alberto conte and afef feki, alcatel-lucent bell labs...

IEEE Wireless Communications • October 201150 1536-1284/11/$25.00 © 2011 IEEE

TRENDS IN LOW-POWER BASE STATIONSThe amount of data exchanged over mobile net-works has increased exponentially over the pastfew years, mainly because of the introduction ofa variety of smartphones, and of the iPhone inparticular. This trend is not expected to slowdown in the near future, as the mobile market iscontinuously enriched with new services andinnovative wireless-enabled devices in the fieldsof consumer and professional electronics (e.g.,photo cameras, e-books), captors and sensors(e.g., fire alarms), and machine-to-machinedevices (e.g., cars, parking, vending machines).

A fundamental element of this (r)evolution isthe omnipresence of high-capacity connectivity,enabled by the fast development of broadbandwireless technologies: today third generation/Universal Mobile Telecommunications System(3G/UMTS), tomorrow fourth generation (4G),in the form of Long Term Evolution (LTE), aswell as the process of densification [1] of wire-less networks, with the deployment of short-range low-power base stations (BSs) to createso-called pico- and femtocells. This last evolu-tion is a major trend in the current growth ofmobile networks, and is based on a well-knownlaw in wireless systems: the gain resulting fromthe reuse of the same radio resource in non-con-tiguous cells drives the increase of the numberof BSs deployed within a given area (i.e., theirdensity).

Cell densification is becoming a key tech-nique in wireless networks deployment, as itenables the increase of capacity and energy effi-ciency in wireless networks [1]. The reason isthat in a non-empty space (i.e., in the vast major-ity of real cases) the power necessary to provideservice to a user increases quicker than the dis-tance between the BS and the user’s mobile ter-minal (MT) (mainly due to the attenuationundergone by the transmission). As an example,the total power consumed by four BSs is gener-ally smaller than the consumption of a high-power BS covering an equivalent area, and thecapacity is about four times higher. Unfortunate-ly, the first wave of densification is occurring asan additional layer to existing macrocell BSs,aiming at absorbing the peaks of capacitydemand, and not as their replacement, an evolu-tion expected to happen in a longer timeframe.Densified wireless networks are thus composedof superposed small (low-power) cells and large(high-power) cells, and are generally referred toas heterogeneous networks.

Currently, heterogeneous networks arenatively overdimensioned: when traffic is low,the probability of unused (i.e., unnecessary) BSsincreases considerably, particularly when the cellsize is small.

NEED FOR SLEEP MODESUnused BSs are one of the main sources ofenergy waste in current wireless networks. Thisis due to the fact that today BSs (regardless oftheir technology: UMTS, LTE, WiMAX, orWiFi) need to permanently signal their presenceand the availability of the cellular service. Inaddition, the current BSs are also designed tocontinuously listen to the radio environment inorder to detect incoming MTs. Continuous emis-sion and reception are always-running processes,which consume energy even when no MT isusing or requesting the BS service (e.g., at night).

With the above mentioned issues in mind, theapplication of sleep modes to BSs constitutes apromising approach to improve energy efficien-cy. Sleep modes allow turning off BSs when andwhere they are not necessary, especially in low-

ALBERTO CONTE AND AFEF FEKI, ALCATEL-LUCENT BELL LABSLUCA CHIARAVIGLIO, DELIA CIULLO, AND MICHELA MEO, POLITECNICO DI TORINO

MARCO AJMONE MARSAN, POLITECNICO DI TORINO AND INSTITUTE IMDEA NETWORKS

ABSTRACT

In this article, we consider the adoption ofsleep modes for the base stations of a cellularaccess network, focusing on the design of basestation sleep and wake-up transients.. We discussthe main issues arising with this approach, andwe focus on the design of base station sleep andwake-up transients, also known as cell wiltingand blossoming. The performance of the pro-posed procedures is evaluated in a realistic testscenario, and the results show that sleep andwake-up transients are short, lasting at most 30seconds.

CELL WILTING AND BLOSSOMING FORENERGY EFFICIENCY

The authors considerthe adoption of sleepmodes for the basestations of a cellularaccess network,focusing on thedesign of base station sleep andwake-up transients.

TE C H N O L O G I E S F O R GR E E E N RAD IO

CO M M UNICAT ION NETWORKS

CONTE LAYOUT 10/11/11 1:28 PM Page 50

IEEE Wireless Communications • October 2011 51

load periods, such as nights, weekends, and holi-days. Up to 20–40 percent of energy can besaved with this approach [2, 3].

