A Green Communication Model For 5G Systems

This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TGCN.2017.2700855, IEEE
Transactions on Green Communications and Networking
1
A Green Communication Model for 5G Systems

Md Munjure Mowla, Student Member, IEEE, Iftekhar Ahmad, Member, IEEE,
Daryoush Habibi, Senior Member, IEEE and Quoc Viet Phung Member, IEEE
Abstract—Small cell networks (SCNs) are envisaged as a key access technologies (Multi-RAT), device-to-device commu-
technology enabling the fifth-generation (5G) wireless communication (D2D), massive multiple-input-multiple-output (M-
nication system to address the challenge of rising mobile data MIMO), network function virtualization (NFV), prioritized
demand. Green communications will be another major attribute
of 5G systems, as power consumption from the Information and spectrum access, and base station (BS) densification [8]. For
Communication Technology (ICT) sector is forecast to increase BS densification in 5G systems, a small cell network (SCN)
significantly by 2030. Accordingly, energy-efficient small cell net- based heterogeneous network (HetNet), is considered as a
work design has attracted significant attention from researchers potential solution [9].
in recent years. In addition, to enable the ubiquitous deployment A HetNet is a mixed wireless infrastructure, with a combi-
of dense small cells, service providers require energy-efficient
backhauling solutions. In this paper, we present an energy- nation of fewer high power macrocells and many low power
efficient communication model for 5G heterogeneous networks small cells (e.g., micro, pico, and femto), that brings the net-
(HetNets). The proposed model considers both the access and work closer to the end user, thereby offering higher signal-to-
backhaul network elements. We formulate and present an an- interference-plus-noise ratio (SINR), which in turn improves
alytical model to calculate the optimum number of small cells link robustness and quality of service (QoS). In a HetNet, high
that need to be kept active at various times of the day in order
to minimize power consumption while meeting users’ quality of reuse of frequency can greatly reduce the bandwidth scarcity
service (QoS) demands. Based on our critical investigation of problem. Another major 5G research challenge is to achieve
backhaul power consumption, we also isolate and present two up to 90% energy savings [6]. Recent evidence suggests that
energy-efficient backhauling solutions for 5G HetNets. Simulated the Information and Communication Technology (ICT) sector
results reveal that the proposed green communication model saves itself consumes 4.7% of global electricity production [10],
up to 48% more power than other existing models.
[11], and it is estimated that 4.4 terawatt-hours (TWh) power
Index Terms—Energy Efficiency, Green Communications, will be consumed by about 100 million SCNs in 2020 [12]. To
Small Cell Networks (SCNs), Traffic Backhauling, 5G. address the challenge of increasing power demand, researchers
have focused on “Green Communications and Networking” to
I. I NTRODUCTION develop energy-efficient solutions for next generation wireless
communication standards [13].
LOBAL wireless data traffic shows no signs of slowing
G down and is expected to maintain its rapid growth in
the near future due to the proliferation of smart devices
Both access and backhaul network power consumption
contribute to total power consumption in 5G HetNet systems.
While SCNs help to reduce the bandwidth scarcity problem
and applications [1]. Emerging applications such as high- in HetNet, increasing numbers of uncoordinated and lightly
resolution video streaming, tactile Internet, remote monitoring, loaded active SCNs can increase the access network power
road safety and real-time control applications, are expected consumption [14]–[26]. There is a resemblance between this
to produce huge amounts of data traffic. In addition, diverse scenario in the 5G HetNet and the electric power industry
proposed services, such as connected cars and moving robots [27], [28]. A smart grid system in the electric power industry
must be supported in efficient scalable ways [2]. The current consists of sources (e.g., diesel generators, renewable sources,
4G wireless communication system is not equipped to meet coal-fired power stations), sinks (i.e., load/demand) and dis-
this explosive growth in traffic demand. Consequently, several tribution networks. The demand curve is not uniform across
consortia, comprising major international mobile operators, various hours of the day. The smart grid system makes a
infrastructure manufacturers, and academic institutions have temporal estimate of demand/load based on its historical data.
engaged in the fifth generation (5G) wireless communica- This estimate is used as a constraint in an optimization model
tion system design, planning, and implementation [3]–[5]. A where the objective is to minimize the operational costs while
public-private partnership for 5G (5G-PPP) has been consti- ensuring that a blackout is avoided during various hours of the
tuted to deliver standards, architecture, and technologies for day. In a HetNet scenario, cells can be considered as sources of
ubiquitous 5G infrastructure [6]. The aim is to develop, by bandwidth, users’ demand for data dictates the demand curve,
2020, a system that can support 1000 times increase in capac- and the communication network (similar to the distribution
ity, 1000 times higher mobile data volume per area, 100 times network) makes sure that the available bandwidth provides
more connected devices, and 5 times reduced end-to-end delay the required quality of service. It is obvious that maximum
than today’s 4G network [7]. To achieve these challenges, bandwidth can be made available by keeping all the small
some potential design considerations include multi-tier radio cells active during all hours of the day, but this would result
in an oversupply of bandwidth and increased operational costs
The authors are with the School of Engineering, Edith Cowan Uni-
versity, Perth, WA 6027, Australia. (E-mail: [email protected]; (i.e., power consumption) during periods of the day when the
[email protected]; [email protected]; [email protected]). demand is not at its peak. As such, in a 5G HetNet, the
2473-2400 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TGCN.2017.2700855, IEEE
2
temporal, as well as the spatial variation (e.g., different traffic backhaul network is created by connecting the base stations
load at different places) of traffic demand can be used to send to the core network through various backhauling solutions
cells into sleep mode. The gain would be power savings and including wired, wireless or mixed architecture of existing
reduction in operational costs. This motivated us to investigate technologies.
and develop an analytical model to calculate the optimum
number of active SCNs that are required to meet the data
demand during various hours of the day. The aim was to reduce
power consumption of 5G access networks without degrading
the QoS.
The true benefit of power savings can be realized if the
savings come from both the access and backhaul segments of
a HetNet. In 5G HetNet systems, traffic generated in small
cells is required to be backhauled to the core network, and
an increasing variety of backhauling solutions are expected to
be deployed, including wired, wireless or mixed architecture
of existing technologies. Using a direct high-capacity wired
connection (e.g., fiber optic cable) from all SCNs to the
core network is challenging and expensive [29]. Very high
frequency (60-80 GHz) millimeter wave (mmWave) technol-
ogy is now considered as a promising alternative wireless
backhauling solution [30]–[33]. Both of these technologies
Fig. 1: Network Architecture of a 5G HetNet.
have their relative strengths and weaknesses, but one major
aim that they both share is to facilitate energy-efficient traffic
forwarding to the core network. The issue of increasing back- Many research projects have been conducted in recent
haul power consumption has attracted considerable attention years to increase the energy efficiency for multi-tier HetNets
from service providers, because of its impact on overall [14]–[26]. Fehske et al. evaluated the influence of power
network power budgets [34]–[42]. Recent evidence suggests consumption by varying the microsite density for a required
that about 56% of providers consider the backhaul as one system level performance in a macro-micro HetNet [14]. The
of the major challenges for future 5G HetNets [43]. In this optimum deployment strategy in a macro-pico type HetNet
research, to complement our proposed energy-efficient 5G was investigated to reduce power consumption in terms of area
access networking solution, we investigate power consumption power consumption with a specified area spectral efficiency
of various backhaul designs and isolate two solutions, one for [16]. An analysis framework to measure the macro offloading
wired passive optical network (PON) and the other for wireless benefits was proposed to increase the energy efficiency for this
millimeter wave (mmWave). In our first solution, we present an HetNet [18]. The authors formulated the spectrum sharing and
access-backhaul design,where we show how to connect passive resource allocation as a three-stage Stackelberg game model
optical network units/terminals with 5G access units to reduce for improving energy efficiency in heterogeneous cognitive
the overall power consumption. In our second solution, we radio networks [19]. To quantify the energy efficiency gain,
present how to integrate mmWave backhaul units with 5G several issues such as frequency plan, mobility, scheduling are
SCN units to reduce power consumption. We also present surveyed in [20], [21].
analytical models to calculate the total power consumption However, all of these studies were conducted for peak traffic
for both of these solutions. In addition, we analyze the QoS demand scenarios in a HetNet system. During peak traffic
parameters, i.e., the delay and the jitter of the proposed green scenarios, some SCNs can be turned off without creating
communication model for a 5G HetNet. coverage holes. The QoS of the SCNs can be assured by
The rest of the paper is organized as follows. Section offloading the traffic either horizontally to neighboring cells
II reviews the related literature and research contribution. or vertically to higher layer cells. Base station sleep mode
Section III introduces the proposed system model for access strategies [22]–[24] and network management schemes [25],
network power consumption. In section IV, backhaul power [26] have been proposed for traffic-aware energy-efficient Het-
consumption modeling for two proposed solutions and four Net designs. Ashraf et al. [22] proposed three different sleep
existing solutions are presented. Section V shows the model mode strategies to reduce power consumption in a HetNet
for the network QoS. Results and analysis are highlighted based on network element characteristics. In [23], Marsan et
in section VI. Finally, section VII draws a conclusion and al. developed an analytical model for optimal switch-off time
highlights potential future research directions. based on the dynamic traffic profile. An architecture using two
sleep strategies was developed in [24], by vertical offloading
of traffic from SCNs to a macrocell. However, none of the
II. R ELATED W ORKS
above-mentioned articles provide the optimal number of SCNs
Figure 1 shows the network architecture of a 5G HetNet, that need to be active during various hours of the day to meet
where an access network is comprised of a macrocell and the traffic demand and minimize power consumption at the
several SCNs (e.g., microcells, picocells, and femtocells). A same time. In addition, existing energy-efficient techniques
3
[14]–[26] do not take into account the power consumption capacity in an access network which may have the potential
of backhaul networks. for reducing overall energy consumption and expenditure. The
Many authors have addressed backhaul issues including cost optimization of optical fiber point-to-point and PON back-
capacity, cost, and power consumption [29]–[42]. A recent hauling for small cell networks was demonstrated [41], [42].
survey revealed the ubiquitous unavailability of optical fiber Although the PON solution was mentioned as a cost-effective
connection around the globe [29]. Several studies have iden- fiber deployment for small cell backhauling, to the best of
tified the millimeter wave (mmWave) technology (E-band or our knowledge, the complete backhaul power consumption
V-band) as a promising alternative backhauling solution on the analysis has not yet been studied. In addition, the influence
basis of its high capacity, low cost, and ubiquitous availability on backhaul power consumption of traffic variation over a 24-
[30]–[33]. The throughput and energy efficiency of a 5G hour period has not yet been investigated. As discussed above,
wireless backhaul network were analyzed using mmWave and although researchers have attempted to address the power
microwave band under two deployment scenarios, i.e., central efficiency issue of various backhaul solutions, there is no work
and distributed [30]. In [31], millimeter wave technology in the literature that investigates the total power consumption
was used for access and backhaul network functions and the from SCNs to the core network, based on a 24-hour traffic
authors highlighted a holistic design requirement approach. profile, when using either the mmWave or optical technology
Mesodiakaki et al. analyzed the backhaul power consumption, as a backhaul solution. The above-mentioned research gaps
with mmWave, microwave, and sub-6 GHz band, for outdoor provided the motivation for this research work.
small cell backhauling under different deployment scenarios The key contributions of this paper are:
[32]. In a related study, the optimal user association problem • we develop and present an analytical model to calculate
for HetNet was demonstrated to maximize the energy effi- the optimal number of required active SCNs during
ciency and spectrum efficiency by using mmWave technology various hours of the day. The proposed solution saves
[33]. Self-backhauling [34] and energy harvesting [35] tech- power by putting redundant small cells into sleep mode
niques are also used to further improve the energy efficiency while maintaining the QoS.
in small cell networks. • we investigate and introduce two energy-efficient back-
When low power, small cell base stations are used to in- haul solutions, to complement our proposed 5G access
crease the capacity of the network, backhaul power consump- network solution.
tion also undermines overall network energy efficiency [36]–
[42]. This indicates a trade-off between the power saved by
using low power small cell base stations and the excess power A. Problem Statement
needed for backhauling their traffic. The impact of the back- The total 5G HetNet power consumption can be split into
haul power consumption model on overall energy efficiency in access network power consumption, and backhaul network
a heterogeneous network was first evaluated by Tombaz et al. power consumption. For the access network, let us assume that
[36], who used optical point-to-point Ethernet for backhauling. total available system bandwidth is W . To meet the increasing
Monti et al. considered various topologies (star, tree, and demands for higher data rates, the HetNet is modeled as
ring) for microwave and optical fiber backhauling [37]. Subse- a q − tier network consisting of 1 macrocell base station
quently, Tombaz et al. explored significant aspects of backhaul from tier one (q = 1) and J number of available SCNs
technologies using three solutions such as two hybrid solutions (e.g., micro, pico, or femto) from other tiers (q > 1). The
(fiber to the node using very high-speed digital subscriber temporal distribution of peak traffic loads for various traffic
line version 2 and fiber to the building with microwave) types over a 24- hour period is available based on historical
and a wireless solution (microwave only) for HetNets [38]. data. It is imperative that all these SCNs can always be
They analyzed the small cell density, backhaul architecture, kept in active mode to ensure the maximum data rates and
and area capacity in terms of area power consumption. For QoS. However, this approach is inefficient in terms of power
the hybrid solutions, a very high-speed digital subscriber line consumption for periods over which the required data rates and
version 2 (VDSL2) modem provided a maximum downlink QoS can be facilitated by the macrocell base station itself.
throughput up to 230 Mbps by using a digital subscriber line Power can be saved if SCNs can be put into sleep mode
access multiplexer (DSLAM) and 4 pair bonding with the during off-peak hours and then enter into active mode when
phantom mode in the presence of copper wire, but it was the extra capacity is required. As such, the research question
vulnerable to electromagnetic interference. Another hybrid for the network planning and service management becomes,
backhauling solution combining fiber to the building (FTTB) for a given day, how many SCNs should be in active mode
with microwave was also suggested as a favorable candidate at various hours of a day so that required data rates and
with a high SCN deployment ratio scenario to support the QoS can be met while access network power consumption
backhauling traffic [38]. A comprehensive estimation model is minimized. Another research challenge is to reduce the
was presented for the total cost of ownership (TCO) cal- backhaul power consumption to further increase the energy
culation of a backhaul network segment when applied to a efficiency of 5G HetNets. Considering the existing backhaul
microwave and 10 GPON architecture with FTTB in [39]. technologies (optical fiber, millimeter wave, microwave, and
Surez et al. further analyzed the power consumption modeling VDSL2), the research question for this paper is: which access-
for the FTTB + 10 Gigabit PON backhauling solution in [40]. backhaul solution is the most energy-efficient to handle 5G
They investigated the impact of SCN density and resource HetNet data demands while maintaining network QoS?
4
1
Conversational
0.9 Streaming
Background
0.8
Normalized Traffic Load

