IEEE ICC Wireless Networking Symposium

Transcription

1 IEEE ICC Wireless Networkig Symposium Traffic-aware Utility based QoS Provisioig i OFDMA Hybrid Smallcells Ravikumar Balakrisha, Berk Caberk ad Ia F. Akyildiz Broadbad Wireless Networkig Laboratory School of Electrical ad Computer Egieerig, Georgia Istitute of Techology, Atlata, GA 30332, USA. {rbalakrisha6,ia}@ece.gatech.edu Dept. of Computer Egieerig Computer ad Iformatics Faculty, Istabul Techical Uiversity, Ayazaga 34469, Istabul-Turkey caberk@itu.edu.tr Abstract Smallcell techology is gaiig sigificace as part of the ext-geeratio cellular systems due to their performace beefits i terms of icreased etwork capacity ad improved idoor ad local coverage. Hybrid access smallcells, which provide service to both idoor as well as eighborig users, adopt adhoc policies to guaratee performace beefits to idoor home users i the presece of exteral eighborig users. Such policies must be able to stabilize user queues as well as to provisio performace beefits i terms of delay ad throughput, especially for the idoor users. As a result, classificatio of user data i terms of traffic type ad user type is required to effectively achieve the differetiated QoS performace. I this paper, a traffic-aware utility fuctio is proposed, which takes ito accout for the user s priority idex ad traffic characteristics to efficietly provide differetiated QoS beefits to users served uder a OFDMA hybrid smallcell. The problem of the traffic-aware utility based schedulig uder power costraits is posed as a optimizatio objective ad a optimal algorithm for the schedulig problem is preseted. The results show that the proposed scheme achieves QoS performace beefits i terms of throughput ad delay. Idex Terms Smallcells, Hybrid access, QoS, Heterogeeous traffic, Utility. I. INTRODUCTION THE tremedous advaces i wireless techologies has led to the proliferatio of Hi-Tech smartphoes, tablet computers that cosume large amout of multimedia ad data traffic i additio to the traditioal voice traffic i cellular etworks. To satisfy the rapidly growig capacity eeds, the existig cellular ifrastructure that cosists of a sigle macrocell layer is beig overlaid with several layers of small base statios ragig from picocells, microcells, femtocells. These smallcells icrease the etwork capacity through the spatial reuse of spectrum ad i additio, improve the idoor cellular coverage. [2]. These are average to low powered devices coected to the operator s core etwork through a proprietary or public backhaul. The applicatio of smallcells to a targeted area such as eterprise, private buildigs result i uique access cotrol policies to hadle differet user types. Smallcells access cotrol policies ca be broadly classified as Ope access policy: Smallcells ca service all mobile users of the carrier withi rage. Closed access policy: Smallcells reserve exclusive access for pre-registered mobile users, also called Smallcell users (SUs). Hybrid access policy: Smallcells utilize adhoc schemes to achieve QoS guaratees for the SUs i the presece of uregistered exteral users (EUs). Although, the ope access policy offers the largest icrease i the etwork capacity, it ca degrade the QoS performace of SUs served uder the smallcells. The QoS degradatio is particularly large whe the umber of EUs icreases or whe the EUs are ruig badwidth-hugry applicatios. The closed access approach is capable of providig better QoS performace for the SUs but the SU performace ca also be sigificatly affected if there are earby EUs that cause strog iterferece to the smallcell etwork. Therefore, a hybrid access scheme that ca provide differetiable service to SUs ad EUs is required to achieve the best of both worlds. A. Related Work Hybrid access schemes have bee ivestigated before i the literature. The authors i [3] cosider hybrid access i femtocells where they propose a fixed probability p for EUs to be able to coect to a femtocell based o the computatio of