Computer Networks 57 (2013) 1518–1528

Contents lists available at SciVerse ScienceDirect

Computer Networks journal homepage: www.elsevier.com/locate/comnet

Green framework for future heterogeneous wireless networks Rajarshi Mahapatra, Antonio De Domenico ⇑, Rohit Gupta, Emilio Calvanese Strinati CEA-LETI, Minatec, 17 rue des Martyrs, 38000 Grenoble, France

a r t i c l e

i n f o

Article history: Received 26 June 2012 Received in revised form 24 November 2012 Accepted 4 February 2013 Available online 10 February 2013 Keywords: Heterogeneous wireless network Green communications Cognitive radio Energy-cognitive cycle

a b s t r a c t Energy-efficient communication has sparked tremendous interest in recent years as one of the main design goals of future wireless Heterogeneous Networks (HetNets). This has resulted in paradigm shift of current operation from data oriented to energy-efficient oriented networks. In this paper, we propose a framework for green communications in wireless HetNets. This framework is cognitive in holistic sense and aims at improving energy efficiency of the whole system, not just one isolated part of the network. In particular, we propose a cyclic approach, named as energy-cognitive cycle, which extends the classic cognitive cycle and enables dynamic selection of different available strategies for reducing the energy consumption in the network while satisfying the quality of service constraints. Ó 2013 Elsevier B.V. All rights reserved.

1. Introduction Wireless networks have seen tremendous growth to support Quality of Service (QoS) requirements of diverse applications with increasing number of users. In order to cope with these requirements, various wireless standards have been proposed, ranging from technologies characterized by networks with few User Equipments (UEs) and small coverage (e.g., bluetooth) to technologies with large coverage and hundreds of users (such as cellular networks). These technologies are diverse in nature and coexist in the same environment to provide ubiquitous network coverage. UEs can connect to any available wireless technologies depending on their requirements. With the increased drive from the industry for a fully connected network, there is a need to design the wireless Heterogeneous Network (HetNet) architecture to interconnect these different wireless technologies. These interworking mechanisms are of prime importance to achieve ubiquitous access and seamless mobility in HetNet environment [1,2].

⇑ Corresponding author. Tel.: +33 648169246. E-mail addresses: [email protected] (R. Mahapatra), [email protected] (A. De Domenico), [email protected] (R. Gupta), [email protected] (E. Calvanese Strinati). 1389-1286/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.comnet.2013.02.007

A typical HetNet scenario consists of several cellular technologies, like 3GPP Long Term Evolution (LTE) and WiMAX, operating in licensed spectrum and other shortrange wireless systems within unlicensed spectrum (such as WLAN and Bluetooth networks) [3]. This over development of wireless technologies using static resource allocation techniques resulted in spectrum scarcity and also low Spectrum Efficiency (SE). In order to cope with this problem, Cognitive Radio (CR) based algorithms have been proposed to increase spectrum usage by opportunistically utilizing spectrum holes [4]. Moreover, the dense deployment of Access Points (APs) required to satisfy the ever-growing requirements of data rate and coverage has resulted in the increase of wireless network’s energy consumption with an impact on the global carbon dioxide (CO2) emissions and has also imposed more challenging operational cost for the operators. These developments have led to tremendous interest from the industry to reduce the power consumption at system level [5]. In such perspective, the Greentouch consortium [6] and funded projects like EARTH [7] and Mobile VCE [8] focus on achieving the infrastructure energy savings of wireless networks at the system level. The main goal of these projects is to develop innovative algorithms and technologies for green operation of wireless networks. However, these pro-

R. Mahapatra et al. / Computer Networks 57 (2013) 1518–1528

jects have solely investigated the optimization of homogeneous wireless systems. Ashraf et al. investigated energy saving procedures that allow APs to dynamically deactivate/activate transmissions functionalities according to UEs presence/absence in its coverage area [9]. Frenger et al. proposed cell DTX [10] to allow cellular APs to switch off radio in subframes where there are no user data transmissions. Cell zooming enables APs to reduce the energy consumption by adjusting the cell size according to the traffic load [11]. Similarly, Alouf et al. [11] investigated control and optimization of energy consumption in WiMAX [12], WLAN [13], and mesh networks [14], respectively. These green engineering solutions showed that adapting the access and transmission parameters of APs to the network scenarios can have a large impact on improving Energy Efficiency (EE) of wireless networks. Nevertheless, these energy saving mechanisms are based on fundamental green trade-offs [15], which essentially focus on the optimization of homogeneous scenarios. Nowadays, due to the diffusion of mobile terminals that are able to access to different Radio Access Technologies (RATs), HetNets are gaining popularity. However, to introduce energy saving in a holistic way, HetNets requires a generic framework that enables intelligent and dynamic selection of different available strategies at APs. Moreover, several researchers have also pointed out that the design of energy saving mechanisms are comparatively easier for homogeneous environment than heterogeneous environment [16,17]. This is mainly due to conflicting design goals of different RATs. EE objective of one particular RAT can have detrimental effects on the other co-existing RATs within the same radio and geographical environment. The co-existence of different technologies may also produce interference, which further complicates the study to understand the system level energy consumption [18]. In order to deal with these challenges our proposal is twofold: 1. We introduce a novel green framework to improve the EE taking into account power consumption of all entities of the HetNet. This framework is an evolution of the classic cognitive cycle [19], which describes the interactions of a smart radio with its environment to improve the spectrum usage and enhance the coexistence in the same bandwidth of uncoordinated networks located in a given geographical area. We propose to extend the cognitive cycle and investigate its implementation in future HetNets, with the main focus on limiting the energy consumption while satisfying QoS constraints. Moreover, to deal with this heterogeneous wireless environment, we propose a novel architectural design, where cognitive intelligence do not reside only at APs but also at UEs. This shared intelligence helps to select the best communication parameters to improve the network EE. 2. We design an energy-efficient architecture, which exploits a generic interface amongst different HetNet technologies. This solution avoids that adaptive mechanisms, implemented in a given network, negatively

