An Application-Aware Spectrum Sharing Approach for Commercial Use of 3.5 GHz Spectrum

Transcription

1 An Applcaton-Aware Spectrum Sharng Approach for Commercal Use of 3.5 GHz Spectrum Haya Shajaah, Ahmed Abdelhad and Charles Clancy Bradley Department of Electrcal and Computer Engneerng Hume Center, Vrgna Tech, Arlngton, VA, 2223, USA {hayajs, aabdelhad, arxv:5.797v [cs.ni] 27 Oct 25 Abstract In ths paper, we ntroduce an applcaton-aware spectrum sharng approach for sharng the Federal under-utlzed 3.5 GHz spectrum wth commercal users. In our model, users are runnng elastc or nelastc traffc and each applcaton runnng on the user equpment (UE) s assgned a utlty functon based on ts type. Furthermore, each of the small cells users has a mnmum requred target utlty for ts applcaton. In order for users located under the coverage area of the small cells enodebs, wth the 3.5 GHz band resources, to meet ther mnmum requred qualty of experence (QoE), the network operator makes a decson regardng the need for sharng the macro cell s resources to obtan addtonal resources. Our objectve s to provde each user wth a rate that satsfes ts applcaton s mnmum requred utlty through spectrum sharng approach and mprove the overall QoE n the network. We present an applcaton-aware spectrum sharng algorthm that s based on resource allocaton wth carrer aggregaton to allocate macro cell permanent resources and small cells leased resources to UEs and allocate each user s applcaton an aggregated rate that can at mnmum acheves the applcaton s mnmum requred utlty. Fnally, we present smulaton results for the performance of the proposed algorthm. Index Terms Applcaton-Aware, Spectrum Sharng, Resource Allocaton wth Carrer Aggregaton, 3.5 GHz Band I. INTRODUCTION The demand for wreless broadband capacty has been recently growng much faster than the avalablty of new spectrum. Because of the ncreasng demand for spectrum by commercal wreless operators, federal agences are now wllng to share ther spectrum wth commercal users. The Commsson and the Presdent have outlned a path to double the avalable spectrum for wreless broadband use, the Presdents Councl of Advsors on Scence and Technology (PCAST) Report dentfes two technologcal advances to ncrease wreless broadband capabltes. Frst, ncreasng the deployment of small cell networks and second usng spectrum sharng technology. The 3.5 GHz Band s an deal band for small cell deployments and shared spectrum use because of ts smaller coverage. The Natonal Insttute of Standards and Technology (NTIA) Fast Track Report [] dentfed the 3.5 GHz Band for potental shared federal and non-federal broadband use. Ths band s very favorable for commercal cellular systems such as LTE-Advanced systems. Small cells are low-powered wreless base statons desgned to play well wth macro networks n a heterogeneous network (HetNet). Small cells are backed up by a macro cell layer of coverage so that f a small cell shuts down n the 3.5 GHz shared band, operators can pck up coverage agan n the macro network. Makng the under-utlzed federal spectrum avalable for secondary use ncreases the effcency of spectrum usage and can provde sgnfcant gan n moble broadband capacty f those resources are aggregated effcently wth the exstng commercal moble systems resources. Many operators are wllng to take advantage of the LTE-Advanced carrer aggregaton feature whch was ntroduced by 3GPP release [2]. Ths feature allows users to employ multple carrers to ensure a wder bandwdth, by aggregatng multple non-contnuous or contnuous component carrers (CCs), and therefore acheve hgher capacty and better performance. In [3], the authors have ntroduced a resource allocaton (RA) optmzaton framework based on carrer aggregaton (CA). The proposed mult-stage resource allocaton algorthm allocates the prmary and secondary carrers resources optmally among users. The fnal optmal rate allocated to each user s the aggregated rate. In ths paper, we ntroduce an applcaton-aware spectrum sharng approach for cellular networks sharng the federal under-utlzed 3.5 GHz