research article cbrs spectrum sharing between … article cbrs spectrum sharing between lte-u and...

Research ArticleCBRS Spectrum Sharing between LTE-U and WiFiA Multiarmed Bandit Approach

Imtiaz Parvez1 M G S Sriyananda1 Esmail Guumlvenccedil2 Mehdi Bennis3 and Arif Sarwat1

1Department of Electrical amp Computer Engineering Florida International University Miami FL 33174 USA2Department of Electrical amp Computer Engineering North Carolina State University Raleigh NC 27513 USA3Department of Communications Engineering University of Oulu 90014 Oulu Finland

Correspondence should be addressed to Arif Sarwat asarwatfiuedu

Received 31 March 2016 Revised 14 June 2016 Accepted 19 July 2016

Academic Editor Miguel Lopez-Benıtez

Copyright copy 2016 Imtiaz Parvez et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

The surge of mobile devices such as smartphone and tablets requires additional capacity To achieve ubiquitous and high data rateInternet connectivity effective spectrum sharing and utilization of the wireless spectrum carry critical importance In this paper weconsider the use of unlicensed LTE (LTE-U) technology in the 35 GHzCitizens BroadbandRadio Service (CBRS) band and developamultiarmed bandit (MAB) based spectrum sharing technique for a smooth coexistence withWiFi In particular we consider LTE-U to operate as a General Authorized Access (GAA) user herebyMAB is used to adaptively optimize the transmission duty cycle ofLTE-U transmissions Additionally we incorporate downlink power control which yields a high energy efficiency and interferencesuppression Simulation results demonstrate a significant improvement in the aggregate capacity (approximately 33) and cell-edgethroughput of coexisting LTE-U and WiFi networks for different base station densities and user densities

1 Introduction

Due to the proliferation of mobile devices and diverse mobileapplications the exponentially increasingmobile data is dou-bled approximately every year [1] The 4G Long-Term Evolu-tion (LTE) has recently emerged as a powerful technology toprovide broadband data rates On the other hand to satisfythe throughput demand of broadband LTE networks in theupcoming years larger bandwidth is needed [2 3] Since thelicensed spectrum is expensive and limited extending theoperation of LTE in the underutilized unlicensed bands isrecently getting significant attention which requires effectivecoexistence with other technologies such as WiFi in thesebands

Recently the Federal Communications Commission(FCC) in the United States has been working on opening a150MHz of spectrum in the 35 GHz band for sharing amongmultiple technologies which is also commonly referred to asthe Citizen Broadband Radio Service (CBRS) However theuse of this spectrum is subject to regularity requirementswhere the incumbent military and meteorological radar

systems have to be protected [4 5] In the CBRS band thereare three kinds of users with hierarchical priority IncumbentAccess (IA) users (tier-1) Prioritized Access License (PAL)users (tier-2) and General Authorized Access (GAA) users(tier-3) as illustrated in Figure 1 In the current scenariothe expansion of unlicensed LTE (LTE-U) as PAL or GAAuser in the CBRS band is an enticing choice because ofhigh penetration at 35 GHz clean channel and wide amountof spectrum [6] The Third-Generation Partnership Project(3GPP) standardization group has been recently working onstandardizing the licensed-assisted access (LAA) technologyin the 5GHz spectrum [7 8] The main goal is to developa global single framework of LAA of LTE in the unlicensedbands where operation of LTE will not critically affect theperformance of WiFi networks in the same carrier In theinitial phase only downlink (DL) operation LTE-A (LTEAdvanced) Carrier Aggregation (CA) in the unlicensed bandwas considered while deferring the simultaneous operationof DL and uplink (UL) to the next phase

Another option for the operation of LTE in the unlicensedspectrum is through a prestandard approach referred to

Hindawi Publishing CorporationMobile Information SystemsVolume 2016 Article ID 5909801 12 pageshttpdxdoiorg10115520165909801

2 Mobile Information Systems

for secondary use by PAL

Federal RLS and ARNS use Federal RLS3 sites only

Tier-1

Tier-2

Tier-3

Pote

ntia

l ban

ds fo

r LTE

-U

depl

oym

ent

3700 MHz3650MHz3550MHz

150MHz channel for use by GAAA minimum of 80 MHz and maximum of

Floating 70 MHz spectrum

Figure 1 CBRS spectrum with 3 types of users

LTE-U where LTE base stations leave transmission gaps forfacilitating coexistence with WiFi networks Development ofLTE-U technology is led by the industry consortium knownas the LTE-U Forum LTE-Umainly focuses on the operationof unlicensed LTE in the regions (eg USA China) wherelisten before talk (LBT) is not mandatory LTE-U definesthe operation of primary cell in a licensed band with oneor two secondary cells (SCells) each 20MHz in the 5GHzunlicensed band U-NII-1 andor U-NII-3 bands spanning5150ndash5250MHz and 5725ndash5825MHz respectively Howeverboth the LTE-U and LAA need licensed band for controlplane Similar to the 5GHz band CBRS band can be utilizedfor LTE-U operation in the absence of IA users such as radarsignal

In our study we consider the coexistence problemof LTE-U andWiFi networks in the CBRS bands SinceWiFi adopts acontention based medium access control with random back-off [9] for channel access and LTE uses dynamic schedulingfor users the unrestrained LTE operation in the same bandwill generate continuous interference on WiFi service Tooperate LTE-U and WiFi simultaneously in the same unli-censed spectrum fair and reasonable coexistencemechanismis indispensable The adverse impact on DL and UL WiFitransmissions due to LTE deployment in the same bandis analyzed in [10ndash12] emphasizing the need for rigorousstudies In this regard discrete mechanisms such as dynamicchannel selection retaining transmission gaps transmissionduty cycle manipulation and LBT have been proposed inthe literature for harmonious coexistence with improvedperformance To select resources dynamically learn from theenvironment and adaptively modify transmission parame-ters for performance improvement variousmachine learningbased techniques [13ndash16] have been introduced

In this paper we introduce a reinforcement learning(MAB) based adaptive duty cycle section for the coexistencebetween LTE-U and WiFi Multiarmed Bandit (MAB) is amachine learning technique designed to maximize the long-term rewards through learning provided that each agentis rewarded after pulling an arm Basically MAB [17 18]problem resembles a gambler (agent) with a finite number ofslot machines in which the gambler wants to maximum hisrewards over a time horizon Upon pulling an arm a rewardis attained with prior unknown distribution The goal is topull arms sequentially so that the accumulated rewards overthe gambling period are maximized However the problem

involves the exploration versus exploitation trade-off that istaking actions to yield immediate higher reward on the onehand and taking actions that would give rewards in the futureon the other hand

In our technique we use a multiarm bandit (MAB)algorithm for selecting appropriate duty cycle Using a 3GPPcompliant Time Division Duplex- (TDD-) LTE and Beaconenabled IEEE 802 systems in the 35 GHz band we simu-late and evaluate the coexistence performance for differentpercentage of transmission gaps We found a significantthroughput improvement for both systems ensuring harmo-nious coexistence The objectives subsequently the gains ofthis study are not limited to throughput enhancements Thebenefits that are achieved in different dimensions with the aidof MAB scheme and the other supporting techniques like PCcan be summarized as follows

(1) Proper coexistence is achieved due to the dynamicexploring and exploitation byMAB So our techniqueis adaptive

(2) The aggregate capacity is improved Due to theapplication ofMAB algorithm optimal or suboptimalsolutions are achieved

(3) Using DL PC higher capacity values are achievedunder dense UE and STA configurations

(4) Higher energy efficiency is also achieved with PCwhich always attempts to reduce the transmissionpower while increasing the energy efficiency

(5) With the use of learning algorithm a high degree ofefficiency is achieved

To the best of our knowledge our work is the first study thatintroduces MAB for improving the coexistence of LTE andWiFi in the unlicensed bands

The rest of the paper is organized as follows Section 2pro-vides a literature review of coexistence of LTE-U andWiFi InSection 3 we provide our systemmodel and problem formu-lation for LTE andWiFi coexistence Section 4 introduces theproposedMABbased dynamic duty cycle selection approachSimulation results with various parameter configurations arepresented in Section 5 Finally Section 6 provides concludingremarks

2 Related Works

21 Coexistence among Unlicensed LTE and WiFi In theliterature several studies can be found that investigate theperformance of LTE and WiFi coexistence in the unlicensedbands In [19] coexistence performance of LTE and WiFihas been investigated in 900MHz considering single floorand multifloor indoor office scenarios It is shown that theperformance of WiFi is heavily affected when WiFi and LTEoperate simultaneously in the unlicensed spectrum

To facilitate harmonious coexistence between LTE-U andWiFi in the same band mainly three techniques have beenproposed in the literature (1) listen before talk (LBT) (2)dynamic channel selection and (3) coexistence gaps InEurope and Japan LBT is mandatory for data offloading in

Mobile Information Systems 3

unlicensed band The usage of LBT has been justified in[20] with different choice of LBT schemes In [21] LBT ispresented considering interradio access technology (RAT)and intra-RAT In this technique energy detection based LBTis proposed to handle inter-RAT interference whereas crosscorrelation based LBT is used to handle intra-RAT interfer-ence However LBT is not mandatory in USA and Chinawhere alternative coexistence techniques can be explored

In [22]Qualcommpresents an effective channel selectionpolicy based on interference level If the interference of theoccupied channel exceeds a certain level LTE-U changes thechannel provided that the interference is measured beforeand during the operation and both at the user equipment(UE) and the network side On the other hand in [6]adaptive bandwidth channel allocation offered by LTE andLeast Congested Channel Search (LCCS) has been suggestedfor channel selection Dynamic channel selection requiresfree or low-interference channel to utilize Since same bandwill be shared by other cellular service providers as well asdifferent technologies such as WiFi finding of clean channelmay not be practical

In [23] blank subframe allocation by LTE has beenproposed where LTE is restrained from transmitting andWiFi keeps on transmission A similar technique has beenproposed in [24] where certain subframes of LTE-U arereserved for WiFi transmission Qualcomm has proposedCarrier Sensing Adaptive Transmission (CSAT) [22] for LTE-U MAC scheduling in which a fraction of TDD duty cycle isused for LTE-U transmission and the rest is used for othertechnologies The cyclic ONOFF ratio can be adaptivelyadjusted based on the activity ofWiFi during the OFF periodIn this paper we focus on the dynamic optimization of coex-istence gaptransmission time along with DL power control

Uplink (UL) power control has been investigated onthe performance of LTE-WiFi coexistence in [25 26] How-ever DL power control in coexistence problem has notbeen explored yet considering uncoordinated LTE and WiFisystems The DL power control enhances performance byreducing interferences which is demonstrated in [27ndash29] Inour study we optimize both the transmission time and DLpower using machine learning technique

Reinforcement algorithm such as Q-learning multiarmbandit and value iteration is effective variant of machinelearning which has been applied for optimization problemsof cellular systems such as channel selection mobility man-agement resource allocation and rate adoption In [13]Q-learning based duty cycle adjustment is presented tofacilitate the sharing of the channel and to increase theoverall throughput In [30] aMAB based distributed channelselection is proposed to use vacant cellular channels in deviceto device (D2D) communication To enhance handoverprocess and increase throughput MAB techniques basedcontext-aware mobility management scheme is studied in[31] In [32] dynamic rate adaptation and channel selectionfrom free primary users have been proposed in cognitiveradio systems usingMAB which yields extensive throughputimprovements

In our studywe propose aMABbased dynamic duty cycleselection for unlicensed LTE systems In particular LTE base