1519

affect performance of neighbouring networks in the same radio environment. This interface is designed to solely transfer energy efficiency related signalling in a local region amongst HetNets. This approach aims at limiting excessive load on the backhaul and enabling fast and reliable system adaptation. The rest of the paper is organized as follows. Section 2 introduces the proposed EE framework for HetNets. Section 3 describes the energy-cognitive cycle. Sections 4 and 5 details the different functionalities of the proposed energy-cognitive cycle. Section 6 proposes a network management scheme that exploits the proposed EE framework of in the macrocell/femtocell environment. Eventually, Section 7 concludes the paper. 2. Energy efficient framework This section introduces our proposed framework for EEaware wireless HetNets. Such framework helps the system to operate in an EE way, while considering network objectives and QoS constraints. As shown in Fig. 1, next generation networks are composed of different types of cells with multiple RATs. In HetNets, UEs can connect to any available neighbouring APs, which can operate with different RATs at a given point of time. The selection of EE operating point is a non-trivial task in such a scenario. To achieve EE operations in a dynamic network environment, we assume the presence of CR algorithms at both the UEs and APs. By exploiting CR capabilities, UEs and APs cooperatively select the appropriate communication parameters to meet the required EEbased objectives. In order to achieve this goal, HetNet components should be aware of the network situation especially from the infrastructure point of view. Such an awareness includes context information like location of active APs and UEs, current traffic load, operating frequencies of APs, availability of different RATs. These information are collected to

Fig. 1. Green framework of heterogeneous wireless network.

1520

Fig. 2. Trade-off architecture.

R. Mahapatra et al. / Computer Networks 57 (2013) 1518–1528

between

centralized,

distributed,

and

ticular case of such a distributed architecture is the localized approach, where energy saving operations are based on purely local information available at a single AP and its associated users. Fundamental trade-offs between centralized, distributed, and localized architecture are depicted in Fig. 2. Having knowledge of the network and radio environment, the green-cognitive network manager (GCNM) (see Fig. 1) selects appropriate operating parameters. This virtual entity resides at each network device and takes decisions by using information collected from both APs and UEs. Furthermore, GCNM exploits seamless cooperation amongst neighbouring APs and UEs. Such type of architecture design allows adaptive mechanisms that aim at achieving the right balance of energy savings amongst network devices. Finally, GCNM enables real time interaction with its environment to adapt appropriate communication parameters.

localized

build a database inside the network awareness manager (NAM) module, shown in Fig. 1. NAM provides the system-wide network state for different HetNet components. The network state defines the current configuration and communication parameters of HetNet. From the architecture perspective, the NAM module as well as optimization functions are distributed in nature and reside in all the entities of the network. Centralized architectures, where system optimization is based on fairly complete information located in a unique database, have been investigated in homogeneous wireless networks [20,21]. However, the proposed architecture is threefold motivated. First, centralized architectures are inefficient for managing operations in networks that are controlled by different providers [22]. Second, it is impractical to have network awareness module reside in a central entity as such required information exchange will make it difficult to design backhaul networks. Third, although a distributed solution is often suboptimal with respect to a centralized one, it is more robust since it does not depend on the reliability of a main node. In order to reduce the signalling overhead, we envision that distributed modules exchange information only with neighbouring entities, which share the same environment and are affected by each other from EE perspective. A par-

3. Energy-cognitive cycle In this section we describe the structure of the proposed energy-cognitive cycle. As shown in Fig. 3, all the tasks of energy-cognitive cycle are divided into two broad categories: the network awareness and the selection & access modules. The first module is further composed by the sensing and the analysis & learn blocks. The second module consists of the decision and the adapt blocks. 1. Sensing: This block implements functionalities that enable APs and UEs to gather the required information about the network status by monitoring the radio environment. 2. Analysis & Learn: By exploiting information in the sensing database, APs and UEs characterize their surrounding environment. Moreover sensing data and static information about the network are jointly processed to estimate the future status of the network.

Radio, Network Hardware & others Reconfiguration

Adapt

Dynamic configuration Information

Static

Sensing

configuration Information

Selection & Access

Communication & configuration parameters

Sensing information Network constraints

Analysis & Learn

Decision Analyzed information

Fig. 3. A detailed representation of energy-cognitive cycle.

Network Awareness

R. Mahapatra et al. / Computer Networks 57 (2013) 1518–1528

Radio, Network Hardware & others

Sensing Information

Communication Parameters

AP Network Awareness

Selection & Access Analyzed

UE Network Awareness

Information

Fig. 4. Energy cognitive cycle at UE and AP.

3. Decision: Based on the feedback received by the network awareness module, UEs and APs take decision about the network energy efficient operating point. Constraints such as AP and UE capabilities and QoS requirements of UEs are used in this decision process to modify the network strategy. 4. Adapt: Thereafter, APs and UEs change their operating parameters according to the decision block information. This reconfigurability enables both APs and UEs to fast adapt their behaviour to the dynamic network environment. The various parameters amongst which adaptation is carried out are activation of APs, operating frequency, radio transmission parameters. It is important to note that the tasks defined by the first two blocks are performed separately in both APs and UEs (as shown in Fig. 4), as such a design enables distributed algorithms. On the contrary, there is exchange of information amongst neighbouring APs to cooperatively enhance the awareness on the network status. However, in the latter blocks (i.e., decision and adapt) APs and UEs cooperate at the algorithmic level sharing optimization parameters to improve the performance of whole network. Such a design avoids that optimization in one part of the network produce detrimental effects on some other part of the network. 4. Network awareness module The goal of network awareness module (see Fig. 3)) is to enable a green HetNet to acquire information about its environment and to predict its future status. This module runs separately on both the APs and UEs and collects information by sensing. The network awareness module of APs creates a HetNet snapshot from an infrastructure perspective, while at the UEs, it is responsible for realizing local snapshot of radio environment. We should note here that the collected awareness information is only related to optimizing the network EE. This awareness information is then subsequently used by the selection and access module to optimize network performance. The separation in network

