QoE-Guaranteed and Power-Efficient Network Operation for Cloud Radio Access Network with Power over Fiber

Transcription

1 QoE-Guaanteed and Powe-Efficient Netwok Opeation fo Cloud Radio Access Netwok with Powe ove Fibe 2016 IEEE. Pesonal use of this mateial is pemitted. Pemission fom IEEE must be obtained fo all othe uses, in any cuent o futue media, including epinting/epublishing this mateial fo advetising o pomotional puposes, ceating new collective woks, fo esale o edistibution to seves o lists, o euse of any copyighted component of this wok in othe woks. This mateial is pesented to ensue timely dissemination of scholaly and technical wok. Copyight and all ights theein ae etained by authos o by othe copyight holdes. All pesons copying this infomation ae expected to adhee to the tems and constaints invoked by each autho's copyight. In most cases, these woks may not be eposted without the explicit pemission of the copyight holde. Citation: Katsuya Suto, Keisuke Miyanabe, Hioki Nishiyama, Nei Kato, Hiotaka Ujikawa, and Ken-Ichi Suzuki, "QoE-Guaanteed and Powe-Efficient Netwok Opeation fo Cloud Radio Access Netwok with Powe ove Fibe," IEEE Tansactions on Computational Social Systems, vol. 2, issue: 4, pp , Feb URL:

2 1 QoE-Guaanteed and Powe-Efficient Netwok Opeation fo Cloud Radio Access Netwok with Powe ove Fibe Katsuya Suto, Keisuke Miyanabe, Hioki Nishiyama, Nei Kato, Hiotaka Ujikawa, and Ken-Ichi Suzuki Gaduate School of Infomation Sciences (GSIS), Tohoku Univesity, Japan Access Netwok Sevice Systems Laboatoies, NTT Copoation, Yokosuka, Japan s: {katsuya.suto, keisuke.miyanabe, bigtee, {ujikawa.hiotaka, Abstact A concept of Cloud Radio Access Netwoks (C- RANs) is becoming a popula solution to suppot the equied communication quality fo new emeging sevice in the futue netwok envionment, i.e., moe than 10Gbps capacity, less than 1ms latency, and connectivity fo numeous devices. In this aticle, we envision a C-RAN based on Passive Optical Netwok (PON) exploiting Powe ove Fibe (PoF), which achieves low installation and opeation costs since it is capable of poviding communication sevices without extenal powe supply fo lage amount of Remote Radio Heads (RRHs). This netwok, howeve, needs to educe the optical tansmission powe of PoF due to the fibe fuse issue. Additionally, the divesification of sevices, devices and pesonality indicates the need to impove use satisfaction, i.e., Quality of Expeience (QoE), based on the use s pespective, which is diffeent fom pevious appoaches that aim to guaantee Quality of Sevices (QoS). Theefoe, we popose a QoE-guaanteed and powe-efficient netwok opeation stategy. Ou poposed opeation is able to educe the tansmission powe while satisfying the QoE constaint by contolling both the schedule of RRH s sleep and optical tansmission powe of PoF. Futhemoe, the effectiveness of ou poposed opeation scheme is evaluated though extensive compute simulations. Index Tems Cloud adio access netwok, quality of expeience (QoE), QoE-guaanteed and powe-efficient netwok opeation, sleep scheduling. I. INTRODUCTION Futue mobile netwoks ae equied to suppot new emeging sevices with a high netwok capacity, as well as educe communication latency and povide connectivity fo numeous devices, e.g., thee-dimensional video steaming/meeting, Augmented Reality (AR), online big data mining, eal-time Intenet of Things (IoT), and so foth [1] [3]. A netwok achitectue utilizing supe-dense small cell deployments and centalized esouce management, i.e., Cloud Radio Access Netwoks (C-RANs), have attacted much attention as a conestone to achieve the afoementioned equiements [4], [5]. In this achitectue, small cells ae capable of poviding high communication capacity between Remote Radio Head (RRH) and uses, and a cental office (CO) connects with RRHs via high-speed fonthaul links and combines the managements of uses and RRHs in ode to fulfill the equied capacity and latency [6], [7]. Although C-RAN is expected as a pomising futue mobile achitectue, it needs highe installation and opeation costs due to the following easons: (i) since the amount of taffic between the CO and RRHs is much highe in contast with the taditional RAN, high-speed links ae equied. Howeve, setting up pivate optical-fibe cables to each RRH indicates a high installation cost: (ii) povision of extenal powe supply to all RRHs will involve high cost, especially in the case of deployment in places lacking extenal powe supply. To this end, Passive Optical Netwok (PON) exploiting Powe ove Fibe (PoF) is ecognized as the key enabling technology [8] [10]. In PON with PoF, an optical splitte is used to enable a single optical-fibe cable to seve multiple RRHs, and an Optical Line Teminal (OLT) that aggegates multiple RRHs supplies powe to RRHs though the optical-fibe cable. Howeve, we need to educe the tansmission powe of OLT fo ou envisioned netwok since supplying powe above a cetain level may cause the fibe fuse effect. To cope with this issue, the ecent investigations on PoF ae tying to develop fibe cables that ae able to tansmit optical signal with few watts and to ceate RRHs that ae opeated with low powe consumption. In contast to these woks, this aticle focuses on the investigation of the powe-efficient netwok opeation in ode to tackle this issue. This appoach is absolutely impeative since the impovement of physical pefomance has limitations. On the othe hand, the emegence of vaious kinds of sevices indicates the need to impove the oveall pefomance fom a use s pespective [11]. Indeed, netwok-opeatos and sevice povides ae now switching thei focus fom netwok Quality of Sevice (QoS) to use Quality of Expeience (QoE). Accoding to many investigations accomplished by [12] [14], the QoE value has a slightly diffeent method of measuement fom QoS value since QoE is detemined by use context such as age, gende, egion, pesonality and so foth [15], [16]. Consequently, it can be said that a QoE-guaanteed netwok opeation is essential fo ou envisioned C-RAN [17], [18]. In this aticle, we aim to demonstate the QoE-guaanteed and powe-efficient netwok opeation. Fist, we descibe the mathematical model of ou envisioned C-RAN such as an enegy havesting model and a battey-poweed model of RRHs. Then, we show a coelation model between the communication distance fom RRH to use and QoE value, which is used fo ou poposed netwok opeation. Futhemoe, we pesent