Tier‐1 IA system

Tier‐2 PALcontroller

Tier‐3 GAAcontroller

Federal SAS Federal database

Interface

Commercial SAS‐2Commercial SAS‐1

Tier‐2 PALRAN user

Tier‐3 GAA Tier‐3 GAA Tier‐3 GAAuser‐1 user‐2 user‐2 CB

RS w

ith li

cens

ed sh

ared

acce

ss (L

SA)

middot middot middot

Figure 2 Users access priority

stations (BSs) measure the utilization of the channel based onchannel status information (CSI) learn the channel utiliza-tion of WiFi (current and previous) select the optimum dutycycle and transmission power and perform transmissionunder this duty cycle which results in effective sharing ofwireless spectrum with WiFi networks Due to this dynamiclearning our technique is adaptive and it improves aggregatecapacity and energy efficiency This is the first time we areapplying MAB for coexisting operation of LTE and WiFi

22 CBRS Spectrum Sharing The CBRS spectrum is com-posed of 150MHz bandwidth divided into two chunks80MHz and 70MHz Based on the architecture of CBRSband the spectrum users are prioritized into three groupswith decreasing interference protection requirements as illus-trated in Figure 2

The IA users in tier-1 such as military radars havethe most protection mainly through geographical exclusionzones [33] that averts other users from transmiting in thevicinity of IA users While the NTIA in April 2015 [5 34]shrunk the earlier exclusion zones in [33] by 77 they stillcover several of the Nationrsquos largest cities [35] The mainchallenge of PAL users in tier-2 have is to protect the IAusers and other PAL users from interference To facilitatethis a spectrum access system (SAS) [36] is utilized whichgrants spectrum access to users based on their locationsThe network providers can purchase PAL licenses in givengeographical areas which consist of census tracts Up to a70MHz of PAL spectrum will be available with chunks of10MHz channels which will be auctioned if there is moredemand from providers than the available spectrum Finallytier-3 users are GAAusers which are allowed to operate in thespectrum that are not used by IA and PAL tiers In areas withno IA and PAL activity GAA users may have access to whole150MHz while in areas with PAL activity but outside of IAexclusion zones at least 80MHz of bandwidth will always beavailable for GAA use

Since spectrum is limited and expensive wireless serviceprovider (LTE WiFi) will be interested to operate in CBRSband as GAA users In the GAA band LTE needs to coexistwith other cellular operators as well as other technologiessuch as WiFi Besides that Licensed Shared Access (LSA)concept [37 38] allows an incumbent spectrum user to share


LTE BS

LTE-U UE

LTE-U UE

LTE BS

WiFi AP

WiFi STA

Desired signalInterference

TE-U UE

WiFi STA

(a) Interference on LTE-U DL and WiFi UL

LTE-U UE

LTE-U UE


LTE BS

LTE BS

WiFi AP

WiFi STA

E-U UE

WiFi STA

(b) Interference on LTE-U UL and WiFi DL

Figure 3 DL and UL interference scenarios for LTE-UWiFi transmissions

spectrum with licensed users with defined rights to accessa portion of spectrum at a given location and time Thisalso requires to develop coexistence mechanism betweenmobile network operators (MNOs) and other technologists(licensedunlicensed) such as WiFi In this study we focuson the coexistence of LTE and WiFi in the 35 GHz CBRSspectrum For this study for simplicity we assume that thecoexistence with IA and PAL users are already maintainedthrough a SAS database and we only consider coexistenceamong LTE-U and WiFi users in the GAA bands

3 System Model and Problem Formulation

To evaluate the coexistence performance of LTE-UwithWiFiin the unlicensed band a collocated LTE-U andWiFi networkscenario is consideredThe sets of LTE-UBSsWiFiAPs LTE-UUEs for BS 119894 andWiFi STAs forAP119908 are given byB

119871B119882

Q119894119871 and Q119908

119882 respectively Q

119871= Q1119871Q2119871 Q119894

119871 Q

|B119871|

119871

and Q119882= Q1119882Q2119882 Q119908

119882 Q

|B119882|

119882 represent the sets of

all UEs and STAs For LTE-U TDD-LTE is considered Forsynchronization of WiFi STAs with the corresponding APs aperiodic beacon transmission is used as in [13]

31 Interference on DL and UL Transmissions Interferencecaused to LTE-UUE and LTE-U BS during DL and UL trans-missions is shown in Figure 3 A TDD frame structure similarto that in [39 Figure 62] is considered for all the BSs andUEswith synchronous operation As shown in Figure 3(a) in thesimultaneous operation of an LTE-U within a WiFi coveragearea the DL LTE-U radio link experiences interference fromother LTE-U DL and WiFi UL transmissions As the sametimeWiFi UL suffers fromnear LTE-U transmission Duringan UL transmission subframe shown in Figure 3(b) LTE-U BS is interfered by the UL transmission of LTE-U UEsas well as the DL transmissions of WiFi Similarly WiFiDL transmission is interfered by other LTE-U ULs wherethe DL received signal of a WiFi STA is interfered by otherLTE-U UL transmissions In the coexistence scenarios with

high density of WiFi users WiFi transmissions get delayeddegrading their capacity performance due to the use of carriersense multiple access with collision avoidance (CSMACA)mechanism [40] This is an additional degradation otherthan the performance reduction experienced due to LTE-Utransmissions operated on the same spectrumand this is validonly for WiFi APs and STAs

32 Duty Cycle of LTE-U In the case of designing a duty cyclefor LTE-Umultiple LTETDD frames are considered For thatpurpose five consecutive LTE frames [39 Figure 62(a)] areused to construct a duty cycle Similar to [13] the LTE-UtransmissionONOFF condition is used to define a duty cyclewhich is shown in Figure 4 (eg 40 duty cycle during thefirst two consecutive LTE-U frames transmission is turnedon and it is turned off during the following three frames) Oneout of these two configurations is used by the UEs and BSin an LTE cell during a duty cycle period According to thisstructure a constant ULDL duty cycle value is maintained

33 Capacity Calculation and Power Control For any BS 119894 isinQ119871 there are N119894 resource blocks (RBs) for the DL For a

given UE 119906 associated with BS 119894 119899119894119906RBs are allocated where

N119894 = sum|Q119894119871|

119906=1119899119894119906 119901119894119904119903 119901119887119904119903 119901119886119904119903 and 119901119902

119904119903are transmit power

values associated with RB 119903 and the transmit power index 119904from the LTE-U BS 119894 LTE-U BS 119887 (119894 = 119887) WiFi AP 119886 andWiFi STA 119902 119894th BS is considered as the desired BS where theBSs indexed by 119887 are the interference generating BSs For anyAP UE or STA total transmit power is equally distributedamong all RBsHowever in every BS the total transmit poweris dynamically changed for every duty cycle according toMAB algorithm ℎ119894

119906119903 ℎ119887119906119903 ℎ119886119906119903 and ℎ119902

119906119903are the channel gain

values from BS 119894 to UE 119906 from BS 119887 to UE 119906 from AP 119886

to UE 119906 and from WiFi STA 119902 to UE 119906 respectively Allchannel gain values are calculated considering path lossesand shadowing In that case interference generated to UE119906 from BSs APs and STAs are given by 119868119906BS 119868

119906

AP and 119868119906

STArespectively Since a synchronized transmission is considered


80 Percentage ofthe duty cyclefor an LTE-U transmission

6040

20

LTE TDD frame LTE TDD frame LTE TDD frame LTE TDD frame LTE TDD frame

Duty cycle

Figure 4 Structure of the duty cycle for LTE-U transmissions

there is no interference from the UL transmission of LTE-U UEs Noise variance is denoted by 1205902 The Signal-to-Interference-plus-Noise Ratio (SINR) expression for UE 119906

served by BS 119894 on RB 119903 at time interval 119896 is given as

SINR119894119906119903[119896]

=119901119894119904ℎ119894119906119903

sum119887isinB119871119894

119901119887119904ℎ119887119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

BS

+ sum119886isinB119882

119901119886119904ℎ119886119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

AP

+ sum119902isinQ119882

119901119902119904ℎ119902119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

STA

+ 1205902 (1)

where 119887 119894 isin B119871

The amount of successfully transmitted data bits119873119861from

119894th LTE-U BS during 119879OFDM time interval 119896 within an activeDL subframes of a duty cycle is given by

119873119894

119861=

K119894

sum119896

sum

119906isinQ119894119871

119877119906

sum119903

119882119894

119906119903log2(1 + SINR119894

119906119903[119896]) 119879OFDM (2)

where119879OFDM is the orthogonal frequency divisionmultiplex-ing (OFDM) symbol duration 119879119894Tx = K119894119879OFDM and K119894 isthe total number of transmit 119879OFDM time intervals for theconsidered duty cycle The total allocated bandwidth for RB119903 for UE 119906 served by BS 119894 is 119882119894

119906119903 The average capacity over

a duty cycle period is used as a performance measure in thisstudy as in [13] The DL capacity 119862

119894of LTE-U BS 119894 is given as

119862119894=

119873119894119861

119879119894Tx + 119879119894

Wait (3)

where 119879119894Wait is the waiting time due to silent subframeallocation

The capacity 119862119894in (3) is used as a performance mea-

sure for each LTE-U BS Since the transmit power of oneBS contributes to the interference power of the other BSneighboring BSs are coupled in terms of interference Thegoal of every BS is to maximize 119862

119894while minimizing the DL

transmit power 119901119894119904 forall119894 isin B

119871 By minimizing the transmit

power values 119901119894119904and 119901119887

119904 the goal is to achieve a comparatively

higher energy efficiency than the case of constantDL transmitpower In the same time a reduction in interference is alsoexpected while guaranteeing a minimum capacity Moreover119875min le 119901

119887

119904le 119875max where 119875min and 119875max are the minimum and

maximum transmit power constraints respectivelyThemin-imum capacity corresponding to a given action is denoted by

119862min119895

The objective is to maximize the average capacity whileminimizing the transmit power which can be written as

maximizesum|B119871|

119894=1119862119894

1003816100381610038161003816B1198711003816100381610038161003816

(4)

minimize 119901119894

119904forall119894 isin B

119871(5)

subject to 119901119894

119904 119901119887

119904 le 119875max

forall119894 119887 isin B119871 119894 = 119887 119904 isin 119878

(6)

119901119894

119904 119901119887

119904 ge 119875min


(7)

119862119894gt 119862

min119895

forall119894 isin B119871 forall119895 isin 119869 (8)

In the case of energy efficiency several parameter config-urations are considered for (8) as

119862119894

119901119894119904

gt119862min119895

119901119894119904

or119862119894

119901119894119904

gt119862min119895

119875min

or119862119894

119901119894119904

gt119862min119895

119875max

(9)

Due to the same denominator 119862119894119901119894119904gt 119862min119895

119901119894119904is simplified

to (8) which can be used as a proportional measure ofenergy efficiencyThe problem is reformulated defining a newobjective to maximize energy efficiency as follows


119894=1(119862119894119901119894119904)

1003816100381610038161003816B1198711003816100381610038161003816

subject to (6) (7) and (9)

(10)

4 MAB Techniques forLTE-U WiFi Coexistence

In a MAB problem an agent selects an action (also knownas arm) and observes the corresponding rewardThe rewardsfor given actionarms are random variables with unknowndistribution The goal of MAB is to design action selection


