16
1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 18, NO. 2, FEBRUARY 2019 Orchestrating Resource Management in LTE-Unlicensed Systems With Backhaul Link Constraints Tuan LeAnh , Student Member, IEEE, Nguyen H. Tran , Senior Member, IEEE , Duy Trong Ngo , Member, IEEE, Zhu Han, Fellow, IEEE , and Choong Seon Hong , Senior Member, IEEE Abstract— Long term evolution (LTE)-unlicensed, an extension of LTE Advanced to unlicensed spectrum, can provide high performance and seamless user experience. To reap the full benefits of the LTE-unlicensed deployment, efficient resource allocation and interference management are critical to ensuring a harmonious coexistence between LTE-unlicensed and WiFi systems. In this paper, we study a resource orchestration scheme for an LTE-unlicensed network where small cells share the same unlicensed spectrum with a WiFi system. An optimization prob- lem for channel and power allocations is formulated to maximize the overall network utility, which is an NP-hard problem. The problem is constrained on meeting the desired data rate demands of the served small-cell users, the capacity-limited backhaul links, and the maximum tolerable interference at the WiFi access point. To solve this challenging problem, a distributed solution based on Lagrangian relaxation is proposed to assist the LTE- unlicensed network in making decisions on channel allocation and transmit power. Furthermore, low-complexity solutions are devised upon applying the one-to-one matching game theory. The simulation results with practical parameter settings show that the proposed algorithms converge to the suboptimal solution after a small number of iterations in the considered examples. Index Terms— Dual decomposition, matching theory, LTE-unlicensed, resource allocation. I. I NTRODUCTION A CCORDING to the latest mobility report from the industry, the number of mobile-connected devices is Manuscript received November 5, 2017; revised May 22, 2018 and October 16, 2018; accepted December 21, 2018. Date of publication January 18, 2019; date of current version February 11, 2019. This work was supported by the National Research Foundation of Korea (NRF) funded by the Korean Government (MSIT) under Grant NRF-2017R1A2A2A05000995. The associate editor coordinating the review of this paper and approving it for publication was S. Kompella. (Corresponding author: Choong Seon Hong.) T. LeAnh and C. S. Hong are with the Department of Computer Science and Engineering, Kyung Hee University, Seoul 446-701, South Korea (e-mail: [email protected]; [email protected]). N. H. Tran is with the School of Computer Science, The University of Sydney, Sydney, NSW 2006, Australia, and also with the Department of Computer Science and Engineering, Kyung Hee University, Global Campus, Seoul 446-701, South Korea (e-mail: [email protected]). D. T. Ngo is with the School of Electrical Engineering and Computing, The University of Newcastle, Callaghan, NSW 2308, Australia (e-mail: [email protected]). Z. Han is with the Electrical and Computer Engineering Department, University of Houston, Houston, TX 77004 USA, also with the Computer Science Department, University of Houston, Houston, TX 77004 USA, and also with the Department of Computer Science and Engineering, Kyung Hee University, Seoul 446-701, South Korea (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TWC.2019.2892431 growing exponentially and it is projected to reach 11.5 billion in 2020 [1]. Additionally, many new applications and services that run on smart mobile devices are being integrated into wireless networks. As a result, data traffic congestion will soon be expected in the current LTE-A systems that utilize the licensed spectrum from 700 MHz to 2.7 GHz. To address the issues of data traffic congestion and spectrum scarcity, LTE-Unlicensed technology is proposed as an extension of the LTE-A solution to improve the network capacity while contin- ually providing excellent user experience to the customers [2]. Using carrier aggregation (CA), it integrates the unlicensed spectrum (e.g., the WiFi’s frequency bands) into the cellular network’s licensed spectrum [3]. Coexistence with WiFi in the unlicensed spectrum is a critical element in establishing a business case for the LTE-Unlicensed technology. Current projects on the LTE- Unlicensed include LTE-U, licensed assisted access (LAA), and MuLTEfire [4], [5]. These specific implementations depend on the frequency allocation policies for the 5GHz band, the regulatory requirements, and the WiFi band allocations of individual countries. A relatively simple mechanism for early deployment, the LTE-U does not require modifications to the existing LTE air interface protocol [6]. LTE-U specifically targets markets without listen-before-talk (LBT) regulation in the unlicensed spectrum such as in the U.S.A., China, and South Korea. In another way, LAA is the 3rd Generation Partnership Project (3GPP)’s effort to standardize the LTE operation in the WiFi bands. It uses a contention protocol known as LBT, which is mandated in some European countries and Japan, to coexist with WiFi devices on the same unlicensed band [4], [5], [7]. In both LTE-U and LAA, the reliable licensed radio links are used for the signaling and control mes- sages, whereas the unlicensed links are for data transmission only. This is different from the MuLTEfire where unlicensed spectrum is used for signaling and control messages. In our work, a coexistence mechanism is designed according to the LBT regulation [6], [8], in which WiFi access point’s interference limit is guaranteed from the LTE-Unlicensed deployment [7]–[11]. A. Related Works and Motivations It is important to meet the transmission quality for both LTE-Unlicensed users and WiFi users in the 1536-1276 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 18, NO. 2, FEBRUARY 2019

Orchestrating Resource Management inLTE-Unlicensed Systems With

Backhaul Link ConstraintsTuan LeAnh , Student Member, IEEE, Nguyen H. Tran , Senior Member, IEEE,

Duy Trong Ngo , Member, IEEE, Zhu Han, Fellow, IEEE,

and Choong Seon Hong , Senior Member, IEEE

Abstract— Long term evolution (LTE)-unlicensed, an extensionof LTE Advanced to unlicensed spectrum, can provide highperformance and seamless user experience. To reap the fullbenefits of the LTE-unlicensed deployment, efficient resourceallocation and interference management are critical to ensuringa harmonious coexistence between LTE-unlicensed and WiFisystems. In this paper, we study a resource orchestration schemefor an LTE-unlicensed network where small cells share the sameunlicensed spectrum with a WiFi system. An optimization prob-lem for channel and power allocations is formulated to maximizethe overall network utility, which is an NP-hard problem. Theproblem is constrained on meeting the desired data rate demandsof the served small-cell users, the capacity-limited backhaul links,and the maximum tolerable interference at the WiFi accesspoint. To solve this challenging problem, a distributed solutionbased on Lagrangian relaxation is proposed to assist the LTE-unlicensed network in making decisions on channel allocationand transmit power. Furthermore, low-complexity solutions aredevised upon applying the one-to-one matching game theory.The simulation results with practical parameter settings showthat the proposed algorithms converge to the suboptimal solutionafter a small number of iterations in the considered examples.

Index Terms— Dual decomposition, matching theory,LTE-unlicensed, resource allocation.

I. INTRODUCTION

ACCORDING to the latest mobility report from theindustry, the number of mobile-connected devices is

Manuscript received November 5, 2017; revised May 22, 2018 andOctober 16, 2018; accepted December 21, 2018. Date of publicationJanuary 18, 2019; date of current version February 11, 2019. This work wassupported by the National Research Foundation of Korea (NRF) funded bythe Korean Government (MSIT) under Grant NRF-2017R1A2A2A05000995.The associate editor coordinating the review of this paper and approving it forpublication was S. Kompella. (Corresponding author: Choong Seon Hong.)

T. LeAnh and C. S. Hong are with the Department of Computer Scienceand Engineering, Kyung Hee University, Seoul 446-701, South Korea (e-mail:[email protected]; [email protected]).

N. H. Tran is with the School of Computer Science, The University ofSydney, Sydney, NSW 2006, Australia, and also with the Department ofComputer Science and Engineering, Kyung Hee University, Global Campus,Seoul 446-701, South Korea (e-mail: [email protected]).

D. T. Ngo is with the School of Electrical Engineering and Computing,The University of Newcastle, Callaghan, NSW 2308, Australia (e-mail:[email protected]).

Z. Han is with the Electrical and Computer Engineering Department,University of Houston, Houston, TX 77004 USA, also with the ComputerScience Department, University of Houston, Houston, TX 77004 USA, andalso with the Department of Computer Science and Engineering, Kyung HeeUniversity, Seoul 446-701, South Korea (e-mail: [email protected]).

Color versions of one or more of the figures in this paper are availableonline at http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TWC.2019.2892431

growing exponentially and it is projected to reach 11.5 billionin 2020 [1]. Additionally, many new applications and servicesthat run on smart mobile devices are being integrated intowireless networks. As a result, data traffic congestion willsoon be expected in the current LTE-A systems that utilizethe licensed spectrum from 700 MHz to 2.7 GHz. To addressthe issues of data traffic congestion and spectrum scarcity,LTE-Unlicensed technology is proposed as an extension of theLTE-A solution to improve the network capacity while contin-ually providing excellent user experience to the customers [2].Using carrier aggregation (CA), it integrates the unlicensedspectrum (e.g., the WiFi’s frequency bands) into the cellularnetwork’s licensed spectrum [3].

Coexistence with WiFi in the unlicensed spectrum isa critical element in establishing a business case for theLTE-Unlicensed technology. Current projects on the LTE-Unlicensed include LTE-U, licensed assisted access (LAA),and MuLTEfire [4], [5]. These specific implementationsdepend on the frequency allocation policies for the 5GHz band,the regulatory requirements, and the WiFi band allocations ofindividual countries. A relatively simple mechanism for earlydeployment, the LTE-U does not require modifications to theexisting LTE air interface protocol [6]. LTE-U specificallytargets markets without listen-before-talk (LBT) regulation inthe unlicensed spectrum such as in the U.S.A., China, andSouth Korea. In another way, LAA is the 3rd GenerationPartnership Project (3GPP)’s effort to standardize the LTEoperation in the WiFi bands. It uses a contention protocolknown as LBT, which is mandated in some European countriesand Japan, to coexist with WiFi devices on the same unlicensedband [4], [5], [7]. In both LTE-U and LAA, the reliablelicensed radio links are used for the signaling and control mes-sages, whereas the unlicensed links are for data transmissiononly. This is different from the MuLTEfire where unlicensedspectrum is used for signaling and control messages. In ourwork, a coexistence mechanism is designed according tothe LBT regulation [6], [8], in which WiFi access point’sinterference limit is guaranteed from the LTE-Unlicenseddeployment [7]–[11].

A. Related Works and Motivations

It is important to meet the transmission quality forboth LTE-Unlicensed users and WiFi users in the

1536-1276 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Page 2: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

LEANH et al.: ORCHESTRATING RESOURCE MANAGEMENT IN LTE-UNLICENSED SYSTEMS 1361

face of co-channel interference [6]. The works ofCavalcante et al. [12], Babaei et al. [13], and Paiva et al. [14]confirm the performance degradation when the LTE cellularnetworks operate in the WiFi band. Meanwhile, the Carrier-Sensed Multiple Access (CSMA) protocol in the WiFi systemis only designed for low interference scenarios. In traditionalWiFi transmission, the interference level must be less than theenergy detection threshold to avoid backoff processes for theWiFi users. Specifically, with limited unlicensed bands, somecellular network operators deploying LTE-Unlicensed may useunlicensed bands of the WiFi access points of other networkoperators. In this case, LTE-Unlicensed transmissions shouldnot cause considerable interference on the existing WiFitransmissions. Also, the interference from the co-channelWiFi users and other LTE-Unlicensed users may deterioratethe LTE-Unlicensed users’ performance.