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
06 08 10 12 14 16 18 20 22 00 02 04
Time (Hour)
Fig. 2: System Model of a 5G HetNet. Fig. 3: Normalized Peak Traffic Load Profile.
III. P ROPOSED S YSTEM M ODEL On the other hand, background traffic (low priority) such
In this section, we present a system model and relevant as email, ftp, and telnet, use network resources when the
analytical model to calculate the access network power con- other two traffic demands are fully satisfied, i.e., looser delay
sumption of a 5G HetNet. Figure 2 shows a q − tier 5G requirements [52]. Total traffic demand is expected to exceed
HetNet system model, where the macrocell base station is the available bandwidth in a macrocell during peak hours,
placed at the center of the network within tier (q = 1), and and SCNs can be used to mitigate this challenge. However,
overlaid by J number of SCN base stations (e.g., micro, a lack of coordination between cells from multiple tiers can
pico and femto) from tiers (q > 1). To ensure spectral result in an oversupply of bandwidth at the cost of increased
efficiency, a macrocell zone is partitioned into three sub- power consumption. A coordinated management technique
zones [44] represented as zone 1, zone 2, and zone 3. In that calculates and use the optimum number of active SCNs
zone 1, high signal to interference noise ratio (SINR) yields can mitigate this problem. To minimize the access network
higher spectral efficiency while in zone 3, spectral efficiency power consumption, we need to calculate the exact bandwidth
remains relatively low. In this paper, we consider the split requirement during various hours of the day.
spectrum strategy [45], [46] for minimizing the cross-tier
interference between multi-tier (q − tier) HetNets, where total A. Analytical Model for Access Network
available spectrum W is divided as, Wm = ηq W for macrocell This section presents bandwidth requirement formulations
base station and Wqj = (1 − ηq )W Hz (q > 1) for SCN of a macrocell, using the traffic load profile for the proposed
system model. These formulations will be used to model the
P tiers, respectively. ηq is spectrum
base stations of different
allocation factor, with q=1 ηq = 1 and ηq > 0. To avoid the access network power consumption. In this paper, we use the
co-tier interference between small cell networks within the following sets for notations:
same tier in 5G HetNets, we considered the static frequency Q set of Q a 5G multi-tier HetNet, index q.
reuse scheme [47], [48]. In this approach, neighboring cells J set of J SCN base stations, index j.
use different spectrum to avoid interference for users in their A set of A Sub-zone area, index a.
respective cells. The available bandwidth Wqj is split into Nsb Z set of Z Sub-zone, index z.
orthogonal sub-bands (frequency reuse factor), and each cell T set of T Traffic class, index t.
transmits on non interfering sub-bands. For example, available U set of U Users, index u.
Wj
resource block for each SCN is now Nsbq Hz. stz (u) binary variable, taking 1, if user u is associated with
In this research, we consider the temporal and spatial zone z with traffic type t; 0 otherwise.
fluctuation characteristics of network traffic. This is useful Nqj binary variable, taking 1, if the small cell j of tier q
to investigate the energy efficiency for network operators in is active; 0 otherwise.
low and high load conditions [49], [50]. Figure 3 shows a The received signal power at a user equipment (UE) from
typical temporal hour-by-hour distribution of normalized peak a base station (BS) can be written as:
traffic loads [1], [51], which captures three types of traffic
Prx = Ptx · r−λ · Ψ (1)
information (i.e., conversational, streaming, and background),
over one week with a resolution of one second in a suburban where Ptx , r, λ and Ψ denote transmit power, propagation
area. Conversational traffic (high priority) such as voice and distance from a UE to BS, path loss component and fading
video conferencing is highly delay sensitive. Streaming traffic parameter respectively. The parameter Ψ can be adapted for
(medium priority), such as streaming audio and video, has further investigation such as small-scale fading or multi-path
relatively less delay sensitivity than conversational traffic. fading.
5
A user equipment u associated with any sub-zone z can downlink rate ηj (u)qos for a base station. The admission
be represented by uz (use u for brevity). Then, the received control for conversational and streaming traffic cases, in terms
signal-to-interference-noise ratio (SINR) for a UE u from of service outage constraints G, can be given as:
macrocell base station can be calculated as:
G = P[ηj (u) < ηj (u)qos ] < ζ ∀j ∈ J (6)
E [Prx,z (u)]
γz (u) = (2)
Iz (u) + σ 2 where ζ denotes the service outage probability threshold (5%
according to the EARTH project [56]). We maintain QoS
where Prx,z (u) denotes the received signal power for a single
(throughput, delay, and jitter) for conversational (high priority)
user u in a sub-zone coverage area z. User is associatied
and streaming traffic (medium priority) only, whereas no QoS
based on the strongest SINR value. We consider split spectrum
is guaranteed for background traffic.
strategy [45], [46] to overcome interference between macrocell
BS and SCNs and assume that interference Iz (u) is constant The coverage radius rz of a sub-zone z can be written as:
within a given time interval τ . Without loss of generality, noise
Prx,z (u)
λ
level σ 2 is assumed to be the same for all users and receiver rz = (7)
Ptx,z (u) · Ψ
noise floor is not considered here.
We use the maximum throughput channel state approach The sub-zone area Az in a macrocell is measured as:
[53] to adapt the modulation and coding scheme after deriving
an estimate of channel behavior based on the channel atten- Az = πrz2 − πrz−1
2
∀z ∈ Z (8)
uation factor. The aim of the maximum throughput approach Therefore, the total number of users Uzt in a sub-zone z can
is to maximize the overall link throughput by selecting the be calculated as:
most appropriate modulation and coding in response to SINR
U
values that allows the maximum throughput based on channel X
Uzt = Az · stz (u) ∀z ∈ Z, ∀t ∈ T (9)
conditions. This is suitable for a high bit-rate system such as
u=1
conversational and streaming traffic with a lower sensitivity to
error probability [54]. For a modulation order, we [55] derive The total bit rate generated in a sub-zone z can be formu-
the symbol error probability and bit error probability. Then, lated as:
T
using maximum likelihood function, we find out the correctly X
dz = Uzt · Rt ∀z ∈ Z (10)
received block which is translated to achievable downlink
t=1
throughput for a user as:
where R is the bit rate.
0 Thus, the required bandwidth for any sub-zone z can be
N z (u) · Rc · Nsym · log2 (L)

