Decision Analysis of Dynamic Spectrum Access Rules

Transcription

1 Decson Analyss of Dynamc Spectrum Access Rules Juan D. Deaton, Chrstan Wernz, Luz A. DaSlva N&HS Drectorate Idaho Natonal Lab Idaho Falls, Idaho USA Bradley Dept. of Electrcal and Computer Engneerng Grado Dept. of Industral and Systems Engneerng Vrgna Tech Blacksburg, Vrgna USA CTVR Trnty College Dubln, Ireland Abstract A current trend n spectrum regulaton s to ncorporate spectrum sharng through the desgn of spectrum access rules that support Dynamc Spectrum Access (DSA). Ths paper develops a decson-theoretc framework for regulators to assess the mpacts of dfferent spectrum access rules on both prmary and secondary operators. We analyze access rules based on sensng and excluson areas, whch n practce can be enforced through geolocaton databases. Our results show that recever-only sensng provdes nsuffcent protecton for prmary and co-exstng secondary users and overall low socal welfare. On the other hand, usng combnng sensng nformaton of the transmtter and recever of a communcaton lnk provdes dramatc ncreases n system performance. The performance of usng these lnk end ponts s relatvely close to that of usng many cooperatve sensng nodes assocated to the same access pont and large lnk excluson areas. These results are useful to regulators and network developers n understandng n developng rules for future DSA regulaton. I. Introducton Recent trends n regulatory paradgms have been shftng from the tradtonal command and control model of spectrum management to one of shared use through Dynamc Spectrum Access (DSA) [1 3]. Both the Federal Communcatons Commsson (FCC) and Ofcom have released ntal regulaton for unlcensed used of TV whte space devces, and the FCC, n a recent Notce of Inqury (NOI), s requestng nformaton on the vablty of DSA technques [3]. In ths effort, the FCC seeks to understand how spectrum access rules based on spectrum sensng and geolocaton databases can be useful n provdng secondary operators wth opportunstc access whle protectng prmary users from harmful nterference. Our paper apples decson theory to analyze the mpact of dfferent shared use spectrum access rules. Ths decson analyss provdes regulators wth a methodology to evaluate Ths work supported by the Idaho Natonal Laboratory (INL) through the Laboratory Drected Research & Development (LDRD) Program under DOE Idaho Operatons Offce Contract DE-AC07-05ID The Unted States Government retans and the publsher, by acceptng the artcle for publcaton, acknowledges that the Unted States Government retans a nonexclusve, padup, rrevocable, worldwde lcense to publsh or reproduce the publshed form of ths manuscrpt, or allow others to do so, for Unted States Government purposes. The vews and conclusons contaned n ths document are those of the authors and should not be nterpreted as representng the offcal polces, ether expressed or mpled, of the Department of Energy or the U.S. Government. spectrum access rules based on the resultng utlty to the varous consttuences that are vyng for spectrum. Our frst contrbuton n ths study s a mult-attrbute utlty model for evaluatng spectrum access rules. Ths mult-attrbute utlty model s based on the fundamental objectves of prmary and secondary operators deployng DSA. Through the Analytcal Herarchy Process (AHP) [4], we weght the relatve mportance between objectves, usng nput from experts n network performance. In our second contrbuton, we propose and evaluate nne spectrum access rules based on spectrum sensng and geolocaton databases. Here, we smulate a scenaro n whch a prmary operator shares spectrum wth a secondary operator. The results of the smulaton are then evaluated through our mult-attrbute utlty model and we calculate the utlty of each rule. Our paper s organzed as follows. Secton II defnes the anatomy of a spectrum access rule R. Secton III, descrbes the decson analytc framework. In ths secton, we propose an objectve herarchy, develop the mult-attrbute utlty model, and dentfy nne spectrum access rules that we set out to evaluate n the ensung analyss. Secton IV presents our system model, smulaton scenaro and the specfc mplementaton of our utlty model. In Secton V, we present the resultng utltes from our smulaton usng the