To address the coexistence issues of multiple LTE-Unlicensed/WiFi systems, interference management and effi-cient unlicensed spectrum utilization are key [6]–[8]. In theblank sub-frame technique, when LTE-Unlicensed/WiFi usersoperate at the same time on the same unlicensed channel,the WiFi users are designed to access the channel during theLTE blank sub-frame [15], [16]. On the other hand, the channelselection technique enables the LTE-Unlicensed users to seekthe empty unlicensed channels as well as allowing the WiFiAPs to search for the least congested channels [17]. Whenthere is no clean channel being available, the LTE-Unlicensednetwork can perform power control at the LTE-Unlicensedbase stations and users to mitigate undue interference at theWiFi system [18]. Here, the LTE-Unlicensed users controltheir transmission power based on information from the pres-ence and proximity estimations of the WiFi users, to avoidstrong interference to the WiFi system. Thus, it is criticalto efficiently allocate the unlicensed band and to adjust thetransmit power of multiple LTE-Unlicensed users, to achievethe highest utilization, while satisfying service demands forboth cellular and WiFi users.

References [7] and [9]–[11] ensure the possible coexistencebetween the LTE-Unlicensed and WiFi systems in the unli-censed spectrum by avoiding interference at the WiFi accesspoints. The requirements of the LAA in LTE-Unlicensedsystems can be met by the underlay cognitive radio solution[19], [20]. Here, LTE-Unlicensed users operate in the unli-censed band of the WiFi system while the overall interferencegenerated by the LTE-Unlicensed users on the same chan-nel is kept below a given threshold. Similarly, the worksof Xu et al. [10] and Chaves et al. [18] investigate theinterference-controlled power allocation problem in LTE-Unlicensed systems. In [18], a power control plan with aninterference-aware power operating point is devised to balancethe LTE-Unlicensed and WiFi performances in the uplinkdata transmission. In [10], a successive cap-limited water-filling method is proposed to handle the interference to WiFisystem. However, the problem of subchannel allocation in theunlicensed channels is not considered in these studies. In [9],a coordinated hierarchical game is formulated to model themulti-operator spectrum sharing in LTE-Unlicensed systems.A multi-leader multi-follower Stackelberg game framework is

developed to analyze the interaction among multiple operatorsand subscribers in the unlicensed spectrum. In this case,the operators profit from operating on unlicensed resourceswhile the users choose which unlicensed bands to trans-mit based on the interference penalty price. Additionally,the adopted game strategies make sure the interference to theWiFi access point be kept under a tolerable level.

Matching game theory [21], [22] has been used for resourceallocation in LTE-Unlicensed [7], [11]. Its aim is to find properand stable unlicensed user partners for the cellular users. Withthe matching games, the competition and negotiation amongdistinct user sets of the LTE-Unlicensed and WiFi systems canbe well modeled, where distributed solutions can be devised.In [11], the interaction between LTE-Unlicensed and WiFiusers is modeled by utilizing the the matching game-basedstudent-project allocation. Preference lists for these two typesof users are generated to avoid the interference between them,while the time division multiple access (TDMA) method isutilized to avoid the interference among the co-channel LTE-Unlicensed users. However, Gu et al. [11] do not considerpower control to mitigate interference at unlicensed users.In [23] a dynamic spectrum sharing solution among multipleoperators in the unlicensed spectrum with time-varying trafficwas proposed. Similarly, in [24] the matching theory frame-work was employed to solve the dynamic resource allocationproblem for the LTE-Unlicensed. Apart from the distinguishedformulated optimization problem, our matching-based solutionis also different from the studies in [23] and [24], in which asequence of the one-to-one matching game is utilized to com-bine with the Lagrangian relaxation-based solution [25], [26]for finding decisions on unlicensed subchannels and transmitpower.

The above mentioned works and other existing workson LTE-Unlicensed do not consider the important issue ofbackhaul protection, which should be rigorously dissected indense small cell deployment situations with different backhaulsolutions for both wireless and wired connections, e.g., non-line-of-sight (NLOS) microwave, wireless mesh networks,point-to-multipoint (PMP) topology, and virtualized wire-less networks [27]–[29]. Note that many current applicationsrequire a large amount of upload capacity such as the Inter-net of Things (IoT), virtual office, Mobile Edge Computing(MEC), web meeting, and connected vehicle safety applica-tions [30], [31]. It is argued that the backhaul network will bea major performance bottleneck in small cell networks [32].The backhaul constraint is also an important issue in thecoexistence of LTE-Unlicensed/WiFi systems [6], [33]. To thebest of our knowledge, our work is the first that considers theissues of backhaul capacity when performing joint channel andpower allocation for the coexistence of LTE-Unlicensed/WiFisystems.

Furthermore, to adapt the inevitable traffic explosion, it isnecessary to enhance the uplink data rate in the future mobilenetworks [34]. This is a direct result of communication sce-narios with significant uplink traffic loads such as the inte-gration of cloud-enabled technologies, the proliferation of IoTsystems, machine-to-machine, and machine-to-cloud platformsin wireless networks. To guarantee the service demands of

Page 3: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

1362 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 18, NO. 2, FEBRUARY 2019

the LTE-Unlicensed users, it is necessary to consider uplinkcommunications [20], [35].

B. Research Contributions

This paper develops an orchestration scheme of channel andtransmit power for the uplink of an LTE-Unlicensed systemto enable coexistence with a WiFi system. In particular, inter-ference threshold protection for the WiFi system is studied.The chunk-based channel allocation approach (i.e., subcarrieraggregation) is used to allocate the unlicensed spectrum forthe uplink transmissions in an orthogonal frequency divisionmultiple access (OFDMA)-based system [36]. The consid-ered joint channel and power allocation complicates anyoptimization-based design due to several coupled constraints:i) power allocation, ii) backhaul limitation, and iii) guarantee-ing coexistence of LTE-Unlicensed/WiFi systems. To obtain asuboptimal solution, we develop distributed algorithms basedon the Lagrangian relaxation approach. In addition, we take thecompetitive behavior of selfish and rational network entitiesinto consideration by modeling the problem as a matchinggame [21], [22]. Our key research contributions are summa-rized as follows.

• We formulate an optimization problem of channel andpower allocations for the uplink LTE-Unlicensed. Thegoal is to maximize the overall network utility while guar-anteeing WiFi access point’s interference limit, protectinglimited backhaul capacity links, and meeting the data raterequirements of the served small cell users.

• To solve the formulated NP-hard problem, we employLagrangian relaxation to decompose it into tractablesubproblems that separate the power allocation and thechannel assignment. An analytical framework is devel-oped to find a locally optimal solution for channel andpower allocations. Two low-complexity solutions are fur-ther devised by combining Lagrangian relaxation withthe matching game. In the proposed designs, small cellusers compete to get matched with the chunk-basedchannels in the formulated one-to-one matching game.After that, we develop distributed algorithms that decidethe assignment of channel and the power allocation forsmall cell users in a distributed manner. We demonstratethat the proposed algorithms converge to the suboptimalsolutions with low computational complexity.

• Simulation results with practical parameters confirm thatthe proposed approach gives suboptimal solutions and itonly takes a small number of iterations to converge.

The remainder of this paper is organized as follows.In Section II, we describe the system model of the coexis-tence of LTE-Unlicensed and WiFi systems, and the problemformulation. The formulated problem solving based on theLagrangian relaxation method is presented in Section III.Two suboptimal solutions by combining Lagrangian relaxationwith one-to-one matching game are devised in Section IV.The computational complexity of the designed algorithms areanalyzed in Section V. Section VI gives simulation results toverify the effectiveness of the suggested algorithms. Conclud-ing remarks for the whole paper are given in Section VII.

Fig. 1. System architecture of an LTE-Unlicensed network.

II. SYSTEM MODEL AND PROBLEM FORMULATION

A. System Model

Let us consider a small cell network (SCN) as shown inFig. 1, in which a set M = {1, 2, . . . ,M} of small cell basestations (SBSs) are deployed to serve a set of small cell users(SUEs) for the uplink transmission. As SBS m serves a set ofNm = {1, 2, . . . , Nm} SUEs, the set of the total number ofSUEs in the SCN is denoted as N = ∪m∈MNm. We assumeall the SUEs have access to sufficient licensed resourcesto maintain a predefined data rate of Rlicensed

nm , ∀n ∈ N ,m ∈ M. Besides, the SBSs and SUEs can also operate in theunlicensed radio spectrum to further enhance the uplink datarate, supporting a minimum data transmission rate Rmin

nm > 0,∀n ∈ N , m ∈ M.

In the unlicensed spectrum, an LTE-Unlicensed manager(LTE-UM) is assumed to operate as a third party in a virtu-alized wireless network model to provide radio resources onthe unlicensed channels to different small cell network oper-ators [9], [29]. The LTE-UM can collect unlicensed channelstate by using the LBT procedure. Besides, the LTE-UM canbundle the unlicensed channel c into chunks before sellingthem to the small cell network provider. Here, the unlicensedchannel c is sensed to be busy if it is being used by theWiFi access point (WiFi-AP). The unlicensed channel c isdivided into a set of Lc orthogonal narrowband flat-fadingsub-bands, with each sub-band spanning a bandwidth ofΔl = Bc

|Lc| Hz. We assume that the sub-bands are deterministicduring the optimization period. Here, Bc is the bandwidthof the unlicensed channel c. These sub-bands are groupedinto a set of Kc chunks, and each chunk is aggregated bya set of Lc,k sub-bands. Depending on the frequency distancebetween sub-bands, the occupied band of licensed channel cand the number of sub-bands in each chunk k, the LTE-UMrequests a payment φc,k from SUEs in return for accessingto the chunk k. Each SUE is only permitted to access atmost one chunk in the unlicensed spectrum and each chunkis allocated to at most one SUE at each SBS. The transmitpowers can be dynamically adapted on each sub-band ofthe chunk. Without loss of generality, we assume that the

Page 4: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

LEANH et al.: ORCHESTRATING RESOURCE MANAGEMENT IN LTE-UNLICENSED SYSTEMS 1363

SBSs are operated on a single unlicensed channel c withnon-overlapping coverage areas. The co-channel interferenceamong small cells is negligible due to the wall penetration lossand low power of SBSs [20], [37]. In our model, it is includedin the thermal noise term.

B. Problem Formulation

At first, we describe all system and design constraints.Then, a resource orchestration problem is formulated forLTE-Unlicensed in coexistence with the WiFi system.

