3.1.1 Wireless Networks

The heterogeneous wireless medium is composed of a set of wireless networks, which can have different access technologies, such as the fourth-generation (4G) cellular networks, WLANs, VLC and WMANs, as illustrated in Figure 3.1. While 4G cellular networks, WLANs and WMANs rely on different RF access technologies, VLC relies on visible light for communications. Particularly, in VLC networks, communications take place by modulating the intensity of light in light-emitting diodes (LEDs) in such a way that it is not observed by the human eyes. Consequently, the illumination energy that is consumed for lighting is also used for communications. More details about VLC networks, the challenging issues in integrating VLC with RF networks and radio resource allocation in an integrated VLC and RF network for green communications are provided in Chapter 6. Chapters 3 through 5 mainly focus on the integration of RF networks for green communications. The cellular networks in can have different cell sizes, for example macro, micro, pico and femto cells. In such a networking environment, different networks are assumed to be operated by different service providers. Also, different networks operate in different radio frequency bands. Every network has a set of BSs or access points (APs), for network , to provide service coverage, radio resource management, user mobility management and access to the Internet.

With overlapped coverage from the BSs/APs of different networks, different service areas can be distinguished. The set of service areas is denoted as . Each service area, , is covered by a unique subset of BSs/APs of various networks. For instance, as shown in Figure 3.1, service area 2 is covered by the cellular network and the WMAN, while service area 3 is covered by all the three networks.

Network () BS/AP () has a maximum available bandwidth of . A cooperative networking environment is considered, where the BSs/APs of different networks are connected through a backbone that enables them to exchange their signalling information.

3.1.2 Mobile Terminals

The networking environment assumes a set of MTs. Network BS/AP has a subset of MTs that lie in its coverage area. Two modes of communications can be distinguished for each MT. The first mode is based on the communication between MT with network BS/AP . The second mode relies on direct communications among MTs in close proximity based on D2D communications. This part of the book (Chapters 3 through 6) mainly focuses on green communications among MTs and BSs/APs. More details about D2D communications and their application to green communications are provided in Chapters 9 and 10.

Each MT is equipped with multiple radio interfaces, and assumes a multi-homing capability for multiple simultaneous associations with different networks. Hence, a given application of the MT, for example, data downloading/uploading or video streaming, can be served on the downlink/uplink using multiple networks. Such an infrastructure presents the following advantages [4]. First, using the multi-homing capability, the available resources at different networks can be aggregated to support bandwidth-hungry applications through multiple threads at the application layer. Second, multi-homing calls enable better mobility supports to ensure that at least one of the used radio interfaces will remain active during the call. Third, the multi-homing support can reduce the call-blocking rate and improve the system capacity.

Each MT has its own home network, but it can also get service from other available networks at its location, using the multi-homing capability. The subset of MTs in the coverage area of network BS/AP is divided into the subset of network subscribers whose home network is network and the subset of network users whose home network is not network .

3.1.3 Radio Resources and Propagation Attenuation

Let () denote the allocated bandwidth from network BS/AP to MT on the uplink (downlink), where the allocated bandwidth is equal to 0 for . Let () denote the allocated transmission power for communication between MT and network BS/AP on the uplink (downlink). In addition, let denote the circuit power that is required to keep the radio interface active for communication between MT and network BS/AP , with for . The maximum power is denoted by for MT , and by for network BS/AP .

The channel power gain between MT and network BS/AP on the uplink (downlink) is denoted by (), and it captures both the channel fast fading and path loss. Let the distance between MT and network BS/AP be and the path loss exponent be . Hence, the path loss between MT and network BS/AP is given by , where and denotes the far-field distance of the transmitting antenna. The one-sided noise power spectral density is denoted by .