1521

awareness module enables better EE optimization involving APs coordination in HetNet environment. Subsequently, it also allows some degree of freedom at the cognitive UE to tune its communication parameters for more efficient operation according to the perceived local conditions. It should be noted that exchange of information between network awareness modules at AP and UE is limited to the channel state information (CSI), simply because such an exchange is only possible on the scarce wireless spectrum, which is a very expensive resource. The awareness module consists of two building blocks: sensing and analysis & learn, which are presented in the following sections. 4.1. AP network awareness AP network awareness is responsible for analyzing the network state from the infrastructure point of view. Sensing information is gathered by cooperation amongst the neighbouring heterogeneous APs and is subsequently used at each AP to realize a snapshot of the network status. Hence, such network-awareness module is a distributed entity that resides in all APs of the HetNet. 4.1.1. Sensing The sensing block of an individual AP collects information about the network usage of the HetNet environment. This operation is performed by analyzing the radio environment and exchanging captured information amongst neighbouring APs. There are three main aspects of sensing that need to be emphasized from an EE point of view. The first aspect is that sensing needs to be carried out in an energy efficient manner as it may involve high overhead in terms of both computing and radio resources. In fact, classic sensing algorithms for cognitive radio networks (such as energy detector [28] and feature detection [29]), results in poor performance from an EE perspective [30]. Recently, cluster-based sensing [31] and compressed sensing [32] have been proposed to achieve network awareness in a more sustainable way. However, green sensing is still an important open research field in CR networks. Then, to avoid resource wastage, network elements should collect only information which specifically helps the algorithms in decision block to improve EE. Finally, one of the main features of the network awareness module is to achieve its objectives with minimum amount of overhead on the HetNet infrastructure. Whenever it is not required for other network functionalities (like inter-cell interference coordination [33]), APs avoid sharing information that have their reliability on a shorttime scale (such as occupied resource blocks, used modulation coding scheme, and allotted power) for two main reasons: first, this information should be continuously updated producing excessive overhead; furthermore, delay introduced by the backhaul may limit its accuracy. Therefore, in the proposed framework, APs only share higher level information such as APs location and duty cycle, operating bandwidth, and average traffic load. Such an approach is confirmed by analyzing also in the fourth column of Table 1, where we resume the sensing measurements required by the main EE enablers proposed in literature.

1522

R. Mahapatra et al. / Computer Networks 57 (2013) 1518–1528

Table 1 Sensing measurements required for the main EE enablers proposed in literature. EE enabler

Type

Key idea

Sensing measurements

Antenna muting [23] Cell discontinuous transmission [10] EE scheduling and power control [24] Bandwidth adaptation [25] Dynamic cell switch off/on [9] Multi-RAT vertical handover [26] Network offloading [27] Cell zooming [11]

Local. Local.

Fast adapt the number of active antennas Introduce fast sleep modes in AP

Momentary cell load, CSI, and QoS constraints Momentary cell load, CSI, and QoS constraints

Local.

Momentary cell load, CSI, and QoS constraints

Coop./ Local. Coop.

Tradeoff delay for energy in the resource allocation process Fast adapt the usage of frequency resources and the power amplifier operation point Switch off/on cells under absence/presence of active users Exploit the EE of different RATs

Coop.

Offload traffic from high power APs to small cell APs

Coop.

Adjust the cell size according to the traffic load

Local.

Localized algorithms, which are listed in the upper part of Table 1, use short-time scale inputs to enable fast adaptation to the momentary cell load and fading instances. On the contrary, cooperative algorithms, exploit local coordination for matching system capacity and coverage to average requirements. In order to efficiently enable the exchange of EE related information amongst different HetNet entities, we introduce the concept of EE signalling network. It is worth to underline here that the proposed signalling network is not related with the control network that coordinates data transfer in the HetNets. The EE signalling network is designed to only deal with the EE aspects of different HetNet components. Finally, in order to manage such dynamic information exchange, amongst different APs, a hierarchi-

Momentary cell load, CSI, and QoS constraints User location and local coverage constraints The EE of available RATs w.r.t. QoS constraints Location of UEs and APs; Traffic requirements and available capacity at APs The average load in the region,location of UEs and APs, and coverage offered by active APs

cal logical structure is needed. Fig. 5 presents the proposed hierarchical structure to efficiently implement cooperative sensing within and across cellular HetNets. The proposed signalling network enables interfaces which allow information exchange of sensing parameters amongst the active APs within the network. The lowest level of this signalling structure is represented by the users connected to a particular AP. All APs of similar RAT are inter-connected by technology-specific gateway, which provides communication interface to enable network awareness amongst APs with similar RAT. The inter RAT interface allows communication amongst different RAT technologies. The different HetNet APs are connected locally with the EE signalling interface to reduce the over-

Fig. 5. Architecture of the proposed EE signalling structure.