3 2 : Base-Band Units Fibe cable CO : Optical Line Teminals coe 50μm : Splitte : Remote Radio Head clad 125μm RRH Antenna : Use OLT ONU LD Tx Rx PD LD Tx Rx PD Rx Tx LD PD Rx Tx Battey Fig. 2. Powe ove Fibe (PoF) in ou consideed netwok. Fig. 1. Achitectue of ou envisioned C-RAN based on PON exploiting PoF. a joint contol method of RRH sleep and tansmission powe of OLTs to educe the tansmission powe while guaanteeing the individual QoE of uses. The emainde of this aticle is oganized as follows. Section II descibes the system model of ou consideed C- RAN achitectue. The coelation between the QoE value and tansmission distance is elucidated in Section III. Ou poposed QoE-guaanteed and powe-efficient netwok opeation scheme is explained in Section IV. We pesent the pefomance evaluation in Section V. Finally, the aticle is concluded in Section VI. II. ENVISIONED CLOUD RADIO ACCESS NETWORK WITH POWER OVER FIBER In this section, we intoduce ou envisioned C-RAN achitectue and the oles of each netwok component. Futhemoe, we descibe the system model of ou envisioned C-RAN. A. Oveview Ou envisioned C-RAN is based on PON exploiting the PoF technology as shown in Fig. 1. As shown in this figue, ou assumed netwok can be divided into thee components, i.e., the CO, the OLTs, and the RRHs. In the emainde of the section, the oles of CO, OLT, and RRH ae descibed. Taditional RAN achitectue utilizes the distibuted esouce management. In this achitectue, a Base-Band Unit (BBU), which is a function to manage the esouce and intefeence, is put on each base station. In contast to this achitectue, C-RAN deploys BBUs in a CO in ode to contol numeous RRHs via an intelligent bain. This achitectue can effectively manage the complex functions of numeous RRHs, e.g., Multiple Input Multiple Output (MIMO), Coodinated MultiPoint (CoMP), and handove. In addition to wieless components, the CO contols the optical components in C-RAN. This means that the CO decides the optical esouce allocation and tansmission powe of OLTs. Ou envisioned C-RAN utilizes the PON to constuct a fonthaul netwok between OLTs and RRHs. Each OLT in ou envisioned netwok has two oles fo the connected RRHs as follows: (i) data communication between the OLT and RRHs and (ii) powe supply to RRHs. Fo data tansmission to multiple RRHs, we assume that Wavelength Division Multiplexing (WDM) is used fo esouce allocation. On the othe hand, we assume that RRHs tansmit data to the OLT based on Time Division Multiplexing (TDM). Fo powe supply to RRHs, we assume the usage of PoF technology which is able to convet optical signal to electical powe. Since PON has a featue of boadcast fom OLT to RRHs [19] and some RRHs eceive data which is not meant fo them, the unnecessay data is conveted to electic powe by PoF technology and used to powe its own opeation. The OLT is, theefoe, used by the CO to supply electic powe to the RRHs by using the unnecessay data. In the envisioned netwok, it is consideed that numeous RRHs ae deployed. As shown in Fig. 2, each RRH is composed of thee modules, i.e., the Optical Netwok Unit (ONU) module, battey module, and antenna module. Futhemoe, Fig. 2 illustates how the ONU module of an RRH communicates with the OLT. The optical fibe used fo communication between the OLT and the ONU module of the RRH is assumed to be a multi-mode fibe having coe diamete of 50µm and clad diamete of 125µm [20]. In case of downlink communication, the OLT eceives data fom the CO at its Rx component [21]. The eceived data is conveted into optical signal at Lase Diode (LD), which is tansmitted ove the fibe to the ONU of RRHs. The RRH, which eceives necessay data, convets the optical signal into electic signal, which is tansmitted to uses via antenna module. In case of eceiving unnecessay data, the optical signal is conveted into electical powe, which is stoed in battey module via Rx component. Hee, the OLT infoms the accommodated RRHs of the addess of tansmission data by sending the signal including the addess infomation in advance. The battey module continuously stoes the conveted powe and supply the stoed electicity to ONU and antenna modules fo thei opeation. Ou supposed RRHs also have a sleep function, which educes powe consumption of RRHs by tuning down some modules. Howeve, the PD and Rx components used to eceive the tansmitted signal fom OLT, and the battey module cannot ente in sleep state all times since they ae equied to havest electic powe [22]. B. System model Ou supposed C-RAN consists of a single CO, L OLTs, and R RRHs, whee the set of OLTs and that of RRHs ae defined as L = {l 1, l 2,..., l L } and R = { 1, 2,..., R },