the carrier to iterferece (C/I) ratio at the locatio of the EUs. I [12], the authors propose a hybrid access scheme for OFDMA smallcells where a limited amout of subchaels v is reserved for EU access. Although, the outage probability is show to otably decrease for EUs i this scheme, icreasig v ca affect the throughput achieved by SUs. I additio, lower outage probability does ot ecessarily equal QoS performace of both SUs ad EUs. Further work has bee coducted uder hybrid access approach, for eg., i [15], the authors propose a adaptive access cotrol strategy based o the average cellular user desity. It is show that the ergodic rate for EUs is otably icreased uder low user-desity case, whereas uder the high user-desity case, the rate gai for EUs is ot sigificat. Ufortuately, the above hybrid access schemes fail to cosider the ature of the higher layer traffic i performig schedulig ad access cotrol for smallcells. Oe of the fudametal performace metric is etwork stability which /13/$ IEEE 6464

2 guaratees the queue size to be bouded for all packet arrivals withi the capacity regio. Schedulig policies like Maximum Delay Schedulig ca stabilize the queues for admissible arrival rates. At the same time, these policies ca result i poor delay performace ad ufair allocatio for the SUs if the EU traffic is bursty. The objective of the paper is to propose ad evaluate a optimal schedulig scheme for QoS provisioig for hybrid smallcells. We summarize the mai cotributios of our work as follows: We defie a specific system model for OFDMA based hybrid smallcell ad review some existig cross-layered schedulig approaches. We propose a cross-layered traffic-aware utility fuctio that provisios QoS for the users uder smallcells. We preset a optimal schedulig algorithm to perform schedulig usig utility as the cost fuctio. We provide results from simulatios highlightig the performace gais obtaied through our schedulig policy. The rest of this paper is orgaized as follows. I Sec. II, the system model for the OFDMA based hybrid smallcells is preseted ad a brief review o some existig schedulig disciplies are discussed. I Sec. III, the utility fuctio is first preseted ad the problem of costraied QoS schedulig usig traffic-aware utility is posed as a optimizatio framework. A optimal schedulig algorithm for the optimizatio framework is provided. Sec. IV illustrates the performace gais of the proposed scheme through simulatio results. Fially, the mai coclusios are summarized i Sec. V. II. SYSTEM MODEL The time-varyig, bursty ad locatio-depedet ature of the wireless chael makes it challegig to achieve schedulig with QoS performace. Therefore, iteractio across the layers ca eable exchage of iformatio i order to make schedulig decisios to provide effective QoS. Particularly, the exchage of system dyamics such as chael coditios, locatio, queue state, applicatio layer requiremets are crucial i achievig QoS satisfactio. Uder such a setup, the time is slotted ad the chael is assumed to be uvaryig for the slot legth. At the begiig of each slot, the scheduler obtais the chael gai from the lower layers through user feedback. Usig this iformatio, the data rates achievable ad power required for the user i the time slot is determied. Based o these parameters, several schedulig algorithms perform resource schedulig to achieve the QoS objectives. The dowlik of a OFDMA hybrid smallcell etwork overlaid o a macrocell coverage area is cosidered as show i Fig. 1 with a smallcell access poit (SAP) servig N users {1, 2,...N}. Out of this, F represets the set of all SUs {1, 2,...F } ad E represets the set of all EUs {1, 2,...E}, ad therefore, N = F E. B represets the total system badwidth cosistig of K subcarriers. Hece the badwidth of each subcarrier is represeted as ΔB = B. The time K is slotted ad each slot has a duratio of T s equivalet to Fig. 1. Network topology of a hybrid smallcell overlaid i a macrocell coverage regio the coheret time of the chael. h,k