spectrum. In our model, the small cells, wth the under-utlzed 3.5 GHz spectrum resources, are located wthn the coverage area of a macro cell. The network operator makes a decson regardng the need for sharng the macro cell s enodeb resources wth small cells users based on the small cell users demand for spectrum resources. We use utlty proportonal farness approach to guarantee a mnmum qualty of servce (QoS) for each user. In our proposed model, small cells users have a mnmum requred utlty value for each of ther applcatons. The network operator decdes to share the macro cell s enodeb resources f the value of any of small cell user s applcaton utlty functon of ts allocated rate,.e. allocated by the small cell s enodeb, does not exceed the user s applcaton mnmum requred utlty value. II. RELATED WORK Carrer aggregaton enables concurrent utlzaton of multple component carrers wth dfferent propagaton characterstcs [4], [5]. Due to the sgnfcant features of CA, an approprate CA management s essental to enhance the performance of cellular networks. A tractable mult-band mult-ter CA models for HetNets are proposed n [6]. Two models are consdered: mult-flow CA and sngle-flow CA, each UE performs cell

2 selecton based on the reference sgnal s mum receved power. A major concern about deployng small cells s ther small coverage areas and low transmt power. The authors n [7], [8] have addressed ths ssue and suggested basng to allow small cells to expand ther coverage areas. Most of the prevous research work have focused on fndng resource allocaton approaches for ntra-system and ntraoperator of a sngle network operator. However, current research on resource allocaton are for more complex network topologes [9], []. Carrer aggregaton n networks that nvolve multple network operators n HetNets need to be further nvestgated. In [], the authors have analyzed the performance of ther proposed carrer aggregaton framework that combnes a statcally assgned spectrum wth spectrum resources from a shared spectrum pool. The RA optmzaton problem can be transformed nto a utlty mzaton problem to mze the user s satsfacton rather than the system throughput, where the user s satsfacton s represented as a functon of the acheved data rate [2]. In [3], [3], [4], we have proposed a multple stage RA wth CA algorthms that use utlty proportonal farness approach to allocate the prmary and the secondary carrers resources optmally among moble users n ther coverage area. However, these algorthms consder optmzaton problems that solve for the allocated rates from the prmary and secondary carrers wthout gvng the user or the network operator the flexblty to decde on the amount of recourses to be allocated to the user by secondary carrers. In ths paper, we address ths ssue and desgn a RA wth CA model that accounts for the users demand of resources and controls whch users are requred to be allocated addtonal resources from the secondary carrers. Ths s mportant for users who do not wsh to pay hgher prce for more resources f they can be satsfed wth certan rates (.e. rates that guarantee certan degree of satsfacton represented by utlty values). A. Our Contrbutons Our contrbutons n ths paper are summarzed as: We present a spectrum sharng approach for sharng the Federal under-utlzed 3.5 GHz spectrum wth commercal users. We present a spectrum sharng algorthm that s based on resource allocaton wth CA to allocate the small cells under-utlzed 3.5 GHz resources to small cells users and allocate the macro cell s resources to both macro cell s users and small cell s users that dd not reach ther applcatons mnmum requred utltes by the small cells allocated rates. We present smulaton results for the performance of the proposed resource allocaton algorthm. The remander of ths paper s organzed as follows. Secton III presents the problem formulaton. In secton IV, we present resource allocaton optmzaton problems that solve for the macro cell and small cells allocated rates. Secton V presents our proposed resource allocaton algorthm. In secton VI, we dscuss smulaton setup and provde