Switching off a portion of the deployed BSsin a given area guarantees energy savings, evenwhen using commonly available chipsets. Thistechnique stands as a complementary approachto the design and development of specific ener-gy-efficient components (e.g., highly efficientpower amplifiers). While the latter approach isacceptable for high-end and high-cost macrocellBSs, the possibility to use commonly availableelectronic components is a fundamental aspectwhen designing low-cost and low-power BSs. Inaddition, since the radio emission is interruptedduring sleep modes, interference between cells isalso reduced, and unnecessary radiation is elimi-nated. For all these reasons, increasing attentionin the research community and telecom industryis paid to such energy-efficient mechanisms.

SLEEP MODE BASICS

The design of a BS sleep mode must satisfythree requirements: minimize the consumedenergy, guarantee the availability of wirelessaccess over the service area, and minimize theperceived degradation of the user experience.

These targets drive the definition of the twoprimary sleep mode mechanisms:• Sleep mode entrance: specifies when, how, and

which BSs enter sleep mode with minimaldegradation of the service offered to endusers

• Sleep mode exit: specifies when, how, andwhich BSs wake up to respond to trafficdemand increases, with minimal degradationof the service offered to end users

SLEEP MODE ENTRANCESleep mode entrance design must address twoaspects: detect when a BS can enter sleep mode,and describe how the transition from active tosleep state is implemented. A simple trigger ofsleep mode start can be obtained by monitoringBS activity to detect when it is unused or under-utilized. A trivial case is when the cell is emptyfor a certain period (i.e., no service is active). Aswe observed, the likelihood of this event increas-es as the size of the cells diminishes. Furthergains in terms of energy saving can be obtainedby considering that, in low load conditions,neighboring BSs can take the responsibility ofthe traffic of a BS that is turning off. Opera-tionally, this technique requires that we:• Identify the set of neighboring BSs that can

absorb the load of the BS to be switched off• Check that the neighboring BSs can effectively

be in charge of the additional load, possibly byapplying specific compensation actions, likeup-tilting their antennas or slightly increasingtheir emitted powerOnce the sleep mode trigger is confirmed, the

transition from active to sleep state can start.This transition must be designed in such a waythat any interruption of ongoing calls/sessions isprevented. Indeed, in several systems, such as3G systems, it is not always possible for the net-work to order all the MTs to hand over out of acell that is going to be switched off, because the

interference among adjacent cells might be toolarge to allow for handovers; in some cases, anabrupt power-off of a BS can induce call inter-ruptions. Call drops can be avoided by adoptinga slow reduction of the BS transmit power. Thistechnique is commonly named BS wilting or pro-gressive switch-off. Figure 1 illustrates the BSwilting concept: three BSs are initially active,and offer service to the MTs that are camping intheir cells. When the central BS starts its switch-off transient, it progressively reduces its trans-mitted power so that its coverage shrinks whilethe area of adjacent cells expands.

During this process, MTs served by the cen-tral BS can hand over to neighboring BSs. Atthe end of the transient, the central BS is insleep mode, and coverage is provided by theneighboring cells.

During BS wilting, if any MT experiences crit-ical degradation without any possibility to migrateto another BS, it can signal to the BS, which inturn can suspend (and even revert) the process.Idle MTs camping under the BS also need to bewarned of the ongoing process, and be invited torelocate away. If they fail to relocate, they shouldalert the BS to stop the wilting process. A cellbarred status is usually signaled by the wiltingBS, alerting the idle MTs and avoiding new MTsmoving in from neighboring cells.

SLEEP MODE EXITSleep mode exit (BS wake-up) should be trig-gered when the data traffic conditions imply ahigh risk of overload in neighboring active BSs,

Figure 1. Example of BS wilting procedure.


