) It s mportant for rado nodes to not only explot avalable resources, but also at the same tme emt the least nterference to others. The RTPC algorthm acheves ths by allowng transmtters to use the mnmum transmt power possble for data transfer. Snce nterference s a recever property, n the RTPC scheme, each recever wll estmate the mnmum transmt power to mantan adequate lnk qualty, based on ts own QoS requrements and path loss estmates. Ths recommended transmt power level s fed back to transmtters by utlzng MAC packet headers (e.g., ACK header). As llustrated n Fg. 5, the recever can sense nterference power changes PI e snce the last measurement, and the receved power of current receved packet P rx. By knowng the target receved power P target, determned by the QoS requrement of the recever (e.g., a level of bt error rate less than 10 6 ), t then can calculate the recommended next transmt power usng Eq. (1). Transmt power s updated on a packet-by-packet bass and P tx (n) forthenth packet s calculated by P tx (n) = P tx (n 1) + (γ target + RSSI(n) P rx (n 1)) +(RSSI(n) RSSI(n 1)) (1) where γ target s the expected target SINR (all terms measuredndbordbm),andp target = γ target + RSSI(n) sthe target receved power, PI e = RSSI(n) RSSI(n 1) s the sensed nterference power change between the nth and (n- 1)th transmsson. In Fg. 5, the TX Power Adjustment block s controlled by energy constrants, whch s not consdered n current study. For mplementaton, the power value (n dbm) can be quantzed to 256 levels stored n an 8-bt feld n the MAC header, whch s pggybacked between the transmtter and recever. In case of pggyback packet loss, a power roll-back mechansm s used to avod deadlock stuatons by ncreasng the (recommended) transmt power by a certan amount (e.g., TRANSMITTER Energy Constrant Fg. 5 TX Power Adjustment Ptx(n) Ptx WIRELESS CHANNEL Reactve transmt power control algorthm RECEIVER Applcaton Requrements RX Power Level Standard Ptarget Ptx(n-1) Prx - PIe Sensor 20% of current power level) each tme a packet s lost untl reachng the maxmum value Tme Aglty (TA) Reactve nterference avodance can also be realzed by controllng transmt probablty or re-schedulng MAC packet transmssons n an nterference-varyng envronment. The Tme Aglty algorthm explores gaps n the tme dmenson by avodng transmssons (and thus potental retransmssons) when channel condtons are bad (.e., nterference level s hgh) and encouragng transmssons when channel condton s good. Ths s realzed by changng transmtters transmt probablty Prob tx as a functon of the nterference power and SINR at the preferred recever. Ths algorthm mplctly allows nodes to adapt to each other s traffc pattern by lstenng on the channel and controllng Prob tx. An example of the algorthm s shown n Fg. 6 where P nterference s the nterference power. Note that the communcaton threshold s assumed to be at BER 10 6 or SINR 12 db wth QPSK modulaton. Smlar to the RTPC scheme, the recever lstens on the channel and updates the recommended transmt probablty Prob tx whch s quantzed to 8 bts and pggybacked n MAC headers. For the algorthm shown n Fg. 6, a SINR near to the threshold (12 db) means that the channel condton s stll good but there may be potental close nterferers around. In order to avod nterferng more severely wth the potental nterferers, the transmt probablty s proportonal to the nverse of sensed nterference power. When the SINR level s less than the threshold, the node can nfer that ether the sgnal strength s too weak, or that the nterference power s too strong, or both. Thus t s preferable to control the transmt Fg. 6 Tme aglty algorthm

6 544 Moble Netw Appl (2006) 11: probablty to be proportonal to the current SINR value (n db) to avod re-transmssons and mutual nterference. Note that n terms of traffc engneerng, when the traffc pattern s easy to learn (e.g. Pareto ON/OFF traffc model [24] wth relatvely long OFF perods), such a tme aglty algorthm can help rados to adapt to each other s traffc pattern and effectvely utlze the avalable degree of freedom n tme. Prob tx s ncreased when the nterferer s traffc load s low (or off), and decreased when the nterferer s traffc load s hgh. Ths algorthm s traffc-type-ndependent, and the dfference s n the degree of dffculty n adaptng to the specfc traffc patterns on the channel. For example, t s easer to adapt to Pareto ON/OFF traffc than CBR traffc wth the same load, due to the extended OFF perod. 