(1) Initialization(2) Set the minimum capacity values 119862min

119895 forall119895 isin 119869 Exploration steps119872 Beta (1 1) 120572119894

119895and 120573119894

119895where forall119895 119895 isin 119869

Select 119889119894119895 forall119895 isin 119869 update 119904 119899

1198940(119889119894119895) V1198940(119889119894119895) and accumulated hypothesisreward 119877

119894(119889119894119895) based on 119862

119894gt 119862min119895

(3) if 120572119894119895(119898) = 120573119894

119895(119898) forall(119897 119898) isin 119872 then

(4) Exploration(5) for119898 = 1 2 3 119872 do(6) Select 119889119894

119895 119889119894119895isin D119894 119895 isin U(1 |D

119894|) cap 119869 and update 119904 (8)

(7) Execute 119889119894119895 119901119894119904 observe 119862

119894and update 119899

119894119898(119889119894119895)

(8) if 119862119894gt 119862min119895

then(9) Reward 119877

119894(119889119894119895) = 119877119894(119889119894119895) + 1

(10) Update 119904 (119904 larr 119904 minus 1) and V119894119898(119889119894119895) (11)

(11) Update 120572119894119895(119898) = 120572119894

119895(119898) + 1

(12) else(13) Reward 119877

119894(119889119894119895) = 119877119894(119889119894119895) + 0

(14) Update 119904 (119904 larr 119904 + 1) and V119894119898(119889119894119895) (11)

(15) Update 120573119894119895(119898) = 120573119894

119895(119898) + 1

(16) end if(17) if 119877

119894(119889119894

119895) = 119877119894(119889119894

119886) 119889119894119895 119889119894

119886isin D119894 119895 = 119886 forall119895 119886 isin 119869

then(18) Select 119889119894

119896 119889119894119896isin D119894 119896 isin U(1 |D

119894|) cap 119869

(19) else(20) Select 119889119894

119896 (12)

(21) end if(22) Exploitation(23) for 119897 = 1 2 3 119871 do(24) Execute the actionA

119894= 119889119894119896 119901119894119904

(25) end for(26) end for(27) end if

Algorithm 1 Multiarm bandit (Thomson sampling)

strategies to maximize accumulate rewards over a given timehorizon However the strategies need to achieve a trade-offbetween exploration (selection of suboptimal actions to learntheir average rewards) and exploitation (selection of actionswhich have provided maximum rewards so far)

In order to dynamically optimize LTE-U transmissionparameters (ie duty cycle and transmit power) a variantof MAB learning techniques called Thomson sampling [4142] algorithm is applied The scenario is formulated asa multiagent problem G = B

119871 A119894119894isinB119871

119862119894119894isinB119871

considering the BSs as players whereA

119894is the action set for

player 119894 During the entire process each BS needs to strikea balance between exploration and exploitation where thereare119872 exploration and 119871 exploitation steps indexed with 1198981 le 119898 le 119872 and 119897 1 le 119897 le 119871 respectively

(i) Agents LTE-U BSsB119871

(ii) Action The action set of agent 119894 A119894is defined as

A119894= 119889119894119895 119901119894119904119895isin119869119904isin119878

119889119894119895 119901119894119904 is the pair of duty cycle

and transmit power elements Configurations of dutycycles are used as part of the action spaceD whereDis common for all players A given BS 119894 selects 119889119894

119895 119889119894119895isin

D according to Algorithm 1 where 119869 = 1 2 |D|119895 isin 119869 and 119869 isin Z+ Probability spaces of positive

integers are denoted byZ+The set of first elements ofthe action vectorD

119894= 1198891198941 1198891198942 119889119894

|D| of BS 119894 is asso-ciated with the duty cycles as 20 40 80respectively The transmit power values set P isrepresented as 119878 = 1 2 |P| 119904 isin 119878 and119878 isin Z+ 119901119894

119904is the transmit power of player 119894 where

P119894= 1199011198941 1199011198942 119901119894

|P| For each action A119894 there is

a distribution Beta (120572119894119895 120573119894119895) forall119895 isin 119869 where 120572119894

119895and

120573119894119895are the shape parameter However in the case of

power control (PC) if119862119894gt 119862

min119895

119904 is decreased by one(119904 larr 119904minus1) reducing the transmit power119901119894

119904by one level

for the next step 119898 + 1 and vice versa Further when119862119894gt 119862min119895

a reward is achieved And for 119862119894gt 119862min119895

120572119894119895is incremented otherwise 120573119894

119895is incremented

(iii) Decision Function The DL capacity of a BS 119894 119862119894is

used as the utility function In order to select a dutycycle a decision function based on the policy UCB1[43] is used where the accumulated rewards achieveddue to values given by 119862

119894are exploited The decision

value for the duty cycle 119889119894119895related to the exploration


step119898 of BS 119894 V119894119898(119889119894119895) is given in (11) while 119889119894

119896based

on the decision is given in (12)

V119894119898(119889119894

119895) = 119909119894119898(119889119894

119895) + radic

2 ln (119898 +1003816100381610038161003816D119894

1003816100381610038161003816)

119899119894119898(119889119894119895)

(11)

119889119894

119896= argmax119889119894

119895isinD119894

(V119894119898(119889119894

119895)) (12)

where 119909119894119898(119889119894119895) = 119877

119894(119889119894119895)119899119894119898(119889119894119895) The argument of

the maximum value is given by arg max(sdot) 119909119894119898(119889119894119895)

119877119894(119889119894119895) and 119899

119894119898(119889119894119895) are the average reward obtained

from 119889119894119895during the exploration step 119898 total rewards

gained form the same 119889119894119895 and the total number of

times 119889119894119895has been played respectively Selection of 119904

is totally independent of the decision function

The multiagent learning problem is addressed using aMAB approach In the contextual MAB problem handled bythe Thomson sampling algorithm [41] current and previousinformation (ie history) is used for the selection of anarm or action Initially 119889119894

119895 forall119895 isin 119869 are played once with

119901119894119904= 119901119894|P| Based on the accumulated reward 119877

119894(119889119894119895) the

parameters 119904 1198991198940(119889119894119895) and V

1198940(119889119894119895) are updated In the learning

process the accumulated reward is used to play the role of theaccumulated hypothesis defined in [44] Subsequently agentsbalance between 119872 exploration and 119871 exploitations stepsDuring the exploration steps 119889119894

119895is selected randomly where

119889119894119895 119889119894119895isin D119894 119895 isin U(1 |D

119894|)cap119869 where a uniformdistribution

with the minimum and maximum values 1199091and 119909

2is given

by U(1199091 1199092) 119904 is decided based on the last available values

of (8) Subsequently the same set of parameters is updatedAt the end of each exploration step based on (8) and theaccumulated rewards an action is selected Then the sameaction is repeatedly played for all the 119871 exploitation steps ofthat particular exploration step as explained in Algorithm 1

5 Simulation Results

For LTE-U TDD-LTE is considered and it is assumedthat all LTE-U UEs are synchronized in both time andfrequency domain as in [13] with the serving BSs A beacon istransmitted periodically for the purpose of synchronizationof WiFi STAs with the corresponding APs To evaluate theperformance an architecture containing two independentlyoperated layers of cellular deployments is considered asshown in Figure 5 Hexagonal cells with omnidirectionalantennas are assumed LTE-U layer encompasses |B

119871| = 7

BSs and |Q119871| UEs where the WiFi layer includes |B

119882| =

7 APs and |Q119882| WiFi STAs In each cell for each APBS

STAsUEs are dropped at random locations All of them areassumed to be uniformly distributed within the cells of theirserving BSs having a mobility speed of 3 kmh and a randomwalk mobility model We consider a nonfull buffer traffic forbothWiFi and LTE networks where the packet arrivals at thetransmitter queues follow a Poisson distribution The traffic

minus50 0 50

Dist

ance

(m)

100

50

0

minus50

Distance (m)

BSAP

WiFi

Area boundariesLTE-U

LTE-U WiFiCells

Figure 5 Cellular coverage layout used in LTE-U and WiFi coex-istence simulations

arrival rates for LTE-U and WiFi are 120582LTE = 120582WiFi = 25

packetsecondThe LTE and WiFi IEEE 80211n medium access control

(MAC) and physical (PHY) layers are modeled in which aPHY layer abstraction is used for Shannon capacity calcula-tions of WiFi and LTE-U The time granularity of each WiFiOFDM symbol duration is 4 120583s which we use to periodicallycapture the number of successfully received bits [13] For bothtechnologies wireless channel is modeled according to [45]when the systems are operated in the 35 GHz band IndoorHotspot (InH) scenario is considered with path loss andshadowing parameters FTP TrafficModel-2 [45] is employedfor either WiFi or LTE-U with a noise spectral power densityof minus95 dBmHz

In each transmission time interval (TTI) DL SINR isreported to the corresponding BS Based on the number ofLTE-U UEs waiting and requesting UL transmission duringone subframe bandwidth is equally shared among them-selves The simulation parameters for LTE-U transmissionsare summarized in Table 1 TDD configuration 1 [39 Figure62(a)] is used for the LTE-U frames having a 50ms totalduty cycle period Minimum required capacity level 119862min

119895is

10Mbps and the set of power levels isP119894= 1199011198941 1199011198942 119901119894

|P| =

8 13 18 23 dBmFor WiFi CSMACA with enhanced distributed channel

access (EDCA) and clear channel assessment (CCA) has been


Table 1 LTE MACPHY parameters

Parameter ValueFrequency 35 GHzTransmission scheme OFDMBandwidth 20MHzDL Tx power 23 dBmUL Tx power PL Based TPCFrame duration 10msScheduling Round RobinUL base power level 119875

0minus106 dBm

TTI 1ms

Table 2 WiFi MACPHY parameters

Parameter ValueFrequency 35 GHzTransmission scheme OFDMBandwidth 20MHzDLUL Tx power 23 dBmAccess category Best effortMAC protocol EDCACCA channel sensing threshold minus82 dBmCCA energy detection threshold minus62 dBmNo of service bits in PPDU 16 bitsNo of tail bits in PPDU 12 bits

Backoff type Fixed contentionwindow

Contention window size U(0 31)

Noise figure 6 [39]Beacon interval 100msBeacon OFDM symbol detection threshold 10 dBBeacon error ratio threshold 15

implemented All WiFi STAs with traffic in their queue willcompete for channel access after receiving a beacon transmis-sion Without reception of a signal beacon transmission orreception will not be initiated The WiFi STA will sense thechannel andwill transmit if it is idle Otherwise transmissionwill be backed off and the next transmission will be initiatedafter a backoff time Random backoff timemechanism is usedfor this study All the parameters for the WiFi transmissionare summarized in Table 2

51 Aggregate Capacity with MAB Aggregate capacity ofstand-alone WiFi coexisting LTE-U (80 duty cycle) andWiFi (with no MAB algorithm) and MAB based coexistenceof LTE-U and WiFi are presented in Figure 7 The aggregatenumbers of WiFi APs and LTE BSs in all scenarios are keptconstant For the WiFi only deployment we replace all theLTE BSs in Figure 5 with WiFi APs It is notable that withthe use of MAB the overall capacity is increased significantlyfrom stand-alone WiFi operation and simultaneous opera-tion of LTE-U and WiFi (without MAB) Also we found thatwith the increase of intersite distance (ISD) in Figure 5 the

MAB

LTE WiFi

Scenario 1

Scenario 2

20 MHz

10 MHz 10 MHz

Figure 6 Scenario with two cases

capacity deceases This is because of higher serving area perAPsSTA within the ISDs