4 3 espectively. Since each OLT connects to multiple RRHs via PON, we define R = {,1,,2,...,, R }, which denotes the set of RRHs that connect to the OLT. Hee, note that the numbe of banches of the splitte in PON, θ, is decided as R. Additionally, the distance of the link fom OLT to RRH is defined as d. In the PON with PoF, the optical tansmission signals fom OLTs ae attenuated because of the fibe attenuation, the powe banch at the splitte, and the loss of photoelectic convesion. Supposed that O, φ(d ), and ρ ae the tansmission powe of OLT, powe loss facto due to tansmission distance, and convesion efficiency, espectively, the eceived powe of RRH, O, can be expessed as O = ρφ(d )O, (1) θ φ(d ) = 10( d φ db/10), (2) whee φ db is the fibe attenuation in db/km [23]. Futhemoe, since the Rx module of the RRHs convets the unnecessay tansmission signal to electic powe and the battey module stoes the conveted powe, the havested powe of RRH, H, can be expessed as H = O (1 ω ) = ρφ(d )O (1 ω ), (3) θ whee ω is the atio of signal addessed to RRH. Let U = {u 1, u 2,..., u U } be the set of uses to which the RRH povides communication sevice, whee the set of total uses is defined as U = {u 1, u 2,..., u U } and each RRH has constaints such as the maximum numbe of accommodated uses M and maximum tansmission ange C. Additionally, based on the set infomation U fo all, we can deive the set of uses that ae vitually connected to the OLT, U, which is expessed as U = U U,1,2.... (4) U, R Since each OLT allocates the othogonal bandwidth to the connected RRHs and the amount of bandwidth to each RRH is decided based on the numbe of accommodated uses of the RRH, the allocated bandwidth fo the RRH, B, is calculated with the atio of accommodated uses of RRH, f, as follows: B = B f = B U, (5) U whee B denotes the total bandwidth assigned to the downlink of OLT in ou envisioned PON. In the envisioned C-RAN, each RRH stoes the havested enegy to the battey module, and the enegy is consumed in its own opeation. Additionally, since the powe consumption of RRHs in sleep state, E sleep, is smalle than that in active state, E active, some RRHs ente in sleep state if the enegy in thei battey is below a cetain level. Let q be the esouce utilization of the link between OLT and the RRH. While the RRH in active state eceives the signal addessed to it with the atio w = q f, the OLT γ : OLTs : Splitte : RRH γ a 1 a 2 a g a g+1 a g+2 a 2g a i Sevice aea A a Rl l 1 l 2 l S l L Fig. 3. Ou consideed RRH deployment. Sub aea a i γ g l1 l2 ls ls+1 ls+2 l2s would neve tansmit signal addessed to the RRH in sleep state (i.e., w = 0). Theefoe, the havested powe in sleep state, H sleep, and that in active state, Hactive, can be expessed as l L γ gs H sleep = O, (6) H active = O (1 q f ). (7) Supposed the time is divided into multiple time slots with length τ and the state of each RRH is decided at each time slot, the amount of battey of RRH in sleep state at the end of time slot t, T sleep can be expessed as γ g active (t), and that in active state, T (t), T sleep (t) = T (t 1) + (Hsleep E sleep)τ, (8) T active (t) = T (t 1) + (Hactive E active )τ whee T (t 1) denotes the amount of battey of RRH at pevious time slot t 1. Additionally, since only RRHs that ae able to opeate duing a time slot can be chosen as active RRHs, we decide whethe to let the RRH ente in active state at time slot t o not, based on the following condition. T (t) + (Hactive E active )τ > 0, (9) whee H active is the maximum havested powe of RRH, which is decided based on the maximum tansmission powe of OLT, O. C. RRHs deployment The est of this section pesents ou consideed stategy to distibute RRHs in the sevice aea. In the conventional stategy, the RRHs connected to the same OLT ae distibuted to a neaby aea due to the simplicity of esouce and intefeence management. Howeve, this esults in highe tansmission powe of OLTs when the taffic (o use) concentates in a specific aea. Consequently, we intoduce a scatteed stategy that distibutes RRHs connected to the same OLT to a distinct (o distant) aea. In the envisioned netwok, this appoach can be ealized because all RRHs ae vitually-managed by the CO without consideation of the connected OLTs. Fig. 3 shows an example of the scatteed deployment stategy. A sevice aea, A, is defined as a squae which has

5 4 dimensions of γ metes wide and deep. Additionally, aea A is divided into R l sub aeas since we choose an RRH fom each OLT, and distibute the chosen RRHs to the sub aeas. In othe wods, each sub aea has L RRHs that connect to the diffeent OLTs. Fo simplicity, supposed that g = R l is an integal numbe, each sub aea, a i, becomes a squae which has dimensions of γ/g metes wide and deep. Theefoe, by using a g-by-g matix, A can be expessed as a 1 a 2 a 3... a g a g+1 a g+2 a g+3... a 2g A =..... (10) a g2 g+1 a g2 g+2 a g2 g+3... a Rl Futhemoe, the sub aea a i is also divided into L cells to which L RRHs that ae chosen fom each R ae distibuted. Hee, supposed that s = L is an integal numbe, each cell becomes a squae which has dimensions of γ/g/s metes wide and deep. In the scatteed deployment, since we choose RRHs fo the k-th cell in sub aea a i, the s-by-s matix, a i, that indicates the location of RRHs, can be expessed as l1 l2 l3... ls ls+1 ls+2 ls+3... l2s a i =..... ls 2 s+1 ls 2 s+2 ls 2 s+3... l L (11) Supposed that the RRHs ae distibuted to the middle of each cell, the distance between neighbo RRHs, e.g., the distance between l1 and l2, is given as γ/g/s. III. THE CORRELATION BETWEEN TRANSMISSION DISTANCE AND QOE VALUE This section pesents the impact of distance between RRH and use on the QoE value in Voice ove IP (VoIP) sevices. Some mathematical expessions and numeical esults ae pesented to show the coelation. While the QoE value is defined as a function of multiple influence factos including some QoS factos [24], [25], we focus on packet loss in ode to deive the fundamental coelation in VoIP sevices. Accoding to the investigation made in [26], the coelation between QoE value, Q, and packet loss pobability, P (d), is modeled by the exponential function as Q = αe δp (d) + β, (12) whee α, β, and δ denote the paametes of the function that ae etieved by means of non-linea egession. Additionally, in ode to model the QoE in diffeent netwok and use envionments, we need to deive the appopiate values of α, β, and δ by using the obsevational data in the actual envionment. The investigation made in [24] has conducted an expeiment to measue the Mean Opinion Scoes (MOS) with diffeent packet loss pobability in the actual situation. Accoding