4. Proactve spectrum etquette protocols 4.1. CSCC etquette protocol The basc CSCC concept was outlned earler n Secton 2(see Fg. 3). In ths approach to spectrum coordnaton, each rado node announces ts parameters to neghborng nodes by broadcastng CSCC messages through a common CSCC channel at the edge of the band. Informaton n the CSCC message, such as node ID, center frequency, bandwdth, transmt power, data rate, modulaton type, data burst duraton, nterference margn (IM), servce type, etc., s used by neghborng nodes to coordnate and share the spectrum n an effcent way. Note that the CSCC protocol mechansm s ndependent of the spectrum coordnaton polcy tself, whch can be mplemented to reflect regonal or applcatonspecfc requrements. Ths s explaned further n Fg. 7 whch shows that a separate CSCC control stack consstng of CSCC PHY and MAC operates n parallel wth the data servce. The spectrum coordnaton (SC) polcy runs on top of the CSCC protocol stack and can be specfed n a completely general way as long as necessary parameters are carred by the CSCC packet. Snce nterference needs to be consdered at recevers rather than transmtters, CSCC announcements may be made by recevers nvolved n actve data sessons by one-hop broadcast, and contenton can be resolved by perodc repetton wth some randomzaton of transmt tme to avod multple collsons. When a node receves a CSCC message, t wll know that there s a data sesson gong on between neghborng Fg. 7 SC Polces CSCC-MAC CSCC-PHY CSCC protocol stack CONTROL NETWORK DATA-MAC DATA-PHY nodes at a specfed frequency slot for some duraton. Then, a coordnaton procedure s ntated ether by swtchng to other bands wth lower nterference or by lmtng transmt power to avod nterference wth exstng rado lnks followng specfed coordnaton polces. The CSCC protocol can help to solve the hdden-recever problem, as llustrated n Fg. 8. R cscc s the coverage range of CCR whch s generally 1 2x the mnmum servce data rado range. When TX2 ntates a data sesson to RX2, t frst notfes RX2 of the transmt power and the estmated data burst duraton T 2 by data packet pggybackng. Then RX2 broadcasts a CSCC message n the CSCC channel to clam the current spectrum,.e., Band#2, for a duraton of T 2. When TX1 receves the CSCC message from RX2, t wll know the spectrum Band#2 s taken by RX2 and TX1 wll ether swtch to other avalable bands or coordnate wth RX2 at Band#2 by reducng ts transmt power,.e., coverage range from R1 to R1. Wthout explct coordnaton from the CSCC protocol (or some other smlar mechansm), node RX2 would become hdden to the nterference from TX1. Smlar to the wellknown hdden termnal problem n IEEE networks [25], the hdden-recever problem exsts n networks wth heterogeneous rados. Intally TX1 covers a range of R1, and RX2 covers a range of R2. There s no way for TX1 to notce the exstence of RX2 only by reactve scannng or sensng, especally when R2 < R1, and therefore the transmsson of TX1 wll nterfere wth RX2 f they share the spectrum. Note TX1/RX1 and TX2/RX2 use dfferent rado technologes for data communcaton and thus they requre a common spectrum coordnaton channel as n the CSCC method proposed here. TX1 then receves CSCC messages from RX2 whch s no longer hdden to TX1, and TX1 can swtch to a dfferent frequency or reduce ts power to avod nterference Spectrum coordnaton polces Spectrum coordnaton polces refer to specfc algorthmc procedures used for adaptaton of frequency or power based on the n-band nterference power. Alternatve coordnaton polces wll also be dscussed Coordnaton by adaptaton n frequency Rado nodes can change operatng frequences to avod nterference by the CSCC protocol. Followng the example of Fg. 8, when TX1 and RX1 have on-gong data communcaton, RX1 broadcasts a CSCC message n the CSCC channel statng t wll take Band#2 for some duraton, as shown n Fg. 9. After a whle, TX2 notfes RX2 that t has data to send, and then RX2 broadcasts a CSCC message statng t wshes to use Band#2 for data transfer. In the event that RX2