3.3.1 Single-User versus Multiuser System

In the literature, several works investigating energy efficiency of MTs focus on a single-user scenario, for example, [16] [24, 29]. Hence, given an MT that is allocated a fixed amount of bandwidth, the main objective is to allocate the uplink transmission power to maximize the MT energy efficiency. The system model under consideration does not capture the effect of interference among different MTs operating in the same frequency band [46]. Also, a single-user system does not capture the competition among different MTs over the radio resources (e.g. bandwidth [28]) to satisfy their target QoS at an improved energy efficiency.

Joint bandwidth and power allocation calls for a multiuser system to model the competition among different MTs over the available bandwidth at the BSs/APs of different networks, given their different channel conditions and path losses. For instance, in Figure 3.2, MTs 1 and 2 (3 and 4) compete on the available bandwidth at BS 2 on the uplink (downlink) to enhance their energy efficiency. However, such a multiuser system further complicates the analysis, as will be presented in Chapter 5. Rather than maximizing energy efficiency for a single MT, the objective now is to enhance energy efficiency for a set of MTs, which leads to the fairness issue, to be discussed later in more details.

3.3.2 Single-Operator versus Multioperator System

Most of the existing works dedicated to the heterogeneous wireless communications environment assume a single-network operator. This is evident from how different MTs are treated in the context of radio resource management. In the literature, all MTs have been treated equally by different networks in the heterogeneous networking environment. However, this does not take into account the fact that different MTs are subscribers of different operators. Hence, with multi-homing services, each operator wants to first support its own subscribers and to ensure that they are satisfied with the maximum required QoS, while at the same time support the subscribers of other networks. For instance, in the context of joint bandwidth and power allocation, a network operator may allocate bandwidth to the mobile subscriber ofanother operator only if this operation does not degrade the energy efficiency of its own subscribers. This calls for service differentiation among network subscribers and users. One way to accomplish such a service differentiation is through a priority mechanism [4]. In this case, every network operator gives a higher priority in allocating its resources to its own subscribers as compared to the other users. However, such a priority mechanism may degrade the overall energy efficiency of the networking environment as it limits the exploitation of the available resources. Further studies are required to assess how to maximize the overall energy efficiency, while at the same time support service differentiation among network subscribers and users in the networking setting.

Other implications for modelling a multioperator system are related to fairness and implementation complexity, as discussed in the following subsections.

3.3.3 Fairness

For energy-efficient uplink communications in a multiuser and multioperator system, a popular problem formulation is to maximize the sum energy efficiency for the users in the system, for example, [28] [29] and [46]. However, this objective provides no fairness guarantee for energy efficiency among different MTs. The sum energy efficiency can be maximized by dramatically improving energy efficiency only for a subset of users. One approach to promote fairness is by maximizing the geometric mean of energy efficiency for all MTs, which introduces proportional fairness among all users [28]. Another notion of fairness is based on a max–min formulation, where the objective is to maximize the energy efficiency performance of the worst MT. An investigation of this objective is presented in Chapter 5.

As for energy-efficient downlink communications, most existing studies make no distinction in the amount of energy saved by different operators, and the main objective is to reduce the total energy consumption in the networking environment [12]. However, this can be satisfied by minimizing the energy consumption of one operator at the expense of increasing energy consumption of another operator. It is desired that green communications result in mutual benefits for all operators. The degree of cooperation among different operators is determined based on the attained utility by each operator. The general problem of how to motivate various operators to cooperate among each other to enhance the overall energy efficiency in the networking environment while maximizing the achieved utility for each operator will be investigated in Chapter 4.

