No Deviation Pte Ltd
CLIENT oriented solutIONS
we focus on delivering tailor-made solutions to allow Clients realise their overall vision
Whether an Integrated Team, or overall scope management, our team with their strong practical and hands-on background has proven experience in developing and implementing strategies and organization to suit the project drivers.
With a focus on implementing a solid foundation and structure based on current industry practices along with the integration of client Personnel to provide practical hands-on experience, No deviation Pte. Ltd. can provide the resources to lead or support your Commissioning and Qualification effort.
Our success is derived from fostering a common goal and defined objectives within the Commissioning and Qualification team and external actors while delivering the guidance and leadership to realise the client’s objectives.
Understand the client’s drivers to develop appropriate support solution
Observe and communicate to develop a sense of empathy
Define client’s needs and develop insights
Implement identified support needs and continually assess the efficiency of the planned actions
We strive to provide Client based support with the above drivers
SKILLED TALENT REQUIREDWe are currently involved in a number of medium and large scale OPEX and CAPEX projects for large Multi-national Corporations (MNC) providing leadership and oversight in a numbers of keyroles within the clients team.To find out more, checkout out
page or contact us by email
We are a client oriented engineering company, providing technical service to end-users In Singapore, Asia and Europe since our estabIishment in 2007. Our main areas of operation are within the Bio-Pharmaceutical / Pharmaceutical industries with a focus in Commissioning & Qualification.
Contact Info
Address: Vision Exchange, 2 Venture Drive
#13-27 Singapore 608526Capex Forensics Ltd. - Construction Dispute Resolution
Construction dispute, advisory and investigatory services
CapEx Forensics specialises in the investigation and analysis of construction projects.We are able to provide capital and revenue expenditure due diligence, cost verification, project snapshots, expenditure reviews and risk analysis.
Construction Dispute, Advisory and Investigatory Services
Our professionals can provide cost certainty throughout your project’s life-cycle by applying our unique blend of experience and knowledge of the construction industry.
CapEx Forensics
CapEx Forensics provides an unique blend of highly experienced experts who&specialise in&construction dispute, advisory and investigatory services. We can help you make&informed and valuable decisions around your capital investments and revenue expenditure.The combined knowledge and expertise of&our professionals can provide clear and practical solutions to the challenging construction issues that face your business and allow you to focus on what you do best.
‘CapEx’ (or Capital Expenditure) is an accounting term used by a company when it acquires or upgrades physical assets such as equipment, property or industrial buildings.‘Forensics’ is the use of science and technology to investigate and establish facts in criminal or civil courts of law.
We know that pursuing or defending a claim can be a drawn-out and expensive process.Working alongside your team. we can help provide a detailed financial overview of your case and, if you decide to continue, help present your dispute in a clear manner that will identify the key matters and the principle sums involved to help achieve settlement.当前位置： >>
A Survey of Energy Efficient Resource Management Techniques for Multicell Cellular Networks
154IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 16, NO. 1, FIRST QUARTER 2014A Survey of Energy Ef?cient Resource Management Techniques for Multicell Cellular NetworksJaya B. Rao, Student Member, IEEE, Abraham O. Fapojuwo, Senior Member, IEEE,Abstract―This paper surveys the recent ?ndings in the area of energy ef?cient radio resource management in cellular networks. The primary objective is to identify and evaluate the key techniques that have the highest energy saving potential to be developed in the context of Green Networks while serving as a guideline for future research endeavours. The focus of the paper is targeted towards multicell networks which are composed of multiple BSs co-existing in the same area sharing the available radio resources. Due to this, greater emphasis is given towards the techniques that take inter-cell interference (ICI) into account while allocating the resources and, in the process, maximize the energy ef?ciency (EE). The resource management solutions presented in the paper are classi?ed under three network domains namely homogeneous, heterogeneous, and cooperative networks. Furthermore, the analytical techniques for characterizing the EE of multicell networks are discussed in terms of the stochastic geometry framework. Finally, the paper outlines the current challenges and open issues in the area of energy ef?cient resource management for multicell cellular networks. Index Terms―Cellular Networks, Energy Ef?ciency, Resource Management, Quality of ServiceI. I NTRODUCTIONTHE RECENT revelation pertaining to global share of energy consumption ascribed to the cellular networks has generated considerable amount of interest in the wireless communications community in both academia and industry [1][2][3]. Currently, it has been estimated that the overall energy consumed by information and communications technology (ICT) industry, which includes cellular networks, already constitutes about 2% of global carbon emissions and is projected to increase much further in the coming years [4]. Moreover, with increasing network operational expenditure (OPEX) coupled with the spur of government led regulatory initiatives have spawned numerous research projects worldwide [5] with speci?c targets to drive down the energy expenditure in all aspects of cellular networks. It has been revealed in [6] that up to 80% of the energy consumption in a cellular network is attributed to the operations and functionality of the Base Stations (BS) in the radio access network (RAN) while the remaining energy is expended in the switching and core networks. Collectively, these factors have heightened the necessity to focus on the energy ef?cient resource managementManuscript received October 18, 2012; revised February 18, 2013. This work was supported in part by a grant from National Sciences and Engineering Research Council of Canada. The authors are with the Department of Electrical and Computer Engineering, University of Calgary, Calgary, AB, T2N 1N4, Canada (e-mail: jbrao@ucalgary. fapojuwo@ucalgary.ca). Digital Object Identi?er 10.1109/SURV..00226solutions that are capable of minimizing the high energy expenditure while at the same time maintain and enhance the quality of service (QoS) offered to the subscribers. In the existing multicell networks optimized for spectral ef?ciency (SE), multiple BSs operate in overlapping coverage areas while sharing the available spectrum as shown in Fig. 1. Under the system con?guration shown in Fig. 1, the ensuing inter-cell interference (ICI) adversely affects the performance for both the uplink and downlink transmissions. Thus, a certain degree of coordination between the BSs is required to manage the resources and minimize the interference level. Note that resource management in this context refers to the task of optimally allocating the radio resources (i.e. transmit power, time slots, bandwidth, antenna con?guration) while ful?lling a certain system performance objective such as maximizing the SE or Energy Ef?ciency (EE) during transmissions. However, this comes at the cost of increased inter-cell coordination over the backhaul links (Fig. 1) to exchange information related to the resource usage, load level and users channel state information (CSI). Furthermore, multicell networks can be structured as heterogeneous or cooperative systems consisting of smaller scale BSs and relay stations (RS) with varied capacity and power attributes where issues concerning admission control and load