represets the chael gai of user trasmittig o subcarrier k. p,k represets the required power for user to trasmit o subcarrier k for a give bit error rate (BER). The oise power over a subcarrier is represeted as σ 2. If the SNR gap is defied as β = 1.5/ l(5 BER), the the trasmissio rate for user o subcarrier k is give as μ,k =ΔBlog(1 + β h,k 2 p,k σ 2 ). (1) Each user ca be assiged several subcarriers with the costrait that the same subcarrier caot be assiged to differet users i the same slot. This is represeted by the biary variable s,k (t) idicatig whether the subcarrier k is assiged to user or ot i slot t. Hece, the subcarrier assigmet costrait is give as N =1 s,k(t) =1. Therefore, the data rate μ (t) assiged to user i slot t is give by the equatio μ (t) = K s,k (t)μ,k (t). (2) k=1 The SAP has queues correspodig to each of the users it serves. The arrival process Λ (t) represets the umber of Δ packet arrivals at queue i time t. Here, λ = E[Λ (t)] ad the arrival rate vector is give as λ =(λ 1,λ 2,...λ N ). Q(t) = (Q 1 (t),q 2 (t),...q N (t)) represets the Queue Legth vector. The Waitig Time of a packet i the queues is represeted by the vector W (t) =(W 1 (t),w 2 (t),...w N (t)). The queue evolves accordig to the Discrete Time Queueig Law as Q (t +1)=max(Q (t) μ (t), 0) + Λ (t). (3) By Little s Law, the waitig time of user i slot t is give by W (t) = Q (t). (4) λ Our objective is to stabilize the queues of all SUs ad EUs whe the arrivals are iside the capacity regio. I additio, we wat to additioally offer QoS performace for 6465

3 differet SU traffic types i terms of throughput ad delay. These objectives together preset a iterestig case of QoS provisioig. Schedulig policies such as proportioally fair (PF) schedulig, Modified Largest Weighted Delay First (M- LWDF) are ot suitable i the presece of heterogeeous traffic sice they do ot provide bouded delay performace. The EXP rule proposed i [8] is show to offer improved QoS performace i terms of throughput ad delay over M-LWDF ad PF schedulig schemes whe there is a mixture of realtime ad o-real-time users i the system. Aother popular approach to QoS schedulig is the utility based approach. Schedulig rules based o maximizig utility, which represets the amout of satisfactio that ca be obtaied by schedulig a resource for a user, have bee proposed i [10], [11], [6]. The utility fuctios here are defied as decreasig fuctios of the packet delay i the queue. I [10], [11], [5], although the schedulig rule is show to achieve throughput optimality, the utility fuctio does ot provide strict bouds o delay. I additio, may of the above policies oly cosider subcarrier allocatio without ay power costraits. Sice smallcells are limited by hardware o the total trasmit power as well as by the iterferece they cause to the exteral etwork, power costraits become sigificat i our problem. III. TRAFFIC-AWARE UTILITY BASED QOS PROVISIONING The eed for achievig diverse QoS requiremets for heterogeeous traffic classes calls for a improved schedulig rule that ca deal with the uique attributes of these traffic. Especially, uder the hybrid smallcell setup, the delay performace for SUs must be bouded i the presece of EUs. The authors i [13], [14] show that if the message size of users exhibit heavy-tail characteristics with a idex α, the the delay has a ifiite mea ad ifiite variace for α<1 ad α<2 respectively. The authors also propose a modified maximum weight-α schedulig policy that allocates chaels for users based o queue size raised to the power α i order to guaratee bouded delay mea ad variace. I this paper, we propose a ovel traffic aware utility based schedulig policy for hybrid smallcells i order to effectively provisio QoS. The scheduler is fed with chael state iformatio ad traffic iformatio i order to make schedulig decisios at every time slot T s based o the computatio of the utility fuctio. The utility fuctio associated with the allocatio of subcarrier k to user is defied as U,k (t) =γ W α (t)μ,k (t), (5) where γ = a. Here, a represets the priority idex ad r ca be tued for SUs ad EUs to achieve the required QoS for each user type. r represets the average data rate achieved for user. α is the expoet of the average waitig time W for packets i queue. This is the traffic coefficiet that takes uique values for differet traffic classes. A. Ituitio The utility fuctio defied i Equatio 5 is aimed at achievig the heterogeeous objectives of QoS perceived by differet user types. For the real-time users, the delay performace is critical ad they have a strict deadlie o the waitig time of the packet. For CBR user types, the throughput as well as delay performace are importat. By usig suitable values for the traffic coefficiet α, differet utility fuctio shapes ca be obtaied for differet users types. I additio, i order to provide fair allocatio for SUs i the presece of bursty EU traffic, the parameter a ca be set to a higher value for SUs. It ca be observed that by settig the value of α to 1 for all traffic types, the utility fuctio shows similarity to the M-LWDF rule. Therefore, for homogeeous best effort traffic classes, it ca be easily established that the traffic-aware utility based schedulig is throughput optimal. B. Power Costraied Utility based Schedulig The subcarrier allocatio with power costraits usig the proposed utility fuctio is performed based o the followig optimizatio objective: subject to max S =1 k=1 K U,k (t) s,k, (6) s,k =1; S i S j i, j N ad i j ; =1 p,k s,k P s ; =1 s,k {0, 1}; =1 =1 p,k s,k P tot, (7) where the optimizatio variable S is the subcarrier allocatio matrix with order N K. S i = {s i,1,s i,2,..., s i,k } is the set of subcarriers allocated to ode i, P s is the maximum allowed subcarrier power ad P tot is the total trasmissio power available at the SAP. C. Miimal Algorithm for Utility-based Subcarrier Assigmet The above optimizatio problem ca be classified ito the Multiple Choice Kapsack Problem (MCKP) with additioal costraits o the maximum weight of each item. The MCKP is defied as a biary kapsack problem with additioal disjoit multiple-choice costraits [9]. The costraits are such that the items are divided ito multiple classes ad oly oe item is to be selected from each of the classes. The MCKP has bee show to be NP-hard sice the KP problem eeds to be solved i the process, evertheless, through dyamic programmig it is show to be solved i pseudo-polyomial time [4]. A miimal algorithm for solvig MCKP is preseted i [7]. First, the itegrality costrait s,k {0, 1} is relaxed to 0 s,k 1 to obtai the Liear Multiple-Choice Kapsack Problem (LMCKP). A simple partitioig algorithm is proposed for solvig the LMCKP ad obtaiig a feasible solutio. Usig the iitial solutio, dyamic programmig is 6466

4 used to solve MCKP. The partitioig algorithm ca compute i O() time, a small subset of items called as the core of classes, to be cosidered for the optimal value. New classes are the added to the core by eed. Applied to the utility-based subcarrier assigmet problem, the classes correspod to the set of subcarriers K. Each item correspods to a ode to be assiged for a subcarrier k. I our problem, we have a additioal costrait i the form of maximum per-subcarrier power. Oly oe ode amog N odes is assiged for subcarrier k give that it satisfies the global ad local power costraits. U,k (t) is the profit while p,k (t) is the weight. The output of the algorithm is the matrix S of dimesio N K with assigmet idicators s,k. The algorithm is preseted i Algorithm 1. The procedure partitioalgo() provides a LP-optimal solutio for the relaxed LMCKP problem. reduceclass() uses upper-boud computatio, domiace tests to prue odes for each subcarrier while reduceset() checks ad updates the CurretBestSolutio if a state improves the lower-boud. The computatioal complexity of the oe-dimesioal MCKP is show to be O( + P tot R k C um k) where R k is the reduced set of subcarriers. As a result, the algorithm solutio has