The WiFi throughput performance with and withoutMAB algorithm is shown in Figure 8 where it is noted thatMAB algorithm improves the WiFi throughput over the twoother scenarios Moreover with the increase of ISD capacitydegrades for all cases The effect of LTE packet arrival rate onaggregate capacity is shown in Figure 9 We found that theaggregate throughput of coexisting LTE and WiFi networksis maximized for 120582

119871= 25 but then it decreases for larger

values of 120582119871due to increased interference levels Also for full

buffer LTE traffic (120582119871= 0) the coexisting system with MAB

has degraded performance compared to coexisting systemwithout MAB

Impact of energy detection threshold on aggregate capac-ity is shown in Figure 10 It is observed that minus62 dBmthreshold provides best performance for all scenarios Sens-ing threshold less than minus62 dBm makes WiFi back off fromtransmission in the presence of LTE transmission and resultsin lower aggregate capacity On the other hand sensingthreshold more than minus62 dBm allows WiFi to transmit in thepresence of LTE operation which reduces aggregate capacitydue to higher interference

For Figure 11 we consider a scenario with two cases asdescribed in Figure 6 In scenario 1 we consider simultaneousoperation of LTE-U and WiFi using MAB on 20MHz band-width On the other hand in scenario 2 stand-alone LTE (ie100 duty cycle) andWiFi are operating on separate 10MHzbandwidth We find that the overall capacity using MAB isimproved significantly when compared with the aggregatecapacity of two stand-alone systems This reflects how thespectral efficiency can be improved usingMAB andmotivatessharing of wireless spectrum among LTE andWiFi networksrather than deploying them separately

The impact of LTE-U UEs and WiFi STAs density onaggregate capacity is given in Figure 12 We find that theaggregate capacity improves for the reductions of users inboth services Comparatively high sensitivity could be seenwhen the density of STAs is changed When the densitiesare reduced particularly the STAs a significant increasein capacity is achieved under reduced interference condi-tions However this reduction is further contributed by theCSMACAmechanism as well Also it is notable that capacitydecreases with the increase of ISD

52 Cell-Edge Performance In Figure 13 5th percentile LTEthroughput for different user densities of STAs is representedWe found that with the increase of STAs 5th percentile UEthroughput reduces due tomore interference caused by STAs


Only WiFiWiFi and LTE-U (80 duty cycle)WiFi and LTE-U (with MAB)

2

4

6

8

10

12

Agg

rega

te ca

paci

ty (b

ps)

times107

50 500250

ISD (m)

Figure 7 Aggregate capacity of coexisting WiFi and LTE-U (80duty cycle) MAB based coexisting LTE-U and WiFi and stand-alone WiFi system for different ISDs


1

2

3

4

5

6

7

8

WiF

i cap

acity

(bps

)

times107

50 500250

ISD (m)

Figure 8 WiFi capacity of coexisting WiFi and LTE-U (80 dutycycle) MAB based coexisting LTE-U and WiFi and stand-aloneWiFi system for different ISDs

However with the increment of UEs the effect of STA densityreducesThismeans that for higher density of UEs and STAsfewer LTE users will experience higher capacity

53 Energy Efficiency Performance Aggregate capacity of|Q119894119871| = 10 and |Q119908

119882| = 10 is presented in Figure 14 for different

power control techniques Four parameter settings are usedfor PC In the first instance noPC is considered In the secondcase PC is used by replacing the parameters in Step (7) of the

LTE traffic arrival rate (packetsecond)

04

06

08

1

12

14

16

18

2

22

Agg

rega

te ca

paci

ty (b

ps)

120582L = 0 120582L = 15 120582L = 25 120582L = 35


times108

Figure 9 Aggregate capacity of coexisting WiFi and LTE-U (80duty cycle) MAB based coexisting LTE-U and WiFi and stand-alone WiFi system for different LTE traffic arrival rates

Only WiFiWiFi and LTE-U (80 duty cycle)WiFi and LTE-U (MAB)

times107

5

6

7

8

9

10

11

12

Agg

rega

te ca

paci

ty (b

ps)

minus62 minus72minus52

Energy detection threshold (dBm)

Figure 10 Aggregate capacity of coexisting sytem ofWiFi and LTE-U (80 duty cycle) MAB based coexisting LTE-U and WiFi andstand-alone WiFi system for various energy detection thresholds

Algorithm 1 with 119862119894119901119894119904gt 119862min119895

119875min where 119875min = 8 dBmFor the third and forth cases parameters are replaced with119862119894119901119894119904gt 119862min119895

119875max and 119862119894 gt 119862min119895

where 119875max = 23 dBmThe set of power levels is defined asP

119894= 1199011198941 1199011198942 119901119894

|P| =

8 11 14 17 20 23 dBm where 119875min = 8 dBm and 119875max =

23 dBm So in the second and third cases a given level ofenergy efficiency is aimed at In the last case according to theexplanation given for (9) the level is dynamically adjusted It


50 500250

ISD (m)

0

2

4

6

8

10

12

Capa

city

(bps

)

LTE-U (10MHz)WiFi (10MHz)LTE-U (10MHz) + WiFi (10MHz) (scenario 2)MAB (20MHz) (scenario 1)

times107

Figure 11 Capacity of 10 STAs orand 10 UEs under stand-aloneWiFi stand-alone LTE coexisting stand-alone WiFi and LTE-U(scenario 1) and MAB based coexisting LTE-U and WiFi (scenario2) for different bandwidths and ISDs

times108

50 500250

ISD (m)

05

1

15

2

25

Agg

rega

te ca

paci

ty (b

ps)

5UEs 5 STAs5UEs 10 STAs

10 UEs 5 STAs10 UEs 10 STAs

Figure 12 Capacity ofMAB based coexistence for different UEs andSTAs ratios and ISDs

is noted that the best and worst performances are found for119875max and 119875min For MAB with PC optimum result is found

In Figure 15 different numbers of UEs are considered toevaluate energy efficiency performance For all the densitiesthe least efficiency is achieved with no PC In the mostdense scenario the best efficiency can be observed under thesecond configuration 119862min

119895119875min [see (9)] As it is expected

with the reduction of densities energy efficiency is increasedHowever after a certain average energy efficiency level nosignificant improvements could be observed

5STA10 STA15 STA

5 1510

Number of UEs

09

1

11

12

13

14

15

5th

perc

entil

e thr

ough

put (

bps)

times107

Figure 13 5th percentile throughput ofMAB based coexisting LTE-U and WiFi for different UEs and STAs ratios

PC configuration

Total (WiFi + LTE-U)WiFiLTE-U

No PC Pmin Pmax PC2

4

6

8

10

12

14

Capa

city

(bps

)

times107

Figure 14 Capacity of 10 UEs and 10 STAs under different PCconfigurations

6 Conclusion

In this paper a MAB based dynamic duty cycle selectionmethod was proposed to facilitate spectrum sharing betweenWiFi and LTE-U in the same unlicensed band Performanceof the proposed algorithm was further enhanced by using aDL PC technique Subsequently the proposed concept wasextended to optimize energy efficiency Considerable gainsin overall throughputs could be achieved via the proposedMAB while ensuring a minimum capacity for LTE-U basedservices in the same band Significant gains in terms of energyefficiency could be achieved where it is observed that the


No PCPC

PminPmax

5 1510

Number of UEs

108

109

Ener

gy effi

cien

cy (b

itsjo

ule)

Figure 15 Energy efficiency under different PC configurations forvarious numbers of UEs (with 10 STAs)

gains under different parameter settings with PC are muchhigher than those with no PC Our future work includesextending our framework to scenarios with IA and PAL usersin the same spectrum

Competing Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

The authors would like to thank Nadisanka Rupasinghe fordeveloping an earlier version of the simulator used in thispaperThis researchwas supported in part by theUSNationalScience Foundation (NSF) under Grants nos ACI-1541108and AST-1443999 and Presidential Fellowship under FloridaInternational University

References

[1] NSN Whitepaper ldquoEnhance mobile networks to deliver 1000times more capacity by 2020rdquo Tech Rep 2013

[2] M SimsekM Bennis and I Guvenc ldquoEnhanced intercell inter-ference coordination inHetNets single vsmultiflow approachrdquoin Proceedings of the IEEE Globecom Workshops (GC Wkshpsrsquo13) pp 725ndash729 Atlanta Ga USA December 2013

[3] M Simsek M Bennis and I Guvenc ldquoLearning basedfrequency- and time-domain inter-cell interference coordina-tion in HetNetsrdquo IEEE Transactions on Vehicular Technologyvol 64 no 10 pp 4589ndash4602 2015

[4] FCC ldquoAmendment of the commissions rules with regard tocommercial operations in the 3550ndash3650MHz bandrdquo Noticeof Proposed Rulemaking and Order 2012 httpsappsfccgovedocs publicattachmatchDA-15-955A1 Rcdpdf

[5] FCC ldquoAmendment of the commissions rules with regard tocommercial operations in the 3550ndash3650MHz bandrdquo Report

and Order and Second Further Notice of Proposed Rulemaking12-354 2015 httpsappsfccgovedocs publicattachmatchFCC-15-47A1pdf

[6] R Zhang M Wang L X Cai Z Zheng X S Shen and L-LXie ldquoLTE-unlicensed the future of spectrum aggregation forcellular networksrdquo IEEE Wireless Communications vol 22 no3 pp 150ndash159 2015

[7] ldquoStudy on licensed-assisted access using LTErdquo Tech Rep RP-141397 3GPP Study Item Edinburgh Scotland 2014

[8] 3GPP ldquoStudy on licensed-assisted access to unlicensed spec-trumrdquo Tech Rep TR 36899 3GPP Athens Greece 2015

[9] L Cai X Shen J WMark and Y Xiao ldquoVoice capacity analysisof WLAN with unbalanced trafficrdquo in Proceedings of the 2ndInternational Conference on Quality of Service in HeterogeneousWiredWireless Networks (QSHINE rsquo05) pp 8ndash9 LakeVista FlaUSA August 2005

[10] F M Abinader E P L Almeida F S Chaves et al ldquoEnablingthe coexistence of LTE and Wi-Fi in unlicensed bandsrdquo IEEECommunications Magazine vol 52 no 11 pp 54ndash61 2014

[11] I Parvez N Islam N Rupasinghe A I Sarwat and I GuvencldquoLAA-based LTE and ZigBee coexistence for unlicensed-bandsmart grid communicationsrdquo inProceedings of the SoutheastCon2016 pp 1ndash6 Norfolk Va USA March-April 2016

[12] N Rupasinghe and I Guvenc ldquoLicensed-assisted access forWiFi-LTE coexistence in the unlicensed spectrumrdquo in Proceed-ings of the IEEE GlobecomWorkshops (GCWkshps rsquo14) pp 894ndash899 Austin Tex USA December 2014

[13] N Rupasinghe and I Gulvenc ldquoReinforcement learning forlicensed-assisted access of LTE in the unlicensed spectrumrdquo inProceedings of the IEEEWireless Communications and Network-ing Conference (WCNC rsquo15) pp 1279ndash1284 New Orleans LaUSA March 2015

[14] M G S Sriyananda I Parvez I Guvenc M Bennis and A ISarwat ldquoMulti-Armed Bandit for LTE-U and WiFi coexistencein unlicensed bandsrdquo in Proceedings of the IEEE Wireless Com-munications and Networking Conference (WCNC rsquo16) DohaQatar April 2016

[15] T Ran S Sun B Rong and M Kadoch ldquoGame theorybased multi-tier spectrum sharing for LTE-A heterogeneousnetworksrdquo in Proceedings of the IEEE International ConferenceonCommunications (ICC rsquo15) pp 3033ndash3038 LondonUK June2015