R. Mahapatra et al. / Computer Networks 57 (2013) 1518–1528

head with respect to the backhaul signalling network and also to allow fast adaptation of EE algorithms. Such signalling structure also allows the APs of different RATs to exchange the communication parameters required for energy savings with each other in technology agnostic way, which is one of the key requirements for designing the proposed signalling architecture. This hierarchical signalling structure enables distributed algorithms within APs to save energy of network through dynamic tuning of communication parameters amongst different HetNet entities. However, frequent exchange of information further increases the overhead while enhancing system performance due to more accurate estimation of network state. Hence, there exists a tradeoff in the network performance between QoS, EE, and network overhead. 4.1.2. Analysis and learn The information collected by the sensing block is now exploited to learn the current state of the network and estimate its future evolution. This block analyses the dynamic information received by the sensing block with the help of static information. The static information consists of deployment, location of APs, and system power consumption models, which are reliable on long time scale. The sensing block provides dynamic information about APs and UEs, such as load, bandwidth, average interference level, and network topology. Having acquired knowledge of these information (static and dynamic), the analysis block processes received feedback and learns about the network environment for EE operations. 4.2. UE network awareness The network awareness module at UE aims at creating a local snapshot of the network at the user level. The UE is aware of its requirements and the other transmission parameters. Cooperation amongst different UE network awareness modules is severely limited because of wireless medium, which is very scarce and expensive resource. Nevertheless, such an operation is feasible as long as its gain offsets the loss due to wireless spectrum usage. We envision that such an exchange of information could be possible within the context of machine-to-machine (M2M) communications in which different devices communicate directly amongst each other. In the past, mobile operators have not exploited mobile terminal capabilities for network management operations, especially for security reasons. For instance, a malicious user, which is able to force a femtocell to accept all end users without having to register them first (i.e., changing the femtocell access scheme), would be able to analyze those user communications [34]. Moreover, by controlling the activity status of neighbouring small cells by remote [9,35], adversarial users may reduce the network available capacity and limit the customer satisfaction. Nowadays, however, tablets, laptop and smartphones have a cost and computation capabilities comparable with those of a small cell APs. Therefore, we believe that the integration of mobile terminals in the HetNet management process could greatly enhance the network performance.

1523

4.2.1. Sensing The role of sensing at UE is to build a local picture of the radio medium in terms of interference and signal strength. This block is also responsible for creating such a model for different RATs and local available APs. It should be noted here that the sensing block at the UE is much less complex compared to that of AP and also there is much less information exchange between the sensing blocks of different UEs. 4.2.2. Analysis and learn Having acquired the sensing information of the different RATs, UE characterizes them using the analysis block. Hence, UE can learn about the best suited RAT with corresponding transmission parameters to meet its requirements. This information is then used along with the analysis information of AP awareness for selecting communication parameters. The network awareness module of APs and UEs provide information to the selection and access module. The network awareness module suggests priority list of configuration and communication parameters from the AP and UE perspective. With this information, the selection and access module will take the decision on configuration and communication according to network constraints and objectives. 5. Selection and access module This section presents how the proposed green framework selects a reconfiguration strategy and subsequently adapts its behaviour in order to efficiently react to received stimuli. Reconfiguration is based on the analysis information from network awareness module (see Fig. 3), network objectives, and constraints. The selection and access procedure is composed of two main parts. First, the decision block decides the reliable strategy to efficiently react to the received stimuli. Thereafter, the adapt block implements the appropriate changes by hardware and software reconfiguration. 5.1. Decision This block aims to find the optimal approach to modify the network characteristics and react to feedback sent by the network awareness module. It should be noted that adaptation may concern a specific geographic area of the HetNet or focus on a given AP. In the former case, either a cooperative approach is followed by all network components or different parts of the network can act according to their own local objectives. On the contrary, in the latter case, reconfiguration involves only a specific cell, which autonomously optimizes its parameters for energy efficient operation. 5.1.1. Local reconfiguration design Local reconfiguration design aims to improve performance mainly from the single cell perspective. Both UEs and serving APs are aware of the network environment status estimated by the network awareness module. Hence,

1524

R. Mahapatra et al. / Computer Networks 57 (2013) 1518–1528

they process this feedback and elaborate the appropriate strategy. However, the decision taken by each UE may differ from strategies chosen by the neighbouring UEs and from the AP objectives and constraints. Therefore, first UEs have to inform the serving AP about their decision, subsequently the AP has the role to find an equilibrium between different goals. 5.1.2. Cooperative reconfiguration design Cooperative reconfiguration exploits available interfaces, which are defined in the proposed signalling architecture, to enable joint optimization in a given geographical area. First, neighbouring APs collect and share mid time-scale statistics regarding their status and the network environment. As previously mentioned, cooperative nodes avoid to exchange information on fast changes in the network, to avoid excessive overhead and enable a stable reconfiguration. Then, iterative distributed algorithms are implemented to reach a common equilibrium and ameliorate the network performance as a whole.

price [4]. The reward describes how a given EE metric changes modifying certain parameters (Table 2 describes the most common EE metrics used in literature); the price represents the cost (such as interference, complexity, and overhead) that this choice implies. The main drawback of using an utility function is that it is generally fixed. Thus, it can be hardly adapted to the evolution of system goals and constraints. An alternative solution can be used by exploiting a look-up-table (LUT), which associates possible inputs received from the selection module with a given solution. The main advantage of a LUT is that it can be periodically updated according to the experienced performance to improve the decision robustness. Further improvements are possible permitting LUTs exchange between terminals that are in the same environment. This cooperative learning process increases the decision speed and accuracy [38].

6. A practical implementation of the proposed EE framework: Advanced Open Access femtocells

5.2. Adapt The adapt block aims to reconfigure the network parameters according to the approach selected by the decision block. Network reconfiguration mainly concerns the physical, MAC, and link layers. However, energy-aware terminals/networks may apply reconfiguration even at higher layers to further improve the system efficiency. For instance, some types of traffic (e.g., real time traffic) is less EE/SE than other types (like non real time traffic). Therefore, trading off end-to-end latency for energy consumption can further improve the system performance, especially for more delay tolerant traffic [15]. Obviously, there are several ways to implement a given input strategy. For instance, the decision block can decide that the generated interference towards the neighbouring cells has to be reduced. Therefore, the adapt block may limit the radiated power and adjust the modulation and coding scheme (MCS), change transmission strategy (e.g., from the interweave to underlay or overlay [4]), modify the cell duty cycle, etc. Such a decision mainly depends on QoS/physical constraints. Some of the different EE enablers that can be used are defined in Table 1. In order to comply with these requirements, decision on reconfiguration can be taken by attempting to maximize a certain EE utility function, which can consider both personal and local wellness. The generic structure of the utility function is often composed of two parts: a reward and a