mult-attrbute utlty model. We conclude ths paper, n Secton VI, wth the summary of our results and a dscusson future work. II. Spectrum Access Rules Recent regulatons n [1] and [2], for DSA descrbe the set of underlyng behavors (sub-rules) that DSA rados use to opportunstcally access prmary spectrum. These regulatons can be categorzed nto three sub-rules: excluson (E), channel assgnment (A), and power and transmsson control (P). Thus, when a DSA rado accesses the spectral medum t uses the spectrum access rule defned by the trplet R = {E, A, P}. The excluson sub-rule E dctates whch channels are avalable. Ths sub-rule has the objectve of preventng harmful nterference to prmary users. Excluson sub-rules determne that channels currently occuped by prmary users must be avoded by secondary users. They also determne how ths nterference avodance s to be accomplshed, e.g., through

2 spectrum sensng, or geolocaton databases. A thorough survey on spectrum sensng s presented n [5]. Addtonal rules for excluson also consdered by regulators support the adopton of cooperatve sensng for determnng channel avalablty [2, 3]. A survey of cooperatve sensng technques and ther tradeoffs have been captured n [6]. The channel assgnment sub-rule A determnes whch channels can be used for opportunstc communcatons, a subset of the channels not excluded by E. Regulaton n [1] and [2] does not specfy channel selecton algorthms; however, t suggests that devce manufactures could use sensng nformaton to select the best channel. If multple secondary operators or devces seek to smultaneously use common spectra, effcent channel assgnment wll be mportant for maxmzng channel reuse. A comparson of dfferent channel assgnment technques for DSA has been examned n [7]. The power and transmsson control sub-rule P dctates the maxmum allowable power lmts, transmt mask and technques for mnmzng power and nterference. Power control s necessary n DSA applcatons to mnmze nterference among co-channel and adjacent channel users. These rules nclude maxmum transmt power and Out of Band Emsson (OOBE) requrements such as adjacent channel attenuaton or Block Edge Mask (BEM) specfcatons. Power and transmsson sub-rules can also work n tandem wth excluson sub-rules. For nstance, the transmt power lmts set by a BEM may take nto account what systems currently operate n adjacent bands. Addtonally, the maxmum power lmts of the secondary transmtters can also be reduced dependng on the proxmty to excluson areas. Power control etquette schemes have been proposed n [8, 9]. III. Decson Analyss In ths secton, we apply technques from decson theory to explore regulators objectves n settng spectrum access rules. In any decson process, there are objectves that the decson maker wshes to accomplsh when selectng among alternatves. These objectves can be expressed as utlty functons, whch are used to evaluate the alternatves and make the fnal decson. In our formulaton, dstnct spectrum access rules are represented as alternatves avalable to the regulator, the decson maker, who must select the set of rules that maxmzes a defned measure of socal welfare. A. Objectve Herarchy The frst and most fundamental objectve of regulators s to enable telecommuncatons that are n the best nterest of the publc [1, 2]. In the context we consder here, ths goal mples that regulators seek to create spectrum access rules as a means to support new and mproved wreless servces. Through spectrum access rules, regulators seek to maxmze the spectrum effcency of underutlzed bands by accomplshng two sub-objectves: (1) mnmzng harmful nterference to prmary operators servce; and (2) maxmzng the utlty of DSA spectrum for secondary operators. Managng the tradeoffs between these two sub-objectves s the man challenge regulators face n evaluatng spectrum access rules. Ideally, spectrum access rules would allow for maxmum secondary usage wthout creatng any harmful nterference to prmary users. We equate maxmzng secondary operator utlty wth maxmzng servce revenue. Ths perspectve s ntutve because customers demonstrate utlty by payng for servces. If secondary operators cannot provde sutable servces wth DSA spectrum, customers wll not use these servces. Maxmzng revenue s captured n two sub-objectves: (1) maxmzng servce volume; and (2) mnmzng customer churn. Maxmzng servce volume equates to supportng the hghest volume of calls or data as possble. Customer churn s defned as the percentage of the customer base that leaves the servce provder annually, usually as a result of servce ssues such as dropped or blocked calls [10 12]. When several secondary compettve operators share the DSA spectrum smultaneously, competton for resources could result n droppng exstng servce requests or blockng future attempts. Therefore, we express customer churn as a functon of dropped or blocked servces. B. Utlty Model Genercally, a mult-attrbute utlty model can be expressed as: n u(x 1, x 2,..., x n ) = w u (x ), (1) where x s the measure for attrbute = 1,..., n, u ( ) s a sngle-attrbute utlty measure for attrbute scaled n the nterval [0,1], and w s the weght for measure, wth w = 1 [13]. Weghts represent the perceved mportance of a specfc utlty attrbute. Our mult-attrbute utlty model for evaluatng spectrum access rules, derved from the objectve herarchy, s gven by the followng equaton: u regulator = Prmary Operator Utlty { }} { w 1 u p drop =1 Secondary Operator Utlty { }} { + w 2 [ v 1 u } {{ s bts } + v 2 (z 1 u s drop + z 2 u s block )]. (2) } {{ } Servce Volume Customer Churn The utlty u p drop represents the total utlty for prmary operators and s based on the servce losses due to harmful secondary nterference. The secondary operator utlty s based on servce volume and customer churn. Servce volume s represented through the utlty u s bts. Customer churn s represented through the two utlty functons u s drop and u s block, whch consder the proportons of dropped and blocked servce attempts, respectvely. Weghts w, v, and z represent the relatve mportance between prmary and secondary operators utltes, servce volume and customer churn, and drops and blocks, respectvely. Weghts can be subjectve, however stakeholders should have nput to determne approprate values for the applcaton. The values of the w reflect the relatve mportance the reg-

3 ulator attaches to servces provded by prmary and secondary operators. As prmary user protecton and prortzaton s already reflected n the spectrum access rules (Secton II), our study s agnostc on the type of servce provded by prmary and secondary operators and thus we set w = w = 0.5. Ths value selecton reflects that secondary servce and prmary servce are equally valued. In other applcatons, the servce value between prmary and secondary servce could be examned from a monetary or socal beneft pont of vew to determne weghts. Determnng the value of v and z should be drven by network operator perspectves who provde secondary servces. We obtan values of v and z through a technque known as Analytcal Herarchy Process (AHP) [4]. Usng AHP, we ntervewed two experts to perform parwse comparsons between attrbutes of relevance to secondary operator performance n a DSA envronment 1. After determnng whch attrbute s more mportant, the more mportant attrbute receves a score from 1-9, wth 1 ndcatng that the two attrbutes are equally mportant. These parwse comparsons are placed n matrx A, wth a j = 1/a j, where each row and column represents a specfc attrbute. Usng the followng equaton: Aw = λ max w, (3) and solvng for λ max, the prncpal egenvalue of A, and w, the prncpal rght egenvector of A, we can normalze the entres of w by dvdng by ther sum and recover the weghted values for our utlty functon. We repeated the above process twce, wth nputs from our ntervews wth two experts n cellular network performance and obtaned two perspectves for the weghts v and z. We asked each expert to compare the relatve mportance of mnmzng customer churn versus maxmzng servce volume and mnmzng sesson blocks versus mnmzng sesson drops. The results of the ntervew are placed n a comparson matrx, from whch the prncpal egenvector s calculated. The results from ths calculaton and resultng weght values are shown n Table I. From Table I, we note that Expert A vews exstng servce requests as more mportant than new servce requests, whereas Expert B vews new and exstng servce requests as equvalent. C. Evaluated Spectrum Access Rules In ths study, we consder the ntent of the FCC rules n [1, 16] and focus on channel excluson