quanttatve results along wth dscusson. Secton VII concludes the paper. III. PROBLEM FORMULATION We consder LTE-Advanced moble system consstng of a macro cell, referred to by the ndex B, wth a coverage radusd B, that s overlad wths small cells. The macro cell s enodeb s confgured at the LTE-Advanced carrer and the small cell s enodeb s confgured to use the 3.5 GHz underutlzed spectrum band. Let S denotes the set of small cells located wthn the coverage area of the macro cell B where S = S. All small cells are connected to the core network. The small cells are assumed to have a closed access scheme where only regstered UEs, referred to by SUEs, are served by the small cells enodebs. On the other hand, all UEs under the coverage area of the macro cell B and not wthn the coverage of any small cell, referred to by MUEs, are served by the macro cell s enodeb. The set of all MUEs under the coverage area of macro cell B s referred to by µ. The set of SUEs assocated to small cell s s referred to by Q s. We assume that the assocaton of the UEs wth ther enodebs remans fxed durng the runtme of the resource allocaton process. We have S s= Q s = Θ and S s= Q s =. Each SUE has a mnmum QoE requrement for ts applcatons that s represented by the utlty of the user s applcaton wth ts allocated rate. Letu req denotes the mnmum requred utlty of SUE Θ. Utlty functons are used to express the user satsfacton wth ts allocated rate [5] [8]. The th user applcaton utlty functon of ts allocated rate r s gven by U (r ) where U s a sgmodal-lke functon used to represent real tme applcatons or logarthmc functon used to represent delay tolerant applcatons. These utlty functons have the followng propertes: U () = and U (r ) s an ncreasng functon of r. U (r ) s twce contnuously dfferentable n r and bounded above. In our model, we use normalzed sgmodal-lke utlty functons, as n [9], that are expressed as ( ) U (r ) = c +e d a(r b), () where c = +ea b and d e a b = so t satsfes U +e a b () = and U ( ) =. The normalzed sgmodal-lke functon has an nflecton pont at r nf = b. In addton, we use the normalzed logarthmc utlty functon, used n [9], that are expressed as U (r ) = log(+k r ) log(+k r ), (2) wherer gves% utlzaton andk s the rate of ncrease of utlty percentage wth allocated rates that vares based on the user applcaton. So, t satsfes U () = and U (r ) =. Fgure shows a heterogeneous network that conssts of one macro cell wth one enodeb and two small cells wthn the coverage area of the macro cell, each of the small cells has one enodeb that s confgured to use the 3.5 GHz underutlzed spectrum. Moble users under the coverage of the macro cell and the small cells are runnng real tme or delay

3 Delay tolerant App Log Real tme App Sg C2 Small cell C Log Macro cell C3 Small cell Fg.. System model for a LTE-Advanced moble system wth one macro cell and two small cells wthn the coverage area of the macro cell. Each of the small cells s confgured to use the 3.5 GHz under-utlzed spectrum. tolerant applcatons that are represented by sgmodal-lke or logarthmc utlty functons, respectvely. IV. RESOURCE ALLOCATION OPTIMIZATION FOR SPECTRUM SHARING WITH THE 3.5 GHZ SPECTRUM In ths secton, we present a resource allocaton framework for cellular networks sharng the federal under-utlzed 3.5 GHz spectrum. In our model, SUEs are allocated resources from the leased under-utlzed 3.5 GHz resources at the small cells enodebs whereas MUEs are allocated resources only by the macro cell s enodeb. Each of the SUEs has a mnmum requred utlty u req for each of ts applcatons. Frst the small cell s enodeb allocates ts avalable leased resources then the network operator decdes whch SUEs stll requre addtonal resources n order to acheve ther mnmum requred utltes and allocate them more resources from the macro cell enodeb based on a resource allocaton wth carrer aggregaton optmzaton problem. The resource allocaton process starts by allocatng each of the small cells resources to SUEs under t coverage area. We use a utlty proportonal farness resource allocaton optmzaton problem