[16] F Shams G Bacci and M Luise ldquoA Q-learning game-theory-based algorithm to improve the energy efficiency of a multiplerelay-aided networkrdquo inProceedings of the 31st General Assemblyand Scientific Symposium of the International Union of RadioScience (URSI GASS rsquo14) pp 1ndash4 XXXIth URSI August 2014

[17] J C Gittins ldquoBandit processes and dynamic allocation indicesrdquoJournal of the Royal Statistical SocietymdashSeries BMethodologicalvol 41 no 2 pp 148ndash177 1979

[18] P Auer N Cesa-Bianchi and P Fischer ldquoFinite-time analysis ofthe multiarmed bandit problemrdquoMachine Learning vol 47 no2 pp 235ndash256 2002

[19] A M Cavalcante E Almeida R D Vieira et al ldquoPerformanceevaluation of LTE and Wi-Fi coexistence in unlicensed bandsrdquoin Proceedings of the IEEE 77th Vehicular Technology Conference(VTC Spring rsquo13) pp 1ndash6 Dresden Gramany June 2013

[20] R Kwan R Pazhyannur J Seymour et al ldquoFair co-existence ofLicensed Assisted Access LTE (LAA-LTE) and Wi-Fi in unli-censed spectrumrdquo in Proceedings of the 7th Computer Scienceand Electronic Engineering (CEEC rsquo15) pp 13ndash18 ColchesterUK September 2015


[21] N Whitepaper ldquoViews on LAA for unlicensed spectrummdashscenarios and initial evaluation resultsrdquo Tech Rep RWS-140026 3GPP RAN1 Standard Contribution Sophia AntipolisFrance 2014

[22] Qualcomm ldquoQualcomm research LTE in unlicensed spectrumharmonious coexistence with WiFirdquo Tech Rep 3GPP RAN1Standard Contribution 2014

[23] E Almeida A M Cavalcante R C D Paiva et al ldquoEnablingLTEWiFi coexistence by LTE blank subframe allocationrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo13) pp 5083ndash5088 IEEE BudapestHungary June2013

[24] T Nihtila V Tykhomyrov O Alanen et al ldquoSystem perfor-mance of LTE and IEEE 80211 coexisting on a shared frequencybandrdquo in Proceedings of the IEEE Wireless Communications andNetworking Conference (WCNC rsquo13) pp 1038ndash1043 ShanghaiChina April 2013

[25] F S Chaves E P L Almeida R D Vieira et al ldquoLTE ULpower control for the improvement of LTEWi-Fi coexistencerdquoin Proceedings of the IEEE 78th Vehicular Technology Conference(VTC Fall rsquo13) pp 1ndash6 September 2013


[27] X Xu G Kutrolli and R Mathar ldquoDynamic downlink powercontrol strategies for LTE femtocellsrdquo in Proceedings of the 7thNext Generation Mobile Applications Services and TechnologiesConference pp 181ndash186 September 2013

[28] ZWangW Xiong C Dong JWang and S Li ldquoA novel down-link power control scheme in LTE heterogeneous networkrdquo inProceedings of the International Conference on ComputationalProblem-Solving (ICCP rsquo11) pp 241ndash245 Chengdu ChinaOctober 2011

[29] T Zahir K Arshad Y Ko and KMoessner ldquoA downlink powercontrol scheme for interference avoidance in femtocellsrdquo inProceedings of the 7th International Wireless CommunicationsandMobile Computing Conference (IWCMC rsquo11) pp 1222ndash1226July 2011

[30] S Maghsudi and S Stanczak ldquoChannel selection for network-assisted D2D communication via no-regret bandit learningwith calibrated forecastingrdquo IEEE Transactions on WirelessCommunications vol 14 no 3 pp 1309ndash1322 2015

[31] M Simsek M Bennis and I Guvenc ldquoMobility managementin HetNets a learning-based perspectiverdquo EURASIP Journalon Wireless Communications and Networking vol 2015 no 1article 26 pp 1ndash13 2015

[32] R Combes and A Proutiere ldquoDynamic rate and channelselection in cognitive radio systemsrdquo IEEE Journal on SelectedAreas in Communications vol 33 no 5 pp 910ndash921 2015

[33] G Locke and L E Strickling ldquoAn assessment of the near-termviability of accommodating wireless broadband systems in the1675ndash1710MHz 1755ndash1780MHz 3500ndash3650MHz and 4200ndash4220MHz 4380ndash4400MHz bandsrdquo Report 2010 httpswwwntiadocgovfilesntiapublicationsfasttrackevaluation 11152010pdf

[34] P R Atkins ldquoNTIA letter office of engineering and technologyFCCrdquo GN Docket No 12-354 2015 httpwwwntiadocgovfilesntiapublicationsntia letter docket no 12-354pdf

[35] L Stefani ldquoThe FCC Raises the Curtain on the CitizensBroadband Radio Servicerdquo CommLawBlog Article May 2015

httpwwwcommlawblogcom201505articlesunlicensed-operations-and-emerthe-fcc-raises-the-curtain-on-the-citi-zens-broadband-radio-service

[36] FCC ldquo35 GHz Spectrum Access System Workshoprdquo Washing-ton DC USA 2014 httpswwwfccgovnews-eventsevents20140135-ghz-spectrum-access-system-workshop

[37] ldquoRSPG opinion on licensed shared accessrdquo Tech Rep RSPG13-538 European Commission Radio Spectrum Policy Group2013

[38] ECC ldquoLicensed shared accessrdquo Tech Rep ECC 205 2014[39] S Sesia I Toufik and M Baker LTEmdashThe UMTS Long Term

Evolution From Theory to Practice John Wiley amp Sons NewYork NY USA 2009

[40] E Perahia and R Stacey LTE The UMTS Long Term EvolutionFromTheory to Practice CambridgeUniversity PressNewYorkNY USA 2008

[41] S Agrawal and N Goyal ldquoAnalysis of thompson samplingfor the multi-armed bandit problemrdquo httpsarxivorgabs11111797

[42] N Gupta O-C Granmo and A Agrawala ldquoThompson sam-pling for dynamic multi-armed banditsrdquo in Proceedings ofthe 10th International Conference on Machine Learning andApplications (ICMLA rsquo11) vol 1 pp 484ndash489Honolulu HawaiiUSA December 2011

[43] P Auer N Cesa-Bianchi and P Fischer ldquoFinite-time analysis ofthe multiarmed bandit problemrdquoMachine Learning vol 47 no2-3 pp 235ndash256 2002

[44] J Langford and T Zhang ldquoThe epoch-greedy algorithm formultiarmed bandits with side informationrdquo in Advances inNeural Information Processing Systems J C Platt D KollerY Singer and S T Roweis Eds vol 20 pp 817ndash824 CurranAssociates 2008

[45] 3GPP ldquoEvolved Universal Terrestrial Radio Access (E-UTRA)further advancements for E-UTRA physical layer aspects(release 9)rdquo Tech Rep TR36814 V900 3GPP 2010

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Distributed Sensor Networks


Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014


ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014


Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia


Biomedical Imaging


ArtificialNeural Systems

Advances in


RoboticsJournal of



Computational Intelligence and Neuroscience

Industrial EngineeringJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in



for secondary use by PAL

Federal RLS and ARNS use Federal RLS3 sites only

Tier-1

Tier-2

Tier-3

Pote

ntia

l ban

ds fo

r LTE

-U

depl

oym

ent

3700 MHz3650MHz3550MHz

150MHz channel for use by GAAA minimum of 80 MHz and maximum of

Floating 70 MHz spectrum

Figure 1 CBRS spectrum with 3 types of users

LTE-U where LTE base stations leave transmission gaps forfacilitating coexistence with WiFi networks Development ofLTE-U technology is led by the industry consortium knownas the LTE-U Forum LTE-Umainly focuses on the operationof unlicensed LTE in the regions (eg USA China) wherelisten before talk (LBT) is not mandatory LTE-U definesthe operation of primary cell in a licensed band with oneor two secondary cells (SCells) each 20MHz in the 5GHzunlicensed band U-NII-1 andor U-NII-3 bands spanning5150ndash5250MHz and 5725ndash5825MHz respectively Howeverboth the LTE-U and LAA need licensed band for controlplane Similar to the 5GHz band CBRS band can be utilizedfor LTE-U operation in the absence of IA users such as radarsignal

In our study we consider the coexistence problemof LTE-U andWiFi networks in the CBRS bands SinceWiFi adopts acontention based medium access control with random back-off [9] for channel access and LTE uses dynamic schedulingfor users the unrestrained LTE operation in the same bandwill generate continuous interference on WiFi service Tooperate LTE-U and WiFi simultaneously in the same unli-censed spectrum fair and reasonable coexistencemechanismis indispensable The adverse impact on DL and UL WiFitransmissions due to LTE deployment in the same bandis analyzed in [10ndash12] emphasizing the need for rigorousstudies In this regard discrete mechanisms such as dynamicchannel selection retaining transmission gaps transmissionduty cycle manipulation and LBT have been proposed inthe literature for harmonious coexistence with improvedperformance To select resources dynamically learn from theenvironment and adaptively modify transmission parame-ters for performance improvement variousmachine learningbased techniques [13ndash16] have been introduced

In this paper we introduce a reinforcement learning(MAB) based adaptive duty cycle section for the coexistencebetween LTE-U and WiFi Multiarmed Bandit (MAB) is amachine learning technique designed to maximize the long-term rewards through learning provided that each agentis rewarded after pulling an arm Basically MAB [17 18]problem resembles a gambler (agent) with a finite number ofslot machines in which the gambler wants to maximum hisrewards over a time horizon Upon pulling an arm a rewardis attained with prior unknown distribution The goal is topull arms sequentially so that the accumulated rewards overthe gambling period are maximized However the problem

involves the exploration versus exploitation trade-off that istaking actions to yield immediate higher reward on the onehand and taking actions that would give rewards in the futureon the other hand

In our technique we use a multiarm bandit (MAB)algorithm for selecting appropriate duty cycle Using a 3GPPcompliant Time Division Duplex- (TDD-) LTE and Beaconenabled IEEE 802 systems in the 35 GHz band we simu-late and evaluate the coexistence performance for differentpercentage of transmission gaps We found a significantthroughput improvement for both systems ensuring harmo-nious coexistence The objectives subsequently the gains ofthis study are not limited to throughput enhancements Thebenefits that are achieved in different dimensions with the aidof MAB scheme and the other supporting techniques like PCcan be summarized as follows

(1) Proper coexistence is achieved due to the dynamicexploring and exploitation byMAB So our techniqueis adaptive

(2) The aggregate capacity is improved Due to theapplication ofMAB algorithm optimal or suboptimalsolutions are achieved

(3) Using DL PC higher capacity values are achievedunder dense UE and STA configurations

(4) Higher energy efficiency is also achieved with PCwhich always attempts to reduce the transmissionpower while increasing the energy efficiency

(5) With the use of learning algorithm a high degree ofefficiency is achieved

To the best of our knowledge our work is the first study thatintroduces MAB for improving the coexistence of LTE andWiFi in the unlicensed bands

The rest of the paper is organized as follows Section 2pro-vides a literature review of coexistence of LTE-U andWiFi InSection 3 we provide our systemmodel and problem formu-lation for LTE andWiFi coexistence Section 4 introduces theproposedMABbased dynamic duty cycle selection approachSimulation results with various parameter configurations arepresented in Section 5 Finally Section 6 provides concludingremarks