3.3.4 Centralized versus Decentralized Implementation

Joint bandwidth and power allocation can be implemented in a centralized or decentralized architecture, as shown in Figure 3.3. In the centralized implementation, a central resource manager is responsible for the resource allocation decisions. The central resource manager has global information of the available resources (e.g. radio bandwidth and maximum transmission power) at BSs/APs of different networks. Also, the central resource manager has information about the required QoS of all MTs in service, the available energy of each MT for uplink communications, and the channel conditions among the MTs and BSs/APs of different networks. Given the collected information, the central resource manager allocates resources from different networks so as to maximize the energy efficiency in the overall networking environment and satisfy the required QoS for all MTs. However, the central resource manager is not a practical solution when different networks are operated by different service providers [4]. A central resource manager who controls the operation of BSs/APs of different networks raises some concerns: (i) the central resource manager creates a single point of failure, hence, if it fails, the whole multi-homing service fails; (ii) the operator in charge of the operation and maintenance of the central resource manager will control the resources of other networks and (iii) changes are required in the structure and operation of different networks to account for the central resource manager.

Given the aforementioned concerns, a decentralized resource management is a more practical and flexible solution. In a decentralized implementation, BSs/APs of different networks make their resource allocation decisions based on their available local information. Better resource allocation decisions can be made through information exchange among BSs/APs of different networks on one side and between BSs/APs and MTs on the other side. However, high signalling overhead is expected in large systems with multiusers and multioperators. Further investigations on how to reduce the associated signalling overhead and guarantee the decentralized solution convergence are presented in Chapter 4 for the green downlink framework and in Chapter 5 for the green uplink communication scenario.

3.3.5 In-device Coexistence Interference

The MT multiple radio interfaces are located in close proximity within the same device. When two radio technologies having adjacent bands operate simultaneously in the same device, the out-of-band (OOB) radiations of the transmitting radio leak into the band of the receiving radio due to the filter non-idealities. These OOB radiations are commonly referred to as IDC interference [127]. One example of using adjacent frequency bands in a multi-homing scenario exists between the LTE band 40 (B40) and WiFi [128] [129]. LTE B40 ranges between 2,300 and 2,400 MHz, while WiFi operates in the industrial, scientific and medical (ISM) band from 2,401 to 2,500 MHz. For the same MT, the data transmission and reception over LTE occur in a band adjacent to the data transmission and reception over the WiFi. Hence, the downlink reception in each radio suffers from an IDC interference due to the uplink transmission in the other radio, as depicted in Figure 3.4. For video calls supported by the multi-homing capability, the MT can utilize one radio interface (LTE) for the uplink stream and the second radio interface (WLAN) for the downlink stream, and hence, the downlink reception on the WLAN can suffer from IDC interference due to the uplink transmission on the LTE. It has been demonstrated that the currently used filter technology is not sufficient to suppress the IDC interference [128]. Hence, multi-homing radio resource management mechanisms should model such interference and account for it in their resource allocation strategies. However, the existing research overlooks this important aspect. Next, we will discuss how to model such an IDC interference [127].

Every filter has a roll-off factor which defines the amount of frequency band secured for the transition between the pass band and stop band. The IDC interference is caused by these filter non-idealities. The non-ideal steepness of the aggressor WLAN spectrum mask allows part of its power spectral density to leak into its neighbouring LTE band. In addition, the non-ideal selectivity of the victim LTE receiver filter captures the emissions from the WLAN band. This leaked power constitutes the IDC interference. As the frequency separation between the LTE and WLAN channels allocated to the same MT decreases, the IDC interference increases, as depicted in Figure 3.5. An overlap factor () can be used to measure the ratio of the leaked WLAN power at the LTE receiver to the total WLAN power [127]. The LTE cellular BS has a set of channels that can be assigned to MTs. Each channel is assigned to a single MT at a time. The WLAN AP has a single channel that is shared by MTs in a TDMA manner using the enhanced version of the Distributed Coordination Function (DCF) [130]. Let represent the LTE receiver filter transfer function operating at channel with frequency . Also, let denote the WLAN power spectral density centred at , which is approximated by its transmit spectrum mask. This is a reasonable assumption since the transmit spectrum mask represents an upper bound of the power spectral density [131]. The overlap factor of the WLAN and LTE filters of MT , given that the LTE radio is assigned to channel , is expressed as