liear time for a small core ad pseudo-polyomial time whe the core is large. Whe the umber of users is ot cosiderably large, it ca also be show that performig adaptive modulatio combied with subcarrier assigmet does ot icrease the algorithm performace sigificatly. This is a reasoable assumptio sice smallcells, o a average, support few tes of users. IV. PERFORMANCE EVALUATION The performace of the costraied utility-based schedulig is evaluated where the utility fuctio is modeled usig OFDMA system parameters ad queue models usig MAT- LAB ad the miimal algorithm routie is implemeted i C. The traffic types modeled for our problem are as follows: A. Traffic Modelig The users served uder SAP are grouped ito three classes usig three differet queuig disciplies as i [1]. These classes are as follows: Costat Bit Rate (CBR) Users: These users have determiistic behaviors, ad are modeled by a D/G/1 queuig system. The average waitig time is calculated as: λ CBR, σ 2 CBR,X W CBR, =, N, (8) 2(1 λ CBR, X ) where λ CBR,, σ 2 ad X CBR,X are the mea arrival rate, the variace of the service time ad the mea service time respectively ad X = E[1/μ ] for OFDMA. Video-Streamig Users: These users are modeled usig Gamma Distributio with shape parameter s ad a G/G/1 queuig system where the average waitig time is λ V id, (σ 2 + s/λ V id,x V id, ) W V id, =, N. (9) 2(1 λ V id, X ) iput : Profit matrix U of dimesio N K, Weight matrix p of dimesio N K {a, b k,s ba,a,s b a,a} partitioalgo(u, p, P tot ); a := fractioal subcarrier, {b k } := LP Optimal sol., s ba,a,s b a,a := fractioal variables i a; Calculate λ =(U b a,a U ba,a)/(p b a,a p ba,a); Calculate λ + k =max N,p,k >p (U bk,k,k U bk,k)/(p,k p bk,k), k=1,..., K, k a; Calculate λ k =mi N,p,k <p (U bk,k b k,k U,k )/(p bk,k p,k ),k=1,..., K, k a; Gradiets L + = {λ + k } ad L = {λ k } for k =1,..., K, k a; sortasce(l + ); sortdesce(l ); Curret Best Solutio z := 0; Iitial Core C := N a ; Curret Set of States Y C := reduceclass(n a ); Vectors i Y C represeted by states (θ k,π k,ν k ); θ k := k C p y k,k + k/ C p b k,k; π k := k C U y k,k + k/ C U b k,k; ν k := partial represetatio of vector y i ; repeat reduceset(y C ); if (Y C = ) the break ; Choose extsubcarrier k from L + R k := reduceclass(k) if R k > 1 the addclass(y C,R k ) ; repeat above steps for L reduceset(y C ); if (Y C = ) the break ; util forever; Fid optimal allocatio S; Algorithm 1: Miimal Algorithm for costraied utilitybased subcarrier assigmet Best-Efffort (BE) Users: The BE users ca be modeled usig a M/G/1 queueig system where the average queue waitig time is expressed as: λ BE, (σ 2 + σ 2 BE,X W BE = T ), m. (10) 2(1 λ BE, X ) where σt 2 is the variace of the iter-arrival time ad ρ BE = λ BE /X m is the utilizatio. B. Simulatio Setup We simulate a smallcell with a SAP servig N users. All users experiece i.i.d Rayleigh fadig ad the required power correspodig to a give BER. The system parameters cosidered are show i Table I. We have 5 users from each traffic class described i the previous sectio with a mix of SUs ad EUs. The arrival rates of CBR users are radomly distributed with rage kbps. Video-Streamig users have arrival rates radomly distributed with rage kbps. The shape parameter for Video-Streamig user is The BE users have radomly distributed arrival rates withi rage 6467

5 Time Average Throughput (i bits per secod) x User Idex Fig. 2. Throughput performace uder Traffic-aware Utility based Schedulig Mea Delay (secods) M LWDF TA Utility (CBR) TA Utility (Video) TA Utility (BE) Worst case User Mea Arrival Rate (kbps) Fig. 3. Mea Delay performace of worst-case user kbps. The maximum delay allowed for CBR ad Video users are fixed 80ms, 150ms respectively. I each slot, the utility ad power matrices, the global ad local power costraits are computed ad fed as iput to the algorithm. The output of the algorithm is the subcarrier assigmet matrix S. The etire simulatio sequece is ru for 1500ms. Iitial results for the traffic-aware