1) Data Transmission Model: When an SUE n ∈ Nm

transmits data on the unlicensed spectrum, its data rate isgiven by

Runlicensednm (ψ,P ) =

k∈Kc

ψknm

l∈Lc,k

rl,knm(P l,k

nm), (1)

where the binary variable ψknm = {0, 1} represents the chunk-

based channel allocation decision for transmission from SUE nto SBS m on chunk k; ψ

Δ= [ψknm]N×M×K , ψk

nm = 1 meansthat SUE n ∈ Nm is assigned to chunk k, and ψk

nm = 0otherwise; rl,k

nm is the data rate of SUE n ∈ Nm on sub-bandl of chunk k which is determined as:

rl,knm(P l,k

nm) = Δl log2(1 + γl,knmP

l,knm), (2)

where γl,knm = gl,k

nm

I(l,k)WiFi,m+σ2

; I(l,k)WiFi,m = g

(l,k)WiFi,m

∫ dcl,k+Δl/2

dcl,k

−Δl/2

J{I|Lc|(w)})dw is the interference from the WiFi systemto SBS m on sub-band l ∈ Lc,k [10], [38]; J{I|Lc|(w)} isthe power spectral density of WiFi c’s signal after |Lc|-Fast-Fourier-Transform (FFT) processing; g(l,k)

WiFi,m is channel gainfrom WiFi AP to SBS m on sub-band l; gl,k

nm represents theinstantaneous channel power gain on sub-band l of chunk kfrom SUE n to SBS m; P l,k

nm represents the transmit poweron sub-band l of chunk k; σ2 is the background noise. In theconsidered fading channel model, gl,k

nm = Gl,knmF

l,knm where

Gl,knm and F l,k

nm are the mean channel power gain from SUEn to SBS m and the fast fading gain in sub-band l of thekth chunk, respectively. In this paper, all channel fading gainsare assumed independent and identically distributed (i.i.d.).We further define P

Δ= {P l,knm, ∀n,m, l, k} as the power

allocation matrix of all SUEs.2) SUE QoS Guarantee: Each SUE requires a minimum

data rate for its individual services. However, the data rateachievable on the licensed channel may not be enough to meetthe user demand. In this case, when Rlicensed

nm ≤ Rminnm , each

SBS m allows its SUE n to access the unlicensed spectrumresources managed by LTE-UM to further improve SUE n’sdata rate [7]. Hence, to satisfy the QoS requirement Rmin

nm forSUE n ∈ Nm, the following constraint must be met:

RnmΔ= Rlicensed

nm +Runlicensednm (ψ,P ) ≥ Rmin

nm . (3)

3) Backhaul Link Constraint: In our model, the backhaullinks of SBSs are capacity-limited. To avoid overcrowding datatraffic at the backhaul links, the data rate aggregated from allSUEs has to meet the following constraint:

n∈Nm

Rnm(ψ,P ) ≤ Zm,bh, ∀m, (4)

where Zm,bh is a predefined parameter representing the maxi-mum backhaul link capacity of SBS m. The value Zm,bh mayvary for different SBSs. The minimum backhaul link capacityof SBS m is assumed to satisfy Zm,bh ≥ ∑

n∈NmRmin

nm , ∀m.4) Interference Introduced by SUEs’ Signal Transmission

and WiFi Protection: The interference at one spectrum poolis caused by the side lobes of the Orthogonal Frequency-Division Multiplexing signal [38]. Besides, the signal on eachchannel of the considered WiFi standard is a rectangular non-return-to-zero signal [10]. In the considered network model,the unlicensed spectrum is divided into multiple orthogonalsub-bands that are utilized in the LTE-Unlicensed network.A coexistence on the same unlicensed channel can lead to themutual interference between the LTE-Unlicensed and WiFi APdue to the non-orthogonality of their respective transmittedsignals. To this end, the interference power from a set oforthogonal sub-bands L to the WiFi system on the channelc is modeled by considering the power spectral density of thesignal, as follows.

When SUE n ∈ Nm emits transmit power P l,knm on sub-band

l of chunk k, the power spectral density of sub-band l is givenas [10]:

Φl,knm(f) = P l,k

nmTs

(sinπfTs

πfTs

)2

, (5)

where Ts is the symbol duration of an ideal Nyquist pulse.The interference at the WiFi AP introduced by the transmis-

sion on sub-band l in chunk k of SUE n ∈ Nm is the integralof the power spectral density of sub-band l across the WiFiAP’s band as [10], [38]:

I(Bc)l,k (dc

l,k, Pl,knm) =

∫ dcl,k+Bc/2

dcl,k−Bc/2

gl,knm,cΦ

l,knm(f)df, (6)

where dcl,k represents the spectral distance between sub-band

l of chunk k and occupied band Bc of the WiFi AP; gl,knm,c =

Gl,knm,cF

l,knm,c where Gl,k

nm,c and F l,knm,c are the mean channel

power gain from SUE n ∈ Nm to the WiFi AP and the fastfading gain in sub-band l of the kth chunk, respectively.

To meet the performance in the WiFi system, it is assumedthat the LTE-Unlicensed network can use the unlicensed bandc whenever the total interference introduced from SUEs to theWiFi AP’s band does not exceed IBc,th

WiFi , i.e.,

m∈M

n∈Nk

k∈Kc

l∈Lc,k

ψknmI

(Bc)l,k (dc

l,k, Pl,knm) ≤ IBc,th

WiFi , (7)

where I(Bc)l,k (dc

l,k, Pl,knm) = P l,k

nmHl,knm,c is determined from (6).

For simplicity, let

H l,knm,c = gl,k

nm,cTs

∫ dcl,k+Bc/2

dcl,k

−Bc/2

(sinπfTs

πfTs

)2

df (8)

be the interference factor at the WiFi AP for sub-band l inchunk k on unlicensed band c from SUE n ∈ Nm.

5) Resource Orchestration for Uplink LTE-Unlicensed inCoexisting LTE-WiFi Systems (ROC): The ROC problem is

Page 5: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

1364 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 18, NO. 2, FEBRUARY 2019

formulated as follows:

max.(ψ,P )

m∈M

n∈Nm

φrRnm(ψ,P )

−∑

m∈M

n∈Nm

k∈Kc

φc,kψknm, (9)

s.t. (3), (4), (7),∑

n∈Nm

ψknm ≤ 1, ∀m ∈ M, k ∈ Kc, (10)

k∈Kc

ψknm ≤ 1, ∀n ∈ Nm, m ∈ M, (11)

k∈Kc

ψknm

l∈Lc,k

P l,knm ≤ Pmax

nm , ∀n ∈ Nm, m ∈ M,

(12)

P l,knm ∈ [P l

min, Plmax], ∀n ∈ Nm, m ∈ M, k ∈ Kc,

(13)

ψknm = {0, 1}, ∀m ∈ M, n ∈ Nm, k ∈ Kc. (14)

Here, the parameter φr is a reward for each data rateunit. In this paper, in order to guarantee a positive net-work profit from each user in the unlicensed channel,we assume the channel utilization cost φc,k is no greaterthan φrRnm, ∀m ∈ M, n ∈ Nm, k ∈ Kc. The network utilityin (9) is the net profit, defined as the total network throughputdiscounted by the total channel utilization costs that the smallcell network provider has to pay to the LTE-UM. Constraint(10) ensures that each chunk is allocated to at most one SUEin its SBS. Constraint (11) guarantees that each SUE can beallocated to at most one chunk. Constraints (12) and (13) meanthat the transmit power of the SUEs on sub-bands are adjustedwithin the permitted range of the total transmitted power andon each sub-band, respectively.

Since the ROC contains binary variables ψ and continuousvariables P , it is a mixed integer non-linear optimizationproblem, which is generally NP-hard. In the following sec-tions, we propose solving the ROC problem in a distributedmanner based on the Lagrangian relaxation and matchinggame approaches.

III. LAGRANGIAN RELAXATION SOLUTION

FOR RESOURCE ORCHESTRATION

Herein, the solution to the ROC is detailed based onthe Lagrangian relaxation method. Then, we propose thedistributed algorithms to find suboptimal solution of the ROCproblem.

A. Lagrangian Relaxation-Based Solution

Let us denote the non-negative multipliers associated withconstraints (3), (4), (7) and (12) by λnm, βm, νc and κnm,respectively. By jointly considering the objective function andconstraints, the partial Lagrangian of the ROC can be derivedas follows:

L(ψ,P ,λ,β, νc,κ)

= φr

m∈M

n∈Nm

k∈Kc

ψknm

l∈Lc,k

rl,knm(P l,k

nm)

−∑

m∈M

n∈Nm

k∈Kc

φc,kψknm

+∑

m∈M

n∈Nm

λnm(∑

k∈Kc

ψknm

l∈Lc,k

rl,knm(P l,k

nm) −Rminnm )

−∑

m∈Mβm(

n∈Nm

(∑

k∈Kc

ψknm

l∈Lc,k

rl,knm(P l,k

nm)) − Zm,bh)

− νc(∑

m∈M

n∈Nk

k∈Kc

l∈Lc,k

ψknmI

(Bc)l,k (dc

l,k, Pl,knm)−IBc,th

WiFi )

−∑

m∈M

n∈Nm

κnm(∑

k∈Kc

ψknm

l∈Lc,k

P l,knm − Pmax

nm ), (15)

where λ = [λnm]1×(NmM), β = [βm]1×M , and κ =[κnm]1×(NmM).

Then, the dual problem of the ROC is given by:

min.(λ�0,β�0,νc≥0,κ�0)

D(λ,β, νc,κ), (16)

where the Lagrange dual function D(λ,β, νc,κ) is

D(λ,β, νc,κ) = max(ψ,P )

L(ψ,P ,λ,β, νc,κ),

s.t. (10), (11), (13), and (14). (17)

Proposition 1: Using the Lagrangian in (15), the maximiza-tion problem (17) can equivalent toROC-D :

max(ψ,P )

m∈M

n∈Nm

k∈Kc

ψknm

[Ωk

nm(P knm) − φc,k

],

s.t. (10), (11), (13), (18)

in which

Ωknm(P k

nm) = φr

l∈Lc,k

rl,knm(P l,k

nm) + λnm

l∈Lc,k

rl,knm(P l,k

nm)

− βm

l∈Lc,k

rl,knm(P l,k

nm)

− νc

l∈Lc,k

P l,knmH

l,knm,c − κnm

l∈Lc,k

P l,knm,

(19)

where P knm = [P l,k

nm]1×|Lc,k| is the power vector of chunk kat SUE n ∈ Nm.

Proof: See Appendix.Disregarding the Lagrangian multipliers λ, β, νc, and κ,

the ROC-D problem is combinatorial optimization problemconcerning ψ for a fixed P . Additionally, it can be seenthat (18) is a concave function with respect to P l,k

nm. Thechannel and power allocations can be found by solving twosubproblems as follows.

1) Power Control Phase: When channel allocation andLagrangian multipliers values are fixed, problem (18) is con-cave with respect to P . Moreover, due to the independentchannel fading in different sub-bands, the ROC-D problemcan be further decomposed into |Lc| subproblems as follows:

maxP

m∈M

n∈Nm

k∈Kc

l∈Lc,k

ψknm

[ωl,k

nm(P l,knm) − φc,k

],

s.t. (13), (20)

Page 6: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

LEANH et al.: ORCHESTRATING RESOURCE MANAGEMENT IN LTE-UNLICENSED SYSTEMS 1365

in which

ωl,knm(P l,k

nm) � (φr + λnm − βm)rl,knm(P l,k

nm)− (νcH

l,knm,c + κnm)P l,k

nm. (21)

Proposition 2: The optimal power allocation of (20) canbe determined based on the Karush-Kuhn-Tucker (KKT) con-ditions [26] as

P l,k∗nm =

[Δl(φr + λnm − βm)

(νcHl,knm,c + κnm) ln 2

− 1

γl,knm

]P lmax

P lmin

, (22)

where [x]ba := max(min(x, b), a).Proof: See Appendix.