In this section, we show the way to develop practical algorithms that improve the system EE exploiting the proposed green cognitive framework. Here, we focus on a twotier network composed of macrocells and femtocells. Femto Base Stations (i.e., HeNBs) operate in the same bandwidth of the Macro Base Station (eNB) and offer service to indoor UEs that are located in their coverage area. The eNB serves outdoor UEs and the indoor UEs that can not be served by active HeNBs. According to the model proposed in [39], HeNBs system power consumption slightly depends on the load, hence femtocell EE is limited in lightly load scenarios. In the perspective of our framework, both HeNBs and UEs are aware of network environment, and reconfigure their parameters to adapt the femtocell network capacity to traffic and deployment characteristics. This network management scheme, named as Advanced Open Access (AOA), increases the macrocell offloading while limiting the number of simultaneously active HeNBs in order to minimize the two-tier network power consumption. In the proposed strategy, HeNBs periodically broadcast their control channel; subsequently, UEs estimate the HeNBs deployment by measuring the Reference Signal Received Power (RSRP). Classically, each selfish user demands service to HeNB with the best RSRP. In AOA scheme, UEs can be served by a poorer HeNB to allow a network reconfiguration that optimizes the system EE. Furthermore, in

Table 2 Main EE metrics investigated in literature. Metric

Ref.

Units

Description

Scenario of interest

Energy Consumption Gain (ECG)

[8]

%

Energy Consumption Rating (ECR)

[36]

Watt/Gbps

The ratio of the energy consumption at the baseline system and at the system under test Ratio of energy consumption over effective system capacity

Useful to compare systems that operate in the same radio context Useful to describe the system;behaviour at full load

Area Power Consumption (APC)

[37]

Watt/K m2)

Performance indicator in rural areas

Power per User

[36]

Watt/User

Ratio of total power consumed w.r.t the coverage area Ratio of total power consumed w.r.t number of subscribers

Performance indicator in;urban areas

R. Mahapatra et al. / Computer Networks 57 (2013) 1518–1528

order to redistribute the traffic load, HeNBs, which are aware of UEs location, can decide to change their access mode, from closed access to open access. Subsequently, the femtocells network associates UEs and HeNBs according to traffic load and network connectivity. Finally, HeNBs that are not serving active UEs self switch off to reduce interference and power consumption. Due to the static characteristic of the indoor femtocell deployment scenarios, the frequency of this process and generated overhead is low. However, when new UEs are detected near the femtocell network or some UEs leave its region, the UE-HeNB association has to be updated accordingly. Note that, cellular coverage for these UEs, which can not be served by active HeNBs, is guaranteed by the macrocell. In the following, we briefly explain the proposed AOA scheme (see Fig. 6) in line with our green framework:  Sensing: HeNBs detect the presence of incoming UEs. Therefore, UEs measure RSRP received from nearby HeNBs.  Analysis and Learn: Each UE identifies the set of HeNBs from which it can be served (i.e., its active set) comparing the strength of the RSRP with a predefined threshold. We indicate the length of active set as the UE degree. UEs feedback their active sets to nearby HeNBs, which share received information and compute the network connectivity matrix.  Decision: Minimize the number of simultaneously active HeNBs.  Adapt: UEs are ordered according to their degree of connectivity from nearby femtocells. Subsequently, our algorithm picks up the unserved UE with the lowest degree and associates it with the active cell that maxi-

Radio, Network Hardware & others

HeNBs detect incoming UEs UEs measure the RSRP

Sensing

UE indentifies nearby HeNB The network computes connectivity matrix

Minimize the number of active HeNBs

UEs are associated to HeNBs

Analysis & Learn

Decision

Adapt

Switch off unused HeNBs

Fig. 6. Main steps of the proposed Advanced Open Access scheme.

1525

mizes a predefined EE metric. If there are not available active HeNBs, the best idle cell is activated. Finally, idle HeNBs are switched off to save power. We assess the effectiveness of the proposed scheme by comparing its performance with the closed access and open access approaches. The main differences between these schemes are the following: 1. In closed access HeNBs deployment, a UE can be served by a HeNB only if that access point is placed in the user’s apartment. 2. In open access HeNBs deployment, UEs can be served by any of the femtocells in its coverage area. However, each user selects the available HeNB associated to the best RSRP. In both closed access and open access schemes, HeNBs are always active, even if they are not serving any user. However, dynamic switch off can be implemented to disable HeNBs that are not serving neighbour UEs. In such a way, HeNBs self-adapt their status to reduce power consumption and interference. On the contrary, in the proposed AOA scheme, HeNBs and UEs cooperatively reconfigure the number of activated HeNBs according to the aggregate capacity in order to enhance the EE performance of the network. We consider a scenario in which 30 cellular users are deployed in the macrocell area. In line with recent studies [40], we assume that 70% of the traffic is generated by indoor UEs. The results are averaged over 50 independent runs. We simulate 103 independent Transmission Time Intervals (TTIs) during each run and update channel fading instances at each TTI. At the beginning of each run, HeNBs and indoor UEs are randomly deployed in a block of 5  5 apartments [41], which is placed at the macrocell edge. Outdoor UEs are randomly deployed in the macrocell region. The traffic generated by cellular users is modelled as a constant bit-rate traffic and the throughput target ðRtg Þ is set equal to 150 Kbps. A proportional fair scheduler is used at both the macrocell and neighboring femtocells [42]. Finally, link adaptation is implemented in downlink transmissions for which modulation and coding schemes are selected according to momentary feedback transmitted by the served UE. In our simulations, we have measured the cellular network energy consumption required to satisfy users’ rate constraints with respect to the different strategies implemented at femtocells. Energy consumption at eNBs and HeNBs is calculated according to a linear model that maps the radiated RF power to the power supply of a BS site and captures the relationship between the BS load and its power consumption [39].

Pin ¼ P0 þ Dp P out ;

0 6 Pout 6 Pmax

ð1Þ

where Dp is the slope of the load-dependent power consumption, P 0 indicates the power consumption at minimum non-zero load, and P max is the RF output power at maximum load. Table 3 shows the used values of Pmax , P 0 , and Dp for eNBs and HeNBs.