technques that consder energy detecton thresholds and a geolocaton database that defnes excluson areas. For power control, we propose an adaptve power control algorthm to mnmze co-channel nterference and conserve transmt power based on [17]. Whle more sophstcated spectrum sensng technques exst, our evaluaton consders spectrum sensng based on energy detecton. Through energy detecton, a channel c s deemed avalable f the recever of lnk measures the receved power 1 Whle n ths paper we only examne only two vewponts, we also note that group decson makng and vewpont aggregaton has also been studed n [14, 15]. Expert A Blocks Drops z Blocks / Drops Expert A Max Servce Mn Churn v Max Servce / Mn Churn Expert B Max Servce Mn Churn v Max Servce / Mn Churn TABLE I Parwse comparson matrces derved from expert ntervews and utlty model weghts for z and v. Blocks and drops for Expert B are not captured snce z = z = 0.5. to be below the detecton threshold α, denoted as I (c) < α. Our study uses a detecton threshold of α = -107 dbm as ndcated n [1]. Usng sensng nformaton, our work consders the channel assgnment sub-rule through Least Interferng Channel (LIC) assgnment, where the LIC s determned by the mn c I (c). We also consder addtonal excluson rules usng cooperatve sensng wth hard combnng. In hard combnng, a channel s determned to be avalable f a certan proporton of the recevers detect power on the channel to be below the detecton threshold. Lke the rulng n [1], we defne channel excluson through excluson areas to protect prmary operators from secondary nterference. Secondary lnks n ths excluson model are not provded wth protecton areas and accept nterference from one another. We consder excluson areas surroundng prmary lnks as dsks of radus d,, where d, j s the Eucldean dstance between the transmtter of lnk and recever of lnk j. If prmary lnk s usng channel c, secondary lnk j s permtted to use channel c f d j, > d, and d, j > d,. The dstance d,, can also be extended by addtonal factors. In our study, we defne the excluson areas such that d j, > κd,, wth κ 1 and we explore values κ {1, 1.5, 2, 2.5}. Power control s necessary n DSA applcatons to mnmze nterference among co-channel lnks. Addtonally, changng network condtons requre adjustng transmt power to mantan the requste lnk Sgnal to Interference and Nose Rato (SINR). Thus, we consder dynamc power control for prmary and secondary users such that all lnks on channel c, L c, teratvely adjust ther transmt power accordng to: ( p (k + 1) = mn p max, β ) p (k), (4) γ where k s the teraton number, p s the power of transmtter, and γ s the SINR of lnk. Foschn n [17] demonstrated that when transmtters use equaton (4) to adjust ther power levels, the transmt powers of the lnks wll converge exponentally to an optmal power assgnment. In our case, optmal power assgnment means usng only the amount of power necessary to mantan requste lnk SINR. If the lnk cannot mantan an SINR of at least β wthout the transmt power of the lnk exceedng maxmum transmt power, p max, the lnk s nfeasble and the power of the lnk s set to zero. Table

4 Sensng Addtonal Channel Excluson Rule 1 Recever Only None Rule 2 Recever and Transmtter None Rule 3 Cooperatve Hard Combnng Rato =.25 Rule 4 Cooperatve Hard Combnng Rato =.5 Rule 5 Cooperatve Hard Combnng Rato = 1 Rule 6 Recever Only Excluson space factor = 1 Rule 7 Recever Only Excluson space factor= 1.5 Rule 8 Recever Only Excluson space factor= 2 Rule 9 Recever Only Excluson space factor= 2.5 TABLE II Spectrum access rules for smulatons. Rules use α = 107 dbm and Least Interferng Channel for channel assgnment. II summarzes the lst of the spectrum access rules that we consder n ths study. IV. System Model and Smulaton Scenaro In ths secton we ntroduce our system model, our smulaton scenaro and also our dervatons of the utlty functons. We adopt the SINR model for defnng nterference between co-channel lnks and determne whch lnks are feasble. The smulaton scenaro defnes how prmary and secondary users share spectrum. Ths secton closes wth a descrpton of how the utltes defned n Equaton (2) are calculated from the smulaton scenaros. A. System Model We defne a set of frequency channels C and a set of communcaton lnks L. Each lnk L comprses