balancing can affect the feasibility of the resource management techniques. The challenge therefore, is on how to perform optimal resource management in the wider multicell context that results in maximizing the EE performance of the entire cellular network in the presence of ICI while maintaining high SE. One of the earliest insights into the potential of the existing cellular networks to operate in an energy aware manner was provided in [7]. Since the existing cellular networks are primarily designed to maximize SE, it is revealed that the potential to achieve high energy savings with the current con?guration is rather limited. To address this, the authors of [7] proposed a list of areas that can be explored from the perspective of the BS hardware, the transmission link and the network layer to maximize the overall EE. For performing resource allocation at the physical layer, the authors of [8] provide a clear theoretical foundation to characterize the EE using concepts from the game theory, information theory and hardware related models which include the device circuitry power attributes. In this context, the EE is de?ned as the ratio of the data rate achieved to the total power consumed over the point-to-point transmission link. A wider scope of energy ef?cient resource allocation techniques applicable at the physical and medium access control (MAC) layers are surveyed in [9]. Most of the techniques reviewed in [9]/$31.00 c 2014 IEEERAO and FAPOJUWO: A SURVEY OF ENERGY EFFICIENT RESOURCE MANAGEMENT TECHNIQUES FOR MULTICELL CELLULAR NETWORKS155Fig. 1. OFDMA based Cellular Network with shared spectrum operation where UEs in each cell are allocated with different RBGs by the associated BSs. The radio access network (consisting of BSs) and the core network (MME, S-GW, P-GW) are interconnected via the backhaul interfaceare relevant to orthogonal frequency division multiple access (OFDMA) system in the context of a single cell setting where the BS is responsible for the resource allocation. A detailed overview and directions for future research initiatives targeted to improve the EE of various wireless systems are also provided in [10]. For each technology reviewed (i.e. OFDMA, Relay Systems and multiple input multiple output (MIMO)), a set of open issues that must be overcome to gain higher energy savings is identi?ed. The various tradeoff relationships inherent in cellular networks pertaining to deployment ef?ciency vs. EE, SE vs. EE, bandwidth vs. power, and delay vs. power are discussed in detail in [11]. Upon outlining the speci?c domains (i.e. single link, single cell, multicell) where the impact of the tradeoff is more pronounced, the authors of [11] identify the realistic and practical issues to be addressed prior to realizing enhanced EE in the network. A more comprehensive survey of the recent techniques proposed with the aim of reducing the energy consumption in cellular networks is given in [12]. The authors highlight deployment of heterogeneous BSs as one of the key techniques from the context of network planning to reduce the power consumption while delivering higher data rates. This paper systematically reviews and evaluates the various studies performed in the area of energy ef?cient resource management in cellular networks, similar to the topic treated in [9]-[12]. However, contrary to [9][10], this paper is fo-cused on energy ef?cient resource management techniques in multicell cellular networks. Interest is on multicell cellular networks due to their practicality, hence the need to assess the effectiveness of resource management techniques in achieving EE. Multicell cellular networks can be con?gured as a single tier homogenous macrocell network or multiple tiers of hierarchically structured heterogeneous network, for example, in a 2-tier network, a number of smaller scale picocell BSs of the pico tier overlays a single macrocell at the macro tier. Multicell cellular networks also permit the incorporation of the cooperative concepts to form cooperative networks, where cooperation is implemented by either relaying or coordination among BSs of the same tier and/or different tiers. The energy ef?cient resource management techniques presented in this paper are categorized into the subdomains of homogeneous, heterogeneous and cooperative networks because of 2 reasons: First, the division enables the resource management schemes to be distinguished in terms of feasibility, complexity and comprehensiveness, depending on the scope for which the schemes are applicable. Second, the division simpli?es the presentation of the open problems and suggestions on the opportunities for future research endeavors. Starting with the homogeneous network subdomain, some of the problems studied include energy ef?cient inter-cell power control and load driven cell size adaptation, both of which are targeted to maximize the network EE while meeting a blocking156IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 16, NO. 1, FIRST QUARTER 2014probability constraint. Heterogeneous networks (HetNets) are generally deployed to enhance the capacity and to close the coverage holes. Given the differences in the HetNet BSs with regards to transmit power level, hardware attributes and signal processing capability, the techniques relevant to the HetNet subdomain must be evaluated separately to better highlight its unique challenges. To this end, certain distinct problems such as joint interference management and resource allocation in full frequency reuse environments and self-organized operation with backhaul capacity disparity are discussed. The energy ef?cient techniques relevant to cooperative networks underscore its signi?cance as a cost effective framework to provide both coverage extension and capacity enhancement. However, unlike the homogeneous and HetNet subdomains, in the cooperative network subdomain, more emphasis is placed on the cost of cooperation which includes overhead signalling, joint processing and inter-BS coordination ? this overhead cost has been largely ignored in previous solutions aimed at maximizing the SE [13]. The organization of the paper is outlined as follows. Section II reviews the resource management techniques from the literature related to homogeneous networks consisting of a single tier macrocell BSs. This is followed by the survey of recent works applicable to heterogeneous networks composed of macro, micro, pico and femtocell BSs in Section III. Some of the related energy ef?cient techniques incorporating multiradio access technologies (RAT) (i.e. WiMAX and Cellular) are also covered under this section. In addition to the discussion on the general schemes, both Sections II and III also review the various load adaptive techniques, as well as energy ef?cient mobility management schemes. Section IV surveys the resource management approaches related to two types of cooperative networks, namely cellular relay systems and coordinated multipoint (CoMP) systems. The main results obtained for the analysis of energy ef?cient multicell networks by applying the stochastic geometry framework are presented in Section V. The current challenges and open issues in the area are given in Section VI, prior to the conclusion in Section VII. II. E NERGY E FFICIENT R ESOURCE M ANAGEMENT I N H OMOGENEOUS C ELLULAR N ETWORKS Homogeneous cellular networks consist of large scale macrocell BSs (operating over licensed spectrum) which are deployed over the service area to provide universal coverage and support seamless mobility. In OFDMA based cellular networks [14], the total available spectrum is partitioned into a number of radio resource blocks (RB) consisting of a set of orthogonal frequency-time units, each occupying a fraction of the overall spectrum. Note that an RB in the frequency domain also constitutes the subchannel over which data transmissions can be performed in the uplink and downlink directions. When performing spectrum allocation, a set of RBs are combined to form a RB group (RBG) which is then assigned to a user at each scheduling time interval. In this context, intra-cell interference is eliminated by grouping the RBs such that