IEEE Wireless Communications • October 201152

or an unacceptable quality of service for endusers. An obvious solution is thus to allow theactive BSs to alert sleeping neighbors as soon astheir load exceeds a given threshold, close to sat-uration. In this case, the sleeping BSs mustreceive a wake-up signal on their backhaul inter-face (i.e., the dedicated link connecting the BSto the core network) and turn on rapidly. In thisapproach, the challenge resides in choosing agood threshold to wake up the BS, neither tooearly, to avoid unnecessary interference andenergy waste, nor too late, to avoid poor perfor-mance. As a general rule of thumb, this kind ofparameter is initially evaluated through extensivenetwork simulations, and then adapted to oper-ating conditions at runtime. The transition fromsleep to active mode must be implemented sothat denial of service and call drops are prevent-ed. This can be achieved through a progressiveswitch-on process, known as BS blossoming, dur-ing which the power of the BS is slowly raised toits nominal target value. Other challenging issuesare the criteria and techniques to detect andlocate the traffic peaks, to wake up the right BSor set of BSs.

USER-FRIENDLY WILTING ANDBLOSSOMING TRANSIENTS

Current and future cellular technologies, such asUMTS and LTE, are essentially interference-lim-ited; hence, the quality of the network operationscan be jeopardized by abrupt changes in trans-mitted power levels. For example, an MT thatreceives a strong signal from a nearby BS typical-ly cannot hear the signal of other BSs, and losesits connection to the network if the nearby BS isswitched off too quickly. Conversely, an MT thatreceives a weak signal from a distant BS loses itsconnection if a nearby BS is switched on too fastbecause of the generated strong interference.

This suggests that the BS wilting (a.k.a. wake-up or switch-on) and blossoming (a.k.a. sleep orswitch-off) transients must be very gradual, toallow MTs to hand over toward the BS that canserve them best after the transient. However,gradual transients can reduce the effectivenessof a BS sleep mode scheme if they are very long,since the power consumed by the BS during thetransients is typically quite close to full opera-tional power.

In this section we consider a small portion ofa realistic cellular network, and examine the caseof gentle BS wilting and blossoming transients,computing their expected duration with an accu-rate signal propagation model based on a tooldeveloped by Alcatel Lucent called Wireless Sys-tem Engineering (WiSE). The tool analyzespropagation in complex environments involvingreal building configurations (including construc-tion materials) by means of ray tracing tech-niques [4].

A preliminary analysis of the duration of gen-tle switch-off transients was originally reportedin [5], based on a much less accurate propaga-tion model.

We look at BS X, whose sleep (or wake-up)decision is made at instant t0 = 0, based onsome sleep mode start decision.

In the case of a BS sleep transient, we needto derive a power reduction profile for X thatjointly guarantees three conditions:• The MTs connected to X can handover to

neighboring BSs, and only very few (if any)are dropped

• The handover overload for all BSs is keptsmall, i.e., the number of simultaneous han-dovers never exceeds the maximum defined bythe operator

• The transient duration is as short as possible,compatibly with the requirements of the net-work and the usersSimilarly, in the case of a BS wake-up tran-

sient, we need to derive a power increase profilefor X that jointly guarantees that the MTs con-nected to neighboring BSs can handover to X,and only very few (if any) are dropped, as wellas the second and third conditions above.

The power reduction/increase profile isdefined by the function PX(t) that specifies thetransmitted power of BS X at time t. For theprofile we use a step function such that fromtime ti to time ti+1 the transmission power of Xis constant and equal to PX(ti), with i = 0,1, ⋅⋅⋅,M and t0 < t1 < ⋅⋅⋅ < tM. In the sleep transientcase, PX(t0) > PX(t1) > ⋅⋅⋅ > PX(tM) = 0 and thetypical normal transmission level of X is PX(t0).At time tM, BS X is switched off. In the switch-on case, PX(t) = 0 for t < t0 and at t0 the wake-up phase starts, with PX(t0) > 0, and PX(t0) <PX(t1) < ⋅⋅⋅ < PX(tM). At time tM–1 the BS X isalready operating at its typical transmissionpower, and at time tM the transient can be con-sidered completed.

The duration of the wilting and blossomingtransient is thus given by tM in both cases.

In this article, as an example, we examine thecase in which the power emitted by the BS beingswitched off is progressively halved, i.e., PX(t1) =PX(t0)/2, PX(t2) = PX(t1)/2, and so on, up to theminimum transmission power PX(tM–1); at thefollowing step the BS is switched off, PX(tM) = 0.For a BS wake-up transient, we examine thecase in which the power emitted by the BS isprogressively doubled, i.e., PX(t1) = 2 PX(t0),PX(t2) = 2 PX(t1), and so on, up to full transmis-sion power PX(tM–1). Clearly, other power reduc-tion/increase profiles might be decided,according to operator-specific criteria, such asminimizing the transient duration, making con-stant the expected number of handovers perpower level, and so on.