2 Related Works

21 Coexistence among Unlicensed LTE and WiFi In theliterature several studies can be found that investigate theperformance of LTE and WiFi coexistence in the unlicensedbands In [19] coexistence performance of LTE and WiFihas been investigated in 900MHz considering single floorand multifloor indoor office scenarios It is shown that theperformance of WiFi is heavily affected when WiFi and LTEoperate simultaneously in the unlicensed spectrum

To facilitate harmonious coexistence between LTE-U andWiFi in the same band mainly three techniques have beenproposed in the literature (1) listen before talk (LBT) (2)dynamic channel selection and (3) coexistence gaps InEurope and Japan LBT is mandatory for data offloading in








Tier‐1 IA system




Interface




RS w

ith li

cens

ed sh

ared

acce

ss (L

SA)








LTE BS

LTE-U UE

LTE-U UE

LTE BS

WiFi AP

WiFi STA


TE-U UE

WiFi STA


LTE-U UE

LTE-U UE


LTE BS

LTE BS

WiFi AP

WiFi STA

E-U UE

WiFi STA






119871B119882

Q119894119871 and Q119908


119871= Q1119871Q2119871 Q119894

119871 Q

|B119871|

119871

and Q119882= Q1119882Q2119882 Q119908

119882 Q

|B119882|








N119894 = sum|Q119894119871|

119906=1119899119894119906 119901119894119904119903 119901119887119904119903 119901119886119904119903 and 119901119902



119906119903 ℎ119887119906119903 ℎ119886119906119903 and ℎ119902




119906

AP and 119868119906




6040

20


Duty cycle




SINR119894119906119903[119896]

=119901119894119904ℎ119894119906119903

sum119887isinB119871119894

119901119887119904ℎ119887119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

BS

+ sum119886isinB119882

119901119886119904ℎ119886119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

AP

+ sum119902isinQ119882

119901119902119904ℎ119902119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

STA

+ 1205902 (1)

where 119887 119894 isin B119871



119873119894

119861=

K119894

sum119896

sum

119906isinQ119894119871

119877119906

sum119903

119882119894

119906119903log2(1 + SINR119894

119906119903[119896]) 119879OFDM (2)





119862119894=

119873119894119861

119879119894Tx + 119879119894

Wait (3)







power values 119901119894119904and 119901119887



119887



119862min119895



119894=1119862119894

1003816100381610038161003816B1198711003816100381610038161003816

(4)

minimize 119901119894


119871(5)

subject to 119901119894

119904 119901119887

119904 le 119875max


(6)

119901119894

119904 119901119887

119904 ge 119875min


(7)

119862119894gt 119862

min119895



119862119894

119901119894119904

gt119862min119895

119901119894119904

or119862119894

119901119894119904

gt119862min119895

119875min

or119862119894

119901119894119904

gt119862min119895

119875max

(9)


119901119894119904is simplified



119894=1(119862119894119901119894119904)

1003816100381610038161003816B1198711003816100381610038161003816


(10)






119895and 120573119894




119894(119889119894119895) based on 119862

119894gt 119862min119895

(3) if 120572119894119895(119898) = 120573119894

119895(119898) forall(119897 119898) isin 119872 then


119895 119889119894119895isin D119894 119895 isin U(1 |D

119894|) cap 119869 and update 119904 (8)

(7) Execute 119889119894119895 119901119894119904 observe 119862

119894and update 119899

119894119898(119889119894119895)

(8) if 119862119894gt 119862min119895


119894(119889119894119895) = 119877119894(119889119894119895) + 1


(11) Update 120572119894119895(119898) = 120572119894

119895(119898) + 1

(12) else(13) Reward 119877

119894(119889119894119895) = 119877119894(119889119894119895) + 0

(14) Update 119904 (119904 larr 119904 + 1) and V119894119898(119889119894119895) (11)

(15) Update 120573119894119895(119898) = 120573119894

119895(119898) + 1

(16) end if(17) if 119877

119894(119889119894

119895) = 119877119894(119889119894

119886) 119889119894119895 119889119894


then(18) Select 119889119894

119896 119889119894119896isin D119894 119896 isin U(1 |D

119894|) cap 119869

(19) else(20) Select 119889119894

119896 (12)


119894= 119889119894119896 119901119894119904





119871 A119894119894isinB119871

119862119894119894isinB119871






A119894= 119889119894119895 119901119894119904119895isin119869119904isin119878



119895 119889119894119895isin



119894= 1198891198941 1198891198942 119889119894



P119894= 1199011198941 1199011198942 119901119894



119895and



min119895


119904by one level











119896based


V119894119898(119889119894

119895) = 119909119894119898(119889119894

119895) + radic

2 ln (119898 +1003816100381610038161003816D119894

1003816100381610038161003816)

119899119894119898(119889119894119895)

(11)

119889119894

119896= argmax119889119894

119895isinD119894

(V119894119898(119889119894

119895)) (12)

where 119909119894119898(119889119894119895) = 119877

119894(119889119894119895)119899119894119898(119889119894119895) The argument of


119877119894(119889119894119895) and 119899









119894(119889119894119895) the

parameters 119904 1198991198940(119889119894119895) and V




119889119894119895 119889119894119895isin D119894 119895 isin U(1 |D



2is given





119871| = 7


119882| =



minus50 0 50

Dist

ance

(m)

100

50

0

minus50

Distance (m)

BSAP

WiFi


LTE-U WiFiCells






119895is


|P| =






0minus106 dBm

TTI 1ms








MAB

LTE WiFi

Scenario 1

Scenario 2

20 MHz

10 MHz 10 MHz














2

4

6

8

10

12

Agg

rega

te ca

paci

ty (b

ps)

times107

50 500250

ISD (m)



1

2

3

4

5

6

7

8

WiF

i cap

acity

(bps

)

times107

50 500250

ISD (m)







04

06

08

1

12

14

16

18

2

22

Agg

rega

te ca

paci

ty (b

ps)

120582L = 0 120582L = 15 120582L = 25 120582L = 35


times108



times107

5

6

7

8

9

10

11

12

Agg

rega

te ca

paci

ty (b

ps)






119875max and 119862119894 gt 119862min119895


119894= 1199011198941 1199011198942 119901119894

|P| =




50 500250

ISD (m)

0

2

4

6

8

10

12

Capa

city

(bps

)


times107


times108

50 500250

ISD (m)

05

1

15

2

25

Agg

rega

te ca

paci

ty (b

ps)








5STA10 STA15 STA

5 1510

Number of UEs

09

1

11

12

13

14

15

5th

perc

entil

e thr

ough

put (

bps)

times107


PC configuration


No PC Pmin Pmax PC2

4

6

8

10

12

14

Capa

city

(bps

)

times107


6 Conclusion



No PCPC

PminPmax

5 1510

Number of UEs

108

109

Ener

gy effi

cien

cy (b

itsjo

ule)



Competing Interests


Acknowledgments


References
























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in










Tier‐1 IA system




Interface




RS w

ith li

cens

ed sh

ared

acce

ss (L

SA)








LTE BS

LTE-U UE

LTE-U UE

LTE BS

WiFi AP

WiFi STA


TE-U UE

WiFi STA


LTE-U UE

LTE-U UE


LTE BS

LTE BS

WiFi AP

WiFi STA

E-U UE

WiFi STA






119871B119882

Q119894119871 and Q119908


119871= Q1119871Q2119871 Q119894

119871 Q

|B119871|

119871

and Q119882= Q1119882Q2119882 Q119908

119882 Q

|B119882|








N119894 = sum|Q119894119871|

119906=1119899119894119906 119901119894119904119903 119901119887119904119903 119901119886119904119903 and 119901119902



119906119903 ℎ119887119906119903 ℎ119886119906119903 and ℎ119902




119906

AP and 119868119906




6040

20


Duty cycle




SINR119894119906119903[119896]

=119901119894119904ℎ119894119906119903

sum119887isinB119871119894

119901119887119904ℎ119887119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

BS

+ sum119886isinB119882

119901119886119904ℎ119886119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

AP

+ sum119902isinQ119882

119901119902119904ℎ119902119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

STA

+ 1205902 (1)

where 119887 119894 isin B119871



119873119894

119861=

K119894

sum119896

sum

119906isinQ119894119871

119877119906

sum119903

119882119894

119906119903log2(1 + SINR119894

119906119903[119896]) 119879OFDM (2)





119862119894=

119873119894119861

119879119894Tx + 119879119894

Wait (3)







power values 119901119894119904and 119901119887



119887



119862min119895



119894=1119862119894

1003816100381610038161003816B1198711003816100381610038161003816

(4)

minimize 119901119894


119871(5)

subject to 119901119894

119904 119901119887

119904 le 119875max


(6)

119901119894

119904 119901119887

119904 ge 119875min


(7)

119862119894gt 119862

min119895



119862119894

119901119894119904

gt119862min119895

119901119894119904

or119862119894

119901119894119904

gt119862min119895

119875min

or119862119894

119901119894119904

gt119862min119895

119875max

(9)


119901119894119904is simplified



119894=1(119862119894119901119894119904)

1003816100381610038161003816B1198711003816100381610038161003816


(10)






119895and 120573119894




119894(119889119894119895) based on 119862

119894gt 119862min119895

(3) if 120572119894119895(119898) = 120573119894

119895(119898) forall(119897 119898) isin 119872 then


119895 119889119894119895isin D119894 119895 isin U(1 |D

119894|) cap 119869 and update 119904 (8)

(7) Execute 119889119894119895 119901119894119904 observe 119862

119894and update 119899

119894119898(119889119894119895)

(8) if 119862119894gt 119862min119895


119894(119889119894119895) = 119877119894(119889119894119895) + 1


(11) Update 120572119894119895(119898) = 120572119894

119895(119898) + 1

(12) else(13) Reward 119877

119894(119889119894119895) = 119877119894(119889119894119895) + 0

(14) Update 119904 (119904 larr 119904 + 1) and V119894119898(119889119894119895) (11)

(15) Update 120573119894119895(119898) = 120573119894

119895(119898) + 1

(16) end if(17) if 119877

119894(119889119894

119895) = 119877119894(119889119894

119886) 119889119894119895 119889119894


then(18) Select 119889119894

119896 119889119894119896isin D119894 119896 isin U(1 |D

119894|) cap 119869

(19) else(20) Select 119889119894

119896 (12)


119894= 119889119894119896 119901119894119904





119871 A119894119894isinB119871

119862119894119894isinB119871






A119894= 119889119894119895 119901119894119904119895isin119869119904isin119878



119895 119889119894119895isin



119894= 1198891198941 1198891198942 119889119894



P119894= 1199011198941 1199011198942 119901119894



119895and



min119895


119904by one level











119896based


V119894119898(119889119894

119895) = 119909119894119898(119889119894

119895) + radic

2 ln (119898 +1003816100381610038161003816D119894

1003816100381610038161003816)

119899119894119898(119889119894119895)

(11)

119889119894

119896= argmax119889119894

119895isinD119894

(V119894119898(119889119894

119895)) (12)

where 119909119894119898(119889119894119895) = 119877

119894(119889119894119895)119899119894119898(119889119894119895) The argument of


119877119894(119889119894119895) and 119899









119894(119889119894119895) the

parameters 119904 1198991198940(119889119894119895) and V




119889119894119895 119889119894119895isin D119894 119895 isin U(1 |D



2is given





119871| = 7


119882| =



minus50 0 50

Dist

ance

(m)

100

50

0

minus50

Distance (m)