7 Moble Netw Appl (2006) 11: Fg. 8 Illustraton of the CSCC protocol and how t helps to solve the hdden-recever problem Rcscc RX1 TX1 CSCC R1R1' Rcscc RX2 R2 TX2 RX2 CSCC Packet Type Band#2 BW Ptx Rate Modu Dura IM Fg. 9 Coordnaton by adaptaton n frequency Frequency Band#1 Band#2 Data from TX1 Data from TX1 Tme TX1/RX1 swtch to Band#1 after recevng the CSCC from RX2 to avod nterference Data from TX2 Tme CSCC Tme CSCC from RX1 CSCC from RX2 has a hgher prorty, t wll take over Band#2 and starts communcaton, whle TX1 s forced to change ts data channel to a clear channel, e.g., Band#1 and notfes RX1 by ether broadcastng a CSCC message or pggybackng n the data packet. Then RX1 wll broadcast a CSCC message to clam Band# Coordnaton by adaptaton n power When the spectral band s heavly loaded and frequency selecton alone cannot be used to avod nterference between smultaneous users, adaptaton of transmt power s an effcent way to reduce nterference. By lstenng to CSCC messages, rado nodes can determne approprate transmt power levels requred to reduce nterference n a specfc frequency band. In ths case the CSCC message carres a feld called the recever s nterference margn (IM). The IM s defned as the maxmum nterference power a recever (the one broadcastng the CSCC message) can tolerate wthout dsturbng ts on-gong data communcaton. When the IM value s changed, t wll be updated to neghborng nodes by CSCC messages. The power adaptaton algorthm s llustrated n Fg. 10. Assume at the data channel #n, the receved power at node from node j s Pr (n) j and ts current sgnal to nterference and nose rado (SINR) s SINR (n) j, the nterference margn can l Fg. 10 (n) Pt k (n) Pt k k j Pr (cscc) k (n) SINR j SINR mn (n) G k (cscc) G k CSCC SINR Pt I (n) (cscc) Coordnaton by adaptaton n power } SINR Tme ( ) Pr n j Data (n) (n) G j Pt j

8 546 Moble Netw Appl (2006) 11: be calculated by I (n) = ( 1 SINR mn 1 SINR (n) j ) Pr (n) j (2) where SINR mn s the mnmum SINR requred to mantan the on-gong communcaton at node, e.g., mantan a mnmum bt error rate of 10 6 for TCP traffc [26]. Node wll (csc c) broadcast a CSCC message wth power Pt at the CSCC channel. The IM I (n) (csc c) and Pt are both contaned n the CSCC message. Assume that node k receves the CSCC message at the control channel, and the path loss gan of the (csc c) control channel from node to node k s G k. Then we have (csc c) (csc c) (csc c) (csc c) Pt G k = Pr k, and Pr k can be reported by the PHY of node k. Assume the CSCC channel s symmetrc, so (csc c) (csc c) (csc c) (csc c) G k = G k = Pr k /Pt. Snce the control channel s usually close to the data channel n frequency, the path loss gan at the CSCC channel s a good estmaton of that at the data channels,.e., G (n) k = G (n) c) k G(csc k. The maxmum transmt power of node k at data channel #n then s bounded by the constrant n order not to dsturb the sgnals receved at node :.e., Pt (n) k Pt (n) k G (n) k I (n) (3) I (n) G (n) k I (n) Pt Pr (csc c) k (csc c) If Pt (n) k s too small for node k to reach ts recever, say node l, t should ether swtch channels seekng a band wth less nterference (.e., more IM avalable), or just keep slent by backng off ts transmssons followng a defned backoff polcy. In the example of Fg. 8, TX1 can calculate ts maxmum transmt power at Band#2 by (4) and reduce ts transmsson range from R1 to R1, keepng the nterference power experenced at RX2 less than ts IM Alternatve polces A wde varety of spectrum coordnaton polces can be appled wthn the CSCC protocol framework. The polces defne rules that rado nodes must follow when they are competng for spectrum resources. A smple access rule s Frst-Come-Frst-Served (FCFS), whch means the frst one comng nto a channel wll clam the spectrum for some duraton by CSCC protocol. Another approach s prortybased, where nodes have dfferent pre-assgned prortes based on ther carred traffc type, and hgh prorty nodes wll take precedence over low prorty ones when there s (4) Fg. 11 A co-exstng IEEE b and a network contenton for the same pece of spectrum. A dynamc prcng aucton polcy [23, 27] n whch users bd on avalable spectrum s another choce. Rado nodes can offer ther prces for usng the spectrum and the allocaton can be done n a dstrbuted way by CSCC protocol to maxmze the system revenue. 5. Co-exstence of IEEE b and a A co-exstng system wth IEEE b hotspots and a cells n the same shared spectrum s consdered to evaluate the effectveness of proposed reactve and proactve spectrum coordnaton polces System framework An example of the co-exstng network s shown n Fg. 11, whch conssts of IEEE b hotspots, wth one Access Pont (AP) and multple clents n each hotspot, and a cells, wth one Base Staton (BS) and multple Subscrber Statons (SS) per cell. W-F hotspots can cover a range of 500 m as wreless local area networks and W-Max cells cover a longer range of 3kmaswreless metropoltan area networks. Both systems are deployed n one geographc area and b hotspots are nsde a cells. Ths s a typcal cogntve rado scenaro where a SS may be clustered wth b hotspots and they overlap n space. We assume that both systems wll share a current or future unlcensed or cogntve rado band, and wll need to co-exst by coordnatng wth each other. Fgure 12 shows a sketch of the channel allocaton for the two systems. W-F rado uses DSSS wth 22 MHz bandwdth, and there are 11 overlappng channels wth center frequences from 2412 to 2462 MHz. OFDM s used n W- Max rados wth 20 MHz bandwdth, and n ths study we assume there are three non-overlappng channels centered at 2412, 2432 and 2452 MHz. To smplfy the smulaton, bandwdth and rate are fxed for both systems, and QPSK