they are non-overlapping between different RBGs. To improve the SE performance of the network, each BS has access to all the RBs (full reuse) which are utilized while transmittingto the users within its cell. It is also feasible to perform power adaptation over the allocated RBGs based on the SINR measurements. In this case, higher data rate is achieved by allocating higher transmit power over the RBG when the channel conditions are favorable (high SINR), and vice versa in adverse channel conditions. A homogeneous network with full frequency reuse operation is shown in Fig. 2 where each macrocell BS assigns a set of RBGs to different users and, in the process, causes ICI at a user location in neighboring cells due to transmissions over the same RBG. Performing energy ef?cient resource management in multicell networks can be rather challenging when executed in a fully coordinated setting due to the dif?culty of acquiring at a central location the time varying CSI and the QoS requirements of all users in the network. The alternative approach is to enable either the macrocell BSs or the users to distributively determine the respective radio resources (i.e. power, subchannels) to be utilized during transmissions based on locally sensed (interference and user activity) information. For gaining insight into practical systems, a multiuser power control problem [15][16] applicable in homogeneous networks is modeled as a non-cooperative game in which each user k individually determines p? k , the transmit power that maximizes the respective EE. The users perceived utility level is modeled as a log function that factors in a pricing parameter in order to discourage the users from transmitting at high power as follows [16]: p? k = arg max log rk (pk ) ? log(pk ? pc )pk(1)where pc denotes power expended in the user equipment (UE) circuitry. The data rate achievable with pk is given by rk (pk ). Given the strictly quasi-concave structure of the utility function, it is shown that there exists a unique Nash Equilibrium (NE) from which no user has any incentive to change the power pro?le. An algorithm called the temporal iterative binary search (TIBS) is provided that determines the optimal power allocation based on the gradient search method to track the channel variation over time with reduced complexity. It is learnt that, in interference limited setting, transmitting beyond the energy ef?cient optimal power level provides insigni?cant gain for SE, but at the cost of incurring a high loss in EE. Furthermore, it is also revealed that the proposed power allocation scheme designed to maximize the EE also improves user throughput which, in essence, minimizes the adverse impact of SE-EE tradeoff. While [16] sheds some light into the dynamics of the SE-EE tradeoff, the severe impact of ICI when considering realistic conditions with multiple subchannels is not highlighted in this work. This is due to the simpli?ed model used in representing the single channel multicell homogeneous environment where all users are assumed to experience equal interference and channel gains. Extending on the non-cooperative game model, a joint subchannel and power allocation problem in the uplink direction is investigated in [17] where the players in this context are the individual BSs, each with the aim of maximizing the sum utility of the users in each cell. The utility function for the users is de?ned as the ratio of a QoS metric (representingRAO and FAPOJUWO: A SURVEY OF ENERGY EFFICIENT RESOURCE MANAGEMENT TECHNIQUES FOR MULTICELL CELLULAR NETWORKS157Fig. 2. Homogeneous cellular network consisting of Macrocell BSs with full frequency reuse. The user of interest k is served in the downlink by the associated macrocell BS i via RBG s1 . The remaining users associated to BS i are allocated RBGs s2 and s3 . The other macro BSs utilizing RBG s1 affect the throughput performance of user k due to inter-cell interferencethe level of satisfaction perceived by achieving a certain data rate) and the total amount of power allocated over the assigned subchannels. The performance assessment of the proposed suboptimal algorithms with and without a pricing factor (to obtain Pareto optimality) indicates that, by deducting the weighted sum of the transmit power from the utility function, it is possible to achieve higher EE due to the reduction realized in the ICI level. On the other hand, the resulting loss in the EE due to the UEs circuitry and signal processing is not captured because the associated power costs, independent of the transmit power, are not accounted for through the utility function in [17]. One of the key problems encountered when analyzing the uplink EE performance in a non-cooperative setting is the convergence of the best response resource allocation solution to a stable NE condition. In a strategic game formulation, each user identi?es a best response strategy out of a set of possible strategies based on the responses of the other players. Thus, guaranteeing an equilibrium point in the game is a relatively dif?cult task, more so when the degrees of freedom available through the strategy set is large. To overcome this issue, apotential game formulation is considered in [18] for an uplink multicell OFDMA system in which the BSs are assumed to be equipped with multiple antennas. By ensuring that the range of the potential function (used to represent the utility of the users) is bounded, it is shown that the non-cooperative subcarrier and power allocation game always converges to the NE condition. However, similar to [16], the authors do not account for the ?xed power costs attributed to the UEs circuitry when de?ning the utility function. As a result, its impact on the uniqueness and the stability of the NE is not analyzed in [18]. In the case of power allocation and scheduling in multicell OFDMA networks, four different power sharing policies that span both the spatial and time dimensions are considered in [19] to maximize the long term sum user EE. These policies translate into four different cell and network level constraints de?ned by either the time average or instantaneous power budgets of the BSs. Using stochastic gradient and greedy primal-dual techniques, the objective function of the optimization problem is incorporated with a virtual queue term that signi?es the penalty incurred in ensuring that the time average158IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 16, NO. 1, FIRST QUARTER 2014TABLE I S UMMARY OF E NERGY E FFICIENT R ESOURCE M ANAGEMENT T ECHNIQUES FOR H OMOGENEOUS C ELLULAR N ETWORKSGeneral Concept Noncooperative power control game (uplink) Noncooperative subchannel and power allocation game (uplink) Potential game formulation (uplink) Spatiotemporal power sharing Energy Ef?cient Scheduling for UE CSI Perfect CSI at Tx Computation Technique Gradient binary search Greedy algorithm (Hungarian)and iterative water?lling Asymptotic approximation Stochastic gradient and greedy primal dual Traverse along the boundary of feasible solution space Convergence Fast convergence to NE Ref.[15][16]Perfect CSI at TxConverges to NE (Pareto optimal)[17]Perfect CSI at Tx Perfect CSI at centralized location Not requiredFast convergence to NE Converges to near optimal solution (concave approximation) Converges rapidly to global optimal solution[18]the time span that is normally associated with the long-term duration is anywhere between tens of minutes to several hours spanning over the duration of a day [20]. This enables implementation of more large scale energy ef?cient resource management schemes that impact the performance of the entire network rather than at the individual cell level. In this context, the short-term load adaptive techniques must account for wireless channel ?uctuations (small scale fading) over the shorter duration while for long-term load adaptive techniques it is suf?cient to consider the average channel behavior by taking into account large scale fading (or average path loss). The resource management techniques applicable in the long-term include network partitioning [21] and switching off of BSs for prolonged periods [22] as compared to transitory approaches such as resource allocation and user handover which are more relevant in the shorter time scale [23]. 