BSAP

WiFi


LTE-U WiFiCells






119895is


|P| =






0minus106 dBm

TTI 1ms








MAB

LTE WiFi

Scenario 1

Scenario 2

20 MHz

10 MHz 10 MHz














2

4

6

8

10

12

Agg

rega

te ca

paci

ty (b

ps)

times107

50 500250

ISD (m)



1

2

3

4

5

6

7

8

WiF

i cap

acity

(bps

)

times107

50 500250

ISD (m)







04

06

08

1

12

14

16

18

2

22

Agg

rega

te ca

paci

ty (b

ps)

120582L = 0 120582L = 15 120582L = 25 120582L = 35


times108



times107

5

6

7

8

9

10

11

12

Agg

rega

te ca

paci

ty (b

ps)






119875max and 119862119894 gt 119862min119895


119894= 1199011198941 1199011198942 119901119894

|P| =




50 500250

ISD (m)

0

2

4

6

8

10

12

Capa

city

(bps

)


times107


times108

50 500250

ISD (m)

05

1

15

2

25

Agg

rega

te ca

paci

ty (b

ps)








5STA10 STA15 STA

5 1510

Number of UEs

09

1

11

12

13

14

15

5th

perc

entil

e thr

ough

put (

bps)

times107


PC configuration


No PC Pmin Pmax PC2

4

6

8

10

12

14

Capa

city

(bps

)

times107


6 Conclusion



No PCPC

PminPmax

5 1510

Number of UEs

108

109

Ener

gy effi

cien

cy (b

itsjo

ule)



Competing Interests


Acknowledgments


References
























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




LTE BS

LTE-U UE

LTE-U UE

LTE BS

WiFi AP

WiFi STA


TE-U UE

WiFi STA


LTE-U UE

LTE-U UE


LTE BS

LTE BS

WiFi AP

WiFi STA

E-U UE

WiFi STA






119871B119882

Q119894119871 and Q119908


119871= Q1119871Q2119871 Q119894

119871 Q

|B119871|

119871

and Q119882= Q1119882Q2119882 Q119908

119882 Q

|B119882|








N119894 = sum|Q119894119871|

119906=1119899119894119906 119901119894119904119903 119901119887119904119903 119901119886119904119903 and 119901119902



119906119903 ℎ119887119906119903 ℎ119886119906119903 and ℎ119902




119906

AP and 119868119906




6040

20


Duty cycle




SINR119894119906119903[119896]

=119901119894119904ℎ119894119906119903

sum119887isinB119871119894

119901119887119904ℎ119887119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

BS

+ sum119886isinB119882

119901119886119904ℎ119886119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

AP

+ sum119902isinQ119882

119901119902119904ℎ119902119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

STA

+ 1205902 (1)

where 119887 119894 isin B119871



119873119894

119861=

K119894

sum119896

sum

119906isinQ119894119871

119877119906

sum119903

119882119894

119906119903log2(1 + SINR119894

119906119903[119896]) 119879OFDM (2)





119862119894=

119873119894119861

119879119894Tx + 119879119894

Wait (3)







power values 119901119894119904and 119901119887



119887



119862min119895



119894=1119862119894

1003816100381610038161003816B1198711003816100381610038161003816

(4)

minimize 119901119894


119871(5)

subject to 119901119894

119904 119901119887

119904 le 119875max


(6)

119901119894

119904 119901119887

119904 ge 119875min


(7)

119862119894gt 119862

min119895



119862119894

119901119894119904

gt119862min119895

119901119894119904

or119862119894

119901119894119904

gt119862min119895

119875min

or119862119894

119901119894119904

gt119862min119895

119875max

(9)


119901119894119904is simplified



119894=1(119862119894119901119894119904)

1003816100381610038161003816B1198711003816100381610038161003816


(10)






119895and 120573119894




119894(119889119894119895) based on 119862

119894gt 119862min119895

(3) if 120572119894119895(119898) = 120573119894

119895(119898) forall(119897 119898) isin 119872 then


119895 119889119894119895isin D119894 119895 isin U(1 |D

119894|) cap 119869 and update 119904 (8)

(7) Execute 119889119894119895 119901119894119904 observe 119862

119894and update 119899

119894119898(119889119894119895)

(8) if 119862119894gt 119862min119895


119894(119889119894119895) = 119877119894(119889119894119895) + 1


(11) Update 120572119894119895(119898) = 120572119894

119895(119898) + 1

(12) else(13) Reward 119877

119894(119889119894119895) = 119877119894(119889119894119895) + 0

(14) Update 119904 (119904 larr 119904 + 1) and V119894119898(119889119894119895) (11)

(15) Update 120573119894119895(119898) = 120573119894

119895(119898) + 1

(16) end if(17) if 119877

119894(119889119894

119895) = 119877119894(119889119894

119886) 119889119894119895 119889119894


then(18) Select 119889119894

119896 119889119894119896isin D119894 119896 isin U(1 |D

119894|) cap 119869

(19) else(20) Select 119889119894

119896 (12)


119894= 119889119894119896 119901119894119904





119871 A119894119894isinB119871

119862119894119894isinB119871






A119894= 119889119894119895 119901119894119904119895isin119869119904isin119878



119895 119889119894119895isin



119894= 1198891198941 1198891198942 119889119894



P119894= 1199011198941 1199011198942 119901119894



119895and



min119895


119904by one level











119896based


V119894119898(119889119894

119895) = 119909119894119898(119889119894

119895) + radic

2 ln (119898 +1003816100381610038161003816D119894

1003816100381610038161003816)

119899119894119898(119889119894119895)

(11)

119889119894

119896= argmax119889119894

119895isinD119894

(V119894119898(119889119894

119895)) (12)

where 119909119894119898(119889119894119895) = 119877

119894(119889119894119895)119899119894119898(119889119894119895) The argument of


119877119894(119889119894119895) and 119899









119894(119889119894119895) the

parameters 119904 1198991198940(119889119894119895) and V




119889119894119895 119889119894119895isin D119894 119895 isin U(1 |D



2is given





119871| = 7


119882| =



minus50 0 50

Dist

ance

(m)

100

50

0

minus50

Distance (m)

BSAP

WiFi


LTE-U WiFiCells






119895is


|P| =






0minus106 dBm

TTI 1ms








MAB

LTE WiFi

Scenario 1

Scenario 2

20 MHz

10 MHz 10 MHz














2

4

6

8

10

12

Agg

rega

te ca

paci

ty (b

ps)

times107

50 500250

ISD (m)



1

2

3

4

5

6

7

8

WiF

i cap

acity

(bps

)

times107

50 500250

ISD (m)







04

06

08

1

12

14

16

18

2

22

Agg

rega

te ca

paci

ty (b

ps)

120582L = 0 120582L = 15 120582L = 25 120582L = 35


times108



times107

5

6

7

8

9

10

11

12

Agg

rega

te ca

paci

ty (b

ps)






119875max and 119862119894 gt 119862min119895


119894= 1199011198941 1199011198942 119901119894

|P| =




50 500250

ISD (m)

0

2

4

6

8

10

12

Capa

city

(bps

)


times107


times108

50 500250

ISD (m)

05

1

15

2

25

Agg

rega

te ca

paci

ty (b

ps)








5STA10 STA15 STA

5 1510

Number of UEs

09

1

11

12

13

14

15

5th

perc

entil

e thr

ough

put (

bps)

times107


PC configuration


No PC Pmin Pmax PC2

4

6

8

10

12

14

Capa

city

(bps

)

times107


6 Conclusion



No PCPC

PminPmax

5 1510

Number of UEs

108

109

Ener

gy effi

cien

cy (b

itsjo

ule)



Competing Interests


Acknowledgments


References
























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





6040

20


Duty cycle




SINR119894119906119903[119896]

=119901119894119904ℎ119894119906119903

sum119887isinB119871119894

119901119887119904ℎ119887119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

BS

+ sum119886isinB119882

119901119886119904ℎ119886119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

AP

+ sum119902isinQ119882

119901119902119904ℎ119902119906119903⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟

119868119906

STA

+ 1205902 (1)

where 119887 119894 isin B119871



119873119894

119861=

K119894

sum119896

sum

119906isinQ119894119871

119877119906

sum119903

119882119894

119906119903log2(1 + SINR119894

119906119903[119896]) 119879OFDM (2)





119862119894=

119873119894119861

119879119894Tx + 119879119894

Wait (3)







power values 119901119894119904and 119901119887



119887



119862min119895



119894=1119862119894

1003816100381610038161003816B1198711003816100381610038161003816

(4)

minimize 119901119894


119871(5)

subject to 119901119894

119904 119901119887

119904 le 119875max


(6)

119901119894

119904 119901119887

119904 ge 119875min


(7)

119862119894gt 119862

min119895



119862119894

119901119894119904

gt119862min119895

119901119894119904

or119862119894

119901119894119904

gt119862min119895

119875min

or119862119894

119901119894119904

gt119862min119895

119875max

(9)


119901119894119904is simplified



119894=1(119862119894119901119894119904)

1003816100381610038161003816B1198711003816100381610038161003816


(10)






119895and 120573119894




119894(119889119894119895) based on 119862

119894gt 119862min119895

(3) if 120572119894119895(119898) = 120573119894

119895(119898) forall(119897 119898) isin 119872 then


119895 119889119894119895isin D119894 119895 isin U(1 |D

119894|) cap 119869 and update 119904 (8)

(7) Execute 119889119894119895 119901119894119904 observe 119862

119894and update 119899

119894119898(119889119894119895)

(8) if 119862119894gt 119862min119895


119894(119889119894119895) = 119877119894(119889119894119895) + 1


(11) Update 120572119894119895(119898) = 120572119894

119895(119898) + 1

(12) else(13) Reward 119877

119894(119889119894119895) = 119877119894(119889119894119895) + 0

(14) Update 119904 (119904 larr 119904 + 1) and V119894119898(119889119894119895) (11)

(15) Update 120573119894119895(119898) = 120573119894

119895(119898) + 1

(16) end if(17) if 119877

119894(119889119894

119895) = 119877119894(119889119894

119886) 119889119894119895 119889119894


then(18) Select 119889119894

119896 119889119894119896isin D119894 119896 isin U(1 |D

119894|) cap 119869

(19) else(20) Select 119889119894

119896 (12)


119894= 119889119894119896 119901119894119904





119871 A119894119894isinB119871

119862119894119894isinB119871






A119894= 119889119894119895 119901119894119904119895isin119869119904isin119878



119895 119889119894119895isin



119894= 1198891198941 1198891198942 119889119894



P119894= 1199011198941 1199011198942 119901119894



119895and



min119895


119904by one level











119896based


V119894119898(119889119894

119895) = 119909119894119898(119889119894

119895) + radic

2 ln (119898 +1003816100381610038161003816D119894

1003816100381610038161003816)

119899119894119898(119889119894119895)

(11)

119889119894

119896= argmax119889119894

119895isinD119894

(V119894119898(119889119894

119895)) (12)

where 119909119894119898(119889119894119895) = 119877

119894(119889119894119895)119899119894119898(119889119894119895) The argument of


119877119894(119889119894119895) and 119899









119894(119889119894119895) the

parameters 119904 1198991198940(119889119894119895) and V




119889119894119895 119889119894119895isin D119894 119895 isin U(1 |D



2is given





119871| = 7


119882| =



minus50 0 50

Dist

ance

(m)

100

50

0

minus50

Distance (m)