1526

R. Mahapatra et al. / Computer Networks 57 (2013) 1518–1528

Table 3 BS Power model parameters. BS type

P max (W)

P 0 (W)

Dp

eNB HeNB

40 0.01

712 10.1

14.5 15

70

60

50

16

14

Aggregate Energy Consumption Gain [%]

Femtocell network Energy Consumption Gain [%]

The power consumption of switched off HeNBs depends on the hardware components that are deactivated during sleep intervals. In particular, more deactivated hardware components result in a longer activation process. Due to the static characteristic of the indoor femtocell deployment scenarios, the system under investigation is not required to react at the fast time scale. Hence, power consumption of deactivated femtocells is neglected in the following analysis. Triangle marked lines, circled marked lines, squared marked lines, diamond marked lines, and star marked lines respectively correspond to the closed access, closed access plus cell switch off, open access, open access plus cell switch off, and AOA plus cell switch off. First, we aim at comparing these strategies in terms of the femtocell ECG (see Table 2), which is defined as ratio of the energy consumption at the baseline system and at the system under test [8]. In Fig. 7, we show the femtocell network ECG with respect to the HeNB deployment ratio ðqd Þ [41]. Deployment ratio captures the density of femtocells in a given geographical area, and is given by the ratio of the number of apartments where a HeNB is installed to the total number of apartments in the macrocell. Closed access and open access scheme are considered as the reference strategies to compute the EE gain. The difference between these two schemes, in terms of power consumption, is related to the number of UEs served by the HeNBs (i.e., the load of the femtocell network): in the open access deployment the probability that a UE can be served by a HeNB is higher than in the closed access deployment. As previously mentioned, HeNB system power consumption slightly depends on the load, thus closed access and open access femtocells result in same performance (see the

squared marked and triangle marked lines). However, femtocell EE can be enhanced by switching off those HeNBs that are not serving active UEs. In this case, high number of closed access HeNBs can be deactivated and energy consumption of closed access femtocells is lower than the open access case, especially for low values of deployment ratio. Clearly, if all 25 apartments in the block are equipped with a femtocell (i.e., qd ¼ 1), the performance of these two strategies are coincident, because a UE can always access the closest HeNB. In our simulations, up to 16% of gain can be achieved with respect to the reference approach by implementing cell switch off mechanism. Finally, adapting the femtocell activity to the traffic load can introduce a much higher gain. In fact, the proposed AOA scheme strongly lowers the femtocell network energy consumption achieving up to 60% of ECG with respect to the reference schemes. Fig. 8 shows the aggregate two-tier network ECG with respect to the femtocell deployment ratio. Our results account for both the macrocell and HeNBs system energy consumption, and they are computed considering the closed access scheme as the reference approach. The EE of the two-tier network is fairly related to the HeNBs access scheme. Both open access and closed access schemes improve the system EE via the macrocell offloading. However, in open access femtocells deployment the probability that a UE can be served by a HeNB is higher than in the closed access case. Thus, the former scheme increases the macrocell offloading and reduces the system energy consumption. Simulation results show that open access strategy achieves up to 10% of gain with respect to the closed access approach. Therefore, although open access femtocells may consume more power than closed access femtocells (see Fig. 7), the former are more energy efficient from the overall cellular network perspective. Nevertheless, when the number of active HeNBs is greater than a certain

Advanced Open Access + Switch off Closed Access + Switch off Open Access + Switch off Closed Access Open Access

40

30

20

Advanced Open Access + Switch off Open Access + Switch off Open Access Closed Access + Switch off Closed Access

12

Advanced Open Access gain 10

8

6

Open Access Offload Gain

4

2

Sleep Mode Gain

10

0 0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Femtocell Deployment Ratio (ρd) Fig. 7. Femtocell network energy consumption with respect to the femtocell deployment ratio qd .

0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Femtocell Deployment Ratio (ρd) Fig. 8. Aggregate energy consumption gain with respect to the femtocell deployment ratio qd .

R. Mahapatra et al. / Computer Networks 57 (2013) 1518–1528 100

Normalized Throughput [%]

7. Conclusions

Open Access Advanced Open Access Closed Access

99 98 97 96 95 94 93 92 91 0

0.1

0.2

0.3

0.4

1527

0.5

0.6

0.7

0.8

0.9

1

Femtocell Deployment Ratio (ρ )

Heterogeneous wireless networks set new challenges for broadband and sustainable communications. To enable fast and effective adaptation to the dynamic wireless environment, in this article, we have proposed a CR-based framework, where awareness and intelligence are shared amongst all network terminals. We have also introduce a novel EE dedicated signalling architecture, which allows to networks characterized by different RATs to exchange communication parameters in technology agnostic way and limited overhead. Future studies will extend this framework to include the user mobility challenges and further investigate the advantages/drawbacks of the proposed architecture from the backhaul network perspective.

d

Fig. 9. Performance experienced by cellular users with respect to the femtocell deployment ratio qd .

value (i.e., qd P 0:4) the aggregate capacity of open access femtocell network exceeds the service request and the EE gain decreases. On the contrary, our proposed scheme permits to adapt the network capacity to the actual traffic load. The proposed AOA scheme achieves up to 10% (14%, respectively) with respect to the classical strategies with (without, respectively) cell switch off. Our proposed algorithm does not permit UEs to select the HeNB associated with the best RSRP. Hence, we may expect that user performance decreases when using AOA femtocells. In Fig. 9, we show the impact of the proposed scheme on the performance perceived by cellular users (both macro and femto users). We show the average Normalized Throughput ðR Þ experienced by end-users with respect to the femtocell deployment ratio for different access schemes. R can be expressed as

R ¼

e R g ¼ 1  PER; Rtg

ð2Þ

e and PER g are the average cell throughput and packwhere R et error rate, respectively. Cell switching off does not impact user performance, thus we compare the performance achieved in the closed access, open access, and AOA femtocell deployment. In the closed access case, only a restricted set of UEs has the right to access femtocells, thus in low/medium density femtocell scenarios several indoor users are constrained to be served by the faraway eNB. Furthermore, femto-to-macro interference decreases the performance of these users. In higher density femtocell scenarios (i.e., qd P 0:5), the probability that an indoor UE may be served by a HeNB increases, thus users experience better performance. In both open access scheme and AOA, there is a high probability that an indoor user can be served by a HeNB even in low density femtocell deployment scenarios. Therefore, these schemes perform in the fairly same way. However, UEs experience better performance compared to the closed access case (up to 7% of the gain) due to the reduced distance between the end-user terminal and its access point and limited effect of interference.