to this wok, the exponential intedependency of QoE value, Q, and packet loss pobability, P (d), is given as Q = 3.01e 4.473P (d) (13) QoE value Distance Fig. 4. The coelation between QoE value and tansmission distance. On the othe hand, it is welnown that the packet loss pobability inceases with highe tansmission distance because of the channel fading due to the path loss [27], [28]. Accoding to (19) descibed in [28], in the data tansmission without etansmission contol, the packet loss pobability function, P (d), is expessed with tansmission ange of RRH, C, distance fom RRH to uses, d, and a positive intege value fo fading paamete, m, as P (d) = (d2 m) m Γ(m) 1/C 2 0 z m 1 e d2 mz dz. (14) We edefine the oiginal function of m descibed in (20) in [28] with the continuous value of d as m = λ µ/d σ. (15) Note that λ, µ, and σ denote the paametes that ae set to model the exact communication envionment. Fig. 4 shows the QoE value with diffeent tansmission distances fom RRH to use. This esult is calculated based on (13)-(15) and the paametes ae set as follows: C = 150 m, λ = 1.5, µ = 150, and σ = 0.5. As shown in Fig. 4, the QoE value is appoximately the same value when the tansmission distance is below 50 m. Fom 50 m, it exponentially deceases with the incease of the tansmission distance. It is obviously undestood that netwok opeatos can guaantee the QoE value by connecting RRHs to the uses whose distance is less than a distance theshold. In othe wods, the distance theshold decides the associated uses fo each RRH to guaantee thei QoE values. Fo instance, in the case of Fig. 4, the distance theshold of RRHs is set to 50 m if the guaanteed QoE value is set to 4 by netwok opeatos. IV. PROPOSED QOE-GUARANTEED AND POWER-EFFICIENT NETWORK OPERATION In this section, we descibe the poposed netwok opeation scheme fo ou envisioned C-RAN. Ou scheme aims to minimize the tansmission powe of each OLT while guaanteeing QoE values of individuals. Pocedue 1 shows ou poposed QoE-guaanteed and powe-efficient netwok opeation. The CO executes this pocedue at each time slot t. In this scheme, the guaanteed QoE value, Q, and tansmission ange of RRHs, C, ae

6 5 Pocedue 1 QoE-guaanteed and powe-efficient netwok opeation 1: Given: guaanteed QoE value, Q, tansmission ange, C 2: /* Decide distance theshold by analyzing the fitting function of Q in aea A */ Run Distance Theshold Decision Function, Ψ DTDF( Q, C) 3: /* Contol the sleep of RRHs and the tansmission powe of OLTs */ Run Sleep and Powe Contol Function, SPCF(Ψ) Function 1 DTDF( Q, C) 1: Deive the paametes, α, β, and δ, in (12) based on measuement data of uses 2: Decide distance theshold, Ψ, based on the fitting function of Q and guaanteed QoE value, Q 3: etun Ψ peviously set by the netwok opeato. Also, this opeation is classified into two functions, i.e., the distance theshold decision function (DTDF) and the sleep and powe contol function (SPCF). While DTDF decides the distance theshold fo guaanteeing the equied QoE value, SPCF jointly contols the state of RRHs, i.e., eithe sleep o active state, and the tansmission powe of OLTs based on the distance theshold. Fist, the CO executes DTDF( Q, C). Since the QoE value depends on use context, egion, and time, we need to deive the adequate function of QoE in eal-time based on (12). Theefoe, the CO fist estimates the appopiate paametes, α, β, and δ, by using the least-squae technique [29], [30]. Hee, the CO uses a lage numbe of measuement data of uses who ae in the focused aea and the data ae obseved and stoed into the CO befoehand. Aftewads, the CO can constuct the elationship between the QoE value and tansmission distance fom RRHs to uses based on the tansmission ange, C, and the deived function of QoE. By using the deived elationship and the guaanteed QoE value, Q, the CO decides the distance theshold, Ψ, in which the QoE value of uses can be guaanteed. Then, the CO executes SPCF(Ψ) by using the deived Ψ. Fist, the CO constucts a set of candidate RRHs, Ṙ, which will ente active state. Hee, the RRHs that satisfy the afoementioned condition (9) ae chosen as the candidates. On the othe hand, since the othe RRHs, i.e., RRHs in R Ṙ, do not have enough powe fo the opeation until the next time slot (t + 1), the CO lets these RRHs ente sleep state. Then, it will continue to decide the state of candidate RRHs until it finishes the state decision fo all candidates o use association to the RRHs fo all uses. In deciding the state of RRHs, the CO fist chooses the RRH, l k, which has the maximum amount of powe fom the set of candidates Ṙ and then emoves fom Ṙ. Afte this, the CO constucts a set of candidate uses, U, in which uses satisfy the distance theshold Ψ, i.e., tansmission distance between uses and RRH l k is shote than Ψ. Then, the CO continues to decide the uses accommodated by RRH l k until the set of candidates becomes empty o Function 2 SPCF(Ψ) 1: Constuct a set of candidate RRHs, Ṙ 2: let RRHs in R Ṙ ente sleep state 3: while Ṙ and U do 4: Choose l k which has maximum amount of powe 5: Ṙ Ṙ { } 6: Constuct a set of candidate uses fo l k, U 7: while U and U l < M do k 8: Choose u who has minimum distance to 9: U U {u l } k 10: U U l + {u k l } k 11: end while 12: if U > 0 then 13: let l k ente active state 14: U U U 15: else 16: let l k ente sleep state 17: end if 18: end while 19: let RRHs in Ṙ ente sleep state 20: while k < L do 21: Decide tansmission powe of, O, accoding to (16) 22: k k : end while the numbe of accommodated uses eaches the maximum value, M. In selecting the uses to be accommodated, the CO chooses the use, u, in ode of inceasing distance fom the closest use. Then, the CO emoves u fom the set of candidate uses U and adds u to the set of accommodated uses U. Afte finishing the use selection, the CO decides the state of RRH l k. The CO should let RRH l k ente active state if RRH accommodates at least one use, i.e., U > 0. On the othe hand, if thee is no uses accommodated by RRH l k, the CO lets RRH l k ente sleep state. Then, once the CO finishes the use association pocess, it lets the est of the candidate RRHs in Ṙ ente sleep state. Finally, the CO decides the tansmission powe of OLT, O, which is expessed with the maximum value of the equied tansmission powe fo the connected RRHs, O, as follows. O = max R O, (16) whee the value of O in case of sleep state can be calculated with the magin of battey, ε, as ( O θ ε T = (t) ) ρφ(d l + E k sleep, (17) ) τ and the value of O in case of active state can be decided as O = θ ρφ(d )(1 q f ) ( ε T (t) ) + E active. (18) τ