ηz (u) = c · 1 − Pblock (3) written as:
Tf
dz
0 wz = ∀z ∈ Z (11)
where Ncz (u), Rc , Nsym , L, Pblock , and Tf represent the ez
number of assigned sub-carriers per user u within sub-zone
Therefore, the total required bandwidth of a macrocell can
z, the adapted modulation and coding scheme rate, downlink
be formulated as:
orthogonal frequency-division multiple access (OFDMA) sym-
bol numbers, modulation order, block error probability, and Z
X
r
OFDMA frame duration. Wm = wz (12)
The spectral efficiency in a sub-zone z can be expressed as: z=1
ηz (u) The main target of our research is to support all traffic

ez = ∀z ∈ Z (4) demands by considering macrocell resource needs first and
Ncz (u) · fsc
then the needs of the SCNs. The motivation behind this
where fsc indicates sub-carrier frequency (15 kHz). approach is that the static power of SCNs is more than the
Let, stz (u) be a binary variable, taking 1, if a user u increased amount of transmission power from the macrocell
is associated with zone z with traffic type t; 0 otherwise. which causes more power consumption. When the required
Assuming user association to a particular sub-zone z, the amount of resource Wm r
is less than the macro-tier spectrum
actual throughput demand for a user u with traffic type t in a Wm , all SCNs switch to sleep mode. But, if Wm r
is greater
base station j can be written as: than or equal to Wm , the SCNs must be activated to support
r
ηj (u) = ηz (u) · stz (u) ∀j ∈ J (5) the excess bandwidth We = Wm − Wm .
The transmission power Ptx,z (u) for a single user u in a
We describe achievable throughput, average end-to-end de- sub-zone z can be derived as:
lay, and average end-to-end jitter as the quality of service for
Iz (u) + σ 2

ez
a 5G HetNet. To maintain the QoS in a 5G HetNet, each user Ptx,z (u) = (2 − 1) · (13)
gz
u has to receive a certain data rate (throughput) ηj (u) within a
base station j. In other words, the QoS demand of a particular where gz is the corresponding channel gain. This transmission
max
user ηj (u) is related to the service outage probability of a base power should be limited within a maximum limit Ptx (43
station j, which should be less than the QoS limited threshold dBm). The transmission power, corresponding to total users,
6
which is known as dynamic power of a macrocell base station, the least number of active SCNs. Mathematically, the problem
is then computed as: can be formulated as:
U Q X
X J
Pqj · Nqj
dynamic
X
Ptx = Ptx,z (u) ∀z ∈ Z (14) Minimize Pm + (19)
u=1 q=2 j=1
For a macrocell base station (MBS), the total power con- Subject to
sumption can be expressed as [50]: max
Ptx,z (u) ≤ Ptx , ∀z ∈ Z, ∀u ∈ U (20)
f ixed dynamic
Pm = Pm + ∆m · Ptx (15)
f ixed
stz (u) ∈ {0, 1}, ∀z ∈ Z, ∀t ∈ T , ∀u ∈ U (21)
where Pm represents the fixed power consumption and ∆m
is the load dependent parameter. T X
Z
Similarly, for a small cell base station (SCN) j from (q > 1)
X
stz (u) = 1, ∀u ∈ U (22)
tier, the total power consumption model can be written as: t=1 z=1
j,dynamic
Pqj = Pqj,f ixed + Hs · ∆s · Pq,tx (16)
T
X
where Hs is the traffic load for SCN base stations and ∆s is γz (u) ≥ γzth (u) · stz (u), ∀z ∈ Z, ∀u ∈ U (23)
the slope of a load-dependent power consumption. Pqj,f ixed t=1
represents the fixed power consumption of an SCN BS.

j,dynamic T T
Pq,tx is the transmission power (38 dBm, 21 dBm and X X
17 dBm) for microcell, picocell and femtocell, respectively. stz (u) ≥ stz0 (u), ∀u ∈ U, ∀z, z 0 ∈ Z, z < z 0 (24)
t=1 t=1
The sleep mode approach is one of the most promising so-
lutions for energy-efficient cellular networking research [14],
ηj (u) ≥ ηj (u)qos , ∀u ∈ U, ∀j ∈ J (25)
[17], [18], [57], which turns off base stations selectively,
during the low traffic load while the coverage is ensured by
Q X
J
the macrocell base station. In the sleep mode operation, the X
small cell base station is considered to shut-down, by putting Wqj Nqj ≥ Wm
r
− Wm , ∀q ∈ Q \ {1}, ∀j ∈ J (26)
q=2 j=1
Hs = 0. Thus, small cell base station power consumption can
be written as, Pqj,sleep = Psleep , where (20) denotes the transmission power that is limited to
max
where Psleep is the power consumed in sleep mode, and a maximum value Ptx fixed by the network operator. It is
this is typically lower than Pqj,f ixed due to deactivation of known as the power budget constraint. (21) indicates a binary
components in SCNs. When an SCN BS changes its working decision variable to determine whether user u with traffic type
mode from sleep to active, additional switching power is t is assigned to the sub-zone z. When stz (u) = 1, the constraint
required to start up the BS power supply, and circuits. This is enforced to transmission power. Conversely, sz (u) = 0
switching power is represented by, Pqj,sw.power = β · Pqj,f ixed . indicates the trivial non-negativity of transmission power. (22)
Typically, the value of β is 0.1% of Pqj,f ixed [58]. represents that each UE u with traffic type t is connected to
Therefore, the access network power consumption of a only one sub-zone z. It is known as the scheduling constraint.
proposed 5G HetNet can be written as: (23) denotes that each UE u with traffic type t connects to a
sub-zone z for which PSINR is greater or equal to its threshold
Q X
J T
AN
X SINR, i.e., γzth (u) · t=1 stz (u). (24) indicates that user-zone
Phetnet = Pm + Pqj · Nqj (17)
association should be with the nearest zone. (25) represents the
q=2 j=1
QoS assurance for conversational and streaming traffic, where
where Pm and Pqj denote the power consumption of single these two types of traffic throughput exceed a QoS limited
macrocell BS and j − th SCN BS (q > 1), respectively. Nqj threshold rate, i.e., a pre-defined fixed value to confirm the
is the number of SCNs j of tier q at a particular hour. QoS. (26) indicates the assigned bandwidth for all SCNs j
The energy efficiency (EE) of a 5G access network can be from tier (q > 1) which must be greater or equal to excess
written as: bandwidth We .
PZ
AN dz The optimization problem in (19) is a mixed integer
EEhetnet = z=1 AN
(18) non-linear programming (MINLP), whose feasible set is
Phetnet
non-convex due to the combinatorial structure of the sub-
channel resource allocation. It cannot be solved by a pseudo-
B. Energy Efficient Optimization Model for Access Network polynomial time algorithm because it is NP-hard [59]. A non-
The power consumption of a 5G access network is linearly zero duality gap can result in its primary and dual solutions
related to the product of BS power consumption and the when the problem is solved by Lagrange dual decompositions
number of base stations. For an estimated traffic load profile, (LDD). However, with the increasing number of sub-channels,
considering the peak load with different types, this model will the duality gap continues to vanish. We make the problem (19)
minimize the access network power consumption by assigning more tractable by relaxing (21) for taking any value [0,1].
7
Algorithm 1 Energy Efficient Resource Management and IV. M ODEL FOR H ET N ET P OWER C ONSUMPTION WITH
Planning BACKHAULING S OLUTIONS
1: Input: User u ∈ U; traffic class t ∈ T ; macrocell sub- This section presents power consumption models for various
zone z ∈ Z. 5G HetNet backhauling solutions. For a HetNet that consumes
2: Output: Required number of active SCNs that minimizes Pqj amount of power for a small cell access network and
power consumption, yet satisfies the traffic demand. Pbh for a backhaul network, the total 5G HetNet power
3: for each hour, h ∈ 1, 2, . . . , 24 do, consumption can be written as:
r
4: Calculate required amount of bandwidth Wm from Q X
J
(12). AN +BH
X
r Phetnet = Pm + Pqj · Nqj + Pbh (27)
5: if Wm ≤ Wm then,
q=2 j=1
6: for each tier q ∈ Q \ {1} do
7: for each BS j ∈ J do The energy efficiency (EE) of a 5G HetNet including the
8: Set Nqj = 0. No SCN is required. access and backhaul power consumption can be written as:
9: end for PZ
AN +BH z=1 dz
10: end for EEhetnet = AN +BH
(28)
Phetnet
11: Terminate.
12: else
r
13: Calculate excess bandwidth We = Wm − Wm . A. Fiber optic point to point Ethernet solution
14: Find the required number of active SCNs, Nqj from The Ethernet/IP interface is frequently used in the 3rd
optimization model (19). Generation Partnership Project (3GPP) to backhaul traffic. In
15: end if this design, the Ethernet switch can be used in a centralized
16: end for or de-centralized manner for the aggregation node. Traffic
generated from all wireless base stations (MBS or SCN) is
collected at one aggregation switch, which then forwards data
to the core network as shown in Fig. 4a. In this solution,
all backhaul links (i.e., base station to aggregation switch
It can be noted that the proposed optimization model is not and aggregation switch to core network) are optical fiber. An
intended for solving the real-time resource allocation problem optical small-form factor pluggable (SFP) interface is used
as the time complexity may prove prohibitive for providers of to connect the port of the Ethernet switch. Therefore, the
a large network. The optimization model considers the peak backhaul power consumption for this solution Pbh 1
is given
hourly traffic load of a day, available from historical data, by [36]:
and provides the optimum number of SCNs that are required
to be active at various hours of the day to meet the traffic 1 1
Pbh = max · (Ns + Nm ) Psw
demand and concurrently minimize the power consumption. Ndl
dl ul
As such, the optimization model can play an important role in + (Ns + Nm )Psw + Nul Psw (29)
network planning and can be run once for the next 24 hours. max
where Ndl denotes the maximum number of downlink
We consider the peak hourly traffic load profile instead of
interfaces available at the aggregation switch for collecting
an average traffic load profile to ensure that the model does
the backhaul traffic of a HetNet. Ns represents the total
not sacrifice the QoS due to the presence of burst traffic and
number of active SCNs for a particular hour of a day. Nm
variation of loads with respect to the average load. We use
is the number of the macro base station of the HetNet.
OPTI MATLAB toolbox [60] to find the optimal solution.
The power consumption of a switch has two main parts,
The detailed procedure for the optimization is summarized in
Psw = αPsw max
+ (1 − α) Ag max
Agmax Psw . The first part models
switch
Algorithm 1.
the back-plane of the switch, which is traffic independent.
In this section, we have shown how to minimize access The other depends on the aggregated backhaul traffic that
network power consumption by calculating the optimum num- passes through the switch Agswitch . The weighting parameter
ber of SCNs during various hours of the day. The total 5G α ∈ [0, 1], is assumed for balancing the relative influence of
HetNet power consumption is contributed by both access and power quantities. Agmax is the maximum amount of traffic
max
backhaul network power consumption. In the next section, that a switch can handle, while Psw represents the maximum
dl ul
we first analyze the power consumption for various existing power consumption of a switch. Psw and Psw denote the
backhauling solutions, i.e., optical point to point Ethernet power consumption by one downlink and uplink interface in
(optical P2P) [36], fiber to the node using very high speed the aggregation switch, respectively. Total aggregated traffic
digital subscriber line version 2 (FTTN + VDSL2) [38], fiber collected at the switch is represented as: Ct = Cm + Cs ,
to the building with microwave (FTTB + MW) [38], and fiber where Cm and Cs denote l the MBS m and SCN aggregated
Ct
to the building with 10 Gbps passive optical network (FTTB traffic, respectively. Nul = U max is the total number of
int
+ 10 GPON) [40]. We then present two backhauling solutions uplink interfaces, where Ct represents the total aggregated
that achieve the best power efficiency during various hours of traffic collected at the switch from MBS traffic Cm and SCN
max
the day. traffic Cs . Uint denotes the maximum transmission rate of
8
(a) Optical Point-to-point Ethernet Solution [36]. (b) FTTN + VDSL2 Solution [38].
(c) FTTB + MW Solution [38]. (d) FTTB + 10 GPON Solution [40].