1) Long-Term Time Scale Solutions: One of the earliest insights into the long-term load adaptive techniques which dynamically vary the network coverage area was provided in [20]. It is argued that, due to the network being designed and dimensioned to cater for worst case load scenarios, keeping the BSs active at full coverage and capacity modes results in excessive power consumption. Based on this, the cell zooming concept was introduced to allow for the cell sizes to change in accordance to the load level. The optimal operation of BSs with cell zooming requires balancing of the tradeoff between two con?icting objectives, namely the minimization of the energy consumption and reduction of the blocking probability (i.e. probability that the users load requirements are not satis?ed) [24]. Therefore, it is implicitly inferred that the strategy that enables both objectives to be ful?lled involves associating the users with a selected number of BSs with high degree of utilization such that the more lightly loaded BSs can be powered-off. On the other hand, as part of the validation performed in [24] when implementing cellzooming, only highly idealized hexagonal cell grid structure is considered. In this case, it is feasible to recon?gure the network by evenly switching off the BSs surrounding each cell during low load conditions while satisfying the blocking probability. In practical networks where there exists a certain degree of irregularity in the inter-BS spacing, the effectiveness of the blocking probability constrained cell zooming technique remains unclear. In order to account for the case where users can have different QoS objectives in terms of minimum data rate requirements, the load can be formulated using a ?ow-level model in the form of an M/M/N queue in which the users are grouped into a set of different classes, each having its own bandwidth requirement [25]. In the load adaptive strategy proposed in [25], the network is ?rst divided into dense and sparse regions based on the corresponding load factors. The radius of sparser zones is then increased to provide higher coverage while satisfying the blocking probability requirements of each user class in that area. This is to ensure that the excess bandwidth in the sparser regions is better utilized by spreading the system resources over a larger area while reducing the number of BSs required. Note that the traf?c model considered in [25] which characterizes the user classes based on different[19][35]power constraint is satis?ed. Given the relative intractability of the problem due to the non-convex formulation, the networkwide problem is decomposed to be solved distributively at each cell. The main observation made in the simulations of a homogenous macrocell network is that, in an interferencelimited environment, the application of the joint space-time power sharing policy ensures that the potential to save power usage is greater than the potential to improve the average user throughput. Although the ?ndings are able to highlight how interference management and space-time power control policies are intertwined in a multicell setting, there are a number of implementation related issues that were not emphasized in [19]. Issues related to the feasibility of slot by slot exchange of load related information via the backhaul links, and the impact of asymmetric load conditions between the BSs on the robustness of the proposed techniques require further consideration. A summary of the energy ef?cient resource management techniques discussed in this paper applicable for homogeneous cellular networks is given in Table I. A. Load Adaptive Techniques When analyzing the various dynamic operational techniques for the purpose of enhancing the network EE, it is vital to differentiate between the time scale at which the different approaches can be implemented based on the ?uctuating load and wireless channel conditions. The load in this context refers to either the user density in the network or the aggregate data rate demanded at a given time instance. In general, the rate of change in the load intensity can be categorized as either long- or short-term based on the time duration considered for which the load level remains relatively stable with a steady mean. While the short-term load is evaluated at smaller time granularity on the order of few time slots (i.e. in the range of seconds or minutes),RAO and FAPOJUWO: A SURVEY OF ENERGY EFFICIENT RESOURCE MANAGEMENT TECHNIQUES FOR MULTICELL CELLULAR NETWORKS159call arrival rates is more appropriate for voice dominated traf?c. In contrast, for assessing the performance of the energy ef?cient resource management schemes with varied user QoS requirements, it is necessary to incorporate more advanced ?ow level models such as alpha-optimal discussed in [26]. Simulation results presented in [25] reveal that it is feasible to achieve up to 25% in energy savings at the network level by adapting the coverage area of BSs in accordance to load variation while meeting the blocking probability requirements in all zones. While the dynamic operation of the BSs minimizes the load dependent power expenditure to a certain extent, due to the high amount of ?xed power costs incurred for keeping the BSs continuously powered on, higher power savings can be achieved only when some BSs are completely powered off. On this basis, the optimization problem to minimize the total energy consumed over the operational duration through optimal BS switching while adhering to a time dependent blocking probability constraint, Prbloc req can be formulated as follows [22]: minAs t+Ds.t.s∈S t bloc P rs (t)Es As (t) dt(2a) (2b)bloc ≤ P rreq , ?b, tbloc where S denotes the set of BSs in the network, P rs (t) denotes the blocking probability of BS s at time t while As (t) represents the on/off activity function of BS s which, in this case, assumes a binary value when taken at discrete time instants over the time duration [t,t+D], where D denotes the monitoring interval, typically one day. The energy consumed by a BS per unit time is given by ES . While the load level varies temporally, the variation over the spatial dimension is not taken into account in [22]. The analysis of the energy consumption model for the given switching strategy in [22] indicates that the higher the BS density and the variation in the load pro?le, the higher is the potential for energy savings. For the feasibility of the BS switching scheme, it must be ensured that the users in the network are redistributed to a selected few active BSs such that more BSs can be poweredoff to save network energy consumption. Based on this requirement, the average energy consumption of the network is minimized while meeting the user-to-BS association (each user associates with only one BS) and bandwidth constraints as follows [27]: ? ? ??1 when x is greater than zero and zero otherwise, captures the presence of users in the cell. The function m(t) indicates the number of users at time t. The parameters ρi (t) and rsi speci?ed in (Eq. (3c)) correspond to the data rate requirement of user i and the SE achieved if the user is served by the BS s, respectively. The bandwidth constraint which limits the number of users that can be associated with each BS is denoted max . In [27], the optimization problem is solved using a by Bs greedy approach that can be implemented centrally under the assumption of having global information on the user channel states and load requirements. However, when the optimization problem is solved distributively where each BS determines the user associations based on locally observed load conditions, higher outage events ensue where a certain fraction of the users is not supported with the requested data rate. Although centralized approaches enable globally optimized performance to be achieved in network EE, the high amount of overhead signaling (e.g. CSI feedback, synchronization) and computational costs involved can have an adverse impact on the effectiveness of such schemes. In light of