9 Moble Netw Appl (2006) 11: OFDM Spectrum DSSS Spectrum CSCC Frequency (MHz) b Channel #1 #2 #3 #4 #5 #6 #7 #8 #9 #10 #11 Fg. 12 Channel allocaton for IEEE b and a modulaton s used wth 2 Mbps data rate for b and 14 Mbps for a rados. We also assume that the CSCC channel s allocated at the left edge of the whole spectrum and s orthogonal to other data channels. In order to capture the nterference effects between the two systems, a physcal-layer nterference model s constructed to calculate the SINR at a recever. Packet recepton s based on smulated packet error rate (PER), whch s calculated from bt error rate (BER) knowng the packet length n bts. The BER s obtaned from the modulaton performance curve [28] by knowledge of SINR. Assume at data channel #n, node transmts to node j wth transmt power Pt (n) j, the path loss gan between them s G (n) j, and the n-band background nose observed at node j s N (n) j, then the SINR at the recever j can be expressed as: SINR (n) j = N (n) j Pt (n) j G (n) j + α (n) lj Pt l G (n) lj l where 0 α (n) lj 1 s the spectrum overlappng rato of node l and j at channel #n. The nterference powers (n watts) from all transmtted sgnals (DSSS and/or OFDM) are summed over overlapped regons (n frequency). Here we assume the transmssons of nodes other than node are addtve nterference Implementaton n ns2 Both reactve (DFS, RTPC and TA) and proactve (CSCC) spectrum etquette polces are mplemented n Network Smulator verson 2.27 (ns2) [5]. For DFS, deal channel swtchng s assumed for b hotpots,.e., the AP n the hotspot selects new channels and all clents n the hotspot wll be notfed by a broadcast message and mmedately swtch to the same new channel whch AP selected. The penalty of swtchng channels s the loss of the current packet f any. The typcal frequency scannng nterval s assumed to be unformly dstrbuted between 100 and 200 ms, whch s the same order of magntude as the transmsson tme for a short data sesson ( 50 packets wth sze of 512 bytes at 2 Mbps). For RTPC, when a MAC packet s ntated at the sender, the current transmt power level (quantzed to an 8-bt nteger number between 0 and 255) s placed nto b (5) Fg. 13 Bt Source ID Destnaton ID Data Burst Sesson Duraton NodeType Prorty BandWdth Modula ton Center Frequency CSCC Message Transmt Power Interference Margn at Data Band CSCC packet format RTS or a frame header. The recever then can obtan the receved power of ths packet and the sender s transmt power from the header. In ths paper, we wll use a constant target SINR of 12 db, whch approxmately corresponds to a BER of 10 6 when usng QPSK modulaton. Then the recever can compute the recommended transmt power from equaton (1) and pggyback n the MAC header to the sender. Maxmum transmt power s used for RTS/CTS due to ther short length and RTPC s appled to both a BS and SS (both downlnk and uplnk). The TA algorthm s mplemented smlar to RTPC. Recevers calculate the recommended transmsson probabltes by Fg. 6, whch are then pggybacked n MAC headers to the transmtters. In cases of packet loss, transmtters wll transmt wth probablty 1 f there s data to send. The CSCC etquette protocol s mplemented wth a dual rado structure n each node. The spectrum coordnaton agent s between network and MAC layers, whch montors both data rado (IEEE b or a) and control rado (1 Mbps type). The control rado s fxed at the CSCC channel. The packet format for CSCC messages s shown n Fg. 13. A Pareto ON/OFF traffc model [24] s used to smulate Internet traffc, and a CSCC message s broadcast per data burst sesson (Pareto ON sesson). Only best-effort traffc wth UDP packets s consdered here. The estmated burst duraton n mllseconds s ncluded n the CSCC message. A FCFS-based polcy s used when there are contentons,.e., the frst node clamng the spectrum wll take t and subsequent transmssons from other nodes must coordnate wth the frst one by swtchng channels or boundng ther transmt powers satsfyng the nterference margn of the frst node. 6. Smulatons Scenaros wth sngle or multple b hotspots are smulated and varous a SS node geographc dstrbu-