Acknowledgements This work has been performed in the framework of the FP7 project TROPIC IST-318784 STP, which is funded by the European Community. The Author(s) would like to acknowledge the contributions of his (their) colleagues from TROPIC Consortium (http://www.ict-tropic.eu). References [1] R. Ferrus, O. Sallent, R. Agusti, Interworking in heterogeneous wireless networks: comprehensive framework and future trends, IEEE Wireless Communications 17 (2010) 22–31. [2] A. Ghosh, R. Ratasuk, B. Mondal, N. Mangalvedhe, T. Thomas, Lteadvanced: next-generation wireless broadband technology, IEEE Wireless Communications 17 (2010) 10–22 (invited paper). [3] A. Damnjanovic, J. Montojo, Y. Wei, T. Ji, T. Luo, M. Vajapeyam, T. Yoo, O. Song, D. Malladi, A survey on 3GPP heterogeneous networks, IEEE Wireless Communications 18 (2011) 10–21. [4] A. De Domenico, E. Calvanese Strinati, M.G. Di Benedetto, A survey on MAC strategies for cognitive radio networks, IEEE Communication Surveys and Tutorials (2012) (First Quarter). [5] E. Calvanese Strinati, A. De Domenico, L. Herault, Green communications: an emerging challenge for mobile broadband communication networks, Journal of Green Engineering 267 (2011) 301. [6] Green Touch Initiative, 2012. . [7] M. Gruber, O. Blume, D. Ferling, D. Zeller, M. Imran, E. Calvanese Strinati, EARTH-energy aware radio and network technologies, in: IEEE 20th International Symposium on Personal, Indoor and Mobile Radio Communications, Tokyo, Japan, 2009, pp. 1–5. [8] C. Han, T. Harrold, S. Armour, I. Krikidis, S. Videv, P. Grant, H. Haas, J. Thompson, I. Ku, C. Wang, et al., Green radio: radio techniques to enable energy-efficient wireless networks, IEEE Communications Magazine 49 (2011) 46–54. [9] I. Ashraf, F. Boccardi, L. Ho, Sleep mode techniques for small cell deployments, IEEE Communications Magazine 49 (2011) 72–79. [10] P. Frenger, P. Moberg, J. Malmodin, Y. Jading, I. Godor, Reducing energy consumption in LTE with cell DTX, in: IEEE 73rd Vehicular Technology Conference, Budapest, Hungary, 2011, pp. 1–5. [11] Z. Niu, Y. Wu, J. Gong, Z. Yang, Cell zooming for cost-efficient green cellular networks, IEEE Communications Magazine 48 (2010) 74–79. [12] S. Alouf, E. Altman, A. Azad, M/G/1 queue with repeated inhomogeneous vacations applied to IEEE 802.16 e power saving, in: ACM SIGMETRICS Performance Evaluation Review, vol. 36, 2008, pp. 451–452. [13] I. Haratcherev, M. Fiorito, C. Balageas, Low-power sleep mode and out-of-band wake-up for indoor access points, in: IEEE GLOBECOM Workshops, 2009, pp. 1–6. [14] L. Cai, H. Poor, Y. Liu, T. Luan, X. Shen, J. Mark, Dimensioning network deployment and resource management in green mesh networks, IEEE Wireless Communications 18 (2011) 58–65. [15] Y. Chen, S. Zhang, S. Xu, G. Li, Fundamental trade-offs on green wireless networks, IEEE Communications Magazine 49 (2011) 30– 37.