Fig. 4: Existing Backhaul Solutions.
F
an interface. It is assumed that all switches are identical, each where Pmodem , PDSLAM , Psw , PSF P and PSF P + are the
BS uses a dedicated downlink interface, and all downlink power consumption of a VDSL2 modem, a DSLAM, a fiber
interfaces are identical and use the same transmission rate. switch, a SFP module and a SFP+ module, respectively. nD
ports
is the number of ports per DSLAM, which is used to calculate
the number of DSLAM, i.e., NDSLAM = d nDNs e. The num-
ports
B. Fiber to the node and very high-speed digital subscriber F
ber of fiber switches is expressed by, Nsw = d NDSLAM +Nm
e,
line version 2 solution nF ports
where nF ports is the number of ports per fiber switch. The fiber
Fig. 4b represents a hybrid solution employing both optical switch aggregates all access network traffic before transmitting
fiber and copper wire to provide high-speed connections. it to the core network through 10 Gbps fiber links and
Here, each SCN is backhauled by a very high-speed dig- SFP+ modules. Finally, the number of uplink interfaces can
ital subscriber line version 2 (VDSL2) modem, which is be calculated as, Nul = max(Nsw F
, d UC
max e), where Ct is
t
ini
in turn connected to a digital subscriber line access mul- the total aggregated throughput from all base stations at a
tiplexer (DSLAM). The DSLAM is usually located in a particular hour and Uintmax
denotes maximum the transmission
remote telephone exchange for connecting multiple customer rate of an uplink interface as previously shown.
digital subscriber line (DSL) interfaces to a high-speed digital
communication channel. VDSL2 supports up to 100 Mbps if C. Fiber to the building and microwave solution
the distance between the DSLAM and the VDSL2 modem is
This mixed solution employs both fiber and microwave
less than 300 m [61]. The DSLAM and the macro BS are
technology to backhaul the high capacity traffic generated from
connected to a fiber switch of 1 Gbps point-to-point optical
the access network side of a HetNet. Two types of aggregation
links. Optical SFPs are used to connect the port of the fiber
points are considered for this solution, as depicted in Fig. 4c.
switch for communication. The backhaul power consumption
2 Here, the SCNs are aggregated to a gigabit Ethernet switch
for this solution Pbh can be calculated as [38]:
(GES) via fast Ethernet (FE) connections and the GES is
2
Pbh = Ns Pmodem + NDSLAM (PDSLAM + 2PSF P ) connected to a fiber switch via a 1 Gbps optical fiber link.
F F
To communicate between the GES and fiber switch, an SFP
+ Nsw Psw + 2Nm PSF P + Nul PSF P + (30) interface is used. Another type of aggregation point (a hub
9
equipped with switches inside) is used for macro base station design, the number of OLT interfaces is calculated as, Ng =
backhauling. The traffic from MBS and SCNs is aggregated d NGES−ON U +Nm
nsplitter
e, where nsplitter
ports is the number of passive
ports
to a fiber switch and microwave hub, respectively. SFP+ splitters. For the final section, the number of uplink interfaces
transceivers and 10 Gbps optical fiber links are used to carry for this design can be calculated as, Nul = max(Ng , d U Cmax e),
t
the aggregated traffic from aggregation points (fiber switch int

where Ct is the total aggregated throughput at a particular
or microwave hub) to the core network. The backhaul power hour and Uint max
denotes the maximum transmission rate of an
consumption for this solutions can be expressed as [38]: uplink interface.
3
Pbh F
= (PGES + 2PSF P ) + Nm Plow−c + Nsw F
Psw In the above subsections, we have analyzed the power
MW MW consumption models for existing solutions such as optical P2P
+ Nhub (Phigh−c + PSF P + ) + Nsw Psw + Nul PSF P +
[36], FTTN + VDSL2 [38], FTTB + MW [38], and FTTB + 10
(31)
GPON [40]. For the optical P2P Ethernet solution, the Ethernet
Ns max max
where PGES = d nGES ePGES is the power consumption switch Psw consumes a maximum amount of power (300
ports
of a gigabit Ethernet switch (GES), while nGES W) depends on traffic load. In the case of FTTN + VDSL2
ports denotes
the number of ports per GES, and PGES max
represents the solution, a backhaul power consumption model consists of
F
maximum power consumption of a GES. Plow−c and Phigh−c a fiber switch Psw (300 W) and a DSL access multiplexer
indicate the low and high power consumption region of a PDSLAM (85 W) which contributes high power. FTTB +
microwave antenna [37]. For the first aggregation, the number MW solution has two high power aggregation points such as
MW F
of hubs used by outdoor MBS traffic can be computed as, a microwave switch Psw (53 W) and a fiber switch Psw
Nhub = d nNMmW e, where nM W
is the maximum number (300 W), respectively. For the FTTB + 10 GPON solution,
sup
sup the power hungry component is the gigabit Ethernet Switch
of microwave links that a hub can support. Therefore, the
PGES (50 W). These analyses demonstrate that high backhaul
number of switches inside the hub can be calculated as a
power consumption is caused by several components which
function of aggregated macro BS traffic only and can be
motivate us to design two new energy-efficient solutions. We
MW
written as, Nsw = max(Nhub , d CCMmW e), where CswMW
is the
sw will describe these solutions in the following two subsections.
maximum capacity of a microwave switch and cm is the MBS
MW
traffic. Psw denotes the power consumption of a microwave
E. Energy efficient passive optical network solution
switch inside the hub. Another level of aggregation point uses
the number of fiber switches for carrying the SCN traffic, In this design, each SCN is connected to an optical network
i.e., NswF
= d nNFGES e, where NGES is the number of GES. unit (ONU) via an optical fiber link shown in Fig. 5a.
ports
ONUs are connected to an OLT through passive splitters. We
Finally, the number of uplink interfaces is calculated from
consider rack/shelf OLT model [62] for 10 GPON technology,
both aggregation points (fiber switches and microwave hubs)
where an OLT is fully equipped with maximum configuration,
as, Nul = max(Nhub + Nsw F
, d UC
max e).
t
int implementing layer-2 aggregation functionality. Each OLT
has a shelf rack, 9 line cards, SFP+ arrays, and 72 GPON
D. Fiber to the building and 10 Gbps passive optical network ports (2.5 Gbps/port). To guarantee a 100 Mbps data rate
solution for each ONU (SCN) units, the maximum number of ONUs
This solution consists of a fiber to the building (FTTB) (SCNs) supported by one GPON port can be calculated as:
scheme with GPON (gigabit passive optical network) technol- max
NBS = 2.5Gbps/100M bps = 25 ONU (SCN). We calculate
ogy as presented in Fig. 4d. The wireless backhaul traffic is the number of OLT ports according to traffic load of a 5G
carried to the core network through a passive optical network HetNet. The aggregated traffic from ONUs to OLT is sent to
architecture. The SCNs are connected to a GES (gigabit the core network by using a 10 Gbps optical fiber link and
Ethernet switch) by fast Ethernet (FE) links. GES traffic is SFP+ modules. Therefore, the backhaul power consumption
pon
sent to the optical network unit (ONU) via a gigabit Ethernet Pbh for this proposed model can be obtained as:
(GE) port. The MBS is connected to an ONU through a 1 pon
Pbh = (NSCN −ON U + Nm )Po
Gbps optical fiber link. Traffic from all ONUs is aggregated
to optical line terminal (OLT) via passive splitters, which are + Ng Pg + Nul PSF P + (33)
in turn connected to the core network. In the uplink, each where NSCN −ON U and Nm represent active SCN number
ONU shares 2.5 Gbps bandwidth and in the downlink, OLT and macro base station numbers at a particular hour during
broadcasts 10 Gbps to the ONUs. Therefore, the backhaul a day. Pg is the power consumption of each GPON port and
4
power consumption Pbh for this solution can be computed Po is the power consumption of an ONU. In this proposed
as [39], [40]: design, the number of GPON ports in an OLT is calculated
4
Pbh = NGES−ON U (PGES + Po ) + Nm Po as, Ng = d NNs +N
max e. Finally, to transmit the aggregated traffic
m
BS
to the core network, the number of uplink interfaces can be
+ Ng Pg + Nul PSF P + (32)
computed as, Nul = max(Ng , d U C max e).
t
int
where PGES , Po , Pg and PSF P correspond to the power
consumption of GES, ONU, OLT port and SFP+ module, re- F. Energy efficient millimeter wave solution
Ns
spectively. NGES−ON U = d nGES e is the number of required In this solution, we use the unlicensed 60 GHz frequency
ports
GES-ONU, while nGES
ports is the port number of GES. In this band for the mmWave technology in order to connect SCNs
10
(a) Proposed PON Solution. (b) Proposed mmWave Solution.