this, selforganized approaches are necessary so that the BSs can autonomously adapt the individual parameters based on locally sensed load conditions. A mechanism termed energy partitioning in which the BSs can distributively recon?gure during off-peak times is discussed in [21],[28]. In this context, the partition constitutes the case where the overall network load of a larger set of BSs is redistributed to a smaller powered-on subset based on the load related knowledge of the powered-off BSs. The distributed algorithm proposed is essentially based on BS pairing procedure where the BSs are grouped within the neighboring set that results in maximal energy saving by powering off while accounting for the load and coverage requirements. It is highlighted that there could be instances that some BS pairs are unable to support the combined load during busy periods. In these cases, a recon?guration procedure is to be executed as part of the distributed algorithm for providing suf?cient load balancing capability. However, the overall energy savings achieved with the recon?guration procedure for the distributed technique is marginal in comparison to the centralized approach. It is only when the load variation is limited to selected localized areas that the distributed approach is expected to perform better in terms of energy savings. A summary of the long-term time scale load adaptive techniques discussed in this paper is provided in Table II. 2) Short-Term Time Scale Solutions: While large scale load adaptive techniques allows to completely power off the BSs for prolonged periods, it would also be feasible to opportunistically adapt the BS functionality at much lower time scales such that additional energy savings can be obtained. With regards to the short-term timedomain techniques which are currently considered for Third Generation Partnership Project Long Term Evolution (3GPP LTE) standard revisions, it is indicated in [23] that by minimizing the transmission of the reference and control signals when there is no downlink traf?c, it is possible to periodically shut down the BS power ampli?er (PA) and reduce the power drain wastage. One of the approachesmin ?xsiSPs sgn ?m(t) ixsi (t)??(3a)s=1 Ss.t.s=1 m(t)xsi (t), ?i = 1, ..., m(t) ρi (t) max ≤ Bs , ?s = 1, ..., S rsi(3b)xsi (t)i(3c) (3d)xsi ∈ {0, 1}, ?s = 1, ..., S, ?i = 1, ..., m(t)where xsi is the binary variable indicating the association of user i to BS s and Ps denotes the power consumed by BS s. The signum of x, denoted by sgn(x), gives a value of160IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 16, NO. 1, FIRST QUARTER 2014TABLE II S UMMARY OF LONG - TERM TIME SCALE LOAD ADAPTIVE TECHNIQUES FOR H OMOGENEOUS AND H ETEROGENEOUS C ELLULAR N ETWORKS Focus of Study Cell zooming Control Mechanism Dynamic adaptation of coverage area and sleep mode BS switching and load balancing Load Pro?le Bandwidth utilization intensity measure User Intensity and bandwidth demand Sinusoidal traf?c pro?le QoS User blocking rate User block and drop probabilities Time dependent user blocking probability User blocking rate User data rate requirement Call drop probability and time to complete handover Ref.[20][23]Autonomous BS adaptation Dynamic network con?guration with QoS Cell size adaptation with QoS User redistribution to active BSs Progressive cell size adaptation[21][27]BS switching[22]Dynamic adaptation of BS coverage area Recon?guration of user-BS association Cell wilting and blossomingMulticlass traf?c intensity User intensity[24][26]User intensity[55]considered in [23] is to power off the PA in time slots that contain signal-free symbols which can enable up to 47% reduction in the PA operational time. The other techniques discussed are to reduce the reference signal transmission (by con?guring the transmission frame structure) and using the discontinuous transmission mode (DTX) that can lower PA operational time by 28% and 7.1%, respectively. Although it is practical to operate the PA dynamically in accordance to the short-term load variation to gain higher energy savings, the impact of additional power consumed during the on-off transient periods on the overall performance requires further study. Moreover, highly agile operation for the PA while maintaining high ef?ciency level generally translates into more expensive hardware for the BSs [29], which must also be considered as part of deployment ef?ciency. In the frequency domain, one effective small scale load tracking approach would be to activate the frequency carriers dynamically such that the energy consumed at the BS is minimized while being able to meet the QoS requirements of the users. In this case, while maintaining a ?xed level of power spectral density, it would be feasible to lower the transmit power by increasing the bandwidth utilized. The carrier aggregation technique discussed in [23] associates a group of carriers to a lower number of PAs such that the other underutilized PAs without any scheduled carriers can be powered off. The energy saving techniques that can be implemented in the spatial domain are the reduction of antenna elements, dynamic con?guration of the cells and inter-cell traf?c of?oading [23]. These techniques which are applicable during low load conditions can power off the PAs associated to the unused antenna elements and cells which are vacant. Higher reduction in power consumption is alsoshown in [23] to be obtained by disabling a certain number of antennas while minimizing the transmission of reference signals when the load is low. The comparison of the different solutions applied in low load conditions indicates that the hybrid approach that combines various techniques in the time, frequency and spatial domains results in the highest energy savings of almost 50% than those of the standalone techniques. In general, the performance gain achieved when implementing short-term load adaptive techniques across multiple domains (time, frequency and spatial) comes at the expense of high overhead signaling and signal processing costs when determining optimal BS con?guration parameters. It is therefore necessary to take a holistic approach in addressing the design challenges when evaluating the overall effectiveness of the short-term time-scale techniques. When considering the mechanics of the BS switching scheme, it is imperative to ensure that the user drop rate is not severely affected during the switch-off duration while transferring the active traf?c load between the BSs. To enable a seamless transfer process, a scheme proposed in [30] calls for the switching-off of the BSs to be performed progressively by reducing the transmit power over discrete steps. In the meantime, adequate amount of time must be provided between the switching intervals to ensure that the user transfer rate is within the capacity limit of the absorbing BSs. It is noted that, in general, the delay incurred during the switching-off process in transferring the users is fairly inconsequential as compared to the time at which the BS will remain powered-off. Therefore, the impact on the resulting energy saving is negligible. The progressive switch-off scheme is shown to be capable of signi?cantly lowering the user drop rate as compared to an abrupt switch-off scheme used otherwise. While the study performed in [30] provides the general awareness into performing load balancing at short time scales in a static setting, the ability to transition the switched off BSs to active modes when there is an increment in the load level or when user mobility is involved has not been discussed. In addition, the possibility to maintain the same QoS for the remaining and transferred users using the limited resources at the active BSs is also not addressed in [30]. 