At the beginning of the sleep transient, X istransmitting at its typical value, PX(t0). We com-pute in this situation the number of MTs con-nected to X that also hear other BSs. Assumingthat at time t0 these MTs start to handover outof X, because X signals that it is about to switchoff, we compute the duration of the time intervalit takes to complete the handovers. Time t1 is setequal to this duration. At t1, X decreases itstransmission power to PX(t1) = PX(t0)/2. Giventhat the power decreased, some MTs that areconnected to X start also hearing other BSs andcan handover out of X. We again derive howmuch time it takes for these handovers, and thisduration defines the amount t2 – t1 that X needsto spend at transmission power level PX(t1). Weiterate this process up to the minimum power

Base station wiltingand blossoming transients must bevery gradual, toallow mobile terminals to handover toward the basestation that canserve them bestafter the transient.


level PX(tM–1). The possible remaining MTs thatcannot hear other BSs but X at power levelPX(tM–1) are dropped when, at tM, X is complete-ly switched off. The risk of dropping MTs isobviously a negative effect of the sleep transient.The occurrence of dropping, however, can bereduced to an acceptable level by appropriatepower reduction profiles with very low values ofthe transmit power in the last steps of the tran-sient. In addition, when strict dropping avoid-ance is required, a more conservative approachcan be adopted by stopping (and possibly revert-ing) the wilting process as soon as an MT raisesa “losing connectivity” alert. This kind of alert iscommonly available on current cellular technolo-gy, for example, event 2D in wideband code-divi-sion multiple access (WCDMA)/UMTStechnology [6] and event A2 in LTE [7].

Instead, at the beginning of the switch-onperiod, X starts transmitting at power level PX(t0).We compute in this situation the number of MTsthat receive the signal of X at higher power thanfor other BSs. At time t0 these MTs start to handover toward X, and we compute the duration ofthe time interval it takes to complete the han-dovers. Time t1 is set equal to this duration. Att1, X doubles its transmission power to PX(t1) = 2PX(t0). Given that the power is increased, somemore MTs start hearing X better than the otherBSs and can hand over to it. We again derivehow much time it takes for these handovers, andthis duration defines the amount of time t2 – t1 Xneeds to spend at transmission power level PX(t1).We iterate this process up to the maximum powerlevel PX(tM–1); during the time interval from tM–1to tM, some more MTs hand over toward X, andat time tM the transient can be considered com-plete. At each power increase step, the strongersignal generated by X risks causing excessiveinterference and call dropping, for MTs that areclose to X but connected to other BSs. For thisreason, the power increase should be smallenough to guarantee that MTs connected toother BSs are not disconnected and have thechance to hand over.

A more formal derivation of the averageduration of sleep transients is presented in [5].

A CASE STUDY

As a case study, we consider a portion of thecentral area of the city of Munich, Germany,which corresponds to a square of 800 × 800 mcomprising 1 macrocell, 8 microcells, and 10femtocells. A map of the considered area,together with a side view, is presented in Fig. 2.

The received power for each location and foreach BS is computed adopting a planning toolbased on ray tracing with up to four reflections.The granularity of the evaluation is a square of 2× 2 m. The evaluation of the received power ismade at an elevation of 1 m, the typical heightof an MT in the pocket. We assume that theMTs are uniformly distributed in the servicearea and that their density is ρ. Unless otherwisespecified, we set ρ = 5 × 10–3 users/m2, corre-sponding to about 3200 users in the consideredarea (i.e., a value that can correspond to inter-mediate load).

Let H be the maximum number of handovers

that can occur simultaneously toward a new BS(due to signaling channel limits), and let TH bethe time interval required to complete one hand-over procedure. Typically, the percentage ofhandovers in a cell can involve around 30 per-cent of the active MTs, but the actual numberdepends on many factors; we therefore let Hspan over a quite large interval, setting H equalto 2, 10, or 20.

When an MT starts a handover, it needs toidentify the new BS. The identification is fast ifthe carrier frequency of the new BS is the sameas that of the old one. A longer time interval isrequired if the carrier frequency changes, but wedo not consider this case. Typical durations forthe identification time and handover time are TI= 0.8 s and TH = 0.5 s, respectively, as indicatedin [8]. Table 1 reports the main parameters foreach cell type.