Fig. 5: Proposed Backhaul Solutions.
to the MBS as illustrated in Fig. 5b. The point-to-point line- SINR = SNR. Therefore, the radio frequency transmission
of-sight (LOS) mmWave backhaul links are denoted by a set power can be written as [33]:
M = {M1 , M2 , . . . , Mk , . . . , MJ }, where each mmWave bh,tx min
backhaul link k is represented by a set of Mk that includes all PM k
= SN RM k
− Gtx − Grx
SCNs which backhaul their traffic through it, i.e., ∀j ∈ Mk . + LP + LP0 + Li + Lsh + La + Nth + NF (36)
A MBS is connected to the optical network unit which in turns
sends traffic to the optical line terminal. SFP+ modules and a where Gtx and Grx represent the transmission and receiving
10 Gbps optical fiber link are used to carry the aggregated antenna gain, respectively. LP = 20log10 ( 4πd
λ ) dB is the toler-
traffic to the core network. Therefore, all SCN traffic is able path loss, while wavelength λ = 5×10−3 m is for 60 GHz
backhauled by mmWave and the rest of the traffic is carried and d is the length of each BH link. LP0 = 20log10 ( 4π×1 λ )
by PON. The backhaul power consumption Pbh mmW ave
for this dB is the path loss at 1 meter distance. We consider imple-
proposed solution can be calculated as: mentation loss Li , shadowing loss Lsh , and attenuation loss
X La for this design. It is assumed that 16 dB/km attenuation
mmW ave bh,f ixed bh,tx
Pbh = (PM k
+ PM k
) + Nm Po occurs due to oxygen absorption in the atmosphere and 18
Mk ∈M dB/km attenuation for the 50 mm/h rainfall, which ensures
+ Ng Pg + Nul PSF P + (34) 99.995% availability. For each backhaul link bandwidth WMk ,
the thermal noise Nth = 10log10 (WMk )−174 (dBm), and the
bh,tx
where PM k
is the load dependent radio frequency transmit noise figure NF (dB) are estimated.
power consumption for any link. The link can be from an
SCN to an SCN aggregation node or an SCN aggregation node
bh,f ixed V. M ODEL FOR N ETWORK Q O S
to an MBS. PM k
denotes the fixed power consumption
by each backhaul link. Pg , Po and PSF P + represent the Future telecommunication systems must be able to support
power consumption of an OLT port, an ONU and SFP+ the massive deployment of switches, routers, and gateways.
modules, respectively. In this proposed solution, the number of Congestion and queues are considered as generic features of
GPON ports in an OLT can be calculated as, Ng = d NNmax m
e. packet-switched networks [64]. To handle the bursty traffic and
BS
Lastly, for transmitting the aggregated traffic from the MBS ensure throughput, the network needs to use packet buffering
to the core network, the number of uplink interfaces can be properly or it will introduce unpredictable and variable delays
computed as, Nul = max(Ng , d U C max e). Effective isotropic
t (jitter) when packets traverse across networks. In this section,
int
radiated power (EIRP) of 60 GHz mmWave band is limited to we analyze the average delay and jitter of a 5G HetNet.
maximum 40 dBm. In addition, we consider an adaptive mod- We consider Poisson packet arrivals with arbitrary length of
ulation and coding technique (AMC) to maintain a minimum packet distribution in a downlink direction (core network to
achievable SINR for successfully transmitting the aggregated user). Any network delay consists of several elements, such
traffic through a backhaul link. The minimum SINR can be as propagation delay, transmission delay, and queuing delay.
written as [63]: The end-to-end packet network delay of a 5G HetNet can be
k
!
Cs
written as, dt = dan + dbh , where dan represents the access
min
SIN RM = 10log10 2 WMk − 1 (35) network delay and dbh denotes the backhaul network delay.
k
We compute the access network delay from [65] as follows:
2
where WMk denotes the bandwidth and Csk is the aggregated σ
∆m L̄L + L̄
capacity generated from all SCNs of the link k. Due to high dan = PZ (37)
path loss at 60 GHz, generated interference is negligible, i.e., 2 z=1 dz (1 − ∆m )
11
where ∆m denotes the traffic load of the macrocell base TABLE I: Simulation Parameters
2
station, σL represents the packet size variance, L̄ denotes the (a) Access Network
mean packet size, and dz indicates the transmission rate for a Parameters Value
sub-zone z. The backhaul network delay can be expressed as fc 2.0 GHz
[66]: Wm 10 MHz
2 dynamic
σ Ptx 43 dBm
L̄ ρ L̄L + L̄ j,dynamic
Pq,tx 17 dBm
dbh = τT + + (38) f ixed
Wbh 2Wbh (1 − ρ) Pm 130 W
where τT denotes the propagation delay, and ρ denotes the Pqj,f ixed 4.8 W
∆m 4.7
relative traffic load, defined as traffic rate λ normalized with ∆s 8
respect to the backhaul channel bandwidth Wbh . We measure Spectrum Allocation Partitioned
the average end-to-end network delay davg t as the total delay Traffic Model Capacity Full Buffer
Environment Sub-urban
experienced by the total number of transferred packets Npkt
and can be expressed as: (b) Backhaul Network
Npkt Parameters Value
1 X (i)
davg
t = d (39) max
Ndl 24
Npkt i=1 t α 0.9
max /P dl /P ul
Psw 300/1/1 W
sw sw
Another major QoS parameter for a telecommunication net- Pmodem /PDSLAM /Psw F 5/85/300 W
work is the delay variation (Jitter). It is defined as the mean PSF P /PSF P + /Plow−c /Phigh−c 1/1/37/92.5 W
max /P M W /P /P
PGES 50/53/2.9/5 W
deviation of the difference in packet spacing at the receiver sw g o
splitters
compared to the sender, for a pair of packets. The minimum nports /nports /nGES
D F
ports /nports /nM W
sup 16/24/12/24/16
max /C M W
Agmax /Uint 24/10/36 Gbps
value of jitter is expected to maintain the network QoS. In this mmW ave /W
sw
fbh 60/1.76 GHz
research, we calculate the average end-to-end jitter jtavg as the d
Mk
100 m
average deviation of total end-to-end delays from the average LP /LP0 /Li /Lsh /La 108/68/4/1/3.2 dB
end-to-end delay:
" Npkt #1/2
avg 1 X (i) avg 2 for a real-time bi-directional constant bit rate (CBR) agent
jt = p (dt − dt ) (40)
Npkt − 1 i=1 with the aid of the user datagram protocol (UDP) protocol.
Streaming traffic (medium priority), such as streaming audio
VI. R ESULTS AND D ISCUSSION and video, has relatively less delay sensitivity. This type of
In this section, we analyze and present the power consump- traffic deployed a unidirectional variable bit rate (VBR) agent
tion, energy efficiency and QoS (throughput, delay, and jitter) of the user datagram protocol (UDP) whereas, background
of the proposed small cell access network and their backhaul traffic (low priority) used the file transfer protocol (FTP)
solutions. sources with a transmission control protocol (TCP) agent. For
this simulation, we assumed the packet size for conversational,
A. Simulation Setup streaming, and background users as 70 Bytes, 1000 Bytes,
and 80 Bytes, respectively. Correspondingly maximum user
We evaluated the system performance through simulations
bit rates were 64 kbps, 4 Mbps, and 0.5 Mbps, respectively. A
in Network Simulator 2 (NS-2) [67]. The proposed 5G ar-
typical temporal hour-by-hour distribution of normalized peak
chitecture considered a circular MBS located at the center
traffic loads is presented in Fig. 2 [1], [51]. Call arrivals were
with a radius of 1000 m [68], where zone 1 covered from
modeled using a Poisson distribution, where the peak traffic
the center up to 300 m, zone 2 covered from 301 m to 600
load matches the profile presented in Fig. 3 for a suburban
m and zone 3 covered from 601 m to 1000 m [69]. All small
area. Detailed simulation parameters for access networks are
cells were assumed to have the same circular shape and size
listed in Table Ia. We also implemented the four existing
and were randomly located within the macrocell coverage
(Fig. 4) and two proposed solutions (Fig. 5) in NS-2 and
area to support the increased network capacity. We used
investigated their performance. The backhaul links are defined
the 3GPP propagation model [70] for implementing realistic
according to the corresponding topology and technology. For
heterogeneous scenarios. The daily forecast was computed
example, the bandwidth of optical fiber, mmWave, microwave
at the beginning of the day, and the transmission power
and VDSL2 technology is listed as 2500 Mbps, 1760 Mbps,
was updated during each hour to guarantee the requested
500 Mbps, and 300 Mbps, respectively. Note that backhaul
capacity. The macrocell transmitted signals through directional
networks have to be setup with access networks. Detailed
antennas and SCNs used omnidirectional antennas. The SCN
simulation parameters for backhaul networks are listed in
placements were assumed to be user correlated with closed
Table Ib.
subscriber group implementation.
In this simulation model, we considered three types of traf-
fic, e.g., conversational, streaming, and background. Conversa- B. Access Network Analysis
tional traffic (high priority), such as voice and video conferenc- We minimize access network power consumption by cal-
ing, is highly delay sensitive. This type of traffic was simulated culating the optimum number of active SCNs as discussed in
12
1
section III. The normalized number of required active SCNs
is depicted in Fig. 6a, where the percentage of active SCN
0.9
numbers from lowest to highest order are 0% (5 am), 30% (4
0.8
pm), 60% (8 pm), and 100% (10 pm). The proposed solution
Normalized Number of SCN
0.7
is compared with the approach used in [71], [72]. A load
0.6
factor [0,1] was considered in their design for each hour in
0.5
a day by estimating the maximum, minimum, and average
0.4
traffic collected over a week, where 0 indicates that load-
0.3
dependent components are in sleep mode, and 1 indicates that
0.2
the cell consumes the highest amount of power. A heuristic
0.1 solution was provided, but no hourly QoS constraints of traffic
0
06 08 10 12 14 16 18 20 22 00 02 04
demand were taken into account in their research [71], [72].
Time (Hour) Conversely, we calculate the optimum number of active SCNs
(a) Normalized Number of Active SCNs. by measuring the bandwidth requirement according to QoS
constraints at various hours of a day, which ultimately leads
1700
Proposed Model to low power consumption without sacrificing QoS. As shown
Access Network Power Consumption (Watt)
Existing Model
1600 in Fig. 6b, which plots access network power consumption
against time, we observe that during low traffic periods (02
1500
am to 10 am), the reduced number of active SCNs deployed,
1400 according to the proposed HetNet model, improves the energy
efficiency by up to 45%, whereas during high traffic periods
1300
(06 pm to 09 pm), it shows almost 18% less power consump-
1200 tion than the existing approach [71]. Even at the peak traffic
period (10 pm), the proposed model consumes 2% less power
1100
than the existing approach, which highlights the significant
1000 energy savings.
06 08 10 12 14 16 18 20 22 00 02 04
Time (Hour) Therefore, the proposed system can be an attractive solution
(b) Power Consumption. for 5G HetNet operators, particularly for network planning
and resource management. For throughput measurement, we
160
Offered Traffic Streaming
maintain the strict priority scheduling for the conversational
140
Received Traffic Streaming
Offered Traffic Conversational
(high priority) and streaming traffic (medium priority) when
120
Received Traffic Conversational
Offered Traffic Background
entering the network. The call admission procedure checks the
Received Traffic Background
availability of the network capacity so that each call can be
Throughput (Mbps)
100
served without being dropped. If the downlink data rate is not
80 sufficient for a certain traffic demand, outage probabilities can
60 occur. We observe that there is a minor difference between
the offered and received traffic, which confirms the QoS for
40
conversational traffic shown in Fig. 6c. For streaming traffic,
20 the received traffic load follows the offered load until 10 am.
0 The gap increases during the peak hours until 10 pm as traffic
06 08 10 12 14 16 18 20 22 00 02 04
Time (Hour) demand increases (Fig. 6c). The number of active SCNs in our
system model helps to maintain the QoS for the high priority
(c) Throughput.
conversational and medium priority streaming traffic. Both
35
Conversational
the conversational and streaming traffic are delay sensitive
30
Streaming
Background
and we calculate the average end-to-end delay for different
services. From Fig. 6d, the average delay for conversational
Access Netwrok Delay (ms)
25
and streaming traffic is reasonably low for most of the time
20
(from 12 pm to 10 pm) when the service demand is relatively
high. Conversational traffic delay is lower than streaming
15
traffic as it is more delay sensitive.
10
5 C. Backhaul Network Analysis