10 548 Moble Netw Appl (2006) 11: Fg. 14 BS DBS-AP Network scenaro for sngle cell case b Hotspot DSS-AP A D 100m SS AP B C a Cell tons are also studed. DFS, RTPC, TA and the CSCC protocol are evaluated and compared n the scenaros consdered Smulaton parameters The parameters used n the smulatons are summarzed n Table Smulaton results Sngle a cell and sngle b hotspot case Each coordnaton algorthm s frst evaluated n a smple network scenaro wth one a cell (one BS and one SS) and one b hotspot (1 AP n the center and 1 4 clents A, B, C and D placed 100 m away from the AP), as shown n Fg. 14. D BS AP s the dstance between a BS and b AP and D SS AP s the dstance between a SS and b AP Effect of DFS for spectrum overlappng In ths smulaton, we assume the center frequency of a cell s fxed at 2412 MHz, whch overlaps the most wth b channel #1, partally overlaps wth b channel #2, #3, or #4, and does not overlap beyond channel #5. DFS enables b devces to avod nterference by swtchng ther operatng channels dynamcally. Fgure 15 shows the beneft of swtchng to dfferent channels. We defne the nterference radus (IR) as the dstance between two systems when ther throughputs begn to degrade due to nterference. When both a DL and lnks are overloaded wth CBR traffc (the most severely nterferng case), IR wll be 1.7 km f b s at channel #1, but IR can be reduced to 1.6, 1.4 and 1.2 km by swtchng b channel to #2, #3 or #4 respectvely. By operatng at channel #5 or beyond, there wll be no nterference between the two systems (IR s zero). Smlar results are observed wth two b traffc flows n Fg. 15(b) Effect of RTPC The same scenaro shown n Fg. 14 s used and D BS AP s fxed at 3 km. RTPC s appled to both b lnks and a uplnk and D SS AP s vared (the closer the a SS to b hotspot, the stronger the nterference). Note that snce the nterference from a BS s fxed, RTPC s not appled to the a downlnk here. Fgure 16 shows the beneft by applyng RTPC: the a SS throughput can ncrease up to 4 tmes at the expense of slght degradaton n b throughput. When the SS node s close to the hotspot (strong nterference), b node tends to more back-offs whch wll beneft a SS (throughput ncrease when D SS AP s small) by less nter- Table 1 Smulaton parameters From CNWLC-811 Wreless b PC Card specfcaton. IEEE a IEEE b MAC protocol TDMA IEEE b BSS mode Channel model AWGN, two ray ground propagaton model Bandwdth/ channels 20 MHz / 3 nonoverlappng channels 22 MHz / 11 overlappng channels Raw Bt Rate 14 Mbps 2 Mbps Rado parameters OFDM (256-FFT, DSSS (QPSK) QPSK) Background nose 174 dbm/hz densty Recever nose fgure 9 db 9 db Recever senstvty 80 dbm (@BER 10 6, 14 Mbps) 82 dbm (@BER 10 5, 2 Mbps) Antenna heght BS 15 m, SS 1.5 m All 1.5 m CSCC coverage 600 meters Maxmum coverage 3 Km (@BS 33 dbm) 500 m (@20 dbm) Transmtter power range BS 0 33 dbm, SS 0 23 dbm 0 20 dbm

11 Moble Netw Appl (2006) 11: ference. In ths case, DFS wll have more beneft when there s no more degree of freedom to explore n the dmenson of power Effect of tme aglty The TA algorthm s mplemented for both systems to fll avalable gaps and avod busy perod n tme doman by settng transmt probabltes to transmtters. Pareto ON/OFF traffc [24] s used for a lnks and the duty cycle (ON to OFF rato) s kept constant at 1: b nodes (usng CBR traffc) wll try to adapt to the a traffc pattern by decreasng transmt probablty when a traffc s ON and ncreasng t when a traffc s OFF by measurng SINR levels. Fgure 17 shows that the TA algorthm can help to mprove the hotspot lnk throughput by up to 30% when the nterferer traffc ON tme s of the order of one second. Although the smple tme (a) a DL throughput Average Lnk Throughput Trace(Mbps) Average Lnk Throughput Trace(Mbps) CH1 CH2 CH3 CH4 CH5 CH Dstance from hotspot to BS (m) (a) Wth one b traffc flow CH1 CH2 CH3 CH4 CH5 CH Dstance from hotspot to BS (m) (b) Wth two b traffc flows Fg. 15 Average b throughput vs. DBS-AP at dfferent channels, when both systems have overloaded CBR traffc (b) b hotspot throughput Fg. 16 Average lnk throughput trace, 4 lnks for hotspot, each has Posson arrval rate wth nter-arrval mean tme 3 ms Fg. 17 Tme aglty by varyng a Pareto traffc ON tme, b nodes use CBR traffc wth load 200 Kbps, and a node load s 1.3 Mbps