The result in (22) confirms our intuition that transmit poweron the sub-bands of chunk k is reduced whenever the backhaulof SBS m is congested (i.e., backhaul price βm increases).This is also the case when the total interference at theWiFi AP is violated, i.e., the interference price νc increases.Additionally, when the total transmit power at the SUE isgreater than a given threshold, the power price increases whichalso leads to reducing the transmit power on sub-bands at thatSUE. Moreover, whenever the data rate of SUE n ∈ Nm isless than a threshold at iteration t, λnm goes up at the iteration(t+1). Hence, the transmit power on each subchannel l ∈ Lc,k

also goes up in next iteration (t+1). Consequently, this leadsto increasing the data rate of the SUE n ∈ Nm on the chunkk at the next iteration (t + 1). These steps are looped untilsatisfying the SUE’s QoS constraint.

2) Channel Allocation Phase: By fixing P and theLagrangian multipliers, problem (18) is combinatorial in thevariable ψ. Moreover, because the mutual interference amongsmall cells is not taken into account in our work, the ROC-Dproblem can be decomposed into M sub-problems, each ofwhich is in the following form.ROC-Dm :

maxψ

n∈Nm

k∈Kc

ψknm

[Ωk

nm(P knm) − φc,k

],

s.t. (10), (11), and (14). (23)

The optimal chunk-based channel allocation in (23)becomes a maximum weighted matching problem. The chunkk∗ is thus allocated to SUE n ∈ Nm according to

ψk∗nm = 1, k∗ = arg max

∀n∈Nm

[Ωk

nm(P knm) − φc,k

]. (24)

By observing (23) and (24), we see that the SBS prefersto allocate the chunk to its SUEs who maximize the utility interms of the benefit Ωk

nm(P knm) minus the channel cost φc,k.

To find the optimal chunk-base channel allocation in (24),the Hungarian method [39] can be centrally performed ateach SBS to solve this bipartite matching problem with acomputational complexity of O (

(KNm)3).

3) Lagrangian Multiplier Update: To determine theLagrangian multipliers λ, β, νc, κ, we solve the primalproblem (16) given the suboptimal solutions P ∗ and ψ∗ from

(22) and (24). By substituting (22) and (24) back into (17),the dual objective is as

D(λ,β, νc,κ) =∑

m∈M

n∈Nm

k∈Kc

ψk∗nm

[ ∑

l∈Lc,k

((φr + λnm

− βm)rl,knm(P l,k∗

nm )−(νcHl,knm,c+κnm)P l,k∗

nm

)

−φc,k

]−

m∈M

n∈Nm

λnmRminnm

+∑

m∈MβmZm,bh +

k∈Kc

l∈Lc,k

νIBc,thWiFi

+∑

m∈M

n∈Nm

κnmPmaxnm . (25)

Since (25) is an affine function of λ, β, νc, and κ,the dual problem is convex. By the projected gradient-descent method [26], the optimal Lagrangian multipliers canbe found according to (27), (28), (29), and (30), where [a]+ =max{a, 0}. Here, the step sizes si(t) (i = 1, 2, 3, 4) are chosensuch that

∞∑

t=0

si(t)2<∞ and

∞∑

t=0

si(t) = ∞, i = 1, 2, 3, 4, (26)

to guarantee convergence of the algorithm [26].

Algorithm 1 ROCH: Resource Orchestration forLTE-Unlicensed System

Initialization: N , M, Kc, L, P (0), ψ(0), λ(0), β(0),ν

(0)c , κ(0).

Repeat:* Algorithm at the WiFi AP:

1: Update and broadcast the interference price νc(t + 1)as (29).* Algorithm at the SBS m ∈ M:

2: Allocate chunk-based channels to SUEs to obtain (24)using the Hungarian searching method.

3: Update and broadcast the congestion price βm(t + 1)as (28).* Algorithm at SUE n ∈ Nm:

4: Update the QoS price λnm(t + 1) and power price κnm

(t+ 1) as (27) and (30), respectively.5: Update the transmit power P k

nm(t+ 1) as (22).Until: |βm(t+ 1)− βm(t)| ≤ ξ1, |νc(t+ 1)− νc(t)| ≤ ξ2,

|λnm(t+1)−λnm(t)| ≤ ξ3, |κnm(t+1)−κnm(t)| ≤ ξ4simultaneously.

B. Proposed Algorithm for the ROC Problem

The above derivation allows us to now propose the distrib-uted algorithm ROCH that solves the originally formulatedproblem ROC using the Hungarian method for the channelallocation. Specifically, the information exchange among theSUEs and SBSs is realized using a feedback mechanism. Thetotal interference on unlicensed band c is estimated at the WiFiAP. Then, the interference price will be updated and broadcastto the SBSs. After that, the interference prices are forwarded

Page 7: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

1366 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 18, NO. 2, FEBRUARY 2019

λnm(t+ 1) =

⎣λnm(t) − s1(t)( ∑

k∈Kc

ψknm(t)

l∈Lc,k

rl,knm(P l,k

nm(t)) −Rminnm

)⎤

⎦+

(27)

βm(t+ 1) =

⎣βm(t) + s2(t)( ∑

n∈Nm

k∈Kc

ψknm(t)

l∈Lc,k

rl,knm(P l,k

nm(t)) − Zm,bh

)⎤

⎦+

(28)

νc(t+ 1) =

[νc(t) + s3(t)

( ∑

m∈M

n∈Nk

ψknmH

l,knm,cP

l,knm(t) − IBc,th

WiFi

)]+

(29)

κnm(t+ 1) =

⎣κnm(t) + s4(t)( ∑

k∈Kc

ψknm(t)

l∈Lc,k

P l,knm(t) − Pmax

nm

)⎤

⎦+

. (30)

by the SBSs to their serviced SUEs. It is assumed that thereexists virtual unlicensed users to manage the unlicensed bandon the WiFi AP [11].

Proposition 3: The sequence of primal-dual variablesupdated by the ROCH algorithm converges to a locally optimalsolution P ∗ and ψ∗ of the ROC problem.

Proof: Since the dual problem ROC-D is convex in thevariable P for a fixed ψ, (22) gives an optimal solution P ∗

for a fixed ψ. As well, (24) gives an optimal solution ψ∗

using the Hungarian method for a fixed P . The channel andpower allocation policies are the unique strategy guaranteeinga optimal solution of the ROC-D in both P and ψ. Forgiven (P ,ψ), the optimal Lagrangian multipliers of the pri-mal problem can be found by the projected gradient-descentmethod [26] according to (27), (28), (29) and (30), shownat the top of this page, with non-negative step sizes s1(t),s2(t), s3(t) and s4(t), respectively. By updating the sequenceof primal-dual variables according to the ROCH algorithm,convergence to a unique solution is guaranteed. The dualitygap is non-zero due to the ROC problem is non-convex.Hence, the ROCH algorithm can only guarantee a locallyoptimal solution P ∗ and ψ∗ of the ROC problem [26].

An optimal channel allocation is obtained by using theHungarian method. The issue with this method is that theSBS and/or LTE-UM would need to have information onall chunks from the SUEs. Here, the channel information ofeach chunk k is the value of (Ωk

nm(P knm) − φc,k), ∀k ∈ Kc.

In addition, the LTE-UM would need to broadcast the channelprice φc,k to all SUEs, ∀k ∈ Kc. A centralized unit is neededto run the Hungarian method. Also, the solution based onthe Hungarian method may require significant overhead formessage exchanges in the network. The Hungarian methodexecution faces selfish and rational SUEs and SBSs that careabout their own utility in the channel allocation phase. In nextsection, the competitive behaviors of the SUEs and SBSs forchannel allocation are carefully examined by taking advantageof the one-to-one matching game [22], [40].

IV. MATCHING GAME-BASED LOW-COMPLEXITY

SOLUTIONS FOR ROC PROBLEM

Hereafter, we design channel allocation by exploitingthe advantages of the matching game theory [22], [40].

Moreover, the both cases of with and without channel uti-lization cost sharing are considered. Based on these assump-tions, we propose two distributed algorithms to solve theROC-Dm problem using the one-to-one matching gameapproach. We first investigate the competitive channel allo-cation without sharing the channel price to the SUEs. Then,we consider the channel allocation with sharing channel pricefrom LTE-Unlicensed to the SUEs. In the formulated games,each SUE competes with others to get matched with themost preferred chunk-based channel in its preference list.Meanwhile, the SBS prefers to match the most preferred SUEon each chunk.

A. Low Computational Complexity Solution Basedon Matching Game Without ChannelUtilization Cost Sharing

1) Matching Game for Channel Allocation Without ChannelUtilization Cost Sharing: First, we study a scenario where theSUEs do not know about channel utilization cost information.We consider a one-to-one matching game at each SBS m ∈M, which is defined by a tuple (Nm,Kc,�Nm ,�Kc). Here,the preference relations of the SUEs and chunks in SBS mare denoted by �Nm= {�nm}n∈Nm and �Kc= {�km}k∈Kc ,respectively. The definition of the modeled matching game isstated as follows:

Definition 1: A matching game with two sets Nm and Kc

for the channel allocation is represented by a function ϕm:Nm → Kc with

(i) n = ϕm(k) ↔ k = ϕm(n), ∀n ∈ Nm, k ∈ Kc;(ii) |ϕm(k)| ≤ 1 and |ϕm(n)| ≤ 1, n ∈ Nm, k ∈ Kc.

In the matching ϕm, SUE n preferring chunk k to k′ isdenoted by k �nm k′ (k, k′ ∈ Kc). Meanwhile, a chunk kpreferring SUE n to n′ is denoted by n �km n′ (n, n′ ∈ Nm).For the two-sided matching game, the aim is to seek a stablematching ϕm such that there exists no blocking pair at boththe SUE and SBS in the matching ϕm. A pair (n, k) is ablocking pair for ϕm if there exists k �nm k′, (k, k′ ∈ Kc)and n �km n′, (n, n′ ∈ Nm).

The utility functions Unm(k) and Ukm(n) that, respectively,form the preference relations �nm and �km of SUEs and theSBS in small cell m are stated as follows:

Page 8: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

LEANH et al.: ORCHESTRATING RESOURCE MANAGEMENT IN LTE-UNLICENSED SYSTEMS 1367

a) Utility function for the SUE: At each SBS m, the SUEestimates a corresponding utility on chunk k based on (19) as

Unm(k) = Ωknm(P k

nm). (31)

With (31), each SUE forms its preference list of all chunks.Then, it only gives a request to its SBS for occupying the mostpreferred chunk in its preference list.

b) Utility function for each chunk at the SBS: In responseto the requests from a set of N req

m SUEs to occupy a certainchunk, each SBS desires to maximize a utility function oneach chunk stated as

Ukm = max{Ωknm(P k

nm) − φc,k}n∈N reqm, (32)

which is the net profit received by SUE n on chunk k. Then,based on (32), each SBS only accepts the SUE on the chunkwith the most preferred SUE in its preference list.

Proposition 4: The ROC-Dm problem can be modeled asan one-to-one matching game in Definition 1.

Proof: In Definition 1, the conditions |ϕm(k)| ≤ 1and |ϕm(n)| ≤ 1 correspond to the constraints (10)and (11), respectively. Moreover, the utility definitions in(31) and (32) also capture the objective function (23) of theROC-Dm. Therefore, the ROC-Dm problem can be mod-eled as a one-to-one matching game. The tuple (Nm,Kc,�Nm

,�Kc) corresponds to the channel allocation by SBS m toits SUEs.