0
In this section, we present simulated results in terms of
06 08 10 12 14 16 18 20 22 00 02 04
Time (Hour)
power consumption and energy efficiency for various backhaul
solutions. Note that we consider the same amount of traffic
(d) Average Delay.
for backhauling at a particular hour during the 24-hour time
Fig. 6: Access Network Analysis periods. Figure 7a illustrates the backhaul power comparison
13
among different solutions during various hours of the day. This 700
figure reveals that the FTTB + MW [38] and the FTTN +
VDSL2 [38] solutions consume relatively higher power than 600
Backhaul Power Consumption (Watt)

other solutions due to their use of electrical switches (e.g., fiber 500
and microwave switches). It is apparent that during low (02
am to 10 am) and medium traffic periods (12 pm to 6 pm), our 400
proposed wireless mmWave solution consumes less power than 300

Optical P2P
FTTN+VDSL2
any other existing solution. The rationale is that we consider 60 FTTB+MW

FTTB+1OGPON
GHz mmWave technology assuming 100 m distance between 200 GPON (Proposed)
mmWave (Proposed)
SCNs, which requires a transmission power less than 10 dBm. 100
The overall backhaul power for this solution mainly depends
on the transmit power consumption. As can be seen from Fig. 0
06 08 10 12 14 16 18 20 22 00 02 04
7a, during low traffic periods (02 am to 10 am), the overall Time (Hour)
backhaul power consumption can reach up to a maximum 20 (a) Power Consumption.

W because fewer SCNs are required. Figure 7a also highlights
4.5
that our proposed wired PON solution consumes less power Optical P2P
4 FTTN+VDSL2
during high traffic periods (08 pm to 00 am) than any other
Backhaul Energy Efficiency (Mbps/Watt)

FTTB+MW
FTTB+1OGPON
solution. At peak traffic load (10 pm), the proposed wired PON 3.5
GPON (Proposed)
mmWave (Proposed)
solution consumes the least backhaul power, but the mmWave 3
solution consumes more power than the FTTB + 10 GPON 2.5
[40] solution due to the increase in transmission power from 2

highly dense SCN deployments. As Fig. 7a shows, the power
1.5
consumption difference between our proposed PON and the
1
existing FTTB + 10 GPON [40] solution reaches almost 50
W during low traffic periods (02 am to 10 am), and 30 W 0.5
during high traffic periods (08 pm to 00 am). The relative 0

06 08 10 12 14 16 18 20 22 00 02 04
power saving between these two configuration reaches up to Time (Hour)
39% during low traffic periods and up to 13% during peak (b) Energy Efficiency.
traffic period (10 pm). The saving between the proposed PON
55
and existing optical P2P Ethernet [36] solution is more than Optical P2P
50% for almost all traffic periods. This indicates that both of 50 FTTN + VDSL2
FTTB + MW
our proposed models are energy-efficient at various hours of 45 FTTB + 10GPON
Backhaul Network Delay (ms)
Proposed PON
the day. 40 Proposed mmWave
However, the optical P2P Ethernet [36] solution provides 35
a linear increase of backhaul power consumption with the 30
increasing number of SCNs as the aggregation capacity of the 25
switch increases linearly. The backhaul power consumption 20
for the FTTB + 10 GPON [40] solution, as plotted in Fig. 7a, 15
remains constant during low (02 am to 10 am) and medium 10
traffic (12 pm to 06 pm) periods due to the high aggregation 5

06 08 10 12 14 16 18 20 22 00 02 04
switch capacity. From these analyses, it is obvious that during Time (Hour)
low and medium traffic periods of the day, the mmWave
(c) Average Delay.
solution performs better than other solutions in terms of power
consumption. Our findings also indicate that the mmWave 0.7
solution can be a better choice for low and medium traffic 0.6
periods and the PON solution is the better option during high
Backhaul Network Jitter (ms)
traffic periods. 0.5
Next, we focus on the backhaul energy efficiency for these 0.4

Optical P2P
FTTB + MW
solutions as shown in Fig. 7b. Backhaul energy efficiency is FTTB + 10GPON
Proposed PON
defined as the ratio of the total throughput (carried traffic 0.3 Proposed mmWave
from base stations to core network) to the backhaul power 0.2

consumption. It can be noted that the throughput remains
the same for the access and backhaul network. The findings 0.1
provide evidence that the mmWave design is the most energy- 0

06 08 10 12 14 16 18 20 22 00 02 04
efficient backhauling solution during low and medium traffic Time (Hour)
periods (02 am to 06 pm). The proposed wired PON solution
(d) Average Jitter.
is suitable for high traffic periods (08 pm to 00 am) which
provides an energy efficiency up to 1.2 (Mbps/Watt). The most Fig. 7: Backhaul Network Analysis.
14
2400
striking result to emerge from the analysis is that the backhaul
energy efficiency in the proposed mmWave increases almost
2200
linearly during low traffic periods (06 am to 12 am), and after
Total Power Consumption (Watt)
2000 that, there is a clear decreasing trend during medium and

1800 high traffic periods. The rationale is that during low traffic
1600
periods, as the throughput increases, the rate of backhaul
power consumption does not increase at the same pace.
1400
However, during medium and high traffic periods, the back-
1200 Exiting AN + (Optical P2P)
Exiting AN + (FTTN + VDSL2)
haul power consumption rate increases rapidly due to high
1000 Exiting AN + (FTTB + MW)
Exiting AN + (FTTB + 10GPON)
SCN deployments. From these analyses, it can be concluded
800
Proposed (AN + PON)
Proposed (AN+ mmWave)
that during low and medium traffic periods (02 am to 06 pm),
06 08 10 12 14 16 18 20 22 00 02 04
the mmWave solution is the most energy-efficient solution
Time (Hour) whereas, during high traffic periods (08 pm to 00 am),
(a) Power Consumption. the proposed PON is the most energy-efficient solution. We
present the average end-to-end backhaul delay for all solutions
0.1
Exiting AN + (Optical P2P) in Fig. 7c, which indicates a clear trend of decreasing delay
0.09 Exiting AN + (FTTN + VDSL2)
Exiting AN + (FTTB + MW) while shifting traffic from low to high periods (06 am to 10
Total Energy Efficiency (Mbps/Watt)
Exiting AN + (FTTB + 10GPON)