WILTING ANDBLOSSOMING TRANSIENT DURATIONS

First, we consider the scenario reported in Fig. 2and evaluate the durations of the sleep andwake-up transients for the BSs of microcell SC1and femtocell F2. We then modify the scenarioby considering the case in which no macrocell ispresent.

We consider first SC1, whose transmissionpower in normal conditions is PSC1(t0) = 1 W,


Figure 2. Case study: Map with cell identifiers (SC = micro cell, F = femtocell). Aerial view (top) and side view (bottom).

SC3

SC6

SC1

SC4

SC5

SC2

Macrocell

SC7

SC8

F8

F8

F5

F4

F3

F2

F1

F10

F11

F12

F7

F9



and we focus on the sleep transient. The inter-ference map when all the BSs transmit at theirmaximum power is reported in Fig. 3. BS SC1’sposition is indicated with a white marker. Thevertical bar in the figure indicates how many BSsan MT can hear in each point.

We assume that the transmission power ofSC1 is halved at each step for a total of M = 8steps, from 1 W to 1/128 W, before going tozero.

Table 2 reports:• The time t1 needed to perform, before starting

the actual transmit power reduction, all thepossible handovers out of X

• The time tM – t1 necessary to hand over theMTs while the power of SC1 decreases

• The total transient duration, tMThe last row of the table indicates the averagenumber of handovers to perform during times t1,tM – t1, and tM. Observe that the procedureallows no user to be dropped (instead of the 72.4that would have been dropped on average withan uncontrolled and abrupt switch-off of SC1) atthe cost of an additional delay before enteringthe sleep mode between 10 and 26 s.

The power decrease profile is detailed in Fig.4. The sleep transient duration is less than 26 sin the worst case (H = 2), a negligible amountwith respect to the typical periods for which aBS is in sleep mode in low traffic conditions.

Of course, different assumptions can be maderegarding the power reduction profile. As anexample, we also computed sleep transient dura-tions for a linear power reduction profile inwhich at each step the power decreases by 0.1W; the transient lasts, in this case, between 15and 29 s.

The computed durations for the cell wiltingtransients, of at most 30 s, depend on the userdensity, and decrease if fewer users are campingin the cell area. We can thus confidently assumethat the time lapse required for BS switch- off inperiods of low traffic is shorter than half a minute.

Consider now the wake-up transient for SC1,with a power increase profile composed of expo-nential steps (power is doubled at each step).The durations of the power level steps is shownin Fig. 5. Again, the transient lasts about half aminute, with the assumed parameters and userdensity.

To study the sleep and wake-up transients forfemtocell BSs, we consider the case of F2. Usual-ly, the number of MTs served by a femtocell canbe up to four or five. We observe in Fig. 3 thatall MTs connected to F2 can also hear other BSs.This means that before the switch-off, we canforce the MTs connected to F2 to hand over toanother BS. Thus, in this case, we do not needto progressively decrease the transmission powerof the femtocell BS, since we can switch off F2 assoon as the MTs are transferred to other BSs.The average number of users connected to F2 isNh(t0) = 0.18. This means that the average timeneeded to switch off the femtocell BS F2 is equalto one handover time: t1 = tM = 0.5 s. There-fore, the switch-off time of a femtocell BS canbe assumed to be on the order of a second, orvery few seconds at most. Similarly, the wake-uptransient does not require a gradual increase ofpower. When the femtocell BS switches on,some MTs start hearing it and are requested tohand over to it. The duration of this process isagain on the order of a few seconds.

NO MACRO CELLWe now study the same scenario, with onlymicro- and femtocells, to consider a setting withlower capacity and less uniform coverage. Forthe sake of brevity, we focus on sleep transientsonly. Since now no macrocell is present, to takeinto account the reduced capacity we use a small-er user density, ρ = 10–3 users/m2.

The interference map for this case is reportedin Fig. 6. We select again SC1 as the microcellBS to enter sleep mode.

Table 3 refers to the power reduction profilein which transmit power is halved at each step.The average number of users that are droppedat the end of the transient when the BS is finallyswitched off is 0.1, which is a negligible value.

Figure 7 reports the power reduction profilesfor both the cases in which, at each step, thepower is halved or decreased by a constantamount equal to 0.1 W. The transients alwayslast about half a minute in this case as well. Notethat with the linear power reduction profile, thelast step is much higher than in the exponentialcase. This implies the risk of disconnecting anon-negligible number of users when the BSpower goes to zero.