7 6 V. PERFORMANCE EVALUATION In this section, we confim the effectiveness of ou poposed QoE-guaanteed and powe-efficient netwok opeation in ou envisioned C-RAN in compaison with the conventional scheme, by using extensive simulations. Futhemoe, we evaluate the pefomance of the poposed scheme in diffeent scenaios, i.e., diffeent use densities and diffeent magins of battey. A. Paamete settings Table I descibes the settings of ou simulations. The simulations ae executed fo 100 seconds, which is divided into 10 time slots. In these simulations, a sevice aea is defined as a squae which has dimensions of 450 metes wide and deep. Uses ae evenly distibuted into this aea. As a system configuation, we conside that thee exist a single CO and 9 OLTs, which ae located in the middle of this aea. Additionally, since each OLT connects to 4 RRHs, thee exist 36 RRHs. The powe consumption of each RRH in active state and that in sleep state ae set to 1.5 W [20] and 0.7 W [31], espectively. Additionally, we assume that each RRH stoes 30 J in its battey at time slot t = 0. As a PoF model, the maximum tansmission powe of each OLT, convesion efficiency, fibe attenuation, and esouce utilization of the communication fom an OLT to any RRHs, ae set to 8 W, 0.6, 0.25 db/km, and 1 % espectively [23], [32]. Futhemoe, the tansmission ange of each RRH and maximum numbe of accommodated uses ae set to 150 m and 15, espectively. As a fading model, λ, µ, and σ ae set to 1.5, 150, and 0.5, espectively. To model the elation between QoE value and packet loss pobability, the following equation is used in the simulations. B. Pefomance compaison Q = 4.05e 8.473P (d) (19) In ode to veify the effectiveness of ou poposed netwok opeation, we compae the pefomance between it and the conventional scheme. While ou poposed scheme changes distance theshold based on the guaanteed QoE level, the TABLE I EVALUATION SETTINGS. Paamete Value Simulation time 100 s Length of time slot 10 s Size of sevice aea m 2 Numbe of OLTs 9 Numbe of RRHs connected to each OLT 4 Powe consumption of RRH in active state 1.5 W Powe consumption of RRH in sleep state 0.7 W Amount of battey at t = 0 30 J Maximum tansmission powe of OLT 8 W Convesion efficiency 0.6 Fibe attenuation 0.25 db/km Resouce utilization 1 % Tansmission ange of RRH 150 m Maximum numbe of accommodated uses 15 Fading paamete m = /d 0.5 TABLE II THE RELATION BETWEEN QOE LEVEL AND DISTANCE THRESHOLD IN THE SIMULATION ENVIRONMENTS. QoE level Distance theshold 137 m 81 m 70 m 60 m 41 m conventional scheme uses a constant distance theshold (i.e., 70 m). In this simulation, we evaluate the pefomance by changing the guaanteed QoE level fom 1 to 5. The adequate distance theshold to guaantee the cetain QoE level is listed as shown in Table II. Additionally, the total numbe of uses and the magin of battey ae set to 100 and 2 J, espectively. Fig. 5(a) demonstates the cumulative sum of aveage tansmission powe of OLTs in diffeent QoE settings. The aveage tansmission powe of OLTs in the conventional scheme does not change accoding to QoE level and is the same as the poposed scheme in Q = 3. In contast to this, the poposed scheme educes the aveage tansmission powe of OLTs when the guaanteed QoE level is lowe than 3. Although highe tansmission powe is equied in the case that the guaanteed QoE level is highe than 3, the diffeence is small compaed to the case of 1 and 2 because the tansmission powe inceases logaithmically with the incease of QoE level. Theefoe, the tansmission powe in the poposed scheme is lowe than the conventional scheme. Fig. 5(b) shows the cumulative sum of uses satisfying the QoE levels. As shown in this figue, the poposed scheme can guaantee the QoE level of much moe uses compaed with the conventional scheme. This is why, in case of highe guaanteed QoE level, the conventional scheme accommodates the uses that ae distant fom the adequate distance theshold although it cannot guaantee the QoE level of such uses. Additionally, in case of lowe guaanteed QoE level, the conventional scheme accommodates only uses that ae within 70 m although it can accommodate moe distant uses while satisfying thei QoE value. Since a lot of RRHs unnecessaily ente in active state at a time slot and the numbe of RRHs that can ente in active state will not be enough at the next time slot, the numbe of uses satisfying QoE value in the conventional scheme is lowe than that in the poposed scheme. Indeed, the poposed scheme associates much moe 55 uses while satisfying thei QoE value. Fig. 5(c) shows the cumulative sum of Guaanteed Uses and Powe Efficiency (GUPE) in diffeent QoE envionments. Since GUPE denotes how many uses can be guaanteed by 1 W tansmission powe, the value of GUPE, η, can be expessed with the numbe of uses satisfying QoE, U guaanteed, and the aveage tansmission powe of OLTs, O, as η = U guaanteed, (20) O Fom Fig. 5(c), it is clea that ou poposed scheme achieves highe GUPE in all QoE settings. The impovement of GUPE ises with incease in the guaanteed QoE level and impovement is 73.7% when the QoE level is 5. This means that the poposed scheme can effectively guaantee QoE level even when uses equest highe QoE level.