a transmtter and recever, whch seek to establsh a wreless communcatons lnk usng a channel c C. All c C have a bandwdth of W. Gven a set of communcatons lnks operatng on a channel c, L c, the SINR of the recever of lnk L c, γ (c), s determned by: γ (c) = g p, (5) N o + I (c) where g j s the gan between the transmtter of lnk j and the recever of lnk. The varable p denotes the power of the transmttng node of lnk, and N o the thermal nose. I (c) s the nterference power at the recever of lnk, expressed as: I (c) = g j p j. (6) j L c, j In ths system model, a feasble lnk s a lnk whose recever SINR, γ, s above a threshold β. We defne a sesson as a par of undrectonal lnks between communcatng nodes and denote γ as the SINR of the uplnk and ˆγ as the SINR of the downlnk. A sesson s feasble f and only f the par of lnks are both feasble,.e., γ β and ˆγ β. B. Smulaton Scenaro In our smulaton, we consder a scenaro n whch a set of prmary and secondary operators share a set of channels, C. Each operator comprses a set of access ponts and correspondng users assocated wth each access pont. Access ponts are assumed to have establshed a control channel to coordnate channel assgnment wth the users. The smulaton randomly places four prmary and four secondary operator access ponts each wth twenty users surroundng the assocated access ponts n a square smulaton area (1000m x 1000m). After placement of the access ponts, users for each access pont are randomly and unformly placed wthn a dstance D max = 700m from ther respectve access pont. We assume a nose floor (N o ) of -110 dbm, p re f = p max = 1W, path loss factor of 4, and ndependent Raylegh fadng. To show a lower bound n our gven scenaro, our smulaton consders a worst case scenaro, where every lnk carres traffc and attempts to be n servce smultaneously. Intally, prmary users are allowed to establsh sessons wth ther correspondng access ponts usng non-nterferng channels c C, wthout secondary users. Followng prmary users, secondary sessons attempt to be admtted nto the network ndvdually and at random. Admsson of secondary lnks begns by determnng channel avalablty usng excluson sub-rule E, followed by the channel assgnment sub-rule A, and then the power control and transmsson sub-rule P. In our scenaro, we use each of the excluson rules defned n Table II followed by LIC for channel assgnment. We also assume that nodes are capable of perfect spectrum sensng and there exsts a common control channel for exchange of sensng nformaton. After channel assgnment of lnk to channel c (the LIC of the recever) power control s ntated by the transmtter of secondary lnk wth ntal power parameter p re f. Lnks n L c then adjust ther transmt power usng equaton (4), untl the power settngs of L c converge. The smulaton tme, T s determned by usng the smulaton tme step when lnk admsson converges (no more lnks can be admtted) or when all lnks are attempted at least once, whchever comes last. Determnng T n ths manner was done to allow every lnk to be attempted and smulatons to converge. C. Utlty Dervatons The utlty u p drop s used to measure the prmary servce losses from harmful secondary nterference. When secondary lnks are admtted, they may cause other sessons to become nfeasble (harmful nterference) through lowerng the SINR of the co-channel lnks below β. If the transmtters of those lnks cannot mantan an SINR of β wthout exceedng p max, those lnks wll drop. Thus, u p drop s a lnear utlty functon that s zero f all sessons are dropped durng T and reaches a value of 1 f no prmary sessons are dropped. Smlarly, the utlty functon u s drop, s based on the number dropped sessons for the secondary operators n the same manner. However, n ths case the lnearly decreasng functon reaches zero when the percentage of secondary sesson drops reaches 4.5% [18]. We use 4.5% as an expected worst case. Thus, u s drop s a lnear utlty functon that s one f no secondary sessons are dropped and zero f more than 4.5% lnks are dropped. To develop u s bts, the utlty for the attrbute bt volume, we examne secondary lnk feasblty over a tme perod T. Ths s a lnear utlty functon that s zero f no secondary sessons are feasble and reaches a value of 1 f all possble secondary sesson are feasble durng tme perod T.