to allocate the small cell resources. The RA optmzaton problem of the small cell s s gven by: r s subject to Q s = Q s U (r s ) r s R s = r s R s, =,2,..., Q s, where r s = {r s,rs 2,...,rs Q s }, Q s s the number of SUEs under the coverage area of the small cell s and R s s the mum achevable rate of the under-utlzed 3.5 GHz leased spectrum avalable at the enodeb of small cell s. The resource allocaton objectve functon s to mze the entre small cell utlty when allocatng ts resources. It also acheves proportonal farness among utltes such that non (3) of the SUEs wll be allocated zero resources. Therefore, a mnmum QoS s provded to each SUE. Ths approach gves real tme applcatons prorty when allocatng the small cell resources. The objectve functon n optmzaton problem (3) Qs s equvalent to r s = logu (r s ). Optmzaton problem (3) s a convex optmzaton problem and there exsts a unque tractable global optmal soluton [9]. From optmzaton problem (3), we have the Lagrangan: Q s Q s L s (r s,p s ) = ( logu (r)) p s s ( r s +z s R s ) (4) = where z s s the slack varable and p s s the Lagrange multpler whch s equvalent to the shadow prce that corresponds to the servce provder s prce per unt bandwdth for the small cell resources [9]. The soluton of equaton (3) s gven by the values r s that solve equaton logu(rs ) r = p s and are the ntersecton of the s tme varyng shadow prce, horzontal lne y = p s, wth the curve y = logu(rs ) r geometrcally. Once the RA process s s performed by the small cell s, each SUE n Q s wll be allocated = r s rate. However, the network operator decdes f any of the SUEs requres addtonal resources n order to reach the mnmum requred utlty u req of ts applcaton by comparng the utlty of the small cell allocated rate that s gven by U ( ) wth the value u req. If the acheved utlty = for certan SUE s less that the mnmum requred utlty, the network operator requests addtonal resources from the macro cell for that SUE. The small cell s enodeb creates a set Q sb of all SUEs that needs to be allocated addtonal resources where Q sb = {SUEs Q s s.t. u req > U ( )}. Once each small cellswthn the coverage area of the macro cell B performs ts RA process based on optmzaton problem (3), the macro cell starts allocatng ts resources to all MUEs wthn ts coverage area as well as the SUEs that were reported, by the network operator, for ther need of addtonal resources. Let Q be the set of SUEs that wll be allocated addtonal resources by the macro cell where Q = S s= Q sb. The set of UEs that wll be served by the macro cell s enodeb;.e. partcpate n the macro cell RA process, s gven by β where β = µ Q. The resource allocaton optmzaton problem of the macro cell B s gven by: r subject to U (r +C ) = r R B = C = { f UE / Q f UE Q r R B, =,2,...,, where r = {r,r 2,...,r }, s the number of UEs that wll be be served by the macro cell s enodeb and R B s the mum achevable rate of the resources avalable at the macro cell s enodeb. The resource allocaton objectve functon s to mze the entre macro cell utlty when allocatng (5)

4 ts resources. The RA optmzaton problem (5) s based on carrer aggregaton. It seeks to mze the multplcaton of the utltes of the rates allocated to MUEs by the macro cell s enodeb and the utltes of the rates allocated to the SUEs n β by small cells enodebs and macro cell s enodeb. Utlty proportonal farness s used to guarantee that non of the UEs wll be allocated zero resources. Real tme applcatons are gven prorty when allocatng the macro resources usng ths approach. The objectve functon n optmzaton problem = logu (r + C ). Optmzaton (5) s equvalent to r problem (5) s a convex optmzaton problem and there exsts a unque tractable global optmal soluton [9]. From optmzaton problem (5), we have the Lagrangan: L B (r,p B ) = ( logu (r +C )) p B ( r +z B R B ) = (6) where z B s the slack varable and p B s the Lagrange multpler whch s equvalent to the shadow prce that corresponds to the servce provder s prce per unt bandwdth for the macro cell resources [9]. The soluton of equaton (5) s gven by the values r that solve