12 550 Moble Netw Appl (2006) 11: Fg. 18 Network throughput by usng CSCC frequency or power adaptaton when both systems have Pareto traffc wth ON/OFF tme = 500 ms/500 ms and traffc load 2 Mbps (a) CSCC frequency adaptaton when D SS-AP =200m (b) CSCC power adaptaton aglty only performs well under lmted crcumstances, ths experment serves as an example of the spectral freedom usage pattern dependence of coordnaton algorthms Evaluaton of CSCC The network s the same as Fgure 14, whch s a typcal hdden-recever scenaro. In the hotspot, traffc goes from AP to node A, and for a, only downlnk (DL) traffc s consdered so that the a SS becomes hdden to b nterferers. All nodes are statc and D BS AP s 1 km. The throughputs for both systems are plotted n Fg. 18. By applyng CSCC frequency adaptaton (see Fg. 18(a)), both a DL and b throughput can almost be doubled snce n ths scenaro there s enough vacant spectrum to use wth CSCC coordnaton. To evaluate CSCC-based power adaptaton algorthm n the hghest nterference case, we consder both systems center frequences fxed at 2412 MHz (they overlap mostly n frequency as shown n Fg. 12). Fgure 18(b) shows 802.l6a DL throughput s mproved by 35% whch vares by D SS AP. Snce the a BS s Fg. 19 Clustered hotspots and a SS (overlap n space)

13 Moble Netw Appl (2006) 11: km away (out of CSCC range), b hotspot throughput s slghtly degraded, but the average network throughput for both systems s stll mproved by about 5 to 15%. When the a SS s out of the hotspot CSCC range, the lnk throughput s the same for the case wth or wthout CSCC, as mght be expected. Snce the BS s always out of the hotspot CSCC range, we would expect greater mprovement for b throughput n cases wth shorter lnks Multple b hotspots wth varyng a SS dstrbuton case In addton to the network scenaro n Fg. 14, four b hotspots (wth 4 clents and 1 AP per hotspot) are placed n one a cell wth 1 km away from the BS, whch s llustrated n Fg b nodes are randomly placed nsde the hotspot wth the dstance to AP less than R max11 meters. The followng geographc dstrbutons of a SS were studed: () randomly (unformly) dstrbuted nsde the a cell wth a radus of 1.5 km; () clustered around each hotspot wth the dstance to each AP less than R c.the clusterng ndex C s defned as the rato of R max11 and R c, whch s between 0 and 1, and obvously the larger the clusterng ndex, the more closely the cluster couples spatally wth hotspots (and thus the hgher the nterference between the two systems). The total number of a SS s kept the same as the total number of b clents n the network and the traffc type s the same as the prevous smulaton. Frst the results for CSCC adaptaton n frequency (denoted as CSCC-F n Fg. 20) are compared wth reactve dynamc frequency selecton (DFS). Both a DL and UL traffcs are consdered. In Fg. 20(a) and (b) are the cases wth unformly-dstrbuted a SS (regme () n Fg. 19); (c) and (d) are the cases wth clusterng-dstrbuted SS (a) Unformly-dstrbuted case, load = 400 Kbps (c) Clusterng-dstrbuted case, load = 400 Kbps (b) Unformly-dstrbuted case, load = 600 Kbps (d) Clusterng-dstrbuted case, load = 600 Kbps Fg. 20 Throughput for unformly (a, b) and clusterng (c, d) dstrbuted a SS nodes (wth 12 nodes n each a channel), when R max11 = 100 m, Rc = 200 m and Pareto traffc wth ON/OFF tme = 500 ms/500 ms

14 552 Moble Netw Appl (2006) 11: nodes (regme () n Fg. 19). The results show CSCC-F can sgnfcantly mprove the average network throughput (up to 50% n unformly dstrbuted case and 140% n the clusterng case). It also performs better than reactve DFS when the a SS node densty s not very hgh, whch means there s vacant spectrum for the two systems to operate n dfferent channels. Comparng Fg. 20(a) wth (b), the mprovement amount s hgher wth more traffc load. When a SS nodes take all avalable spectrum bands (.e., 36 nodes takng all 3 avalable a channels), the coordnaton n frequency may be nsuffcent due to lack of avalable spectrum, whle adaptaton n power wll be explored. To evaluate coordnaton va power adaptaton, we assume the hghest nterference case wth fxed center frequency at 2412 MHz for both systems (no adaptaton n frequency). The CSCC-based power adaptaton algorthm (denoted as CSCC-P n the Fgures) s compared wth reactve ones,.e., RTPC (Reactve Transmt Power Control) and TA (Tme Aglty). The results for unform dstrbuton of a SS nodes n regme () are shown n Fg. 21 wth average hotspot and a DL/UL throughputs plotted separately. In ths case the SS nodes are sparsely dstrbuted n the cell and there s a lower probablty of hdden recevers. Fgure 21- (a) Offered load = 600 Kbps (a) Average hotspot throughput (b) Average a DL/UL throughput Fg. 21 Throughput for a SS random dstrbuton n regme () wth varyng hotspot radus Rmax11, and the number of a SS nodes : b nodes = 2:1, load 600 Kbps (b) Offered load = 1 Mbps Fg. 22 Throughputs for power adaptaton wth clusterng-dstrbuted a SS n regme (), wth