utility(ta-utility) based scheme are preseted i this sectio. Fig. 2 shows the time average throughput achieved by the users. The fairess ca be adjusted betwee the SUs ad EUs i order to provide higher throughput for SUs. Fig. 3 presets the mea delay vs arrival rate performace of our solutio compared with the M-LWDF scheme. It is clear that the TAutility scheme achieves lower ad bouded mea delay for delay-sesitive traffic types of CBR ad V ideo compared to the M-LWDF scheme for arrivals withi the capacity regio. The BE traffic experieces large delays compared to M-LWDF scheme but this is still acceptable due to BE s elastic ature. V. CONCLUSIONS I this paper, the problem of QoS provisioig i hybrid smallcells with SUs ad EUs has bee cosidered. The eed for a improved schedulig to provide stability ad QoS performace i the presece of heterogeeous traffic has bee explaied. A ovel traffic-aware utility maximizatio approach TABLE I SIMULATION PARAMETERS System Badwidth 1.92MHz Number of Subcarriers 64 Subcarrier Badwidth 30KHz BER Required 10 3 Max. SAP Tx. Power 1W Max. subcarrier Tx. Power 0.05W Total Number of Users 15 Slot Legth 10ms uder power costraits has bee proposed ad is posed as a optimizatio objective. I order to obtai the optimal solutio, a miimal algorithm that results i a miimal core of allocatio vectors is preseted. I the ed, the performace of the proposed scheme is illustrated. The proposed work ca be cosidered as a framework i order to desig efficiet admissio cotrol algorithms i a hybrid smallcell etwork. Further proof of stability for the proposed scheme through etwork cotrol techiques ca establish the feasibility of the scheme for other traffic types. REFERENCES [1] B. Caberk, I. Akyildiz, ad S. Oktug, A qos-aware framework for available spectrum characterizatio ad decisio i cogitive radio etworks, i IEEE 21st Iteratioal Symposium o Persoal Idoor ad Mobile Radio Commuicatios, IEEE PIMRC, sept. 2010, pp [2] V. Chadrasekhar, J. Adrews, ad A. Gatherer, Femtocell etworks: a survey, IEEE Commuicatios Magazie, vol. 46, pp , Sep [3] D. Choi, P. Moajemi, S. Kag, ad J. Villaseor, Dealig with loud eighbors: The beefits ad tradeoffs of adaptive femtocell access, i IEEE Global Telecommuicatios Coferece, IEEE GLOBECOM, Dec [4] K. Dudziski ad S. Walukiewicz, Exact methods for the kapsack problem ad its geeralizatios, Europea Joural of Operatioal Research, vol. 28, pp. 3 21, [5] M. Katoozia, K. Navaie, ad H. Yaikomeroglu, Utility-based adaptive radio resource allocatio i ofdm wireless etworks with traffic prioritizatio, IEEE Trasactios o Wireless Commuicatios, vol. 8, pp , Ja [6] W.-H. Park, S. Cho, ad S. Bahk, Scheduler desig for multiple traffic classes i ofdma etworks, i IEEE ICC, Ju [7] D. Pisiger, A miimal algorithm for the multiple-choice kapsack problem. Europea Joural of Operatioal Research, vol. 83, pp , [8] A. S. S. Shakkottai, Schedulig algorithms for a mixture of real-time ad o-real-time data i hdr, i Proc. ITC, [9] P. Siha ad A. A. Zolters, The multiple-choice kapsack problem, Operatios Research, vol. 27, o. 3, pp. pp , [10] G. Sog ad Y. Li, Utility-based resource allocatio ad schedulig i ofdm-based wireless broadbad etworks, IEEE Commuicatios Magazie, vol. 43, pp , Dec [11] G. Sog, Y. Li, ad L. Cimii, Joit chael- ad queue-aware schedulig for multiuser diversity i wireless ofdma etworks, IEEE Trasactios o Commuicatios, vol. 57, pp , Jul [12] A. Valcarce, D. Lopez-Perez, G. de la Roche, ad J. Zhag, Limited access to ofdma femtocells, i Persoal, Idoor ad Mobile Radio Commuicatios, 2009 IEEE 20th Iteratioal Symposium o, sept. 2009, pp [13] P. Wag ad I. Akyildiz, O the origis of heavy-tailed delay i dyamic spectrum access etworks, IEEE Trasactios o Mobile Computig, vol. 11, pp , Feb [14], Network stability of cogitive radio etworks i the presece of heavy tailed traffic, i IEEE SECON, Ju [15] P. Xia, V. Chadrasekhar, ad J. Adrews, Femtocell access cotrol i the tdma/ofdma uplik, i IEEE GLOBECOM, Dec