equaton logu(r+c) r = p B and are the ntersecton of the tme varyng shadow prce, horzontal lne y = p B, wth the curve y = logu(r+c) r geometrcally. Once the macro cell enodeb s done performng the RA process based on optmzaton problem (5), each UE n β wll be allocated = r +C rate. = V. THE MACRO CELL AND SMALL CELLS RA OPTIMIZATION ALGORITHM In ths secton, we present our resource allocaton algorthm. The proposed algorthm conssts of SUE, MUE, small cell enodeb and macro cell enodeb parts shown n Algorthm, 2, 3 and 4, respectvely. The executon of the algorthm starts by SUEs and MUEs, subscrbng for moble servces, transmttng ther applcatons utltes parameters to ther correspondng enodebs. Frst, each small cell s enodeb calculates ts allocated rate to each SUE n Q s. It then checks whether the achevable utlty of that rate s less or greater than the SUE s mnmum requred utlty u req. If for any SUE U ( ) < u req, the small cell s enodeb sends the applcaton parameters and the allocated rate for that SUE to the macro cell s enodeb requestng addtonal resources. Otherwse, t allocates the rate to that SUE. Once the macro cell s enodeb receves the set Q sb from each small cell n S wthn ts coverage area. It starts the RA process to allocate ts avalable resources to each UE n β based on a RA wth carrer aggregaton optmzaton problem. Once the RA process of the macro cell s performed, the macro cell allocates rate = r +C to the th UE n β. VI. SIMULATION RESULTS Algorthm, 2, 3 and 4 were appled n C++ to multple utlty functons wth dfferent parameters. Smulaton results showed convergence to the global optmal rates. In ths secton, we consder a macro cell wth one enodeb. Wthn the the Algorthm The th SUE Q s Algorthm Send applcaton utlty parameters k, a, b, r and u req to the SUE s n band small cell s enodeb. Receve the fnal allocated rate from the small cell s enodeb or from the macro cell s enodeb. Algorthm 2 The th MUE µ Algorthm Send applcaton utlty parameters k, a, b and r to the macro cell s enodeb. Receve the fnal allocated rate from the macro cell s enodeb. Algorthm 3 Small Cell s enodeb Algorthm Intalze Q sb = ; =. Receve applcaton utlty parameters k, a, b, r u req from all SUEs n Q s. and Solve r s Qs = arg r s = logu (r s) ps ( Q s = (rs ) R s ). Let = r s be the rate allocated by the s small cell s enodeb to each user n Q s. Calculate the SUE utlty U ( ) Q s for SUE to Q s do f U ( ) < u req then Q sb = Q sb SUE{} Send SUE parameters k, a, b, r and to the macro cell s enodeb else Allocate rate = to SUE end f end for Algorthm 4 The Macro Cell s enodeb Algorthm Intalze C = ; =. for s to S do Receve applcaton utlty parameters k, a, b, r and fo SUEs n Q sb from small cell s enodeb. C = Q sb end for Create user group Q = S s= Q sb Create user group β = µ Q Solve r = arg r p B ( = (r ) R B ). Allocate = r +C to each UE n β = logu (r + C )

5 TABLE I USERS AND THEIR APPLICATIONS UTILITIES User s Index User s Type Applcatons Utltes Parameters UE = {} SUE Sg2 a = 3, b = 2, u req =.8 UE2 = {2} SUE Sg3 a =, b = 3, u req =.8 UE3 = {3} SUE Log2 k = 3, r =, u req =.5 UE4 = {4} SUE Log3 k =.5, r =, u req =.5 UE5 = {5} MUE Sg a = 5, b = UE6 = {6} MUE Sg3 a =, b = 3 UE7 = {7} MUE Log k = 5, r = UE8 = {8} MUE Log3 k =.5, r = U(r) Sg Sg2 Sg3 Log Log2 Log r Fg. 2. The users utlty functons U (r ) used n the smulaton (three sgmodal-lke functons and three logarthmc functons) R s (a) The rates allocated by the small cell s enodeb to users n Q s wth < R s <. U( ) R s =5 R s =7 UE UE2 UE3 UE4 (b) Users QoE represented by the utlty of user s applcaton of ts allocated rate U ( ) when R s = 5 and R s = 7. Fg. 3. The small cell s enodeb allocated rates wth < R s < and users QoE when R s = 5 and R s = 7. coverage area of the macro cell there exsts one small cell s. Four SUEs are located under the coverage area of the small cell s wth UEs ndexes {,2,3,4}. The SUEs user group s gven by Q s = {,2,3,4}. Four MUEs are located under the coverage area of the macro cell s enodeb but not wthn the small cell. The MUEs user group s gven by µ = {5,6,7,8}. Each