3) UE Battery Aware Resource Management Techniques: Given the sluggish advancements made in improving the battery capacity [31], it is critical to be more comprehensive in considering the multiple parameters related to the UE s battery (e.g. state of charge and state of health [32]), in addition to QoS and load conditions when formulating the resource management schemes for multicell networks. A set of UE centric energy management schemes that take into account differences in QoS supported in multi-radio access technology (RAT) environments has been recently surveyed in [33]. Alternatively, when analyzing the impact of BS switching on the UE, it is important to recognize that transmission distance in the uplink direction increases while reaching the closest active BS when some of the existing BSs are powered off. One way to characterize the additional uplink power costs is to assume a worst case scenario in which the BSs in the network are powered off randomly and each UE determinesRAO and FAPOJUWO: A SURVEY OF ENERGY EFFICIENT RESOURCE MANAGEMENT TECHNIQUES FOR MULTICELL CELLULAR NETWORKS161if it can associate with any active BS while adhering to the UE s maximum transmit power and outage constraints. Based on this model, closed form expressions pertaining to the average uplink power has been derived in [34] as a function of the fraction of active powered-on BSs in the network. It is found that, in interference free environments, the uplink power increases to a peak value as the number of active BSs increases prior to decreasing and saturating at a ?xed level. The uplink power is also determined when ICI is factored in, in which case the power increases monotonically with respect to the number of active BSs. However, the analysis and validation for the uplink power with interference is performed only for a one-dimensional network model where the BSs and UEs are located randomly along a line. It would be interesting to investigate the overall power expended by the UE when considering more realistic scenarios accounting for shadowing and fading channel conditions along with the interference in a general two-dimensional network setting. For transmissions in the downlink direction, the energy expended during reception can be minimized if the UE is scheduled to receive data over a certain limited number of time slots and frequency bands while powering off the internal circuitry in the remaining slots. Following this approach, a multiuser resource allocation problem is studied in [35] to minimize the total circuit power consumption for all users over a ?nite number of time slots. Although the differences in the UE battery capacities are accounted for in the problem formulation, the QoS requirements which, in general, can be different from user to user, are not considered in [35]. Based on the nonlinear integer structure of the problem, a heuristic algorithm is proposed that is able to determine close to optimal solution that exists at the boundary of the feasible search space. For a cell consisting of 300 users with 100 frequency bands, it has been numerically validated that the solution can be found within 15 iterations and that the UE energy expenditure is reduced by more than 60% as compared to the conventional round robin scheduling scheme [14]. In the case when the users are mobile while the BSs are transitioning into low power operational modes, the energy consumed in the UE can vary in both the spatial and time dimensions over the short-term time duration. In addition to expending higher energy to reach the nearest BS, satisfying the delay requirement will be a challenge in this scenario, more so when transmitting delay sensitive traf?c in the uplink direction. Based on this insight, a store-carry and forward relaying approach is proposed in [36] in which the UEs transmit to the BSs through other intermediary nodes (i.e. other UEs or vehicles) that are mobile. The ?ndings derived from simulations indicate that when minimal delay is of importance, a direct transmission from user to BS is preferred, translating into the case where more BSs are operated in the active mode (full capacity). On the other hand, by relaxing the delay requirement, multihop transmission using mobile relays is favored while resulting in higher energy savings through BSs operating in low power modes. When the delay tolerances are high, the received data is stored for longer periods in the mobile nodes, allowing it to travel further distances until the closest active BS is reached. It is important to note that the analysis performed in [36] is based on theassumption that the delay requirement for all users is similar. In practical scenarios, it is necessary to investigate the potential for energy savings achieved in the short-term time scale through the mobile relaying scheme when users have diverse QoS requirements and located randomly in the network. It is expected that the computational complexity of the proposed scheme increases signi?cantly when determining the optimal scheduling links (between users, mobile relays and BSs) over space-time dimensions as the number of transmission nodes increases. B. Energy Ef?cient Mobility Management User Mobility Management (MM) presents another set of new challenges as well as opportunities to minimize the energy consumption in cellular networks both at the RAN level as well as at the UE. In this context, there exist several avenues for realizing higher energy savings when viewed from the perspective of resource management in multicell cellular networks. MM, in general, can be decomposed into two procedures, namely location management (LM) and handover (HO) management (HM). The LM procedure, when performed in the UE idle mode (i.e. when the UE is not receiving or transmitting data) enables the cell selection mechanism to be implemented where the UE is associated to a serving BS and its location in the network is tracked. The HO procedure, applicable only in the UE active mode, consists of 3 phases as follows [37]: i) serving cell monitoring and evaluation, ii) cell search and measurement reporting and iii) handover decision and execution. In this case, the HM procedure can trigger a HO event to seamlessly transfer the user from a serving to a target BS during an ongoing call. Note that, to ensure service continuity in the presence of mobility, the available radio resources must be judiciously allocated for network scanning, measurements reporting and UE tracking such that the overall energy expended by the network is minimized. To improve the EE of HO decision making process, it would be necessary for the network to be context aware [38] taking into account multiple parameters including user preferences, user-BS location information, UE battery capacity, monetary costs, and the BS load conditions. Several existing standards, such as IEEE 802.21 (Media Independent Handover) [39] and IEEE P], can be incorporated to provide the necessary framework to facilitate intra and inter-RAT HO and to support interworking through optimal radio resource sharing. On this basis, an energy ef?cient cost-function driven HO procedure that applies the media-independent information service (MIIS) speci?cation of 802.21 is investigated in [41]. In this case, the MIIS de?nes the network control messages and the retrieval of network speci?c information (from a remote server) that is invaluable in assisting with the HO process. To support context awareness, the cost function is speci?ed as a weighted sum of multiple factors such as bandwidth, delay and power consumption. The scheme proposed in [41] simpli?es the procedure associated with network discovery (scanning) and network association by using the MIIS, resulting in lower amount of overhead and improved UE battery life performance. The energy consumption measurements made on a multimode UE in a testbed setting using the 802.21162IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 16, NO. 1, FIRST QUARTER 2014were recently reported in [42]. On average, it is revealed that network scanning drains the maximum energy amounting to 8.6 J, while location updating and signalling consume about 1.8 J and 10 mJ, respectively. These values show a reduction of more than 40% in overall energy consumption when compared to a network that does not use the services provided through 802.21. Load balancing in the short-term time scale provides another important avenue to improve the network EE in which the users are associated to a set of BSs based on the available resources and the QoS requirements. While