8 7 Cumulative tansmission powe of OLTs Cumulative uses satisfying QoE Cumulative guaanteed uses-powe efficiency Conventional scheme Poposed scheme Guaanteed QoE level (a) Cumulative sum of aveage tansmission powe of OLTs Conventional scheme Poposed scheme Guaanteed QoE level (b) Cumulative sum of uses satisfying QoE. Conventional scheme Poposed scheme Guaanteed QoE level (c) Cumulative sum of guaanteed uses-powe efficiency. Fig. 5. Pefomance compaison in diffeent QoE envionments. C. Impact of use density on the pefomance of the poposal Hee, we investigate the pefomance of ou poposed scheme in diffeent use densities. In this simulation, we change the numbe of uses fom 50 to 300 in 50 incements. Also, we set thee diffeent QoE levels, i.e. Q = 2, Q = 3, and Q = 4, whee (19) is used as a QoE function. Additionally, the magin of battey is set to 2J. Fig. 6(a) shows the aveage tansmission powe of OLTs in diffeent use densities. It is shown that the tansmission powe logaithmically ises and that thee is an uppe limitation of tansmission powe. This phenomenon happens fo the following easons. Since the numbe of accommodated uses Aveage tansmission powe of OLTs Ratio of uses satisfying QoE Guaanteed uses-powe efficiency Q = Numbe of uses (a) Aveage tansmission powe of OLTs. 0.7 Q = Q =2 Numbe of uses (b) Ratio of uses satisfying QoE Numbe of uses (c) Guaanteed uses-powe efficiency. Fig. 6. Pefomance evaluation in diffeent use densities. is limited, the numbe of RRHs that should ente in active state inceases with the incease in use density. Theefoe, each OLT needs to incease its tansmission powe in ode to give enough powe to the RRHs whose amount of battey is low. On the othe hand, since we set a lowe bound of battey, which decides whethe RRH can ente in active state, the tansmission powe is limited. Theefoe, the tansmission powe dastically inceases when the node density is low and the gowth of tansmission powe is gadual with the incease in use density. Fig. 6(b) demonstates the atio of uses satisfying QoE level in diffeent use densities in a setting equal to that of Fig. 6(a). In all cases, the atio of uses satisfying QoE degades with the

9 8 incease in use density. This is because, although the numbe of RRHs in active state should incease to guaantee the QoE level in case of highe use density, some uses cannot be accommodated by RRHs since the numbe of accommodated uses and the numbe of RRHs that can ente in active state at each time slot ae limited. Fig. 6(c) depicts the changes of GUPE in diffeent use densities. As shown in this figue, the inceasing ate in low QoE level setting (e.g., Q = 2) is highe than that in high QoE level setting (e.g., Q = 4). This indicates that, in case of high use density, the highe we set QoE, the moe difficult it is to guaantee QoE level and achieve high powe-efficiency. D. Impact of magin of battey on the pefomance of the poposal The est of this section demonstates the impact of opeation paamete ε, which indicates the magin of battey, on the pefomance of ou poposed scheme. In this simulation, we change the value of ε fom 0 to 10 in 1 incement. Additionally, (19) is used as QoE function and the numbe of uses is set to 100. As shown in Fig. 7(a), the aveage tansmission powe of OLTs can be inceased by setting a highe magin of battey. This is because this paamete denotes the amount of battey in an RRH that has minimum amount of battey at next time slot and the tansmission powe of OLTs is decided based on its amount of battey. On the othe hand, the atio of uses satisfying QoE level gadually inceases with the incease in the magin of battey, as shown in Fig. 7(b). This is caused by the incease of aveage tansmission powe. Theefoe, the poposed scheme can contol the tansmission powe of OLTs and atio of guaanteed uses by changing the magin of battey. Fig. 7(c) demonstates GUPE in diffeent magins of battey. Fom this figue, it is clea that thee exists an optimal point fo maximizing GUPE and the point depends on the QoE level, e.g., ε = 0 when Q = 2, ε = 1 when Q = 3, and ε = 6 when Q = 4. Consequently, we can conclude that the poposed scheme can effectively guaantee the QoE level by setting the optimal value of ε. VI. CONCLUSION In this aticle, we addessed the challenge of QoEguaanteed and powe-efficient netwok opeation fo C-RAN based on PON exploiting PoF. To addess this challenge, we deived a mathematical model to evaluate the pefomance of ou envisioned C-RAN. Additionally, we constucted a novel famewok to evaluate the coelation between the QoE value and tansmission distance fom RRH to use. This model showed the existence of the distance theshold, in which a cetain QoE value of uses can be guaanteed. Based on the distance theshold, we poposed an adequate netwok opeation scheme. Ou poposed scheme jointly contols the sleep scheduling of RRHs and the tansmission powe of OLTs to educe the tansmission powe while satisfying the QoE value. Extensive simulations demonstated the effectiveness of ou poposed scheme. Aveage tansmission powe of OLTs Ratio of uses satisfying QoE Guaanteed uses-powe efficiency Q = Magin of battey (a) Aveage tansmission powe of OLTs Q = Magin of battey (b) Ratio of uses satisfying QoE Magin of battey Q =2 (c) Guaanteed uses-powe efficiency. Fig. 7. Pefomance evaluation when the magin of battey changes. REFERENCES [1] S. Chen and J. Zhao, The equiements, challenges, and technologies fo 5G of teestial mobile telecommunication, IEEE Communications Magazine, vol. 52, no. 5, pp , May [2] E. Dahlman, et al., 5G wieless access: equiements and ealization, IEEE Communications Magazine, vol. 52, no. 12, pp , Dec [3] M. D. Sanctis, et al., Satellite communications suppoting Intenet of emote things, IEEE Intenet of Things Jounal, DOI: /JIOT , Oct [4] J. Wu, et al., Cloud adio access netwok (C-RAN): a pime, IEEE Netwok, vol. 29, no. 1, pp , Jan.-Feb [5] Z. Zhou, et al., Enegy-efficient esouce allocation fo D2D communications undelaying cloud-ran based LTE-A netwoks, IEEE Intenet of Things Jounal, DOI: /JIOT , Nov [6] G. Chang, C. Liu, and L. Zhang, Achitectue and applications of a vesatile small-cell, multi-sevice cloud adio access netwok using