Algorithm 2 MCAW: Matching-Based Channel AllocationWithout Channel Utilization Cost Sharing

Initialization: Nm,Kc, ϕm = ∅.* Discovery and utility evaluation:

1: Each SUE n ∈ Nm builds �nm based on (31).* Find stable matching ϕ∗

m:2: while

∑∀k,n

bn→k = 0 do

3: Algorithm at SUE n ∈ Nm:4: Send a bid for its SBS to chunk

k∗ = arg maxk∈nm

Unm(k).

5: Algorithm at the SBS m ∈ M:6: Construct �km based on (32).7: Update n∗ = ϕm(k)|n∗ = arg max

n∈km

Ukm(n).8: Update the reject list and �nm.9: end whileResults: Convergence to stable matching ϕ∗

m.

Using Proposition 4, we now present a distributed algorithmMCAW in Algorithm 2 for channel allocation in a single SBSwithout channel utilization cost sharing. Our goal is to finda suboptimal channel allocation under a stable matching ϕ∗

m.At first, each SUE forms its preference list based on (31)[cf. Line 1]. In the matching process, each SUE gives a requestbn→k = 1 to occupy chunk k that takes the highest utility[cf. Lines 2, 3 and 4]. On the SBS side, the SBS collectsand creates a preference list on chunks [cf. Line 6]. The SBSdecides to assign chunk-based channels to SUEs that give thehighest utility value on each chunk [cf. Line 7]. The SUEsthat are replaced or rejected by the SBS will be updated in the

reject list of the SBS. Then, the SUEs’ preference relations arealso updated by removing the chunks that are rejected by theSBS [cf. Line 8]. It is noted that the acceptance or rejectionprocess of applicants (i.e., the SUEs) is executed similarlyto the conventional deferred acceptance algorithms [21], [22].Hence, the MCAW algorithm converges to the stable matchingϕ∗

m where there exists no blocking pair between the SUEsand SBSs.

Theorem 1: The stable matching ϕ∗m gives a locally optimal

solution of the ROC-Dm problem.Proof: See Appendix.

2) Algorithm Design for the ROC Problem Based on theMatching Game Without Channel Utilization Cost Sharing:A distributed algorithm utilizing the matching game to finda suboptimal solution to the ROC problem is presentedin the ROCM-W algorithm (Algorithm 3). Since there isno interference in data transmission among small cells inthe considered model, one assignment of a chunk does notgenerate interference to other small cell users. Consequently,the preference of other users does not change in the channelallocation phase given a fixed transmit power level. Hence,the ROCM-W guarantees a convergence to the stable matchingat each outer-iteration of Algorithm 3.

Algorithm 3 ROCM-W: Low-Complexity Solution for ROCProblem Based on Matching Game Without Channel Utiliza-tion Cost Sharing

Initialization: N , M, Kc, L, P (0), ψ(0), λ(0), β(0), ν(0)c ,

κ(0).Repeat:* Power allocation phase:

- Algorithm at the WiFi AP:1: Update and broadcast the interference price νc(t + 1) as

(29).- Algorithm at the SBS m ∈ M:

2: Update and broadcast the congestion price βm(t + 1) as(28).- Algorithm at SUE n ∈ Nm:

3: Update the QoS price λnm(t + 1) and the power priceκnm(t+ 1) as (27) and (30), respectively.

4: Update the transmit power P knm(t+ 1) as (22).

* Channel allocation phase:5: Allocate chunk-based channels to SUEs (i.e., ψk

nm(t+ 1))using the MCAW algorithm.

Until|βm(t + 1) − βm(t)| ≤ ξ1, |νc(t + 1) − νc(t)| ≤ ξ2,|λnm(t+1)−λnm(t)| ≤ ξ3, |κnm(t+1)−κnm(t)| ≤ ξ4are simultaneously satisfied.

To implement a distributed channel and the power allo-cation, we assume that the channel and power allocationsoccur on different time-scales. In the ROCM-W algorithm,a distributed channel allocation based on the MCAW algorithmis implemented instead of the centralized Hungarian methodin the ROCH counterpart. In the former case, SBSs and SUEsshare partial information to obtain a suboptimal solution.

Page 9: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

1368 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 18, NO. 2, FEBRUARY 2019

Theorem 2: The sequence of primal-dual variables updatedby the ROCM-W algorithm converges to a suboptimal solutionof the ROC problem.

Proof: The problem ROC-D is a convex in the variableP for a given ψ. Hence, at each iteration t of the ROCM-Walgorithm, the solution in (22) gives an optimal solution P (t)for a given ψ(t). Then, given P (t), the preference lists of�Nm (t) and �Kc (t) are determined based on (31) and (32)at each iteration t, respectively. Hence, the channel allocationphase based on MCAW algorithm guarantees a convergence tothe stable matching ϕ∗

m(t) at each iteration t. Since ϕ∗m(t) only

gives a locally optimal solution in the channel allocation phaseat iteration t as proved in Theorem 1, the Slater condition is notsatisfied, and hence there is a nonzero duality gap. For givenallocationsψ(t) and P (t), the optimal value of the Lagrangianmultipliers of the primal problem can be obtained by theprojected gradient-descent method [26] according to (27), (28),(29) and (30) with non-negative step sizes of s1(t), s2(t), s3(t)and s4(t), respectively. Hence, the solution of the ROCM-Walgorithm keeps improving their schedules according to theweights resulting from P and dual variables λ, β, νc, and κ.Eventually, the channel allocation in the ROCM-W algorithmconverges to a unique solution. Therefore, by updating thesequence of primal-dual variables according to the ROCM-Walgorithm with a sufficiently small step sizes, the convergenceto a suboptimal solution (ψ∗,P ∗) is guaranteed.

B. Low Computational Complexity Solution Basedon a Matching Game With ChannelUtilization Cost Sharing

In this subsection, we study a scenario in which the SBSshares utilization cost on chunk-based channels to the SUEs.A matching game is applied to find channel allocations. Unlikethe one-to-one matching game in Section IV-A, the demandfunction of the SUE is expressed as

Unm(k) = Ωknm(P k

nm) − φc,k. (33)

Responding to the demands to occupy chunks the SUEs foroccupying chunks, each SBS seeks to match each chunk withthe SUE that maximizes the SBS’s utility function as

Ukm = max{Ωknm(P k

nm) − φc,k}n∈N reqm. (34)

After that, a locally optimal solution is obtained for theROC-Dm problem based on the matching game with channelutilization cost sharing as in the MCAW algorithm. A locallyoptimal solution with channel utilization cost sharing canbe found using the MCAS algorithm which is obtained bysubstituting (31) and (32) in the MCAW algorithm by (33)and (34), respectively.

By replacing the MCAW with the MCAS algorithm in thechannel allocation phase of the ROCM-W algorithm, we havethe ROCM-S algorithm.

Theorem 3: The sequence of primal-dual variables updatedby the ROCM-S algorithm converges to a suboptimal solutionof the ROC problem.

Proof: The proof for Theorem 3 is similar to that forTheorem 2 and so it is omitted for brevity.

Algorithm 4 ROCM-S: Low-Complexity Solution for ROCProblem Based on Matching Game With Channel UtilizationCost Sharing

Initialization: N , M, Kc, L, P (0), ψ(0), λ(0), β(0), ν(0)c ,

κ(0).Repeat:* Power allocation phase:

- Algorithm at the WiFi AP:1: Update and broadcast the interference price νc(t + 1) as

(29).- Algorithm at the SBS:

2: Update and broadcast the congestion price βm(t + 1) as(28).- Algorithm at the SUE:

3: Update the QoS price λnm(t+1) and power price κnm(t+1) as (27) and (30), respectively.

4: Update transmit power P knm(t+ 1) as (22).

* Channel allocation phase:5: Allocate chunk-based channels to SUEs (ψk

nm(t + 1)) toobtain (24) using the MCAS algorithm.

Until convergence:6: |βm(t+1)−βm(t)| ≤ ξ1, |νc(t+1)−νc(t)| ≤ ξ2, |λnm(t+

1) − λnm(t)| ≤ ξ3, |κnm(t + 1) − κnm(t)| ≤ ξ4 aresatisfied, simultaneously.

V. COMPUTATIONAL COMPLEXITY ANALYSIS

To find the optimal (P ,ψ) in the ROCH algorithm,the Lagrangian multipliers need to be updated via the sub-gradient method. Specifically, N QoS demand prices, Mcongestion prices, one interference price, and N power pricesare updated at each iteration with step sizes (27), (28), (29) and(30), respectively. Thus, the computational complexity of thesub-gradient method is O

((N2M

)2)

[26], here N = NmM .Additionally, the channel allocations need to be determinedbased on the Hungarian method in each SBS. Since thecomputation in the channel allocation phase is distributed toeach SBS for each outer loop in the ROCH algorithm, thecomputational complexity in the channel allocation phase isO(MN3

mK3). Thus, the total computational complexity of the

ROCH algorithm is O (MN3

mK3 +N4M2

).

In the suboptimal solution using ROCM-W, the channelallocation computation is updated using the MCAW algorithmat each SBS. The MCAW algorithm is hard to determine theexact computational complexity since it depends on privateinformation. Nevertheless, we can determine an upper boundon the maximum number of communication message passingsbetween the SUEs-SBSs in the MCAW algorithm by analyzingthe worst-case scenario. This scenario occurs whenever all ofthe SUEs own the same preference relations in their preferencelist. Moreover, the list order of of chunks at the SBS side ison exact contrary to the list order at the SUEs side. Thus,when a SBS m has K chunks, in each loop of the MCAWalgorithm, Nm − 1 SUEs will be rejected by the SBS m andthis rejection is repeated for K−1 rounds. The computational

Page 10: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

LEANH et al.: ORCHESTRATING RESOURCE MANAGEMENT IN LTE-UNLICENSED SYSTEMS 1369

complexity for the channel allocation at SBS m in eachouter loop of the ROCM-W algorithm is bounded aboveby O(N2

m(K − 1)). Therefore, the computational complexityfor the ROCM-W algorithm is O(MN2

m(K − 1) + N4M2).By similar arguments, the total complexity of the ROCM-Salgorithm is also O(MN2

m(K − 1) +N4 M2).

VI. PERFORMANCE EVALUATION

In this section, computer simulations are presented to verifythe effectiveness of the proposed algorithms.

A. Simulation Setup

We consider one WiFi AP and five SBSs (M = 5) inwhich each SBS has a coverage radius of 30 m. The SBSsare located in a small indoor area to serve |N | = 10 SUEs,in which each SBS m has Nm = 2 SUEs. We setup anLTE-Unlicensed system using 12 sub-bands operating on thesingle frequency of the WiFi AP where each sub-band has abandwidth of 180 kHz. These sub-bands are further dividedinto 6 chunks to be assigned to the SUEs. The channel powergains are supposed to be i.i.d. Rayleigh random variables withunit mean. We adopt the log-distance pathloss model of [41].In the pass-loss model of the SUE-to-WiFi AP path-loss fordistance a, we have La = 15.3 + 37.6 log10(a) + ρ. Thewall penetration loss ρ is taken as 10 dB. In the path-lossmodel between the SBS and its SUEs with the distance a,we have La = 38.46 + 20 log10(a). The maximum toleranceinterference power on the channel c at the WiFi AP is setas −75 dBm. The noise power at any receiver is set to−110 dBm. Each SUE has a minimum required data rateof 2.048 Mbps and a maximal power constraint of 100 mW.We set φr = 10 per each unit data rate Mbps. Besides,we set the data rate Rlicensed

nm = 0, ∀m ∈ M, ∀n ∈ Nm.Moreover, we use error tolerances ξi = 10−3, i = {1, 2, 3, 4}to terminate the proposed algorithms.