0.08
Proposed (AN + PON) pm) because of the use of a large number of active SCNs.
0.07 Our proposed PON solution provides less delay at the peak
0.06 traffic period (10 pm) while the proposed mmWave solution
0.05
also shows comparatively similar behavior to other existing
solutions. The FTTN + VDSL2 [38] solution exhibits a greater
0.04
average delay for the bandwidth constraints of the DSL
0.03
technology. The average delay variation (jitter) is illustrated
0.02 in Fig. 7d for all solutions except the FTTN + VDSL2 [38]
0.01
06 08 10 12 14 16 18 20 22 00 02 04
solution as the delay variation for this solution is very large.
Time (Hour) Our findings also suggest that both of our proposed solutions
(b) Energy Efficiency. (PON and mmWave) exhibit less average jitter (below 0.1 ms)
during all time periods, which confirms better QoS.
80
Proposed (AN + PON)
70
D. Communication Model Analysis
End-to-End Network Delay (ms)
60
The previous two sections show the access and backhaul
50
network power consumption, respectively. In this section, we
evaluate the total network power consumption and energy
40
efficiency (i.e., access plus backhaul) of the proposed system
30
models against existing models.
As shown in Fig. 8a, during low traffic periods (02 am
20
to 10 am) our proposed green communication model shows
10 almost 48% less power consumption than the optical P2P [36]
06 08 10 12 14 16 18 20 22 00 02 04
Time (Hour) and FTTB + 10 GPON [40] solutions. This saving reaches
almost 80% on average when compared with the FTTN +
(c) Average Delay.
VDSL2 [38] and FTTB + MW [38] solutions. Simulation
0.4 results also indicate that during peak load (10 pm), our green
Proposed (AN + PON)
0.38 Proposed (AN+ mmWave) communication model consumes almost 33% less power than
existing solutions except for the FTTB + 10 GPON [40]. From
End-to-End Network Jitter (ms)
0.36
Fig. 8b, it is apparent that during low traffic periods, our
0.34
proposed solutions are 35% more energy-efficient than other
0.32 existing solutions. Even at the peak traffic load (10 pm), our
0.3
proposed solutions are 11% more energy-efficient than other
solutions. Our finding from these figures is that our proposed
0.28
solution consumes less power and is more energy-efficient than
0.26 other existing solutions.
0.24
We also analyze the total end-to-end average network delay
06 08 10 12 14 16 18 20 22 00 02 04
Time (Hour)
and jitter for both of our proposed solutions as shown in
Figs. 8c and 8d. During high traffic loads (08 pm to 00 am),
(d) Average Jitter.
both solutions show the average delay below 30 ms.The total
Fig. 8: Communication Model Analysis. network jitter (delay variation) is not excessive (less than 0.4
15
ms) during all traffic periods. As shown in Fig. 8a, the network [9] J. Andrews, S. Buzzi, W. Choi, S. Hanly, A. Lozano, A. Soong, and
consumes the maximum power during the peak period. This J. Zhang, “What will 5G be?” IEEE J. Sel. Areas Commun., vol. 32,
no. 6, pp. 1065–1082, Jun. 2014.
is because a maximum number of active SCNs are required [10] B. Lannoo et al., “Overview of ICT energy consumption,” iMinds
during the peak period. Passive optical and mmWave backhaul Deliverable D8.1, Tech. Rep. FP7-288021, Feb. 2013.
components also need to operate at a higher capacity to meet [11] E. Gelenbe and Y. Caseau, “The impact of information technology on
energy consumption and carbon emissions,” Ubiquity, vol. 2015, no. 1,
the demand during the peak period. The combined affect of pp. 1–15, Jun. 2015.
this high capacity access and backhaul networking together [12] A. Ekti, M. Shakir, E. Serpedin, and K. Qaraqe, “End-to-end downlink
helps to achieve a lower average delay and jitter as shown in power consumption of heterogeneous small-cell networks based on the
Figs. 8c and 8d. probabilistic traffic model,” in Proc. IEEE WCNC, Apr. 2014, pp. 1138–
1142.
[13] Z. Hasan, H. Boostanimehr, and V. K. Bhargava, “Green cellular net-
VII. C ONCLUSION works: A survey, some research issues and challenges,” IEEE Commun.
Surveys Tuts., vol. 13, no. 4, pp. 524–540, 2011.
Network densification using small cells is expected to play [14] A. Fehske, F. Richter, and G. Fettweis, “Energy efficiency improvements
a major role in the bandwidth scarcity problem of 5G systems. through micro sites in cellular mobile radio networks,” in Proc. IEEE
However, network densification introduces another dimension GLOBECOM Workshops, Nov. 2009, pp. 1–5.
[15] F. Richter, G. Fettweis, M. Gruber, and O. Blume, “Micro base stations
of problem in the form of higher power consumption. The in load constrained cellular mobile radio networks,” in Proc. IEEE 21st
aim of this research was to investigate the impact of network Int. Sym. PIMRC, Sep. 2010, pp. 357–362.
densification on power consumption, from both the access and [16] H. Claussen, L. Ho, and F. Pivit, “Effects of joint macrocell and
backhaul networking perspectives. Our investigation led to the residential picocell deployment on the network energy efficiency,” in
Proc. IEEE 19th Int. Sym. PIMRC, Sep. 2008, pp. 1–6.
development of an analytical model for 5G access networks [17] Y. Hou and D. Laurenson, “Energy efficiency of high QoS heterogeneous
that considers the temporal variation of traffic and calculates wireless communication network,” in Proc. IEEE 72nd VTC Fall, Sep.
the optimum number of required active SCNs during various 2010, pp. 1–5.
[18] D. Calin, H. Claussen, and H. Uzunalioglu, “On femto deployment
hours of the day. We also introduced two backhauling solu- architectures and macrocell offloading benefits in joint macro-femto
tions which complement the proposed energy-efficient access deployments,” IEEE Commun. Mag., vol. 48, no. 1, pp. 26–32, Jan.
networking solution and help to achieve lower overall energy 2010.
[19] R. Xie, F. R. Yu, H. Ji, and Y. Li, “Energy-efficient resource allocation
consumption. From the outcome of our investigation, it can for heterogeneous cognitive radio networks with femtocells,” IEEE
be concluded that our green communication model consumes Trans. Wireless Commun., vol. 11, no. 11, pp. 3910–3920, Nov. 2012.
a significant amount of less power than existing solutions. [20] K. Davaslioglu and E. Ayanoglu, “Quantifying potential energy effi-
An important conclusion drawn from this research includes ciency gain in green cellular wireless networks,” IEEE Commun. Surveys
Tuts., vol. 16, no. 4, pp. 2065–2091, 2014.
that the increased power consumption problem attributed by [21] T. Yang, F. Heliot, and C. H. Foh, “A survey of green scheduling schemes
network densification, can be mitigated to a great extent by for homogeneous and heterogeneous cellular networks,” IEEE Commun.
using strategies, where redundant elements are put into sleep Mag., vol. 53, no. 11, pp. 175–181, Nov. 2015.
[22] I. Ashraf, F. Boccardi, and L. Ho, “Sleep mode techniques for small
mode. Our proposed solution is an example of such strategies. cell deployments,” IEEE Commun. Mag., vol. 49, no. 8, pp. 72–79,
Another finding from this research includes that a single Aug. 2011.
backhauling technology (mmWave or Optical) is not sufficient [23] M. A. Marsan, L. Chiaraviglio, D. Ciullo, and M. Meo, “Multiple daily
base station switch-offs in cellular networks,” in Proc. IEEE 4th ICCE,
to warrant minimum power consumption during all traffic Aug. 2012, pp. 245–250.
periods of the day. Our future research will investigate the [24] S. Zhang, J. Gong, S. Zhou, and Z. Niu, “How many small cells can be
optimum instant time when traffic backhauling should switch turned off via vertical offloading under a separation architecture?” IEEE
from one technology to another to minimize the overall power Trans. Wireless Commun., vol. 14, no. 10, pp. 5440–5453, Oct. 2015.
[25] Z. Li, D. Grace, and P. Mitchell, “Traffic perception based topology
consumption. We will also explore the possibility of using the management for 5G green ultra-small cell networks,” in Proc. IEEE 1st
BS caching technique to enable longer sleep periods so that Int. CCS Workshop, Sep. 2014, pp. 1–5.
more power savings can be achieved. [26] X. Li, X. Zhang, and W. Wang, “An energy-efficient cell planning
strategy for heterogeneous network based on realistic traffic data,” in
Proc. IEEE ComManTel, Jun. 2014, pp. 122–127.
R EFERENCES [27] A. Y. Saber and G. K. Venayagamoorthy, “Plug-in vehicles and renew-
[1] Cisco Visual Networking Index: Global Mobile Data able energy sources for cost and emission reductions,” IEEE Trans. Ind.
Traffic Forecast Update, 2016-2021, White Paper. [Online]. Electron., vol. 58, no. 4, pp. 1229–1238, Apr. 2011.
Available: http://www.cisco.com/c/en/us/solutions/service-provider/ [28] U. K. Debnath, I. Ahmad, D. Habibi, and A. Y. Saber, “Improving
visual-networking-index-vni/index.html battery lifetime of gridable vehicles and system reliability in the smart
[2] P. Agyapong, M. Iwamura, D. Staehle, W. Kiess, and A. Benjebbour, grid,” IEEE Syst. J., vol. 9, no. 3, pp. 989–999, Sep. 2015.
“Design considerations for a 5G network architecture,” IEEE Commun. [29] J. Allen and F. Chevalier, “Report for vodaphone - mobile backhaul
Mag., vol. 52, no. 11, pp. 65–75, Nov. 2014. market: Phase 2 report,” Analysys Mason, Cambridge, UK, Tech. Rep.
[3] Mobile and Wireless Communications Enablers for the Twenty-Twenty 39013-215, May 2014.
Information Society. [Online]. Available: https://www.metis2020.com/ [30] X. Ge, H. Cheng, M. Guizani, and T. Han, “5G wireless backhaul
[4] 5th Generation Non-Orthogonal Waveforms for Asynchronous networks: challenges and research advances,” IEEE Netw., vol. 28, no. 6,
Signalling. [Online]. Available: http://www.5gnow.eu/ pp. 6–11, Nov. 2014.
[5] 5G Innovation Centre. [Online]. Available: http://www.surrey.ac.uk/5gic [31] C. Dehos, J. L. González, A. De Domenico, D. Kténas, and L. Dussopt,
[6] 5G Infrastructure Public Private Partnership. [Online]. Available: “Millimeter-wave access and backhauling: the solution to the exponential
https://5g-ppp.eu/ data traffic increase in 5G mobile communications systems?” IEEE
[7] Q. Li, H. Niu, A. Papathanassiou, and G. Wu, “5G network capacity: Commun. Mag., vol. 52, no. 9, pp. 88–95, Sep. 2014.
key elements and technologies,” IEEE Veh. Technol. Mag., vol. 9, no. 1, [32] A. Mesodiakaki, F. Adelantado, A. Antonopoulos, E. Kartsakli,
pp. 71–78, Mar. 2014. L. Alonso, and C. Verikoukis, “Energy impact of outdoor small cell
[8] X. Ge, S. Tu, G. Mao, C. X. Wang, and T. Han, “5G ultra-dense cellular backhaul in green heterogeneous networks,” in Proc. IEEE 19th Int.
networks,” IEEE Wireless Commun., vol. 23, no. 1, pp. 72–79, Feb. 2016. CAMAD Workshop, Dec. 2014, pp. 11–15.
16
[33] A. Mesodiakaki, F. Adelantado, L. Alonso, M. D. Renzo, and C. Verik- [55] M. M. Mowla, I. Ahmad, D. Habibi, and Q. V. Phung, “An energy
oukis, “Energy- and spectrum-efficient user association in millimeter- efficient resource management and planning system for 5G networks,”
wave backhaul small-cell networks,” IEEE Trans. Veh. Technol., vol. 66, in Proc. IEEE 14th CCNC, Jan. 2017, pp. 216–224.
no. 2, pp. 1810–1821, Feb 2017. [56] Energy Aware Radio and neTwork tecHnologies. [Online]. Available:
[34] L. Chen, F. R. Yu, H. Ji, G. Liu, and V. C. Leung, “Distributed https://www.ict-earth.eu/
virtual resource allocation in small-cell networks with full-duplex self- [57] M. Arshad, A. Vastberg, and T. Edler, “Energy efficiency gains through
backhauls and virtualization,” IEEE Trans. Veh. Technol., vol. 65, no. 7, traffic offloading and traffic expansion in joint macro pico deployment,”
pp. 5410–5423, Jul. 2016. in Proc. IEEE WCNC, Apr. 2012, pp. 2203–2208.
[35] L. Chen, F. R. Yu, H. Ji, B. Rong, X. Li, and V. C. Leung, “Green [58] K. Samdanis, P. Rost, A. Maeder, M. Meo, and C. Verikoukis, Green
full-duplex self-backhaul and energy harvesting small cell networks with Communications: Principles, Concepts and Practice. UK: John Wiley
massive mimo,” IEEE J. Sel. Areas Commun., vol. 34, no. 12, pp. 3709– & Sons, 2015.
3724, Dec. 2016. [59] Y. F. Liu and Y. H. Dai, “On the complexity of joint subcarrier and
[36] S. Tombaz, P. Monti, K. Wang, A. Vastberg, M. Forzati, and J. Zander, power allocation for multi-user OFDMA systems,” IEEE Trans. Signal
“Impact of backhauling power consumption on the deployment of Process., vol. 62, no. 3, pp. 583–596, Feb. 2014.
heterogeneous mobile networks,” in Proc. IEEE GLOBECOM, Dec. [60] J. Currie and D. I. Wilson, “OPTI: Lowering the Barrier Between
2011, pp. 1–5. Open Source Optimizers and the Industrial MATLAB User,” in Proc.
[37] P. Monti, S. Tombaz, L. Wosinska, and J. Zander, “Mobile backhaul Foundations of Computer-Aided Process Operations, Jan. 2012, pp. 8–
in heterogeneous network deployments: Technology options and power 11.
consumption,” in Proc. IEEE 14th ICTON, Jul. 2012, pp. 1–7. [61] W. Vereecken, W. V. Heddeghem, M. Deruyck, B. Puype, B. Lannoo,
[38] S. Tombaz, P. Monti, F. Farias, M. Fiorani, L. Wosinska, and J. Zander, W. Joseph, D. Colle, L. Martens, and P. Demeester, “Power consumption
“Is backhaul becoming a bottleneck for green wireless access networks?” in telecommunication networks: overview and reduction strategies,”
in Proc. IEEE ICC, Jun. 2014, pp. 4029–4035. IEEE Commun. Mag., vol. 49, no. 6, pp. 62–69, Jun. 2011.
[39] M. Mahloo, P. Monti, J. Chen, and L. Wosinska, “Cost modeling of [62] European Commission, “Technical Assessment and Comparison of
backhaul for mobile networks,” in Proc. IEEE ICC, Jun. 2014, pp. 397– Next-generation Optical Access System Concepts,” OASE Deliverable
402. 4.2.2, Tech. Rep. FP7- ICT-GA249025, Jun. 2013.
[40] L. Surez, M. A. Bouraoui, M. A. Mertah, M. Morvan, and L. Nuaymi, [63] K.-C. Huang and Z. Wang, Millimeter wave communication systems.
“Energy efficiency and cost issues in backhaul architectures for high NJ:John Wiley & Sons, 2011.
data-rate green mobile heterogeneous networks,” in Proc. IEEE 26th [64] M. Marchese, QoS over heterogeneous networks. UK:John Wiley &
Int. Sym. PIMRC, Aug. 2015, pp. 1563–1568. Sons, 2007.
[41] C. S. Ranaweera, P. P. Iannone, K. N. Oikonomou, K. C. Reichmann, and [65] H. Beyranvand, W. Lim, M. Maier, C. Verikoukis, and J. A. Salehi,
R. K. Sinha, “Design of cost-optimal passive optical networks for small “Backhaul-aware user association in FiWi enhanced LTE-A heteroge-
cell backhaul using installed fibers [invited],” IEEE J. Opt. Commun. neous networks,” IEEE Trans. Wireless Commun., vol. 14, no. 6, pp.
Netw., vol. 5, no. 10, pp. A230–A239, Oct. 2013. 2992–3003, Jun. 2015.
[42] C. Ranaweera, C. Lim, A. Nirmalathas, C. Jayasundara, and E. Wong, [66] F. Aurzada, M. Scheutzow, M. Reisslein, N. Ghazisaidi, and M. Maier,
“Cost-optimal placement and backhauling of small-cell networks,” J. “Capacity and delay analysis of next-generation passive optical networks
Lightw. Technol., vol. 33, no. 18, pp. 3850–3857, Sep. 2015. (NG-PONs),” IEEE Trans. Commun., vol. 59, no. 5, pp. 1378–1388, May
[43] A. H. Jafari, D. López-Pérez, H. Song, H. Claussen, L. Ho, and J. Zhang, 2011.
“Small cell backhaul: challenges and prospective solutions,” EURASIP [67] Network Simulator 2. [Online]. Available: http://www.isi.edu/nsnam/ns/
J. Wireless Commun. Netw., vol. 2015, no. 1, p. 206, 2015. [Online]. [68] F. Richter and G. Fettweis, “Cellular mobile network densification
Available: https://link.springer.com/article/10.1186/s13638-015-0426-y utilizing micro base stations,” in Proc. IEEE ICC, May 2010, pp. 1–
[44] Z. Shen, J. Andrews, and B. Evans, “Adaptive resource allocation in 6.
multiuser OFDM systems with proportional rate constraints,” IEEE [69] C. Tarhini and T. Chahed, “On capacity of OFDMA-based IEEE802.16
Trans. Wireless Commun., vol. 4, no. 6, pp. 2726–2737, Nov. 2005. WiMAX including adaptive modulation and coding (AMC) and inter-
[45] V. Chandrasekhar and J. G. Andrews, “Spectrum allocation in tiered cell interference,” in Proc. IEEE 15th LANMAN Workshop, Jun. 2007,
cellular networks,” IEEE Trans. Commun., vol. 57, no. 10, Oct. 2009. pp. 139–144.
[46] W. C. Cheung, T. Q. S. Quek, and M. Kountouris, “Throughput opti- [70] 3GPP, “Further Advancements for E-UTRA: Physical Layer Aspects
mization, spectrum allocation, and access control in two-tier femtocell (Release 9),” Tech. Rep. 3GPP TR 36.814 V9.0.0, Mar. 2010. [Online].
networks,” IEEE J. Sel. Areas Commun., vol. 30, no. 3, pp. 561–574, Available: http://www.qtc.jp/3GPP/Specs/36814-900.pdf
Apr. 2012. [71] M. Deruyck, W. Joseph, and L. Martens, “Power consumption model
[47] A. S. Hamza, S. S. Khalifa, H. S. Hamza, and K. Elsayed, “A survey on for macrocell and microcell base stations,” Trans. Emerging Tel.
inter-cell interference coordination techniques in OFDMA-based cellular Techno., vol. 25, no. 3, pp. 320–333, 2014. [Online]. Available:
networks,” IEEE Commun. Surveys Tuts., vol. 15, no. 4, pp. 1642–1670, http://onlinelibrary.wiley.com/doi/10.1002/ett.2565/abstract
2013. [72] M. Deruyck, E. Tanghe, W. Joseph, and L. Martens, “Characterization
[48] B. M. Hambebo, M. M. Carvalho, and F. M. Ham, “Performance evalu- and optimization of the power consumption in wireless access networks
ation of static frequency reuse techniques for ofdma cellular networks,” by taking daily traffic variations into account,” EURASIP J. Wireless
in Proc. IEEE 11th ICNSC, Apr. 2014, pp. 355–360. Commun. Netw., vol. 2012, no. 1, p. 248, Aug. 2012. [Online]. Available:
[49] E. Oh, B. Krishnamachari, X. Liu, and Z. Niu, “Toward dynamic energy- https://link.springer.com/article/10.1186/1687-1499-2012-248
efficient operation of cellular network infrastructure,” IEEE Commun.
Mag., vol. 49, no. 6, Jun. 2011.
[50] G. Auer, V. Giannini, C. Desset, I. Godor, P. Skillermark, M. Olsson,
M. Imran, D. Sabella, M. Gonzalez, O. Blume, and A. Fehske, “How
much energy is needed to run a wireless network?” IEEE Wireless
Commun., vol. 18, no. 5, pp. 40–49, Oct. 2011.
[51] M. Ahmed, I. Ahmad, and D. Habibi, “Service class resource manage-
ment for green wireless-optical broadband access networks (WOBAN),” Md Munjure Mowla received M.Sc Engineering
J. Lightw. Technol., vol. 33, no. 1, pp. 7–18, Jan. 2015. degree in Electrical & Electronic Engineering from
[52] 3GPP, “Quality of service (QoS) concept and architecture,” Rajshahi University of Engineering & Technology
Tech. Rep. ETSI TS 123.107 V10.2.0, Rel 10, Jan. 2011. (RUET), Bangladesh, in 2013. He is currently pursu-
[Online]. Available: http://www.etsi.org/deliver/etsi ts/123100 123199/ ing the Ph.D. degree with the School of Engineering,
123107/10.02.00 60/ts 123107v100200p.pdf Edith Cowan University, Australia. His research in-
[53] R. Fantacci, D. Marabissi, D. Tarchi, and I. Habib, “Adaptive modulation terests include network planning, optimization, and
and coding techniques for OFDMA systems,” IEEE Trans. Wireless resource management for small cells heterogeneous
Commun., vol. 8, no. 9, pp. 4876–4883, Sep. 2009. network and backhaul network in 5G technologies.
[54] A. Ladanyi, D. Lopez-Perez, A. Juttner, X. Chu, and J. Zhang, “Dis-
tributed resource allocation for femtocell interference coordination via
power minimisation,” in Proc. IEEE GLOBECOM Workshops, Dec.
2011, pp. 744–749.
17
Iftekhar Ahmad is currently working as a Senior Quoc Viet Phung is currently working as a post doc-
Lecturer with the School of Engineering, Edith toral research fellow with the School of Engineering,
Cowan University, Australia. He received the Ph.D. Edith Cowan University, Australia. He received the
degree in communication networks from Monash Ph.D. degree in communication networks from Edith
University, Australia, in 2007. His research interests Cowan University, Australia, in 2010. His research
include 5G technologies, green communications, interests include 5G technologies, green communi-
QoS in communication networks, software-defined cations, wireless sensor networks and computational
radio, wireless sensor networks and computational intelligence.
intelligence.
Daryoush Habibi (M95- SM99) ) graduated with a