Table 1. Cell parameters.

Size Type Power [W] Height[m] Azimuth Tilt

Macrocell Tri-sector 39.81 (sect.) 33 (90,210,330) –5

Microcell Isotropic 1 4 0 0

Femtocell Isotropic 0.02 2 0 0

Figure 3. BSs interference areas.

[m]100

100

[m]

200 300 400 500 600 700 8001

200 2

400

300

3

500

600 4

700

800 5

CONTE LAYOUT 10/11/11 1:28 PM 4


CHALLENGES FOR ADVANCEDOPTIMIZATIONS

Cell sleep mode is not a last-minute novelty:rudimental sleep modes have been in use incellular network management for several years.They are usually based on static configurationof on-off periods, defined through the analysisof traffic history (e.g., night vs. day, or businessvs. residential hours), careful network plan-ning, as well as post-treatment of alerts andtraces to detect possible problems. Today,research efforts focus on the definition ofadvanced optimization algorithms to furtherimprove the performance in terms of energysavings and intercell interference reduction.Some recent relevant works in this field consid-er the switch-off of some BSs [2, 3, 9, 10, 11]or active transmitters [12]. Other approaches,instead, focus on planning with energy andspectral efficiency optimization as the primarydesign target [13, 14].

One of the major open challenges resides inthe possibility to dynamically detect and locatecapacity needs, and to turn on/off BSs accord-ingly in (quasi) real time. This leads to a difficult(and currently unsolved) optimization problem:how to set the best combination of active andsleeping BSs in a given area in order to satisfythe needs of users, while at the same timeachieving maximum energy savings.

Another important issue to be solved con-cerns the difficulty of guaranteeing continuouscoverage and the avoidance of coverage holeswhen switching off a base station. This isindeed very difficult, as radio propagation isstrongly affected by the environment and thuspoorly predictable. Even the progressive cellwilting process described previously is noterror proof. As an example, consider a housewhere a femtocell BS enters sleep mode afterthe last MT is switched off. Later, the houseowner may wish to switch on his MT in a loca-tion where no other BS can be heard. This isclearly a deadlock: the femtocell BS is off andcannot detect the MT, and the MT has nomeans to signal i ts problem. To solve thisissue, a new approach is emerging [15] thatproposes to move from an avoidance problemto a handling problem: coverage holes’ exis-tence is accepted, and MTs shall be allowed totransmit a bl ind wake-up message (a.k.a.reverse-paging message) to ask for any BS toswitch on.

Coverage hole handling (through blind wake-

up) will open the door to a new generation ofsleep mode algorithms, applicable also to homo-geneous deployments, and capable of achievingvery substantial energy savings. Unfortunately,blind wake-up is not available on current cellularsystems, and its definition will go through a longstandardization cycle before being effectivelyimplemented in real products.

BS sleep modes must also deal with closedaccess cells, that is, “private” cells that can beused only by a limited and well identified set ofusers. In this case, the mechanisms previouslydescribed must be enhanced with an identifica-tion of the MT in order to avoid switching on aprivate BS to serve an MT that has no right touse it.

Figure 4. Switch-off transient for SC1. Exponential power decrease profile, withH = 2, 10, 20.

Time (s)50

0.2

0

BS t

rans

mis

sion

pow

er (

W)

0.4

0.6

0.8

1

1.2

10 15 20 25

H=2H=10H=20

Figure 5. Switch-on transient for SC1. Exponential power increase profile, withH = 2.

Time (s)50

0.2

0

BS t

rans

mis

sion

pow

er (

W)

0.4

0.6

0.8

1

1.2

10 15 20 25

Table 2. Switch-off transient for SC1: averageswitch-off times and number of handovers.

H t1 [s] tM – t1 [s] tM [s]

2 16 9.3 25.3

10 3.5 7.8 11.3

20 2 7.8 9.8

No. of handovers 63 9.4 72.4


CONCLUSIONSIn this article, we have analyzed cell wilting andblossoming techniques that consist in a progres-sive base stations switch-off and -on, respective-ly. Specifically, we have considered a realisticcellular network scenario, and we have comput-ed the duration of BS sleep and wake-up tran-sients using an accurate signal propagationmodel. Our main finding shows that these tran-sients are very short, thus allowing BSs to beswitched off and on in short time, with no signif-icant reduction of the energy savings achievablewith sleep mode approaches.