B. Simulation Results

Here, we first show the results given by the ROCH algorithmin a single snapshot based on the above settings. After that,the results of the proposed algorithms over multiple snapshotswill be presented.

1) Evaluation of the ROCH Algorithm Within a SingleSnapshot of the Chunk-Based Channel Assignment and PowerAllocations: We evaluate the proposed ROCH algorithm with5 SBSs and 10 SUEs as shown in Fig. 2. We set φc,k equalsto 5 for all chunks. The capacities of the backhaul links areset as Z th

1 = 6 Mbps, Z th5 = 16 Mbps, and 8 Mbps for the

remaining backhaul links. The maximum interference powerat the WiFi AP is kept at −75 dBm. The results of solvingthe ROC problem by the ROCH algorithm are presentedin Figs. 4 and 3. Fig. 3 shows that the proposed algorithmconverges in fewer than 40 iterations. The channel allocationand transmit power are adapted based on the ROCH algorithmas shown in Figs. 4a and 4b. All backhaul links are guaranteedto be less than pre-defined thresholds as shown in Fig. 4c.The WiFi AP interference threshold is also guaranteed as

Fig. 2. Simulation model.

Fig. 3. Convergence of the total utility of all SUEs with K = 6 chunks.

shown in Fig. 4d. In the proposed algorithm, small cell basestations 1, 2, and 4 force their SUEs to reduce transmissionrate to avoid overloading at their backhaul links. However,to maximize total network utility, the SUEs in small cells 3 and5 increase their respective data rate. To protect the WiFi AP,the data traffic in small cells 3 and 5 only increase wheneverthe total interference power at the WiFi AP is kept below thepredefined threshold. Besides, small cell 4 has a higher ratethan does small cell 3. This is because the SUEs in small cell3 produce a higher interference power level than that of theSUEs in small cell 4 when they are allocated the same transmitpower, i.e., the distance between the WiFi system and smallcell 3 is shorter than that of small cell 4 in Fig. 2. Intuitively,to maximize the total network utility, the proposed algorithmprefers increasing transmission rate at small cell 4 to smallcell 3.

Page 11: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

1370 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 18, NO. 2, FEBRUARY 2019

Fig. 4. A snapshot of the chunk-based channel assignment and power allocations resulting from the the ROCH algorithm with 5 SBS, 10 SUEs, and6 chunks.

2) Evaluation of the Proposed Algorithms in Multiple Snap-shots: Here we compare four schemes: ROCH, ROCM-W,ROCM-S, and ROC-Greedy. For the ROC-Greedy scheme,we utilize the ROCM-S algorithm in which the channelallocation phase uses a greedy algorithm that the SUE alwaysselects a chunk in its preference list with the highest utility.All the presented results are averaged over five hundred ofindependent simulation runs, each of which realizes randomlocations of the SUEs inside the SBSs’ coverage area andrandom channel power gains.

Fig. 5 compares the total network utility achieved by differ-ent schemes, in which Nm = 4, ∀m, Z th

m = 20, ∀m, φc,k = 5(k = {1, 2, 3, 4, 5}), φc,6 = 25, and IB,th

WiFi = −75 dBm.Cumulative distribution functions with different schemes arecaptured from five hundred independent simulation runs. It isobserved that the ROCM-W and ROCM-S schemes performmore similarly to the ROCH scheme than to the ROC-Greedyscheme. This is because of using the matching game only

obtains a locally optimal solution of channel allocation in eachouter loop of the updated primal-dual iteration meanwhile theHungarian method obtains an optimal solution. Besides, sincethe updated channel allocation in the matching game-basedschemes is better than in the greedy scheme, the dual gapsby the ROCM-W and ROCM-S algorithms are smaller thanthose by the ROC-Greedy algorithm. Moreover, the ROCM-Sscheme outperforms the ROCM-W scheme. This confirms thatthe proposal side (i.e., SUEs) in matching operations withthe channel utilization cost sharing in the ROCM-S schemeis better than the without channel utilization cost sharingin the ROCM-W scheme. Clearly, the proposed algorithmscan efficiently operate with the partially sharing networkinformation, giving a near optimal solution.

Fig. 6 presents the average network utility versus thebackhaul links capacity for Nm = 4, ∀m, φc,k = 5, k ={1, 2, 3, 4, 5}, φc,6 = 25, and IB,th

WiFi = −75 dBm. As thebackhaul links capacity increases, the average network utilities

Page 12: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

LEANH et al.: ORCHESTRATING RESOURCE MANAGEMENT IN LTE-UNLICENSED SYSTEMS 1371

Fig. 5. CDF of the network utility with different methods.

Fig. 6. Average network utility versus the limited backhaul rate with differentmethods.

of all proposed schemes go up due to the growing data trafficof the SUEs at SBSs. These average network utilities saturatefor the high-capacity backhaul due to the constraint on theWiFi AP’s interference threshold. It is clear from Fig. 6 thatthe utility value by the ROCM-S algorithm is very close tothat by the ROCH algorithm for low-capacity backhaul. Dueto the total reward getting from SUEs is limited by the low-capacity backhaul links, the total network utility does notchange in the first term of (9). In this case, the SUEs will beassigned to low-price chunks to improve the network utilityfrom the channel allocation phase of the ROCM-S algorithm.Here, chunk 6 with the highest channel utilization cost in thematching processes of the ROCM-S algorithm is less likely tobe chosen by SUEs for low-capacity backhaul. Besides, dueto the remaining chunks are the same price, the result of theROCM-S is similar to that of the ROCH algorithm when thereis no SUE allocating on chunk 6.

Fig. 7. Average network utility versus the number of SUEs.

Moreover, Fig. 6 depicts that the network utilities by theROCM-S and ROCM-W schemes achieve the respective gainsof 17.6% and 20.1% over the ROC-Greedy scheme for abackhaul link capacity of 18 Mbps. Moreover, the averagenetwork utilities of the ROCM-W and ROCM-S schemesachieve 4% and 2.1% lower than that of the ROCH algo-rithm for backhaul link capacity of 18 Mbps, respectively.These results confirm that the proposed algorithms usingmatching theory with different strategies of cost sharing canobtain good performances in terms of utility values withlow computational complexities in the considered simulationsetup.

In Fig. 7, we show the average network utility dependingon the number of SUEs for different schemes with Z th

m =18 Mbps, ∀ m, φc,k = 5, k = {1, 2, 3, 4, 5}, φc,6 = 25and IB,th

WiFi = −75 dBm. The SUEs are deployed uniformlyinside the SBSs’ coverage. As seen the network utilities ofall proposed schemes increase with more SUEs due to thegrowing data traffic of SUEs at SBSs. Also show that thenet utility values of the ROCM-W and ROCM-S schemes areclose to the ROCH scheme for a small number of SUEs.Particularly, using the proposed matching processes in thechannel allocation phase in the ROCM-S scheme can obtainthe results as the ROCH scheme for five SUEs. This isbecause there is only one SUE in each SBS in this scenario.Therefore, there are no rejections of SUEs at the SBS fromthe MCAS algorithm in the channel allocation phase of theROCM-S algorithm for occupying chunks. This confirms thatthe matching process in the MCAS algorithm can obtainthe same result as the Hungarian algorithm in the channelallocation phase through numerical results.

Additionally, Fig. 7 shows that the average network utilityof the ROCM-S and ROCM-W schemes can reach up to 17.5%and 22.7% gains over the ROC-Greedy algorithm for 25 SUEs,respectively. With more SUEs, there is more competition foroccupying the chunks among the SUEs. As such, the net utilityin the channel allocation phase as well as the net utility in boththe ROCM-S and ROCM-W schemes are reduced. In Fig. 7,

Page 13: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

1372 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 18, NO. 2, FEBRUARY 2019

Fig. 8. Average network utility following the channel utilization costs ofchunk 6.

the average network utility of the ROCM-W and ROCM-Sschemes is 5.3% and 1.8% lower than that of the ROCHalgorithm for 25 SUEs, respectively.

To estimate optimality gap of the proposed approach, we runa globally optimal exhaustive search to find optimal solutionof the ROC problem. An optimal solution can be found byexamining all possibilities of chunk assignments, followed bysolving the associated convex power-allocation problems. Dueto the combinatorial nature of the problem, we limit to a small-to-medium sized problem with maximum number of 3 SUEsfor each SBS. Fig. 6 shows that the ROCH scheme yields thesolution close to that of the ROC-Optimal scheme, with a gapof 4.08% for a network with 3 SUEs a per each SBS.

The proposed algorithms are further investigated withrespect to the channel utilization cost for Z th

m = 18 Mbps,∀m, and Nm = 4 SUEs, ∀m, IB,th

WiFi = −75 dBm. To evaluatethe effects of the channel utilization cost, we fix φc,k = 5(k = {1, 2, 3, 4, 5}). Then, the channel utilization cost ofchunk 6 is increased from 5 to 25 units. As can be seen fromFig. 8, the average network utilities of the proposed schemesdecrease as the channel utilization cost of chunk 6 increases.Intuitively, the channel utilization cost of chunk 6 affects theROC problem in both the ROCM-S and ROCM-W algorithmsas follows. We first consider the network utility when φc,6 = 5for all chunks. At this operation point, the preference relationsfor the proposal side in both cases of with and without channelutilization cost sharing are not affected by channel utilizationcost. As such, the SUEs’ proposals for occupying chunks arethe same preference list in the matching operation in both theROCM-S and ROCM-W algorithms. Similarly, the preferencerelations at the acceptance side are also the same in bothscenarios. The average network utilities by the ROCM-S andROCM-W algorithms are the same as shown in Fig. 8. How-ever, the gap between the ROCM-S and ROCM-W schemeswidens as the channel utilization cost of chunk 6 increases.It is confirmed that the SUEs can have a better observationto choose low cost channels. The average network utility with

the ROCM-S algorithm is more efficient with the ROCM-Wscheme in terms of the channel utilization cost.

Particularly, in Fig. 8, we also show that the average networkutility of the ROCM-S and ROCM-W schemes can reach upto 18.5% and 17.0% gains, respectively, over results of theROC-Greedy scheme for the channel utilization cost of 20 forchunk 6. Additionally, the average network utility of theROCM-W and ROCM-S schemes also respectively achieveapproximations of 2.8% and 1.6% compared to the ROCHscheme for φc,6 = 20.

VII. CONCLUSIONS

In this paper, we have studied joint chunk-based channel andpower allocation for an LTE-Unlicensed network with limitedbackhaul link capacity. We have formulated an optimizationproblem that maximizes the overall uplink network utilitywhile guaranteeing the data rate requirements of the servedSUEs, avoiding congestion at the backhaul links, and protect-ing the WiFi AP. A distributed framework based on Lagrangianrelaxation has then been proposed to analyze the interactionsamong the SUEs, SBSs, and WiFi AP. Under this framework,the distributed algorithms have been introduced to enable theLTE-Unlicensed network to obtain a suboptimal decision aboutchunk-based channel assignment and transmit power alloca-tions. Moreover, two distributed solutions with the limitedinformation sharing and low computational complexity havealso been proposed based on the one-to-one matching game.Simulation results have illustrated that the proposed algorithmsconverge after a few iterations. Additionally, the results of theproposed algorithms using the matching game are shown toapproach the suboptimal solution. In addition, the results havealso pointed out that the ROCM-S scheme outperforms theROCM-W scheme at the cost of some extra communication.