load balancing is generally performed at the network level (network controlled), it is possible for the UEs to assist in executing the HO decisions. On this basis, a UE assisted network selection scheme that balances the load between the BSs is proposed in [43]. By factoring in the UE s battery capacity, mobility parameters (user distance to BS and velocity) as well as the load conditions, a Linear Programming (LP) problem is formulated to determine the best BS that maximizes a weighted bene?t to cost ratio. Performance evaluation of the proposed scheme shows that the number of HOs required when both the UE and network cooperate while performing load balancing are signi?cantly lower than that of the conventional network controlled approach. Moreover, the energy consumed through the UE assisted approach [43] is also substantially lower as compared to the existing policy that associates the users to BSs providing maximum bandwidth. However, the gain realized in energy savings in [43] is achieved only at the expense of a certain loss in throughput performance due to the tradeoff relationship between the two metrics. The authors of [44] implemented a HO triggering algorithm to increase the energy savings in the RAN in addition to the UE, considering the power consumed for overhead signalling as well as the power drained in the BS hardware. The simulation results indicate that up to 46% improvement in EE can be achieved during low load conditions when only the transmit power is evaluated. By incorporating the hardware power, the EE gain reduces to 13%. On the other hand, in high load scenarios, the EE gains related to transmit power and total power (transmit plus hardware power) drops to insigni?cant levels of 5% and 1%, respectively, indicating the limitation in the energy aware MM scheme. III. E NERGY E FFICIENT R ESOURCE M ANAGEMENT I N H ETEROGENEOUS C ELLULAR N ETWORKS Heterogeneous cellular networks (HetNets) consisting of multiple tiers of small scale micro, pico and femtocell BSs and another tier of large scale macrocell BSs are primarily deployed to enhance the capacity of the network, especially in concentrated areas with high traf?c demand (hotspots). While the downlink data rates achievable in the small cells are higher due to the lower BS-to-user path loss, increasing the small cell density results in high interference over the shared channels and diminished SE and EE performance. A typical HetNet deployment, shown in Fig 3, highlights the severity of ICI in a full frequency reuse operation where all BSs (macro, pico and femto) have access to all the spectral resources that can be allocated to the respective associated users at any given time. The deployment of the HetNets also comes at the expenseof additional power consumption in the form of the ?xed BS hardware related expenditure. In general, for the sustainability of the HetNets, the improvement achieved in the capacity must justify the increase in the energy expenditure at the BS level (macro, pico and femto) in order to obtain a net gain over the EE. In [45], the HetNet deployment studied consists of macrocell BSs located in a hexagonal grid structure while the small scale microcells are randomly placed within each macrocell. Given this model, it was determined in [45] that, for different load conditions, there exists an optimal number of microcells for a reference cell area that maximizes the EE. Beyond the optimal value, further addition of micro BSs does not provide suf?cient gain in the network capacity to justify the increase in power consumption. For OFDMA based HetNets, the authors of [46] evaluated the EE performance achieved in the downlink for a network consisting of a number of microcells overlaying a macrocell at the cell edge. With the assumption of perfect CSI, the subchannels allocated by any BS to the users are set to be mutually exclusive and non-interfering. The resource allocation problem is then formulated to maximize the sum EE of all BSs subject to the individual BS total power and subchannel exclusivity constraints as follows [46]: max n n bS s=1 S N S K n cn k,s xk,sxk,s ,pk,s NPs(4a)n=1 s=1 k=1s.t.n=1 s=1 K S k=1 s=1n pn k,s xk,s ≤ Pk s = 1, ..., S(4b) (4c) (4d)xn k,s ≤ 1, n = 1, ..., N xn k,s ∈ {0, 1}where n, s and k respectively denote the subchannel, BS and user indices. Each cell is assumed to consist of K users. In Eq. (4.0), cn k,s is the index of the optimization subchannel n assigned to user k by BS s, xn k,s is a variable indicating whether or not subchannel n is assigned to user k by BS s, and b is the bandwidth of a subchannel. The variables here are the binary subchannel selection xn k,s and the transmit power pn . By assuming that the total power k,s S consumed is ?xed (where s=1 Ps is a constant), the problem is reformulated to maximize the sum data rate of all BSs. The optimal solution then is obtained using the Lagrange dual decomposition (LDD) method. A suboptimal approach with reduced complexity in which the subchannel and power allocation solution is determined sequentially is also proposed in [46]. Note that, since the maximum transmit power available at all BSs is not adapted during each resource scheduling interval, the true potential EE of the network is not realized in [46]. Furthermore, the resource allocation problem does not account for the ?xed hardware related power costs incurred at the BSs while performing downlink transmissions, and thus provides only a limited assessment of the overall network performance. The deployment of femtocells has gained much attention recently mainly due to the capability to improve the SE and coverage in indoor environments while incurring much lowerRAO and FAPOJUWO: A SURVEY OF ENERGY EFFICIENT RESOURCE MANAGEMENT TECHNIQUES FOR MULTICELL CELLULAR NETWORKS163Fig. 3. 3-Tier Heterogeneous cellular network consisting of Macro, Pico and Femtocell BSs with shared spectrum operation. The user of interest k is associated to picocell BS i and is served in the downlink over RBG s1 . The other macro, pico and femtocell BSs utilizing RBG s1 in the same scheduling time interval interfere with the downlink transmission intended for user kcosts as compared to large scale BSs. However, the primary concern is on optimizing the transmit power of the femtocells over the shared spectral resources such that the occurrence of ICI during uplink and downlink transmissions is mitigated while simultaneously maximizing the EE performance. The results of numerical simulations provided in [47] of a network consisting of a macrocell overlaid with randomly deployed femtocells reveal that, as the number of femtocells increases, the average transmit power of each femtocell can be reduced to improve the EE. For instance, in the presence of 20 femtocells, it is shown that 80% of the femtocells transmit with 80mW as compared to the maximum power level of 125mW when the transmit power is tuned based on the received power sensed from the macrocell over the shared subchannels. When comparing the SE achieved for the different number of femtocells deployed per macrocell, it is found that the SE reduces from 6 b/s/Hz when all femtocells transmit with the highest allowed power to about 5 b/s/Hz when power control mechanism is invoked [47]. In this case, it is clear that power control trades off a relatively low loss in the SE for a much higher EE performance. However, the EE performance might be lower in realistic scenarios when the additional energy costs for sensing the activity level over multiple subchannels as well as overhead signaling costs incurred over the backhaul interface are explicitly accounted for in the calculations. In [48], spectrum partitioning is performed between the macrocell and all femtocells as a solution to minimize the occurrence of inter-tier interference. The total spectrum allocatedto the femtocells is further divided based on an access ratio. On this premise, a spectrum allocation strategy is proposed to optimize the access ratio and the number of femtocells in order to maximize the network EE subject to a constraint that balances the achieved throughput in the downlink direction for both tiers. But