This indicates that the use of BS s leepmodes can be very effective in reducing ener-gy consumption and might become pervasivein the wireless networks of the future. We canenvision that in a not-so-distant future, driv-ing back home from a party, late at night, lis-tening to our favorite web radio or themessages recorded on our answering machine,or email messages being read by a voice syn-thesizer, over a wireless cellular connection,the BSs in charge of our MT will progressivelywake up along our route as we drive by, andgo back to sleep after we drive past them.Since traffic is low, late at night, keeping BSson implies an unreasonable energy waste, andthe switch-on and switch-off transients dura-tions are much shorter than the times betweencars passing by, so energy savings are notcompromised.

ACKNOWLEDGMENTThe research leading to these results hasreceived funding from the European Union Sev-enth Framework Programme (FP7/2007–2013)under grant agreement no. 257740 (Network ofExcellence “TREND”).

REFERENCES[1] F. Richter and G. Fettweis, “Cellular Mobile Network

Densification Utilizing Micro Base Stations,” ICC 2010,Cape Town, South Africa, May 2010.

[2] M. Ajmone Marsan et al., “Optimal Energy Savings inCellular Access Networks,” 1st Int’l. Wksp. Green Com-mun., Dresden, Germany, June 2009.

[3] M. Ajmone Marsan et al., “Energy-Efficient Manage-ment of UMTS Access Networks,” ITC 21 — 21st Int’l.Teletraffic Congress, Paris, France, Sept. 2009.

[4] S. Fortune et al., “WiSE Design of Indoor Wireless Sys-tems: Practical Computation and Optimization,” Com-puting in Sci. and Eng., vol. 2, no. 1, 1995, pp. 58–68.

[5] M. Ajmone Marsan et al., “Switch-Off Transients in Cel-lular Access Networks with Sleep Modes,” GreenComm4, Kyoto, Japan, June 2011.

[6] TS 25.331, 3GPP; Tech. Spec. Group Radio Access Net-work; Radio Resource Control (RRC); protocol spec..

[7] TS 36.331, 3GPP; Tech. Spec. Group Radio Access Net-work; Evolved Universal Terrestrial Radio Access (E-UTRA); Radio Resource Control (RRC); protocol spec..

[8] H. Holma and A. Toskala, WCDMA for UMTS, Wiley,2004.

[9] E Oh, B. Krishnamachari, “Energy Savings throughDynamic Base Station Switching in Cellular WirelessAccess Networks,” IEEE GLOBECOM, Miami, FL,Dec.2010.

[10] Z. Niu et al., “Cell Zooming for Cost-Efficient GreenCellular Networks,” IEEE Commun. Mag., Nov. 2010,vol. 48, no. 11, pp. 74–79.

[11] K. Dufkova et al., “Energy Savings for Cellular Networkwith Evaluation of Impact on Data Traffic Perfor-mance,” Euro. Wireless 2010, Lucca, Italy, Apr. 2010.

[12] L. Saker and S. E. Elayoubi, “Sleep Mode Implementa-tion Issues in Green Base Stations,” PIMRC 2010, Istan-bul, Turkey, Sept. 2010.


Figure 6. No macro cell. BSs interference areas.

[m]100

100

[m]

200 300 400 500 600 700 8001

200 2

400

300

3

500

600 4

5

700

800

Figure 7. No macro cell. Switch-off transient for SC1. Exponential and linearpower decrease profiles, with H = 2.

Time (s)50

0.2

0

BS t

rans

mis

sion

pow

er (

W)

0.4

0.6

0.8

1

1.2

10 15 20 25 30 35

ExponentialLinear

Table 3. No macro cell: average switch-off timesand number of handovers.

H t1 [s] tM – t1 [s] tM [s]

2 16 14.4 30.4

10 10.9 3.5 14.4

20 10.4 2 12.4

No. of handovers 62.33 21.66 83.99



[13] A. J. Fehske, F. Richter, and G. Fettweis, “Energy Effi-ciency Improvements Through Micro Sites in CellularMobile Radio Networks,” GreenComm ’09, Honolulu,HI, Dec. 2009.

[14] F. Richter et al., “Micro Base Stations in Load Con-strained Cellular Mobile Radio Networks,” PIRMC 2010W-Green Wksp., Istanbul, Turkey, Sept. 2010.