BSAP

WiFi


LTE-U WiFiCells






119895is


|P| =






0minus106 dBm

TTI 1ms








MAB

LTE WiFi

Scenario 1

Scenario 2

20 MHz

10 MHz 10 MHz














2

4

6

8

10

12

Agg

rega

te ca

paci

ty (b

ps)

times107

50 500250

ISD (m)



1

2

3

4

5

6

7

8

WiF

i cap

acity

(bps

)

times107

50 500250

ISD (m)







04

06

08

1

12

14

16

18

2

22

Agg

rega

te ca

paci

ty (b

ps)

120582L = 0 120582L = 15 120582L = 25 120582L = 35


times108



times107

5

6

7

8

9

10

11

12

Agg

rega

te ca

paci

ty (b

ps)






119875max and 119862119894 gt 119862min119895


119894= 1199011198941 1199011198942 119901119894

|P| =




50 500250

ISD (m)

0

2

4

6

8

10

12

Capa

city

(bps

)


times107


times108

50 500250

ISD (m)

05

1

15

2

25

Agg

rega

te ca

paci

ty (b

ps)








5STA10 STA15 STA

5 1510

Number of UEs

09

1

11

12

13

14

15

5th

perc

entil

e thr

ough

put (

bps)

times107


PC configuration


No PC Pmin Pmax PC2

4

6

8

10

12

14

Capa

city

(bps

)

times107


6 Conclusion



No PCPC

PminPmax

5 1510

Number of UEs

108

109

Ener

gy effi

cien

cy (b

itsjo

ule)



Competing Interests


Acknowledgments


References
























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






119895and 120573119894




119894(119889119894119895) based on 119862

119894gt 119862min119895

(3) if 120572119894119895(119898) = 120573119894

119895(119898) forall(119897 119898) isin 119872 then


119895 119889119894119895isin D119894 119895 isin U(1 |D

119894|) cap 119869 and update 119904 (8)

(7) Execute 119889119894119895 119901119894119904 observe 119862

119894and update 119899

119894119898(119889119894119895)

(8) if 119862119894gt 119862min119895


119894(119889119894119895) = 119877119894(119889119894119895) + 1


(11) Update 120572119894119895(119898) = 120572119894

119895(119898) + 1

(12) else(13) Reward 119877

119894(119889119894119895) = 119877119894(119889119894119895) + 0

(14) Update 119904 (119904 larr 119904 + 1) and V119894119898(119889119894119895) (11)

(15) Update 120573119894119895(119898) = 120573119894

119895(119898) + 1

(16) end if(17) if 119877

119894(119889119894

119895) = 119877119894(119889119894

119886) 119889119894119895 119889119894


then(18) Select 119889119894

119896 119889119894119896isin D119894 119896 isin U(1 |D

119894|) cap 119869

(19) else(20) Select 119889119894

119896 (12)


119894= 119889119894119896 119901119894119904





119871 A119894119894isinB119871

119862119894119894isinB119871






A119894= 119889119894119895 119901119894119904119895isin119869119904isin119878



119895 119889119894119895isin



119894= 1198891198941 1198891198942 119889119894



P119894= 1199011198941 1199011198942 119901119894



119895and



min119895


119904by one level











119896based


V119894119898(119889119894

119895) = 119909119894119898(119889119894

119895) + radic

2 ln (119898 +1003816100381610038161003816D119894

1003816100381610038161003816)

119899119894119898(119889119894119895)

(11)

119889119894

119896= argmax119889119894

119895isinD119894

(V119894119898(119889119894

119895)) (12)

where 119909119894119898(119889119894119895) = 119877

119894(119889119894119895)119899119894119898(119889119894119895) The argument of


119877119894(119889119894119895) and 119899









119894(119889119894119895) the

parameters 119904 1198991198940(119889119894119895) and V




119889119894119895 119889119894119895isin D119894 119895 isin U(1 |D



2is given





119871| = 7


119882| =



minus50 0 50

Dist

ance

(m)

100

50

0

minus50

Distance (m)

BSAP

WiFi


LTE-U WiFiCells






119895is


|P| =






0minus106 dBm

TTI 1ms








MAB

LTE WiFi

Scenario 1

Scenario 2

20 MHz

10 MHz 10 MHz














2

4

6

8

10

12

Agg

rega

te ca

paci

ty (b

ps)

times107

50 500250

ISD (m)



1

2

3

4

5

6

7

8

WiF

i cap

acity

(bps

)

times107

50 500250

ISD (m)







04

06

08

1

12

14

16

18

2

22

Agg

rega

te ca

paci

ty (b

ps)

120582L = 0 120582L = 15 120582L = 25 120582L = 35


times108



times107

5

6

7

8

9

10

11

12

Agg

rega

te ca

paci

ty (b

ps)






119875max and 119862119894 gt 119862min119895


119894= 1199011198941 1199011198942 119901119894

|P| =




50 500250

ISD (m)

0

2

4

6

8

10

12

Capa

city

(bps

)


times107


times108

50 500250

ISD (m)

05

1

15

2

25

Agg

rega

te ca

paci

ty (b

ps)








5STA10 STA15 STA

5 1510

Number of UEs

09

1

11

12

13

14

15

5th

perc

entil

e thr

ough

put (

bps)

times107


PC configuration


No PC Pmin Pmax PC2

4

6

8

10

12

14

Capa

city

(bps

)

times107


6 Conclusion



No PCPC

PminPmax

5 1510

Number of UEs

108

109

Ener

gy effi

cien

cy (b

itsjo

ule)



Competing Interests


Acknowledgments


References
























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





119896based


V119894119898(119889119894

119895) = 119909119894119898(119889119894

119895) + radic

2 ln (119898 +1003816100381610038161003816D119894

1003816100381610038161003816)

119899119894119898(119889119894119895)

(11)

119889119894

119896= argmax119889119894

119895isinD119894

(V119894119898(119889119894

119895)) (12)

where 119909119894119898(119889119894119895) = 119877

119894(119889119894119895)119899119894119898(119889119894119895) The argument of


119877119894(119889119894119895) and 119899









119894(119889119894119895) the

parameters 119904 1198991198940(119889119894119895) and V




119889119894119895 119889119894119895isin D119894 119895 isin U(1 |D



2is given





119871| = 7


119882| =



minus50 0 50

Dist

ance

(m)

100

50

0

minus50

Distance (m)

BSAP

WiFi


LTE-U WiFiCells






119895is


|P| =






0minus106 dBm

TTI 1ms








MAB

LTE WiFi

Scenario 1

Scenario 2

20 MHz

10 MHz 10 MHz














2

4

6

8

10

12

Agg

rega

te ca

paci

ty (b

ps)

times107

50 500250

ISD (m)



1

2

3

4

5

6

7

8

WiF

i cap

acity

(bps

)

times107

50 500250

ISD (m)







04

06

08

1

12

14

16

18

2

22

Agg

rega

te ca

paci

ty (b

ps)

120582L = 0 120582L = 15 120582L = 25 120582L = 35


times108



times107

5

6

7

8

9

10

11

12

Agg

rega

te ca

paci

ty (b

ps)






119875max and 119862119894 gt 119862min119895


119894= 1199011198941 1199011198942 119901119894

|P| =




50 500250

ISD (m)

0

2

4

6

8

10

12

Capa

city

(bps

)


times107


times108

50 500250

ISD (m)

05

1

15

2

25

Agg

rega

te ca

paci

ty (b

ps)








5STA10 STA15 STA

5 1510

Number of UEs

09

1

11

12

13

14

15

5th

perc

entil

e thr

ough

put (

bps)

times107


PC configuration


No PC Pmin Pmax PC2

4

6

8

10

12

14

Capa

city

(bps

)

times107


6 Conclusion



No PCPC

PminPmax

5 1510

Number of UEs

108

109

Ener

gy effi

cien

cy (b

itsjo

ule)



Competing Interests


Acknowledgments


References
























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






0minus106 dBm

TTI 1ms








MAB

LTE WiFi

Scenario 1

Scenario 2

20 MHz

10 MHz 10 MHz














2

4

6

8

10

12

Agg

rega

te ca

paci

ty (b

ps)

times107

50 500250

ISD (m)



1

2

3

4

5

6

7

8

WiF

i cap

acity

(bps

)

times107

50 500250

ISD (m)







04

06

08

1

12

14

16

18

2

22

Agg

rega

te ca

paci

ty (b

ps)

120582L = 0 120582L = 15 120582L = 25 120582L = 35


times108



times107

5

6

7

8

9

10

11

12

Agg

rega

te ca

paci

ty (b

ps)






119875max and 119862119894 gt 119862min119895


119894= 1199011198941 1199011198942 119901119894

|P| =




50 500250

ISD (m)

0

2

4

6

8

10

12

Capa

city

(bps

)


times107


times108

50 500250

ISD (m)

05

1

15

2

25

Agg

rega

te ca

paci

ty (b

ps)








5STA10 STA15 STA

5 1510

Number of UEs

09

1

11

12

13

14

15

5th

perc

entil

e thr

ough

put (

bps)

times107


PC configuration


No PC Pmin Pmax PC2

4

6

8

10

12

14

Capa

city

(bps

)

times107


6 Conclusion



No PCPC

PminPmax

5 1510

Number of UEs

108

109

Ener

gy effi

cien

cy (b

itsjo

ule)



Competing Interests


Acknowledgments


References
























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





2

4

6

8

10

12

Agg

rega

te ca

paci

ty (b

ps)

times107

50 500250

ISD (m)



1

2

3

4

5

6

7

8

WiF

i cap

acity

(bps

)

times107

50 500250

ISD (m)







04

06

08

1

12

14

16

18

2

22

Agg

rega

te ca

paci

ty (b

ps)

120582L = 0 120582L = 15 120582L = 25 120582L = 35


times108



times107

5

6

7

8

9

10

11

12

Agg

rega

te ca

paci

ty (b

ps)






119875max and 119862119894 gt 119862min119895


119894= 1199011198941 1199011198942 119901119894

|P| =




50 500250

ISD (m)

0

2

4

6

8

10

12

Capa

city

(bps

)


times107


times108

50 500250

ISD (m)

05

1

15

2

25

Agg

rega

te ca

paci

ty (b

ps)








5STA10 STA15 STA

5 1510

Number of UEs

09

1

11

12

13

14

15

5th

perc

entil

e thr

ough

put (

bps)

times107


PC configuration


No PC Pmin Pmax PC2

4

6

8

10

12

14

Capa

city

(bps

)

times107


6 Conclusion



No PCPC

PminPmax

5 1510

Number of UEs

108

109

Ener

gy effi

cien

cy (b

itsjo

ule)



Competing Interests


Acknowledgments


References
























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




50 500250

ISD (m)

0

2

4

6

8

10

12

Capa

city

(bps

)


times107


times108

50 500250

ISD (m)

05

1

15

2

25

Agg

rega

te ca

paci

ty (b

ps)








5STA10 STA15 STA

5 1510

Number of UEs

09

1

11

12

13

14

15

5th

perc

entil

e thr

ough

put (

bps)

times107


PC configuration


No PC Pmin Pmax PC2

4

6

8

10

12

14

Capa

city

(bps

)

times107


6 Conclusion



No PCPC

PminPmax

5 1510

Number of UEs

108

109

Ener

gy effi

cien

cy (b

itsjo

ule)



Competing Interests


Acknowledgments


References
























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




No PCPC

PminPmax

5 1510

Number of UEs

108

109

Ener

gy effi

cien

cy (b

itsjo

ule)



Competing Interests


Acknowledgments


References
























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



research article cbrs spectrum sharing between … article cbrs spectrum sharing between lte-u and...

Documents