this strategy comes at the expense of forgoing the maximum potential SE and EE achievable from high spectrum reuse as a result of partitioning the spectrum between the tiers. The alternative approach for spectrum management between multiple tiers would be to coordinate the scheduling of the shared RBs over the frequency and time dimensions. In this manner, the transmissions over the shared RBs can be controlled while meeting the target QoS and lowering the downlink power. A resource management procedure outlined in [49] enables the femtocells to sense the surrounding interference level from the macrocell BS and other femtocells and report to a centralized unit (CU) the channel measurement information and the QoS requests of the users being served. Based on this information, the CU computes the schedule for the users that can be supported over the available RBs. For each of the allocated RB, the transmission power is reduced while satisfying a certain received signal-to-interference-plus-noise ratio (SINR) threshold and a packet error rate requirement. However, similar to [47], the study does not account for the overhead signaling costs incurred for CSI measurement and messaging to the CU, both of which can constitute a signi?cant portion of the overall energy consumed especially in small164IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 16, NO. 1, FIRST QUARTER 2014TABLE III S UMMARY OF E NERGY E FFICIENT R ESOURCE M ANAGEMENT T ECHNIQUES FOR H ETEROGENEOUS C ELLULAR N ETWORKSGeneral Concept Energy Ef?cient Resource Allocation in HetNets Intra-tier interference mitigation to improve EE performance Joint BS operation and load distribution Optimal BS switching with resource sharing Energy Ef?cient Admission Control Energy Ef?cient Resource Allocation with limited backhaul capacity CSI UE estimates CSI over all subchannels UE estimates CSI over all subchannels Not required Computation Technique LDD (optimal) and suboptimal subchannel and power allocation Subchannel scheduling and MSC adaptation Greedy Algorithm (Hungarian) Convergence Faster convergence with suboptimal method with 20% performance degradation The convergence rate to suboptimal solution is moderate Fast convergence to suboptimal solution Slow convergence rate that scales with number of BSs, users, channels Converges to optimal admission control policy in polynomial time Converges to optimal power allocation at superlinear rate Ref.[46][49][59]Not requiredGreedy algorithm[60]Not requiredLP for optimal admission controlwhen the load intensity decreases. It would also be feasible to increase the density of the smaller cells at low deployment costs in order to provide suf?cient redundancy to the network with overlapping coverage. During low load conditions, these BSs can then be powered-off while transferring the residual load to either the macro BSs or other heterogeneous cells. On the contrary, having to deal with diverse BS attributes can lead to additional complexity in certain tasks related to optimal load distribution and adaptation of the coverage parameters. Moreover, the presence of femtocells which introduce a set of other issues concerning user access policies (i.e. open, closed, hybrid) can signi?cantly impact the performance of neighboring and overlapping cells which are transitioning into different load adaptive modes. In this respect, if all femtocells support open access policy, it would be feasible to optimally of?oad the users from the larger cells that are transitioning to sleep modes while satisfying users QoS requirements [51]. On the other hand, when active femtocells support different user access policies, it becomes formidable to perform optimal of?oading and load redistribution while achieving higher EE performance by powering off more BSs [52]. 1) Long-Term Time Scale Solutions: Similar to homogeneous networks, the long-term time scale solutions for HetNets involve redistribution of users to minimal number of BSs and powering off of lightly loaded BSs of different tiers when the load level is low for prolonged periods. One of the earliest insights into the bene?ts of utilizing small cells for the purpose of lowering the network power consumption through dynamic operation was provided in [53]. Using a newly introduced energy consumption gain (ECG) metric [53] denoting the ratio of the network energy consumption when all cells have maximum coverage radius to that consumed with minimum radii, the authors evaluated the network performance when a fraction of the lightly loaded BSs are powered off to save energy. It is indicated that while the energy per bit consumption ratio (in Joules per bit) decreases exponentially by powering off a number of BSs, the ECG increases linearly with the number of small size cells. The authors of [53] however did not consider the ?xed power costs attributed to the BS s PA and other components (e.g. cooling, DSP circuitry) in the de?nition of ECG, which can change the ECG scaling behavior when comparing the overall network energy consumption with respect to the number of small cells deployed. In the context of green HetNet operation, a straightforward approach would be to dynamically power down the small size cells at low load conditions while the macro BSs expand their cell radii to close the coverage gaps. In addition, several other mechanisms are proposed in [54], which include the timed sleep mode, user location prediction and reverse channel sensing that enables for the adaptation of the BSs to lower their energy costs. Note that the load in this case is de?ned only through the user density at a given time, without considering the aggregate throughput demanded by all users, which is more relevant in OFDMA based networks. It is shown via simulations that up to 40% savings in energy consumption can be realized through the proposed dynamic[61]Perfect CSI at centralized locationDinkelbach method[84]cell networks [50]. The numerical evaluation of the proposed technique in a LTE deployment scenario is shown to provide better performance in terms of both data rate and EE of the femtocell network for various QoS parameters than that of the conventional decentralized method in which each femtocell allocates the resources in a purely distributive manner. A summary of the resource management techniques applicable for HetNets is provided in Table III. A. Load Adaptive Techniques In contrast to homogenous networks, the use of HetNets to adaptively vary the BS functionality in accordance to the ?uctuating load conditions brings about certain unique advantages as follows. The smaller scale BSs are able to track the load variation with ?ner granularity while incurring lower circuitry related energy expenditure. Although the PA ef?ciency in the smaller scale BSs are generally lower than that of the macro BSs [3], the predominance of the load dependent RF front end and the signal processing components allow for the heterogeneous BSs to be more ?exible in adjusting the operation level. This also facilitates the optimal adaptation of the available radio resources by having a greater leverage over the varying load conditions in the process of improving the overall network EE. In addition, it would be more practical to deploy the smaller cells at locations that have occasionally high data rate demand so that the BSs can be easily transitioned to low power modesRAO and FAPOJUWO: A SURVEY OF ENERGY EFFICIENT RESOURCE MANAGEMENT TECHNIQUES FOR MULTICELL CELLULAR NETWORKS165mechanisms as compared to the conventional always-on static approach. While the sleep mode techniques are primarily designed to minimize the BS power costs, it is important to note that the various approaches must ensure that the service quality in the coverage area is not compromised at all times. Emphasizing on this concern, an enhanced sleep mode mechanism that can be implemented through the coordination of multiple surrounding BSs is provided in [55]. A triggering phase for the sleep mode is described by a procedure in which a particular BS that senses reduced traf?c load over a certain period enters into the low power mode. This is achieved by gradually reducing the