UE whether t s SUE or MUE s runnng ether real tme applcaton or delay tolerant applcaton. Each of the SUEs applcatons utltes has a mnmum requred utlty that s gven by u req that s equvalent to the C value for that user whereas MUEs do not have mnmum requred utltes for ther applcatons. The UEs ndexes, types and applcatons utltes parameters are lsted n table I. Fgure 2 shows the sgmodal-lke utlty functons and the logarthmc utlty functons used to represent the SUEs and MUEs applcatons. A. Small Cell Allocated Rates and Users QoE In the followng smulatons, the small cell s carrer total rate R s takes values between and wth step of. In Fgure 3, we show the small cell s allocated rates for users n Q s wth dfferent values of the small cell s carrer total rate R s and the users QoE wth the small cell allocated rates when R s = 5 and R s = 7. In Fgure 3(a), we show that users runnng real tme applcatons are gven prorty when allocatng the small cell s resources due to ther sgmodal-lke utlty functon nature. We also observe that non of the UEs s allocated zero resources because we used a utlty proportonal farness approach. We also show how the proposed rate allocaton algorthm converges for dfferent values of R s. In Fgure 3(b), we show the QoE for the four SUEs whch s represented by ther applcatons utltes of the small cell allocated rates U ( ) when R s = 5 and R s = 7. We notce that n the case of R s = 5, the utltes of the small cell allocated rates for UE2, UE3 and UE4 dd not reach the mnmum requred utltes for these SUEs whereas n the case of R s = 7 the utlty of the small cell allocated rate for UE4 dd not reach the mnmum requred utlty for that SUE. Therefore, based on the proposed algorthm the network operator wll request addtonal resources for these UEs from the macro cell s enodeb and these UEs wll be allocated addtonal resources based on a resource allocaton wth carrer aggregaton scenaro. B. Macro Cell Allocated Rates and Users QoE In the followng smulatons, the macro cell s carrer total rate R B takes values between and wth step of and R s s fxed at 5. As dscussed n VI-A, n the case of R s = 5 the network operator requests addtonal resources for three SUEs (.e. UEs n Q sb = {2,3,4}) as they dd not reach ther mnmum requred utltes. Therefore, the macro cell s enodeb performs a resource allocaton wth carrer aggregaton process to allocate resources to the UEs n user group β where β = {2,3,4,5,6,7,8}. In Fgure 4, we show the fnal allocated rates for the UEs n β and these users

6 r all = r +b R B (a) The aggregated rates = r + C allocated by the macro cell s enodeb to users n β when R s = 5. U(r all ) UE2 UE3 UE4 UE5 UE6 UE7 UE8 (b) Users QoE represented by the utlty of user s applcaton of ts allocated rate U ( ) when R B = 8 and R s = 5. Fg. 4. The total aggregated rates = r + C allocated by the macro cell s enodeb to users n β wth < R B < when R s = 5 and the users QoE when R B = 8 and R s = 5. QoE wth the fnal allocated rates when R B = 8. In Fgure 4(a), we show the macro cell s fnal allocated rates converges for dfferent values of R B. Agan we observe that non of the users s allocated zero resources and that real tme applcatons are gven prorty when allocatng the macro cell s resources. In Fgure 4(b), we show the QoE for the seven UEs n β whch s represented by ther applcatons utltes of the fnal allocated rate U () when R s = 5 and R B = 8. We notce that the utltes of the fnal allocated rates for the three SUEs nq sb (.e. UE{2,3,4}) exceed the mnmum requred utltes for these SUEs because of the addtonal resources allocated to these users by the macro cell s enodeb. VII. CONCLUSION In ths paper, we proposed a spectrum sharng approach for sharng the Federal under-utlzed 3.5 GHz spectrum wth commercal users. We used sgmodal-lke utlty functons and logarthmc utlty functons to represent real tme and delay tolerant applcatons, respectvely. We presented resource allocaton optmzaton problems that are based on carrer aggregaton. The proposed resource allocaton algorthm ensures farness n the utlty percentage. Users