3.1

Let and denote the interfering WLAN uplink power allocated by MT and the losses due to the antenna isolation at MT , respectively. The antenna isolation depends on the physical separation between the WLAN and LTE antennas. The IDC interference that affects LTE channel when it is allocated to user is expressed as

3.2

The WLAN reception is also affected by the LTE transmission in the same MT. Under the typical case where each radio has similar transmitter and receiver filters, the overlap factor of interference at the WLAN receiver takes the form

3.3

However, the user may be assigned more than one LTE channel. All these channels contribute to the IDC interference at the WLAN receiver , which is given by

3.4

where and denote the interfering LTE uplink power allocated by MT on channel and the channel assignment variable that indicates if MT is assigned to LTE channel () or not (), respectively.

In order to show the level of IDC interference that the LTE and WLAN might suffer from, consider the simultaneous operation of LTE and WLAN radio technologies in one mobile, where the WLAN operates on a 22-MHz time division duplex (TDD) channel, while the LTE operates on a 5-MHz TDD channel. The design of the WLAN spectrum mask is obtained from the IEEE 802.11 standard [131]. The used LTE spectrum mask is similar to a 5-MHz ISM filter, but shifted to create a pass band for the LTE channels, as carried out by the 3GPP [128]. Interference parameters are summarized in Table 3.1. Figure 3.6 depicts the interference on the WLAN channels when the aggressor LTE is centred at 2,397.5, 2,387.5 and 2,377.5 MHz. The interference is compared with a typical noise floor in WLAN receivers. As expected, the WLAN channels close to the aggressor LTE suffer from very high interference. The severity of interference across the WLAN channels decreases as we move away from the LTE band until it falls beneath the noise floor. Similarly, the interference caused by the WLAN operation on the LTE channels is depicted in Figure 3.7. The behaviour of the IDC interference on the LTE channels is similar to the WLAN case. The IDC interference can reach undesirable values due to the fact that the interfering and victim antennas lie in close vicinity in one device.

Parameter	Value
Antenna isolation ()	20 dB
LTE uplink power ()	23 dBm
WLAN uplink power ()	20 dBm

Such an IDC model in the context of a green multi-homing radio resource allocation framework is further investigated in Chapter 4.

3.3.6 Computational Complexity

The joint bandwidth and power allocation approach has a higher computational complexity than a power-only allocation scheme. From a resource allocation perspective, the increased computational complexity is mainly due to the increased number of decision variables (bandwidth and power allocation for all radio interfaces in the joint approach versus only power allocation for all radio interfaces) and constraints. In particular, in uplink resource allocation, the joint bandwidth and power allocation should include the network-side constraints (e.g. the total allocated bandwidth satisfying the BS/AP transmission capacity) in the problem formulation, whereas the power-only allocation mainly deals with the MT-side constraints in terms of the MT available energy and required QoS.

For computational efficiency, the joint bandwidth and power allocation problem can be decomposed into two steps, corresponding to the network-side and MT-side, respectively [54]. In the first step, bandwidth is allocated from the BSs/APs of different networks, on the uplink/downlink, to the MTs, given the required QoS of MTs. This step mainly focuses on the network-side resource allocation (bandwidth) and constraints while satisfying the required QoS, and it can be implemented in the network layer. On the basis of the allocated bandwidth, power allocation is performed in the second step to enhance energy efficiency of the MTs in the uplink and the BSs/APs in the downlink. For uplink resource allocation, this step mainly focuses on the MT-side constraints (MT total available energy) and can be implemented in the physical layer. The resource allocation approach iteratively executes these two steps until energy efficiency is maximized, a strategy that will be discussed in Chapters 4 and 5.