10 9 adio-ove-fibe technologies, Poc. of IEEE ICC 2013, pp , Budapest, Jun [7] M. Geasimenko, et al., Coopeative adio esouce management in heteogeneous cloud adio access netwoks, IEEE Access, vol. 3, pp , Ap [8] Y. Guo, et al., Demonstation of a symmetic 40 Gbit/s TWDM-PON ove 40 km passive each using 10 G bust-mode DML and EDC fo upsteam tansmission, IEEE/OSA Jounal of Optical Communications and Netwoking, vol. 7, no. 3, pp. A363-A371, Ma [9] N. Shibata, et al., Dynamic IQ data compession using wieless esouce allocation fo mobile font-haul with TDM-PON, IEEE/OSA Jounal of Optical Communications and Netwoking, vol. 7, no. 3, pp. A372-A378, Ma [10] K. Miyanabe, et al., A cloud adio access netwok with powe ove fibe towad 5G netwok: QoE-guaanteed design and opeation, IEEE Wieless Communications, vol. 22, no. 4, pp , Aug [11] I. Bisio, et al., Gende-diven emotion ecognition though speech signals fo ambient intelligence applications, IEEE Tansactions on Emeging Topics in Computing, vol. 1, no. 2, pp , Jul [12] J. Zhang and N. Ansai, On assuing end-to-end QoE in next geneation netwoks: challenges and a possible solution, IEEE Communications Magazine, vol. 49, no. 7, pp , Jul [13] M. Dong, et al., Quality-of-Expeience (QoE) in emeging mobile social netwoks, IEICE Tansactions on Infomation and Systems, vol. E97-D, no. 10, pp , Oct [14] M. Dong, et al., QoE-ensued pice competition model fo emeging mobile netwoks, IEEE Wieless Communications, vol. 22, no. 4, pp , Aug [15] K. Mita, A. Zaslavsky, and C. Ahlund, Context-awae QoE modelling, measuement, and pediction in mobile computing systems, IEEE Tansactions on Mobile Computing, vol. 14, no. 5, pp , Dec [16] S. Baaković and L. Skoin-Kapov, Suvey and challenges of QoE management issues in wieless netwoks, Jounal of Compute Netwoks and Communications, vol. 2013, Aticle ID , 28 pages, Dec [17] P. Agyapong et al., Design consideations fo a 5G netwok achitectue, IEEE Communications Magazine, vol. 52, no. 11, pp , Nov [18] Y. H. Cho, et al., A QoE-awae popotional fai esouce allocation fo multi-cell OFDMA netwoks, IEEE Communications Lettes, vol. 19, no. 1, pp , Jan [19] K. Saito, et al., A MPCP-based centalized ate contol method fo mobile stations in FiWi access netwoks, IEEE Wieless Communications Lettes, vol. 4, no. 2, pp , Jan [20] D. Wake, et al., Optically poweed emote units fo adio-ove-fibe systems, Jounal of Lightwave Technology, vol. 26, no. 15, pp , Aug [21] J. Wei, and K. Zhaoyuan, Design of WDM RoF PON based on OFDM and optical heteodyne, IEEE/OSA Jounal of Optical Communications and Netwoking, vol. 5, no. 6, pp , Jun [22] H. Nishiyama, et al., A coopeative ONU sleep method fo educing latency and enegy consumption of STA in smat-fiwi netwoks, IEEE Tansactions on Paallel and Distibuted Systems, vol. 26, no. 10, pp , Oct [23] R. S. Penze, et al., Fibe poweed extende fo XG-PON/G-PON applications, IEEE/OSA Jounal of Optical Communications and Netwoking, vol. 6, no. 3, pp , Ma [24] M. Fiedle, T. Hossfeld, and P. Tan-Gia, A geneic quantitative elationship between quality of expeience and quality of sevice, IEEE Netwok, vol. 24, no. 2, pp , Ma.-Ap [25] J. Neckeboek, H. Buneel, and M. Moeneclaey, Application laye ARQ fo potecting video packets ove an indoo MIMO-OFDM link with coelated block fading, IEEE Jounal on Selected Aeas in Communications, vol. 28, no. 3, pp , Ap [26] T. Hoßfeld, et al., Testing the IQX hypothesis fo exponential intedependency between QoS and QoE of voice codecs ilbc and G.711, Technical epot, Univesity of Wuzbug, Gemany, Ma [27] Q. T. Zhang, Outage pobability in cellula mobile adio due to Nakagami signal and intefees with abitay paametes, IEEE Tans. on Vehicula Technology, vol. 45, no. 2, pp , May [28] M. Xiaomin, et al., MAC and application-level boadcast eliability in vanets with channel fading, Poc. of ICNC 2013, pp , San Diego, CA, Jan [29] K. Slavakis and G. B. Giannakis, Online dictionay leaning fom big data using acceleated stochastic appoximation algoithms, Poc. of ICASSP 2013, pp , Floence, May [30] D. Bebeidis, et al., Online censoing fo lage-scale egessions, Poc. of 2014 Asiloma Confeence on Signals, Systems and Computes, pp , Pacific Gove, CA, Nov [31] L. Shi and S-S. Lee, Enegy-efficient PON with sleep-mode ONU: pogess, challenges, and solutions, IEEE Netwok, vol. 2012, no. 2, pp , Ma.-Ap [32] X. Xie, et al., 1.8 watt RF powe and 60% powe convesion efficiency based on photodiode flip-chip-bonded on diamond, Poc. of CLEO: QELS-Fundamental Science, JTh5B.9, Katsuya Suto eceived M.S. degee in Infomation Science fom the Gaduate School of Infomation Sciences (GSIS), Tohoku Univesity, Japan, in Cuently, he is pusuing the Ph.D. degee in the GSIS at Tohoku Univesity. His eseach inteests ae in the aeas of big data mining achitectue, esilient netwok design, and wieless netwoking. He was a ecipient of the pestigious Dean s Awad fom Tohoku Univesity in Mach He eceived the Best Pape Awad at the IEEE 79th Vehicula Technology Confeence (VTC 13-sping), the IE- ICE Academic Encouagement Awad in 2014, the IEEE VTS Japan 2015 Young Reseache s Encouagement Awad, and the Best Pape Awad at the IEEE/CIC Intenational Confeence on Communications in China in 2015 (ICCC 15). He is a student membe of IEEE and IEICE. Keisuke Miyanabe is pusuing the M.S. degee in the Gaduate School of Infomation Sciences at Tohoku Univesity. He eceived the IEEE VTS Japan 2014 