located under the coverage area of the small cells are allocated resources by the small cells enodebs whereas both the macro cell users and the small cells users that dd not reach ther mnmum requred utltes by ther small cells allocated rates are allocated resources by the macro cell s enodeb based on carrer aggregaton. We showed through smulatons the the proposed algorthm converges to the optmal rates. We also showed that small cells users can acheve ther mnmum requred QoE by usng the proposed spectrum sharng approach. REFERENCES [] Natonal Telecommuncatons and Informaton Admnstraton (NTIA), An assessment of the near-term vablty of accommodatng wreless broadband systems n the MHz, MHz, MHz, MHz, and MHz bands (Fast Track Report). Onlne, October 2. [2] G. Yuan, X. Zhang, W. Wang, and Y. Yang, Carrer aggregaton for LTE-advanced moble communcaton systems, n Communcatons Magazne, IEEE, vol. 48, pp , 2. [3] H. Shajaah, A. Abdel-Had, and C. Clancy, Utlty proportonal farness resource allocaton wth carrer aggregaton n 4g-lte, n Mltary Communcatons Conference, MILCOM IEEE, pp , Nov 23. [4] S. Parkvall, E. Dahlman, A. Furuskar, Y. Jadng, M. Olsson, S. Wanstedt, and K. Zang, Lte-advanced - evolvng lte towards mt-advanced, n Vehcular Technology Conference, 28. VTC 28-Fall. IEEE 68th, pp. 5, Sept 28. [5] X. Ln and H. Vswanathan, Dynamc spectrum refarmng wth overlay for legacy devces, Wreless Communcatons, IEEE Transactons on, vol. 2, pp , October 23. [6] X. Ln, J. Andrews, R. Ratasuk, B. Mondal, and A. Ghosh, Carrer aggregaton n heterogeneous cellular networks, n Communcatons (ICC), 23 IEEE Internatonal Conference on, pp , June 23. [7] A. Ghosh, N. Mangalvedhe, R. Ratasuk, B. Mondal, M. Cudak, E. Vsotsky, T. Thomas, J. Andrews, P. Xa, H. Jo, H. Dhllon, and T. Novlan, Heterogeneous cellular networks: From theory to practce, Communcatons Magazne, IEEE, vol. 5, pp , June 22. [8] A. Damnjanovc, J. Montojo, Y. We, T. J, T. Luo, M. Vajapeyam, T. Yoo, O. Song, and D. Mallad, A survey on 3gpp heterogeneous networks, Wreless Communcatons, IEEE, vol. 8, pp. 2, June 2. [9] T. Yang, L. Zhang, and L. Yang, Cogntve-based dstrbuted nterference management for home-enb systems wth sngle or multple antennas, n Personal Indoor and Moble Rado Communcatons (PIMRC), 2 IEEE 2st Internatonal Symposum on, pp , Sept 2. [] A. Attar, V. Krshnamurthy, and O. Gharehshran, Interference management usng cogntve base-statons for umts lte, Communcatons Magazne, IEEE, vol. 49, pp , August 2. [] J. McMenamy, I. Macaluso, N. Marchett, and L. Doyle, A methodology to help operators share the spectrum through an enhanced form of carrer aggregaton, n Dynamc Spectrum Access Networks (DYSPAN), 24 IEEE Internatonal Symposum on, pp , Aprl 24. [2] Z. Jang, Y. Ge, and Y. L, Max-utlty wreless resource management for best-effort traffc, Wreless Communcatons, IEEE Transactons on, vol. 4, pp., Jan 25. [3] H. Shajaah, A. Khawar, A. Abdel-Had, and T. Clancy, Resource allocaton wth carrer aggregaton n lte advanced cellular system sharng spectrum wth s-band radar, n Dynamc Spectrum Access Networks (DYSPAN), 24 IEEE Internatonal Symposum on, pp , Aprl 24. [4] H. Shajaah, A. Abdelhad, and T. C. Clancy, A prce selectve centralzed algorthm for resource allocaton wth carrer aggregaton n LTE cellular networks, arxv:48.45, Accepted n WCNC, 25. [5] J.-W. Lee, R. R. Mazumdar, and N. B. Shroff, Downlnk power allocaton for mult-class wreless systems, IEEE/ACM Trans. Netw., vol. 3, pp , Aug. 25. [6] S. Shenker, Fundamental desgn ssues for the future nternet, Selected Areas n Communcatons, IEEE Journal on, vol. 3, no. 7, pp , 995. [7] G. Tychogorgos, A. Gkelas, and K. K. Leung, Utlty-proportonal farness n wreless networks, n Personal Indoor and Moble Rado Communcatons (PIMRC), 22 IEEE 23rd Internatonal Symposum on, pp , Sept 22. [8] A. Abdel-Had and C. Clancy, A Robust Optmal Rate Allocaton Algorthm and Prcng Polcy for Hybrd Traffc n 4G-LTE, n PIMRC, 23. [9] A. Abdel-Had and C. Clancy, A Utlty Proportonal Farness Approach for Resource Allocaton n 4G-LTE, n ICNC Workshop CNC, 24.