3.3.7 Number of MT Radio Interfaces versus Number of Available Networks

In heterogeneous wireless networks, researchers mainly assume that an MT connects to all existing networks in a multi-homing manner (i.e. the number of MT radio interfaces is equal to the number of available BSs/APs from different networks). This assumption reduces the computational complexity of the problem from a resource allocation perspective as we deal with a nonlinear program (rather than dealing with a mixed-integer nonlinear program (MINLP) as in the case with unequal number of radio interfaces and available BSs/APs). However, such a vision overlooks the fact that the MT is equipped with only a limited number of radio interfaces for each network type, for example, one cellular and one WiFi radio interface, as shown in Figure 3.8. Hence, the MT has to select one cellular BS and one WiFi AP from all the available ones to get its required radio resources in a multi-homing manner. In order to account for the limited number of interfaces for MTs, a joint user assignment and bandwidth and power allocation framework is required to support MTs with multi-homing capabilities.

The work in [132] has investigated this challenge in the context of network capacity maximization. Consider a heterogeneous wireless medium that is composed of two networks, namely cellular network and WLAN, that is, , where 1 represents the cellular network and 2 represents the WLAN. The set of BSs of the cellular network is denoted by and the set of APs for the WLAN is represented by . The set of available BSs and APs is denoted by , where . Each MT, is equipped with cellular and WLAN radio interfaces, and and denote the set of cellular and WLAN radio interfaces, respectively. MT asks for a best effort service. Although some of smartphones with dual SIM card are equipped with more than one cellular radio interface, these interfaces cannot be used simultaneously to run two calls, as all cellular interfaces share the same transceiver. Therefore, the MT can have multiple radio interfaces of the same type, but the MT can only utilize one radio interface of one type at a given moment of time. Consequently, the MT can utilize one radio interface from and one radio interface from and can aggregate the offered bandwidth from these two radio interfaces to support its ongoing call. Hence, the problem reduces to specify: (i) a binary user assignment decision variable that indicates if MT can connect to BS/AP using its radio interface () or not (), (ii) a bandwidth allocation decision variable that indicates the amount of bandwidth allocated to MT from BS/AP via radio interface and (iii) a power allocation decision variable that indicates the amount of power allocated (in the uplink or downlink) on the link connecting MT to BS/AP via radio interface . The binary decision variable constraints (e.g. (3.5) and (3.6)) ensure that an MT connects to a cellular BS using only a (single) cellular radio interface, that is,

3.5

3.6

3.7

3.8

Similarly, the constraints (3.7) and (3.8) guarantee that an MT connects to a WLAN AP using only a (single) WLAN radio interface. In addition to the aforementioned constraints, the following constraints should be accounted for: (i) capacity limitation of BSs/APs, which accounts for the limitation on the available bandwidth at the BS/AP and (ii) total power consumption constraint (at MT for uplink radio resource allocation or at BS/AP for downlink radio resource allocation). For energy efficiency maximization, the objective function can be expressed by (1.27) for the network operators (i.e. in the downlink radio resource allocation) and (1.28)–(1.32) for the mobile users (i.e. in the uplink radio resource allocation).

The radio resource allocation problem described above is a fractional MINLP due to the binary nature of the MT assignment problem and the continuous nature of the radio resource (bandwidth and power) allocation problem. Hence, higher computational complexity is expected than the approach discussed in the previous subsection. A possible solution relies on exhaustive search, where the MT tries all combinations of cellular network BS and WLAN AP and selects the pair that maximizes the energy efficiency while satisfying the required QoS. However, in a multiuser multioperator system, such an approach incurs high computational complexity and requires high signalling overhead in a decentralized architecture. Furthermore, the heuristic randomization approach proposed in [132] is no longer valid for energy efficiency maximization. The heuristic approach of [132] randomly assigns MTs to BSs and APs and equally allocates the bandwidth available at each BS/AP among the assigned users. However, for energy efficiency maximization, the MT locations and distances from different BSs/APs should be accounted for, which makes the randomization approach not useful in this context. Hence, more efficient techniques should be developed to properly select BSs and APs and allocate the required radio resources.