1528

R. Mahapatra et al. / Computer Networks 57 (2013) 1518–1528

[16] J. Hoydis, M. Kobayashi, M. Debbah, Green small-cell networks, IEEE Vehicular Technology Magazine 6 (2011) 37–43. [17] E. Oh, B. Krishnamachari, X. Liu, Z. Niu, Towards dynamic energyefficient operation of cellular network infrastructure, IEEE Communications Magazine (2011) 56–61. [18] D. Lopez-Perez, I. Guvenc, G. De La Roche, M. Kountouris, T. Quek, J. Zhang, Enhanced intercell interference coordination challenges in heterogeneous networks, IEEE Wireless Communications 18 (2011) 22–30. [19] S. Haykin, Cognitive radio: brain-empowered wireless communications, IEEE Journal on Selected Areas in Communications 23 (2005) 201–220. [20] Nokia Siemens Networks, Self-Organizing Network (SON) Introducing the Nokia Siemens Networks SON Suite-an efficient, future-proof platform for SON. [21] C. Stevenson, G. Chouinard, Z. Lei, W. Hu, S. Shellhammer, W. Caldwell, IEEE 802.22: the first cognitive radio wireless regional area network standard, IEEE Communications Magazine 47 (2009) 130– 138. [22] S. Feng, E. Seidel, Nomor Research, Self-Organizing Networks (SON) in 3GPP Long Term Evolution, 2009. . [23] EARTH, Project Deliverable, D3.2, Green Network Technologies, 2011. [24] E. Calvanese Strinati, P. Greco, Green resource allocation for OFDMA wireless cellular networks, in: IEEE International Symposium on Personal Indoor and Mobile Radio Communications, Instanbul, Turkey, 2010. [25] A. Ambrosy, O. Blume, H. Klessig, W. Wajda, Energy saving potential of integrated hardware and resource management solutions for wireless base stations, in: IEEE 22nd International Symposium on Personal Indoor and Mobile Radio, Communications, 2011, pp. 2418–2423. [26] A. Serrador, L.M. Correia, Energy efficiency gains using VHOs in heterogeneous networks, in: Future Heterogeneous Network, Prague, Czech Republic, 2012. [27] K. Lee, I. Rhee, J. Lee, S. Chong, Y. Yi, Mobile data offloading: how much can wifi deliver? in: ACM 6th International Conference on Emerging Networking Experiments and Technologies, 2010. [28] H. Urkowitz, Energy detection of unknown deterministic signals, Proceedings of the IEEE 55 (1967) 523–531. [29] W.A. Gardner, Signal interception: a unifying theoretical framework for feature detection, IEEE Transactions on Communications 36 (1988) 897–906. [30] Z. Hasan, H. Boostanimehr, V. Bhargava, Green cellular networks: a survey, some research issues and challenges, IEEE Communications Surveys and Tutorials 13 (2011) 524–540. [31] J. Wei, X. Zhang, Energy-efficient distributed spectrum sensing for wireless cognitive radio networks, in: IEEE Conference on Computer Communications (INFOCOM Workshops), 2010, pp. 1–6. [32] D. Donoho, Compressed sensing, IEEE Transactions on Information Theory 52 (2006) 1289–1306. [33] G. Boudreau, J. Panicker, N. Guo, R. Chang, N. Wang, S. Vrzic, Interference coordination and cancellation for 4G networks, IEEE Communications Magazine 47 (2009) 74–81. [34] I. Bilogrevic, M. Jadliwala, J. Hubaux, Security issues in next generation mobile networks: Lte and femtocells, in: 2nd International Femtocell Workshop, Luton, UK, 2010. [35] R3-110030, Dynamic H(e)NB Switching by Means of a Low Power Radio Interface for Energy Savings and Interference Reduction, 3GPP TSG RAN WG3 Meeting, Dublin, Ireland. [36] A.P. Bianzino, A.K. Raju, D. Rossi, Apple-to-apple: a framework analysis for energy-efficiency in networks, ACM SIGMETRICS Performance Evaluation Review 38 (2011) 81–85. [37] A. Fehske, F. Richter, G. Fettweis, Energy efficiency improvements through micro sites in cellular mobile radio networks, in: 2nd IEEE Workshop on Green Communications, Honolulu, USA, 2009. [38] L. Giupponi, A. Galindo-Serrano, P. Blasco, M. Dohler, Docitive networks: an emerging paradigm for dynamic spectrum management, IEEE Wireless Communications 17 (2010) 47–54. [39] G. Auer, V. Giannini, C. Desset, I. Godor, P. Skillermark, M. Olsson, M. Imran, D. Sabella, M. Gonzalez, O. Blume, A. Fehske, How much energy is needed to run a wireless network?, IEEE Wireless Communications 18 (2011) 40–49 [40] Informa, Mobile Broadband Access at Home, 2008. [41] 3GPP-TSG-RAN4#51, Alcatel-Lucent, picoChip Designs, Vodafone, R4-092042, Simulation Assumptions and Parameters for FDD HENB RF Requirements, 2009.

[42] K. Norlund, T. Ottosson, A. Brunstrom, Fairness measures for best effort traffic in wireless networks, in: 15th IEEE International Symposium on Personal Indoor and Mobile Radio, Communications, vol. 4, 2004, pp. 2953–2957.

Rajarshi Mahapatra received his Ph.D in Electronics and Communication Engineering from IIT Kharagpur. Recently, he has completed his Postdoc in CEA-LETI, France. In his postdoc research, he was engaged in FP7 Call4 BeFEMTO and Greentouch. He is currently working as Professor in the Dept. of ECE, Graphic Era University, Dehradun. He has served as a member of TPC for several national and international conferences and peer-reviewed journals in the area of wireless network. He has published about 20 peer reviewed paper in several international journals and conferences. His current research interests include cognitive radio, dynamic spectrum access, energy consumption in wireless networks and optical access networks.

Antonio De Domenico obtained his Masters degree from the University of Rome ‘‘La Sapienza,’’ Rome, Italy, in 2008, and his Ph.D in Engineering Science in 2012 on energyefficient communications in heterogeneous cellular networks. He is presently involved in ICT-FP7 EARTH and ICT-FP7 BeFEMTO projects that aim to successfully develop future mobile wireless technologies to enable a cost-efficient provisioning of ubiquitous broadband services.

Rohit Gupta received his PhD from University of Washington in Seattle where he researched on various aspects of radio layer for dense wireless deployments. He has also worked at Ericsson research during his PhD on LTE COMP schemes. After finishing his PhD, he worked on EARTH project for two years at CEA-LETI on energy-efficient architectures of next generation cellular systems. Currently, he is working as a Product Architect at Signalion for test eNodeB project.

Emilio Calvanese Strinati obtained his Masters degree in 2001 from the University of Rome ‘‘La Sapienza’’ and his Ph.D in Engineering Science in 2005 on Radio link control for improving the QoS of wireless packet transmission. He started working at Motorola Labs in Paris in 2002. Then in 2006 he joints the Centre for Atomic Energy (CEA) in Grenoble as a research engineer. Since 2004 Emilio Calvanese Strinati is giving lectures at ENST, the University of Rome ‘‘La Sapienza’’ and, INPG-Grenoble on physical and MAC layer topics. Its main research topics are information theory, advanced coding schemes, cooperative communications, scheduling, resources allocation and green ICT for wireless mobile networks. From 2007, he becomes a PhD supervisor. Emilio Calvanese Strinati has published around 40 papers in international conferences and books chapters, and is the main inventor or co-inventor of more than 20 patents. Currently, he is the Head of telecommunication strategy in CEA.