5 Rule Comparson Utlty for Expert A Rule Comparson Utlty for Expert B Utlty 0.6 Utlty u p drop u p drop 0.2 u s bts u s drops 0.2 u s bts u s drops u s blocks u s blocks Rule Number Rule Number Fg. 1. Utlty comparson of spectrum access rules usng attrbute weghts from Expert A. Rule 10 shows utopa pont for perspectve. The utlty u s block represents the measure of blocked secondary sesson admssons. Durng admsson, secondary users can be blocked for two reasons. Frst, the channel could be unavalable because of excluson,.e., the spectrum access subrule E prevents the channel from beng used. Second, the requste SINR of the lnk may not be reached because of excessve co-channel nterference. Ths lnear utlty functon s zero f all secondary lnks are blocked and reaches a value of 1 f no lnks are blocked. V. Results Fgures 1 and 2 show the resultng utlty for attrbute weghts from Expert A and Expert B, respectvely. Rule 10 represents the utopa pont, the maxmum utlty due to each attrbute f there were no conflcts between prmary and secondary user objectves. The most salent feature n both Fgures 1 and 2 s the poorest performng rule, Rule 1. Placng ths n context, Rule 1 s excluson sub-rules are based only on recever sensng,.e. channels are avalable f the measured power s below a detecton threshold. The poor performance s due to two reasons. Frst, Rule 1 provdes nsuffcent protecton for prmary users, causng a relatvely large proporton of prmary sessons to drop,.e. low utlty for u p drop. Second, Rule 1 also causes many secondary lnks to drop,.e. low utlty for u s drop. Thus, Rule 1, the receveronly sensng excluson sub-rule, provdes the least amount of utlty for prmary and secondary operators, at the expense of admttng a large number of new sessons. Fgures 1 and 2 show that Rules 2, 5, and 9 result n the hghest overall utltes. Rule 2 and Rule 5 both use cooperatve sensng by hard combnng. Rule 2 requres the transmtter of the lnk to sense the LIC of the recever to be below the detecton threshold. Compared to recever-only sensng, ths s a dramatc mprovement n provdng protecton from harmful nterference to both prmary and secondary users,.e. hgh u s drop and u p drop. Rule 5 only allows channels to be avalable Fg. 2. Utlty comparson of spectrum access rules usng attrbute weghts from Expert B. Rule 10 shows utopa pont for perspectve. f all users assocated to the same access pont sense the LIC of the recever to be below the detecton threshold, provdng the best protecton for prmary and secondary users,.e. u s drop and u p drop. However, Rule 5 provdes very lttle throughput for secondary users,.e. low u s bts and low u s block, by beng overly conservatve about opportunstc use. Rule 9 provdes the largest excluson regons, allowng for sgnfcant amount of protecton for prmary users through preventng secondary use. In summary, based on the two weghted values from Experts A and B, Rules 2, 5, and 9 have relatvely equvalent overall utltes. It s also evdent that as secondary users are less restrcted, and acheve hgher u s bts and u s block, they can cause more servce drops to both prmary and secondary users and thus lower overall utltes. Ths phenomenon s magnfed n the utlty functon snce throughput, u s bts s less mportant n the perspectve of both experts. We next examne the causes of blockng. Fgure 3 shows the cumulatve number of blocks for each rule durng the smulaton and classfy them accordng to the cause of sesson blockng. An SINR Block s a sesson block caused because the SINR of the lnk cannot be reached. SINR Blocks are a result of lmted transmtter power or hgh levels of nterference. Channel Access Blocks are due to the excluson subrule, preventng access to the spectrum. From Fgure 3, the rules wth the hghest utltes also have the hghest proportons of channel access blocks. One could also argue that Channel Access Blocks are less expensve than SINR Blocks, n that on the latter the transmtter must use energy and tme to attempt to access the channel. VI. Concluson and Future Work Applyng decson theory, ths study created a mult-attrbute utlty model for evaluatng dfferent dynamc spectrum access rules. Usng ths utlty model, we developed a scenaro for evaluatng network performance of both secondary and prmary operators and evaluated dfferent spectrum access rules.