[15] I. Haratcherev, C. Balageas, and M. Fiorito, “Low Con-sumption Home Femto Base Stations,” 1st IEEE PIMRC2009 Wksp. Indoor and Outdoor Femto Cells, Tokyo,Japan, Sept. 2009.

BIOGRAPHIESALBERTO CONTE ([email protected]) is a teammanager in the Networks Research Domain at Alcatel-Lucent Bell Labs, Paris, France. After receiving M.S. degreesin telecommunication engineering from the Politecnico diTorino in Italy and Institut Eurécom in Sophia Antipolis,France, he joined the software department department atBell Labs, France, initially working on embedded systemsand VoIP softswitches. He then shifted his research inter-ests to wireless technologies for both home/enterprise(WiFi) and public cellular systems. He and his team pio-neered the definition of end-to-end mobile solutions basedon WiMAX technology, and he played a key role as Alcatel-Lucent representative at the WiMAX Forum. His currentresearch focuses on 4G networks and their optimizations,including high-density self-x deployments, energy savingssolutions, and ubiquitous access enablers to improve enduser experience. He is an active lecturer at Ecole NationaleSupérieure des Télécommunications (ENST), INSA-Lyon, andSupelec (France), where he teaches WiMAX technology toM.S. degree students. He is also a member of the Alcatel-Lucent Technical Academy.

AFEF FEKI ([email protected]) is a research engi-neer at Alcatel Lucent Bell Labs, France. She holds an engi-neering degree in telecommunications as well as an M.Sc.in optics and radio frequency, both from the PolytechnicInstitute of Grenoble in France, and she received a Ph.D. intelecommunications from Telecom ParisTech with highhonours. Her research interests at Bell Labs include 4Gwireless networks, femto and small cells deployment, andgreen networks, as well as self configuration and manage-ment of radio networks. Previously, she worked at OrangeLabs R&D France where she was involved in various pro-jects on cognitive radio, dynamic spectrum access issues,and the optimization of radio resource management. Shehas published over 10 journal and conference papers, andholds several patents. She is co-chair of the International

Workshop on Indoor and Outdoor Femto Cells (IOFC), TPCco-chair of WiOpt 2012, and TPC member for other confer-ences such as WCNC 2012 and VTC 2011.

LUCA CHIARAVIGLIO ([email protected]) received hisMaster’s degree in computer engineering and Ph.D. degreein electronics and communications engineering, both fromPolitecnico di Torino in 2007 and 2011, respectively.Between August 2009 and April 2010 he was a visitingscholar at the WING group of Boston University under thesupervision of Prof. Ibrahim Matta. His main research inter-ests are in the field of energy-aware design for backboneand access networks.

DELIA CIULLO ([email protected]) received her Master’sdegree in telecommunications engineering and Ph.D.degree in electronics and communications engineering,both from Politecnico di Torino in 2007 and 2011, respec-tively. In 2009 she was a visiting student at the CNRGgroup of MIT under the supervision of Prof. Eytan Modi-ano. She is currently a postdoctoral researcher at Politecni-co di Torino. Her research interests are in the fields ofenergy-aware networks, scaling properties in wireless net-works, and P2P-VOD systems.

MICHELA MEO ([email protected]) received her Laurea degreein electronics engineering in 1993, and her Ph.D. degree inelectronic and telecommunications Engineering in 1997,both from the Politecnico di Torino, Italy. Since November1999 she has been an associate professor at Politecnico diTorino. She has co-authored more than 150 papers andedited six special issues of international journals. Herresearch interests are in the field of green networking, per-formance evaluation and modeling, traffic classificationand characterization, and P2P.

MARCO AJMONE MARSAN [F] ([email protected]) is a fullprofessor in the Electronics Department of Politecnico diTorino and a part-time chief researcher at IMDEA Networks,Madrid, Spain. From September 2002 to March 2009 hewas director of the Institute for Electronics, Information andTelecommunications Engineering of the National ResearchCouncil. From 2005 to 2009 he was Vice-Rector forResearch, Innovation and Technology Transfer at Politecnicodi Torino. In 2002 he was awarded a honorary doctoraldegree in telecommunications networks from the BudapestUniversity of Technology and Economics. He is the chair ofthe Italian Group of Telecommunications Professors, andthe Italian Delegate to the ICT Committee of the SeventhFramework Programme of the European Union.


alberto conte and afef feki, alcatel-lucent bell labs...

Documents