numbers of a SS : b nodes = 1:1, and Pareto traffc wth ON/OFF tme = 500 ms/500 ms (a) shows that when the hotspot sze s larger, ts throughput s severely affected by the nterference from a DL/UL, but the CSCC protocol can help mprove hotspot throughput by % when R max11 s greater than 350 m accompaned by a slght degradaton of a average throughput. The CSCC protocol performs better than the reactve RTPC and TA because the reactve schemes can also mprove the hotspot throughput but tend to degrade a throughput more severely. The results for clusterng-dstrbuted a SS n regme () are shown n Fg. 22. X-axs s the clusterng ndex C = R max11 /R c, and Y-axs s the average network throughput of both systems. The R max11 s fxed at 50 m and C s vared by changng R c. By applyng CSCC-P, average network throughput can be mproved up to 20% when the clusterng ndex s greater than about 0.2 and the amount of

15 Moble Netw Appl (2006) 11: mprovement ncreases wth C, whch means hgher nterference between the two systems. The amount of throughput mprovement ncreases wth the offered traffc load (600 Kbps vs. 1 Mbps). The CSCC protocol also performs better than reactve methods n cases wth sgnfcant spatal clusterng, manly due to the fact that t can deal wth the hdden-recever problem dscussed earler. In summary, when the network scenaro s smple and there s suffcent free space n frequency, power and tme, smple reactve algorthms may be adequate for reducng nterference and mprovng system throughput. Coordnaton schemes utlzng frequency adaptaton (CSCC-F and DFS) can sgnfcantly mprove the network throughput when there s vacant spectrum and the mprovement wll depend on the avalablty of vacant spectrum. When avalable spectrum s somewhat more congested the CSCC-based power adaptaton algorthm can beneft hotspot throughput when the hotspot sze s large wth unformly dstrbuted a SS. In spatally clustered scenaros, the CSCC protocol can sgnfcantly mprove average network throughput over reactve schemes when the clusterng ndex s large. 7. Conclusons and future work Spectrum co-exstence of IEEE b and a networks has been studed usng both reactve and proactve spectrum coordnaton polces to coordnate and reduce nterference. Specfcally, reactve algorthms such as DFS, RTPC and TA and proactve CSCC etquette protocols are studed. The hdden-recever scenaro n whch reactve algorthms may not work well was dentfed, and t was shown that the CSCC approach can help to solve ths problem. Proposed reactve and proactve coordnaton polces were smulated n representatve WF-WMax co-exstence scenaros, and system performance based on average throughput was evaluated and compared. Varous a SS node densty and geographc dstrbutons were studed leadng to an dentfcaton of spatal clusterng regmes where CSCC coordnaton can sgnfcantly mprove system throughput by solvng the hdden-recever problem. Our results demonstrate that CSCC power adaptaton can help mantan servce qualty at the expense of a modest decrease n throughput n the hdden-recever scenaro consdered. Overall system throughput can be sgnfcantly mproved over reactve schemes dependng on the degree of spatal clusterng. In future work, alternatve spectrum coordnaton algorthms and addtonal system performance metrcs (such as delay and control overhead) wll also be studed n context of b/802.16a co-exstence. A prototype mplementaton for expermental verfcaton s also planned. References 1. C. Eklund, R.B. Marks, K.L. Stanwood and S. Wang, IEEE Standard : A techncal overvew of the wrelessman ar nterface for broadband wreless access, IEEE Communcatons Magazne (June 2002). 2. D. Raychaudhur and X. Jng, A spectrum etquette protocol for effcent coordnaton of rado devces n unlcensed bands, n Proc. PIMRC 2003, Bejng, Chna (Sept. 2003). 3. J. Mtola III, Cogntve rado: An ntegrated agent archtecture for software rado, PhD thess, Royal Insttute of Technology (KTH), Sweden (May 2000). 4. D. Raychaudhur, Adaptve wreless networks usng cogntve rados as a buldng block, n: MobCom 2004 Keynote Speech, Phladelpha, PA (Sept. 2004). 5. The network smulator ns2. http: // 6. A. De Vany, Property rghts n electromagnetc spectrum, n The New Palgrave Dctonary of Economcs ed. P. Newman (1998). 7. G. Faulhaber and D. Farber, Spectrum management: Property rghts, markets and the commons, n: Proceedngs of the Telecommuncatons Polcy Research Conference, Alexandra, VA (Oct. 2003). 8. Y. Benkler, Some economcs of wreless communcatons, Harvard Journal of Law and Technology 16 (2002). 