6 Count Rule Comparson Blockng by Type SINR Block Channel Access Block Rule Number Fg. 3. Blockng by type of rules. Rules 2, 5, and 9 have the hghest proportons of channel access blocks. We consdered rules based on sensng and excluson spaces. Our results show that as secondary users are less restrcted, they can cause servce drops to both prmary and secondary users and thus overall lower utltes. Addtonally, rules wth hgher proporton of channel access blocks result n the largest overall utltes. Future work wll examne the rsk preferences of regulators and secondary network operators. Ths paper consdered rskneutral operators and regulators, whch led us to use lnear utlty functons. A new mult-attrbute utlty wth rsk-prone or rsk-averse preferences would result n convex or concave utlty functons, respectvely [19]. Addtonally, t s not clear that the attrbutes throughput and customer churn should be modeled as beng ndependent. For nstance, f there s no throughput there are no drops due to the non-exstence of lnks performng servce. Thus, nvestgaton nto a utlty model whch s not utlty-ndependent should also be examned. In concluson, ths work provdes a frst look n developng a decson analyss framework for regulators to evaluate the techncal merts of spectrum access rules. References [1] Federal Communcatons Commsson, In the Matter of: Unlcensed Operaton n the TV Broadcast Bands (ET Docket No ) and Addtonal Spectrum for Unlcensed Devces Below 900 MHz and n the 3 GHz Band (ET Docket No ), FCC :Second Memorandum Opnon and Order, September [2] Ofcom, Implementng geolocaton, Consultaton, November [3] Federal Communcatons Commsson, :In the Matter of: Promotng More Effcent Use of Spectrum Through Dynamc Spectrum Use Technologes (ET Docket No ), FCC : Notce of Inqury, November [4] T. Saaty, How to make a decson: the analytc herarchy process, European Journal of Operatonal Research, vol. 48, pp. 9 26, [5] T. Yucek and H. Arslan, A survey of spectrum sensng algorthms for cogntve rado applcatons, IEEE Communcatons Surveys Tutorals, [6] S. Mshra, A. Saha, and R. Brodersen, Cooperatve sensng among cogntve rados, IEEE Internatonal Conference Communcatons (ICC), vol. 4, [7] J. Deaton, S. Ahmad, U. Shukla, R. Irwn, L. DaSlva, and A. MacKenze, Evaluaton of dynamc channel and power assgnment for cogntve networks, Sprnger Journal on Wreless Personal Communcatons, vol. 57, [8] D. Satapathy and J. Peha, A novel co-exstence algorthm for unlcensed fxed power devces, IEEE Wreless Communcatons and Networkng Conference (WCNC), vol. 3, [9], A novel co-exstence algorthm for unlcensed varable power devces, IEEE Internatonal Conference on Communcatons (ICC), [10] J.D. Power and Assocates, Incdence of Dropped Calls Increases Consderably among Customers Who Are Most Lkely to Swtch Wreless Provders, September [11] M. Lombardo, personal communcaton. [12] M. Shomaker, personal communcaton. [13] W. Edwards, R. Mles, and D. Von Wnterfeldt, Advances n Decson Analyss. Cteseer, [14] E. Forman and K. Penwat, Aggregatng ndvdual judgments and prortes wth the analytc herarchy process, Elsever European Journal of operatonal research, vol. 108, no. 1, [15] R. Ramanathan and L. Ganesh, Group preference aggregaton methods employed n AHP: An evaluaton and an ntrnsc process for dervng members weghtages, Elsever European Journal of Operatonal Research, vol. 79, no. 2, [16] Federal Communcatons Commsson, In the Matter of: Unlcensed Operaton n the TV Broadcast Bands (ET Docket No ) and Addtonal Spectrum for Unlcensed Devces Below 900 MHz and n the 3 GHz Band (ET Docket No ), FCC : Second Report and Order and Memorandum Opnon and Order, November [17] G. Foschn and Z. Mljanc, Dstrbuted autonomous wreless channel assgnment algorthm wth power control, IEEE Transactons on Vehcular Technology, vol. 44, no. 3, pp , August [18] A. Golub and P. Carton, The Battle Contnues Among Wreless Industry Leaders, ChangeWave Area Report, May [19] R. Keeney and H. Raffa, Decsons wth multple objectves. Cambrdge Unversty Press, 1993.