Bachelor of Engineering (Electrical) with First Class
Honours from the University of Tasmania in 1989
and a PhD from the same University in 1994. His
employment history includes Telstra Research Lab-
oratories, Flinders University, Intelligent Pixels Inc.,
and Edith Cowan University, where he is currently
a Professor and the Executive Dean of Engineering.
His research interests include engineering design for
sustainable development, reliability and quality of
service in communication systems and networks,
smart energy systems, and environmental monitoring technologies. He is a
Fellow of Engineers Australia, and Editor-in-Chief of the Australian Journal
of Electrical and Electronic Engineering.

A Green Communication Model For 5G Systems

Uploaded by

Copyright:

Available Formats

A Green Communication Model For 5G Systems

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Green Communication Model For 5G Systems

Uploaded by

Copyright:

Available Formats

This article has been accepted for publication in a future issue of this journal, but has not been

A Green Communication Model for 5G Systems

Normalized Traffic Load

ηz (u) The main target of our research is to support all traffic

represents the fixed power consumption of an SCN BS.

(c) FTTB + MW Solution [38]. (d) FTTB + 10 GPON Solution [40].

the aggregated traffic from aggregation points (fiber switch int

(a) Proposed PON Solution. (b) Proposed mmWave Solution.

5 C. Backhaul Network Analysis

Backhaul Power Consumption (Watt)

proposed wireless mmWave solution consumes less power than 300

any other existing solution. The rationale is that we consider 60 FTTB+MW

backhaul power consumption can reach up to a maximum 20 (a) Power Consumption.

Backhaul Energy Efficiency (Mbps/Watt)

solution consumes more power than the FTTB + 10 GPON 2.5

[40] solution due to the increase in transmission power from 2

during high traffic periods (08 pm to 00 am). The relative 0

However, the optical P2P Ethernet [36] solution provides 35

a linear increase of backhaul power consumption with the 30

increasing number of SCNs as the aggregation capacity of the 25

switch increases linearly. The backhaul power consumption 20

for the FTTB + 10 GPON [40] solution, as plotted in Fig. 7a, 15

remains constant during low (02 am to 10 am) and medium 10

traffic (12 pm to 06 pm) periods due to the high aggregation 5

traffic periods. 0.5

Next, we focus on the backhaul energy efficiency for these 0.4

from base stations to core network) to the backhaul power 0.2

provide evidence that the mmWave design is the most energy- 0

2000 that, there is a clear decreasing trend during medium and

Exiting AN + (FTTB + 10GPON)

Daryoush Habibi (M95- SM99) ) graduated with a

You might also like