9. D. Satapathy and J.M. Peha, Etquette modfcaton for unlcensed spectrum: Approach and mpact, n: Proceedngs of the IEEE Vehcular Technology Conference (1998), pp T.W. Hazlett, Spectrum flash dance: El Noam s proposals for open access to rado waves, Journal of Law and Economcs 41(2) (1998) Y. Benkler, Overcomng Agoraphoba: Buldng the commons of the dgtally networked envronment, Harvard Journal of Law and Technology 11 (1998) G. Rosston et al., Comments of 37 Concerned Economsts. In Submtted to the U.S. Federal Communcatons Commsson In the Matter of Promotng the Effcent Use of Spectrum through Elmnaton of Barrers to the Development of Secondary Markets, WT Docket , (2001). 13. G. Magure and J. Mtola, Cogntve rado: Makng pcs personal, IEEE PCS Magazne 6(4) (Aug. 1999) FCC Notce of Inqury, n the Matter of Addtonal Spectrum for Unlcensed Below 900 MHz and n the 3 GHz Band, ET Docket No , 17 FCC Rcd (2002). 15. FCC NPRM on Cogntve Rado, http: //hraunfoss.fcc.gov/edocs publc/attachmatch/fcc a1.pdf. 16. XGWORKING GROUP, The XG Vson. Request For Comments, verson 1.0. Prepared by: BBN Technologes, Cambrdge, Massachusetts, USA (July 2003). http: // 17. K. Challapal, S. Mangold and Z. Zhong, Spectrum agle rado: Detectng spectrum opportuntes, n: 6th Annual Internatonal Symposum On Advanced Rado Technologes, Boulder, Colorado (March 2004). 18. S. Mangold, Z. Zhong, K. Challapal and C.T. Chou, Spectrum agle rado: Rado resource measurements for opportunstc spectrum usage, n: IEEE Globecom 2004, Dallas, TX, USA (Nov. 2004). 19. S.-L. Wu, Y.-C. Tseng, C.-Y. Ln and J.-P. Sheu, A mult-channel MAC protocol wth power control for mult-hop moble ad hoc networks, The Computer Journal 45 (2002) Y. Xng, R. Chandramoul, S. Mangold and S. Sankar, Dynamc spectrum access n open spectrum wreless networks, wll appear n Specal ssue on 4G Wreless Systems, n: IEEE Journal on Selected Areas n Communcatons (2006). 21. C.T. Chou, Hyol Km, N. Sa Shankar and K.G. Shn, What and how much to gan from Spectrum Aglty, n: IEEE/ACM Trans.

16 554 Moble Netw Appl (2006) 11: on Networkng (June 2004). 22. J. Monks, V. Bharghavan and W. Hwu, A power controlled multple access protocol for wreless packet networks, n Proc. IEEE Infocom (Aprl 2001) pp C. Evc and B. Fno, Spectrum management, prcng, and effcency control n broad-band wreless communcatons, n: Proceedngs of the IEEE (Jan. 2001) vol. 89, no H. AhleHagh, WR. Mchalson and D. Fnkel, Statstcal characterstcs of wreless network traffc and ts mpact on ad hoc network performance, n: Proceedngs of the 2003 Appled Telecommuncaton Symposum (2003). 25. S. Khurana, A. Kahol and A.P. Jayasumana, Effect of hdden termnals on the performance of IEEE MAC protocol, n: Proceedngs of the IEEE LCN Conference (1998) pp G. Xylomenos, G.C. Polyzos, P. Mahonen, and M. Saaranen, TCP performance ssues over wreless lnks, n: IEEE Communcatons Magazne (Aprl 2001). 27. O. Iler, S. Mau and N.B. Mandayam, Prcng for enablng forwardng n wreless ad hoc networks, n: Proceedngs of WCNC 04, Atlanta, GA (March 2004). 28. R.M. Ward Jr., Smulated BER Results of Proposed OFDM Structure n Multpath.IEEE802.16a.3 01/8 document (2001). Dpankar Raychaudhur s Professor, Electrcal & Computer Engneerng Department and Drector, WINLAB (Wreless Informaton Network Lab) at Rutgers Unversty. As WIN- LAB s Drector, he s responsble for a cooperatve ndustry-unversty research center wth focus on next-generaton wreless technologes. WINLAB s current research scope ncludes topcs such as RF/sensor devces, UWB, spectrum management, future 3G and WLAN systems, ad-hoc networks and pervasve computng. He has prevously held progressvely responsble corporate R&D postons n the telecom/networkng area ncludng: Chef Scentst, Iospan Wreless ( ), Assstant General Manager & Dept. Head-Systems Archtecture, NEC USA C&C Research Laboratores ( ) and Head, Broadband Communcatons Research, Sarnoff Corp ( ). Dr. Raychaudhur obtaned hs B.Tech (Hons) from the Indan Insttute of Technology, Kharagpur n 1976 and the M.S. and Ph.D. degrees from SUNY, Stony Brook n 1978, 79. He s a Fellow of the IEEE. Xangpeng Jng receved the B.S. degree n Electrcal Engneerng from Pekng Unversty, Bejng, P. R. Chna n 2000 and the M.E. degree n Electrcal Engneerng from Cty College of Cty Unversty of New York, New York n He s currently a Ph. D. canddate n the WINLAB (Wreless Informaton Network Laboratory), Rutgers Unversty, NJ. He has been wth the MobNet Group snce Hs research nterests nclude spectrum etquette protocols, co-exstence between wreless communcaton systems, cogntve rado technologes, and adaptve wreless ad hoc networks.