In this work, the intra-tier interference among small cellsand admission control mechanisms to guarantee a feasiblesolution of the proposed work have not been investigated.Nevertheless, our solution is still applicable to practical net-work settings without intra-tier interference among small cells,i.e., the small cells are deployed to operate on a single unli-censed channel with non-overlapping coverage areas or withnon-overlapping sub-channels among small cells.

APPENDIX

A. Proof of Proposition 1

The Lagrangian in (15) can be rewritten as (35), shown atthe top of the next page. After grouping ψk

nm and substitutingI(Bc)l,k (dc

l,k, Pl,knm) by P l,k

nmHl,knm,c, we have:

L(ψ,P ,λ,β, νc,κ)

=∑

m∈M

n∈Nm

k∈Kc

ψknm

[φr

l∈Lc,k

rl,knm(P l,k

nm)

+λnm

l∈Lc,k

rl,knm(P l,k

nm) − βm

l∈Lc,k

rl,knm(P l,k

nm)

− νc

l∈Lc,k

P l,knmH

l,knm,c − κnm

l∈Lc,k

P l,knm − φc,k

]

Page 14: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

LEANH et al.: ORCHESTRATING RESOURCE MANAGEMENT IN LTE-UNLICENSED SYSTEMS 1373

L(ψ,P ,λ,β, νc,κ) =∑

m∈M

n∈Nm

k∈Kc

ψknmφr

l∈Lc,k

rl,knm(P l,k

nm) −∑

m∈M

n∈Nm

k∈Kc

ψknmφc,k

+∑

m∈M

n∈Nm

k∈Kc

ψknmλnm

l∈Lc,k

rl,knm(P l,k

nm) −∑

m∈M

n∈Nm

λnmRminnm

−∑

m∈M

n∈Nm

k∈Kc

ψknmβm

l∈Lc,k

rl,knm(P l,k

nm) +∑

m∈MβmZm,bh

−∑

m∈M

n∈Nm

k∈Kc

ψknm

l∈Lc,k

νcHl,knm,cP

l,knm + νcI

B,thWiFi

−∑

m∈M

n∈Nm

k∈Kc

ψknmκnm

l∈Lc,k

P l,knm +

m∈M

n∈Nm

κnmPmaxnm . (35)

−[ ∑

m∈M

n∈Nm

λnmRminnm −

m∈MβmZm,bh − νcI

B,thWiFi

−∑

m∈M

n∈Nm

κnmPmaxnm

], (36)

Let us define

Λknm(λ,β, νc,κ)

Δ=[ ∑

m∈M

n∈Nm

λnmRminnm −

m∈MβmZm,bh − νcI

B,thWiFi

−∑

m∈M

n∈Nm

κnmPmaxnm

](37)

and

Ωknm(P k

nm)Δ= φr

l∈Lc,k

rl,knm(P l,k

nm)

+λnm

l∈Lc,k

rl,knm(P l,k

nm) − βm

l∈Lc,k

rl,knm(P l,k

nm)

− νc

l∈Lc,k

P l,knmH

l,knm,c − κnm

l∈Lc,k

P l,knm. (38)

where P knm = [P l,k

nm]1×|Lc,k| is the power vector of SUE n ∈Nm on sub-bands of chunk k.

Due to the fixed Lagrangian multipliers λ,β, νc and κ,Λk

nm(λ,β, νc,κ) is a constant value in ψ and P . Hence,the results of ψ and P from the optimization problem (17) areequivalent to results in (18). Therefore, the objective function(18) follows after eliminating the terms that do not contain ψand P .

B. Proof of Proposition 2

Taking the first-order derivative of (21) with respect to P l,knm,

we have:

∂ωl,knm(P l,k

nm)/∂P l,knm = (φr + λnm

− βm)Δlγl,knm/((1 + γl,k

nmPl,knm) ln 2)

− (νcHl,knm,c + κnm). (39)

By setting ∂ωl,knm(P l,k

nm)/∂P l,knm = 0, we have:

(φr + λnm − βm)Δlγl,knm/((1 + γl,k

nmPl,knm) ln 2)

− (νcHl,knm,c + κnm) = 0. (40)

Then, (40) is equivalent to

γl,knmP

l,knm =

(φr + λnm − βm)Δlγl,knm

ln 2(νcHl,knm,c + κnm)

− 1. (41)

By dividing both sides of (41) by γl,knm, we obtain (22).

C. Proof of Theorem 1

Denote by τ the τ -th epoch time of the while loopin the MCAW algorithm.

∑k∈Kc

U(τ)km =

∑k∈Kc

max{Ωk

nm(P k,(τ)nm ) − φ

(τ)c,k}n∈N req

mdenote the total utility value of

the ROC-Dm problem that is captured at the end of eachthe τ -th epoch time. Let ϕ(τ)

m be the formed matching atiteration τ .

On the one hand, the decisions made by the SBS on chunksin the MCAW algorithm can be seen as a sequential acceptanceor rejection operation in the matching ϕ(τ)

m as:

ϕ(0)m → ϕ(1)

m . . .→ ϕ(τ)m → ϕ(τ+1)

m . . . (42)

Since the MCAW algorithm is executed similarly to theconventional deferred acceptance algorithm, it converges toa stable matching ϕ∗

m, implying no blocking pair at boththe SUE and SBS in the matching ϕ∗

m [21], [22]. Hence,the MCAW algorithm terminates at the stable matching ϕ∗

m.On the other hand, in the MCAW algorithm, the applicant

SUE n ∈ Nm always seeks a chunk k in its preference list�(τ+1)

nm to guarantee k∗ = arg maxk∈(t+1)

nm

Unm(k), ∀τ , where

k ∈�(t+1)nm . At the acceptance side, the SBS m executes

acceptance or rejection process on each chunk to guarantee{nm}(τ+1) �k {n′m}(τ) (n, n′ ∈ Nm), which meansthat U

(τ+1)km ≥ U

(τ)km , ∀k ∈ Kc, ∀m ∈ M, ∀τ . Hence,∑

k∈KcU

(τ+1)km ≥ ∑

k∈KcU

(τ)km , ∀m ∈ M, k ∈ Kc, ∀τ .

This leads to an increment∑

k∈KcU

(τ+1)km (τ + 1)-th match-

ing. Therefore, given the proposals from the SUEs, everyacceptance/rejection process from ϕ

(τ)m → ϕ

(τ+1)m produces

a dominance for allocating chunks to the SUEs as follows:

ϕ(τ)m → ϕ(τ+1)

m ⇔∑

k∈Kc

U(τ+1)km >

∑k∈Kc

U(τ)km ,

∀m ∈ M, k ∈ Kc, ∀τ (43)

Therefore, from (42) and (43), we can see that the objectivefunction (

∑k∈Kc

Ωknm(P k,(τ)

nm ) − φ(τ)c,k) is dominated by the

Page 15: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

1374 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 18, NO. 2, FEBRUARY 2019

matching ϕ(τ)m . In addition, the strategies of SUEs in stable

matching ϕ∗m give a locally optimal solution of the channel

allocation in the ROC-Dm problem.

REFERENCES

[1] T. Nakamura et al., “Trends in small cell enhancements in LTEadvanced,” IEEE Commun. Mag., vol. 51, no. 2, pp. 98–105, Feb. 2013.

[2] R. Zhang, M. Wang, L. X. Cai, Z. Zheng, X. Shen, and L.-L. Xie,“LTE-unlicensed: The future of spectrum aggregation for cellular net-works,” IEEE Wireless Commun., vol. 22, no. 3, pp. 150–159, Jun. 2015.

[3] G. Yuan, X. Zhang, W. Wang, and Y. Yang, “Carrier aggregation forLTE-advanced mobile communication systems,” IEEE Commun. Mag.,vol. 48, no. 2, pp. 88–93, Feb. 2010.

[4] (2014). E. Q. T. I. LTE-Forum, Formed by Verizon in Coop-eration With Alcatel-Lucent and Samsung. [Online]. Available:http://www.lteuforum.org/lte-u-forum/

[5] R. Ratasuk, N. Mangalvedhe, and A. Ghosh, “LTE in unlicensedspectrum using licensed-assisted access,” in Proc. IEEE GlobecomWorkshops (GC Wkshps), Austin, TX, USA, Dec. 2014, pp. 746–751.

[6] X. Wang, S. Mao, and M. X. Gong, “A survey of LTE Wi-Fi coexistencein unlicensed bands,” GetMobile, Mobile Comput. Commun., vol. 20,no. 3, pp. 17–23, 2017.

[7] Y. Gu, Y. Zhang, L. Cai, M. Pan, L. Song, and Z. Han, “LTE-unlicensedcoexistence mechanism: A matching game framework,” IEEE WirelessCommun., vol. 23, no. 6, pp. 54–60, Dec. 2016.

[8] B. Chen, J. Chen, Y. Gao, and J. Zhang, “Coexistence of LTE-LAA andWi-Fi on 5 GHz with corresponding deployment scenarios: A survey,”IEEE Commun. Surveys Tuts., vol. 19, no. 1, pp. 7–32, 1st Quart., 2017.

[9] H. Zhang, Y. Xiao, L. X. Cai, D. Niyato, L. Song, and Z. Han,“A multi-leader multi-follower Stackelberg game for resource manage-ment in LTE unlicensed,” IEEE Trans. Wireless Commun., vol. 16, no. 1,pp. 348–361, Jan. 2017.

[10] W. Xu, B. Li, Y. Xu, and J. Lin, “Lower-complexity power allocationfor LTE-U systems: A successive cap-limited waterfilling method,” inProc. IEEE 81st Veh. Technol. Conf. (VTC), Glasgow, U.K., May 2015,pp. 1–6.

[11] Y. Gu, Y. Zhang, L. X. Cai, M. Pan, L. Song, and Z. Han,“Exploiting student-project allocation matching for spectrum sharing inLTE-unlicensed,” in Proc. IEEE Global Commun. Conf. (GLOBECOM),San Diego, CA, USA, Dec. 2015, pp. 1–6.

[12] A. M. Cavalcante et al., “Performance evaluation of LTE andWi-Fi coexistence in unlicensed bands,” in Proc. IEEE Veh. Technol.Conf. (VTC), Dresden, Germany, Jun. 2013, pp. 1–6.

[13] A. Babaei, J. Andreoli-Fang, and B. Hamzeh, “On the impact ofLTE-U on Wi-Fi performance,” in Proc. IEEE 25th Annu. Int. Symp.Pers., Indoor, Mobile Radio Commun. (PIMRC), Washington, DC, USA,Sep. 2014, pp. 1621–1625.

[14] R. C. Paiva, P. Papadimitriou, and S. Choudhury, “A physical layerframework for interference analysis of LTE and Wi-Fi operating inthe same band,” in Proc. Asilomar Conf. Signals, Syst. Comput.,Pacific Grove, CA, USA, Nov. 2013, pp. 1204–1209.

[15] E. Almeida et al., “Enabling LTE/WiFi coexistence by LTE blank sub-frame allocation,” in Proc. IEEE Int. Conf. Commun. (ICC), Budapest,Hungary, Jun. 2013, pp. 5083–5088.

[16] H. Zhang, X. Chu, W. Guo, and S. Wang, “Coexistence of Wi-Fi andheterogeneous small cell networks sharing unlicensed spectrum,” IEEECommun. Mag., vol. 53, no. 3, pp. 158–164, Mar. 2015.