Student Pape Awad. His eseach inteest is in the aea of Fibe Wieless netwok. Hioki Nishiyama is an Associate Pofesso at the Gaduate School of Infomation Sciences (GSIS), Tohoku Univesity, Japan. He eceived his M.S. and PhD in Infomation Science fom Tohoku Univesity, Japan, in 2007 and 2008, espectively. He has published moe than 160 pee-eviewed papes including many high quality publications in pestigious IEEE jounals and confeences. He was awaded Best Pape Awads fom many intenational confeences including IEEE s flagship events, such as the IEEE Global Communications Confeence in 2014 (GLOBECOM 14), GLOBECOM 13, GLOBECOM 10, and the IEEE Wieless Communications and Netwoking Confeence in 2014 (WCNC 14), WCNC 12. He was also a ecipient of the Special Awad of the 29th Advanced Technology Awad fo Ceativity in 2015, the IEEE Communications Society Asia-Pacific Boad Outstanding Young Reseache Awad 2013, the IEICE Communications Society Academic Encouagement Awad 2011, and the 2009 FUNAI Foundation s Reseach Incentive Awad fo Infomation Technology. He cuently seves as an Associate Edito fo IEEE Tansactions on Vehicula Technology, an Associate Edito fo Spinge Jounal of Pee-to- Pee Netwoking and Applications, and the Secetay of IEEE ComSoc Sendai Chapte. His eseach inteests cove a wide ange of aeas including satellite communications, unmanned aicaft system (UAS) netwoks, wieless and mobile netwoks, ad hoc and senso netwoks, geen netwoking, and netwok secuity. One of his outstanding achievements is Relay-by-Smatphone, which makes it possible to shae infomation among many people by using only WiFi functionality of smatphones. He is a Senio Membe of the IEEE, as well as a membe of Institute of Electonics, Infomation and Communication Enginees (IEICE).

11 10 Nei Kato eceived his Bachelo Degee fom Polytechnic Univesity, Japan, in 1986, M.S. and Ph.D. Degees in infomation engineeing fom Tohoku Univesity, in 1988 and 1991 espectively. He joined Compute Cente of Tohoku Univesity as an assistant pofesso in 1991, and was pomoted to full pofesso position with Gaduate School of Infomation Sciences, Tohoku Univesity, in He became a Stategic Advise to the Pesident of Tohoku Univesity in 2013, and the Diecto of Reseach Oganization of Electical Communication (ROEC), Tohoku Univesity in He has been engaged in eseach on compute netwoking, wieless mobile communications, satellite communications, ad hoc & senso & mesh netwoks, smat gid, and patten ecognition. He has published moe than 300 papes in pee-eviewed jounals and confeence poceedings. He cuently seves as a Membe-at-Lage on the Boad of Govenos, IEEE Communications Society, the Chai of IEEE Ad Hoc & Senso Netwoks Technical Committee, the Chai of IEEE ComSoc Sendai Chapte, the Edito-in-Chief of IEEE Netwok Magazine, the Associate Edito-in-Chief of IEEE Intenet of Things Jounal, and an Aea Edito of IEEE Tansactions on Vehicula Technology. He has seved as the Chai of IEEE ComSoc Satellite and Space Communications Technical Committee ( ), the Chai of IEICE Satellite Communications Technical Committee ( ). His awads include Minou Ishida Foundation Reseach Encouagement Pize (2003), Distinguished Contibutions to Satellite Communications Awad fom the IEEE ComSoc, Satellite and Space Communications Technical Committee (2005), the FUNAI infomation Science Awad (2007), the TELCOM System Technology Awad fom Foundation fo Electical Communications Diffusion (2008), the IEICE Netwok System Reseach Awad (2009), the IEICE Satellite Communications Reseach Awad (2011), the KDDI Foundation Excellent Reseach Awad (2012), IEICE Communications Society Distinguished Sevice Awad (2012), Distinguished Contibutions to Disaste-esilient Netwoks R&D Awad fom Ministy of Intenal Affais and Communications, Japan, seven Best Pape Awads fom IEEE GLOBECOM/WCNC/VTC, and IEICE Communications Society Best Pape Awad (2012). Besides his academic activities, he also seves on the expet committee of Telecommunications Council, Ministy of Intenal Affais and Communications, and as the chaipeson of ITU-R SG4 and SG7, Japan. Nei Kato is a Distinguished Lectue of IEEE Communications Society and Vehicula Technology Society. He is a fellow of IEEE and IEICE. Hiotaka Ujikawa is an Enginee at NTT Access Netwok Sevice Systems Laboatoies, NTT Copoation, Kanagawa, Japan. He eceived his B. E. and M. E. degees in compute science fom Waseda Univesity, Japan in 2007 and 2009, espectively. In 2009, he joined NTT, whee he has been engaged in the eseach and development of optical access systems. His cuent eseach inteests include dynamic bandwidth allocation and dynamic sleep scheduling fo enegy efficient access systems. Ken-Ichi Suzuki is a Senio Reseach Enginee, Supeviso at NTT Access Netwok Sevice Systems Laboatoies, NTT Copoation. He eceived the B.E. and M.E. degees in electonic engineeing fom Utsunomiya Univesity in 1988 and 1990, espectively, and eceived the Ph. D. degee in infomation science and technology fom Hokkaido Univesity in In 1990, he joined NTT laboatoies, whee he has been woking on eseach and development of optical communication systems including passive optical netwok (PON) based optical access systems. His cuent eseach inteests ae 10G-EPON systems/technologies and opticalamplifie-based long-each PON systems/technologies. Since 2006, he has been woking fo IEEE standadizations and is a woking goup vote. He seved as an Executive Secetay of IEEE P SIEPON woking goup duing and seves as a Vice Chai since He seves as a Diecto of Optical Access Adhoc woking goup in HATS confeence in Japan fo EPON inteopeability tests since He also seves as a TPC Secetay of CQR2013. He is a membe of the Institute of Electonics, the Infomation and Communication Enginees (IEICE) of Japan, Optical Society of Ameica (OSA), and IEEE.