[17] F. M. Abinader et al., “Enabling the coexistence of LTE and Wi-Fi inunlicensed bands,” IEEE Commun. Mag., vol. 52, no. 11, pp. 54–61,Nov. 2014.

[18] F. S. Chaves et al., “LTE UL power control for the improvement ofLTE/Wi-Fi coexistence,” in Proc. IEEE Veh. Technol. Conf. (VTC),Las Vegas, NV, USA, Sep. 2013, pp. 1–6.

[19] T. LeAnh et al., “Matching theory for distributed user association andresource allocation in cognitive femtocell networks,” IEEE Trans. Veh.Technol., vol. 66, no. 9, pp. 8413–8428, Sep. 2017.

[20] H. Zhang, C. Jiang, N. C. Beaulieu, X. Chu, X. Wang, and T. Q. S. Quek,“Resource allocation for cognitive small cell networks: A cooperativebargaining game theoretic approach,” IEEE Trans. Wireless Commun.,vol. 14, no. 6, pp. 3481–3493, Jun. 2015.

[21] D. Gale and L. S. Shapley, “College admissions and the stability ofmarriage,” Amer. Math. Monthly, col. 69, no. 1, pp. 9–15, 1962.

[22] A. E. Roth and M. A. O. Sotomayor, Two-Sided Matching: A Study inGame-Theoretic Modeling and Analysis. Cambridge, U.K.: CambridgeUniv. Press, 1992.

[23] F. Teng, D. Guo, and M. L. Honig, “Sharing of unlicensed spectrumby strategic operators,” IEEE J. Sel. Areas Commun., vol. 35, no. 3,pp. 668–679, Mar. 2017.

[24] Y. Gu, C. Jiang, L. X. Cai, M. Pan, L. Song, and Z. Han, “Dynamicpath to stability in LTE-unlicensed with user mobility: A match-ing framework,” IEEE Trans. Wireless Commun., vol. 16, no. 7,pp. 4547–4561, Jul. 2017.

[25] T. LeAnh, N. H. Tran, D. T. Ngo, and C. S. Hong, “Resource allocationfor virtualized wireless networks with backhaul constraints,” IEEECommun. Lett., vol. 21, no. 1, pp. 148–151, Jan. 2017.

[26] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:Cambridge Univ. Press, 2009.

[27] M. Jaber, M. A. Imran, R. Tafazolli, and A. Tukmanov, “5G backhaulchallenges and emerging research directions: A survey,” IEEE Access,vol. 4, pp. 1743–1766, 2016.

[28] O. Tipmongkolsilp, S. Zaghloul, and A. Jukan, “The evolution of cellularbackhaul technologies: Current issues and future trends,” IEEE Commun.Surveys Tuts., vol. 13, no. 1, pp. 97–113, 1st Quart., 2011.

[29] C. Liang and F. R. Yu, “Wireless network virtualization: A survey, someresearch issues and challenges,” IEEE Commun. Surveys Tuts., vol. 17,no. 1, pp. 358–380, Mar. 2015.

[30] Cisco. (2014). Cisco Global Cloud Index: Forecast andMethodology, 2013–2018. [Online]. Available: https://www.digital4.biz/upload/images/10-2013/131028130134.pdf

[31] Ericsson. (Nov. 2012). Ericsson Mobility Report, on the Pulseof the Networked Society. [Online]. Available: http://www.ericsson.com/res/docs/2012/ericsson-mobility-report-november-2012.pdf

[32] J. Ghimire and C. Rosenberg, “Revisiting scheduling in heterogeneousnetworks when the backhaul is limited,” IEEE J. Sel. Areas Commun.,vol. 33, no. 10, pp. 2039–2051, Oct. 2015.

[33] H. Cui, V. C. M. Leung, S. Li, and X. Wang, “LTE in the unli-censed band: Overview, challenges, and opportunities,” IEEE WirelessCommun., vol. 24, no. 4, pp. 99–105, Aug. 2017.

[34] J. Oueis and E. C. Strinati, “Uplink traffic in future mobile networks:Pulling the alarm,” in Proc. Int. Conf. Cogn. Radio Oriented WirelessNetw. Philadelphia, PA, USA: Springer, 2016, pp. 583–593.

[35] D. Liu et al., “User association in 5G networks: A survey and anoutlook,” IEEE Commun. Surveys Tuts., vol. 18, no. 2, pp. 1018–1044,2nd Quart. 2016.

[36] H. Zhu and J. Wang, “Chunk-based resource allocation in OFDMAsystems—Part I: Chunk allocation,” IEEE Trans. Commun., vol. 57,no. 9, pp. 2734–2744, Sep. 2009.

[37] T. Q. Quek, G. de la Roche, I. Güvenç, and M. Kountouris, Small CellNetworks: Deployment, PHY Techniques, and Resource Management.Cambridge, U.K.: Cambridge Univ. Press, 2013.

[38] T. Weiss, J. Hillenbrand, A. Krohn, and F. K. Jondral, “Mutual interfer-ence in OFDM-based spectrum pooling systems,” in Proc. IEEE 59thVeh. Technol. Conf. (VTC), vol. 4, May 2004, pp. 1873–1877.

[39] H. W. Kuhn, “The Hungarian method for the assignment problem,”Naval Res. Logistics Quart., vol. 2, nos. 1–2, pp. 83–97, Mar. 1955.

[40] Y. Gu, W. Saad, M. Bennis, M. Debbah, and Z. Han, “Matching theoryfor future wireless networks: Fundamentals and applications,” IEEECommun. Mag., vol. 53, no. 5, pp. 52–59, May 2015.

[41] H. Wang and Z. Ding, “Power control and resource allocation for outagebalancing in femtocell networks,” IEEE Trans. Wireless Commun.,vol. 14, no. 4, pp. 2043–2057, Apr. 2015.

Tuan LeAnh received the B.Eng. and M.Eng.degrees in electronic and telecommunication engi-neering from the Hanoi University of Technology,Vietnam, in 2007 and 2010, respectively, and thePh.D. degree from Kyung Hee University, Seoul,South Korea, in 2017.

From 2007 to 2011, he was a Member of Tech-nical Staff with the Network Operations Center,Vietnam Telecoms National Co., Vietnam Posts andTelecommunications Group (VNPT). Since 2017,he has been a Post-Doctoral Researcher with the

Department of Computer Science and Engineering, Kyung Hee University.His research interests include queueing, optimization, control and game theoryto design, analyze, and optimize the resource allocation in communicationnetworks, including cognitive radio, wireless network virtualization, URLLC,mobile edge computing, long term evolution-unlicensed, and vehicular com-munications.

Page 16: Orchestrating Resource Management in LTE-Unlicensed Systems …networking.khu.ac.kr/layouts/net/publications/data/2019... · 2019-04-01 · 1360 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,

LEANH et al.: ORCHESTRATING RESOURCE MANAGEMENT IN LTE-UNLICENSED SYSTEMS 1375

Nguyen H. Tran (S’10–M’11–SM’18) received theB.S. degree from the Ho Chi Minh City Universityof Technology in 2005 and the Ph.D. degree inelectrical and computer engineering from KyungHee University (KHU) in 2011.

From 2012 to 2017, he was an Assistant Profes-sor with the Department of Computer Science andEngineering, Kyung Hee University. Since 2018, hehas been with the School of Computer Science, TheUniversity of Sydney, where he is currently a SeniorLecturer. His research interest is applying analytic

techniques of optimization, game theory, and machine learning to cutting-edge applications such as edge computing, datacenters, resource allocationfor 5G networks, and the Internet of Things. He was a recipient of the BestKHU Thesis Award in Engineering, in 2011, and several best paper awards,including the IEEE ICC 2016, the APNOMS 2016, and the IEEE ICCS 2016.He received the Korea NRF Funding for Basic Science and Research from2016 to 2023. He has been an Editor of the IEEE TRANSACTIONS ON GREEN

COMMUNICATIONS AND NETWORKING since 2016.

Duy Trong Ngo (S’08–M’15) received the B.Eng.degree (Hons.) in telecommunication engineeringfrom The University of New South Wales, Australia,in 2007, the M.Sc. degree in electrical engineering(communication) from the University of Alberta,Canada, in 2009, and the Ph.D. degree in electri-cal engineering from McGill University, Canada,in 2013.

Since 2013, he has been a Senior Lecturer withthe School of Electrical Engineering and Computing,The University of Newcastle, Australia. He currently

leads the research effort in design and optimization for the next-generationwireless communications networks. His research interests include cloudradio access networks, multi-access edge computing, simultaneous wirelessinformation and power transfer, and vehicle-to-everything communicationsfor intelligent transportation systems.

Zhu Han (S’01–M’04–SM’09–F’14) received theB.S. degree in electronic engineering from TsinghuaUniversity in 1997 and the M.S. and Ph.D. degreesin electrical and computer engineering from theUniversity of Maryland, College Park, MD, USA,in 1999 and 2003, respectively.

From 2000 to 2002, he was a Research and Devel-opment Engineer with JDSU, Germantown, MD,USA. From 2003 to 2006, he was a Research Asso-ciate with the University of Maryland. From 2006 to2008, he was an Assistant Professor with Boise State

University, Boise, ID, USA. He is currently a Professor with the Electricaland Computer Engineering Department, University of Houston, Houston, TX,USA, and also with the Computer Science Department, University of Houston.His research interests include wireless resource allocation and management,wireless communications and networking, game theory, big data analysis,security, and smart grid. He is currently an IEEE Communications SocietyDistinguished Lecturer. He was a recipient of the NSF Career Award in 2010,the Fred W. Ellersick Prize of the IEEE Communication Society in 2011,the EURASIP Best Paper Award for the Journal on Advances in SignalProcessing in 2015, the IEEE Leonard G. Abraham Prize in the field ofcommunications systems (the Best Paper Award in the IEEE JSAC) in 2016,and several best paper awards in IEEE conferences.

Choong Seon Hong (AM’95–M’07–SM’11)received the B.S. and M.S. degrees in electronicengineering from Kyung Hee University, Seoul,South Korea, in 1983 and 1985, respectively, andthe Ph.D. degree from Keio University, Tokyo,Japan, in 1997.

In 1988, he joined KT, where he focused onbroadband networks as a Member of the TechnicalStaff. In 1993, he joined Keio University. He waswith the Telecommunications Network Laboratory,KT, as a Senior Member of Technical Staff and as

the Director of the Networking Research Team until 1999. Since 1999, he hasbeen a Professor with the Department of Computer Science and Engineering,Kyung Hee University. His research interests include future Internet, ad hocnetworks, network management, and network security. He is a member ofACM, IEICE, IPSJ, KIISE, KICS, KIPS, and OSIA. He has served as theGeneral Chair, a TPC Chair/Member, or an Organizing Committee Memberfor international conferences such as NOMS, IM, ICC, GLOBECOM,APNOMS, E2EMON, CCNC, ADSN, ICPP, DIM, WISA, BcN, TINA,SAINT, ICOIN, and BigComp. Also, he was an Associate Editor of theIEEE TRANSACTIONS ON SERVICES AND NETWORKS MANAGEMENT.He is currently an Associate Editor of the International Journal of NetworkManagement, the Journal of Communications and Networks, and the FutureInternet Journal and an Associate Technical Editor of IEEE CommunicationsMagazine.