Published in IET Communications

Received on 1st March 2009

Revised on 8th June 2009
doi: 10.1049/iet-com.2009.0153
In Special Issue on Vehicular Ad Hoc and Sensor Networks
ISSN 1751-8628
Green information and communication
technology: energy efciency in a
motorway model
W. Feng H. Alshaer J.M.H. Elmirghani
Electronic and Electrical Engineering, University of Leeds, Leeds LS2 9JT, UK
E-mail: elwf@leeds.ac.uk
Abstract: The focus in Green ICT (information and communication technology) is on software and hardware
solutions to reduce energy consumption. Here the authors consider the design of vehicular ad hoc networks
(VANETs) to reduce energy consumption because of communications. Based on a realistic transport trafc
scenario on a motorway, M4 in UK, this study computes the energy consumption in a motorway ad hoc
wireless network. This is based on a geographical adaptive delity (GAF) topology management protocol,
where the energy consumption has been analysed for equal-grid, adjustable-grid and genetic algorithm
models. Considering nodes in the VANET network and their transmission ranges, this study provides a network
model to compute the energy consumption of VANETs and compare their energy efciency performance. As a
result, the proposed genetic algorithm model has been found to be more energy efcient than the adjustable-
grid model which is better than the equal-grid model.
1 Introduction
Green ICT (information and communication technology) is a
newfocus with the goal of saving energy by optimising mobile
network communications and ultimately protecting the
natural climate [1]. In terms of the global carbon emissions,
some analysts have reported that ICT accounts for 22.5%
of all harmful emissions, which is equal to the global
aviation industry [2]. According to a report by the Ministry
of Internal Affairs and Communications of Japan [3], ICT
equipment (routers, PCs and other network systems)
consumed about 45 000 000 MWh in 2006 which is 3% of
the country total energy consumption and this gure has
increased about 20% over the past 5 years. This paper
studies optimal and efcient motorway vehicular
communications in terms of energy consumption and
vehicular ad hoc networks (VANETS) connectivity. A
vehicle can communicate with another within its
transmission range by using a vehicle-to-vehicle (V2V)
ad hoc mode. A vehicle can communicate with a transport
trafc centre by using vehicle-to-road (V2R) or vehicle-to-
infrastructure (V2I) communication mode via a base station
(BS) [or access point (AP)], installed along the road, when
it is covered by BS (AP) transmission range. Transport
security and leisure information such as emergency messages
and transport status notications can be propagated (relayed)
from one transceiver to another installed along the road by
using road-to-road (R2R) communication mode. This calls
for new mechanisms and techniques that can optimise the
power consumption of the whole vehicular communication
system. Although wireless transceivers on-board vehicles are
less constrained because they can benet from the vehicle
battery, vehicular communications can involve other wireless
devices such as road transceivers that have to be autonomous
and efcient in terms of energy consumption as they are fed
with solar energy. It is noteworthy that the energy
consumption of transceivers installed along the road can be
roughly between 25 and 40 mAh.
A vehicular node can cooperate with its neighbours and
exchange data within its transmission range which can be
smaller than the distance separating it from the destination.
The transmission energy required to guarantee high quality
of communication increases with distance, and can
subsequently dominate the total energy consumption of a
transceiver [4]. Two transmission techniques are broadly
850 IET Commun., 2010, Vol. 4, Iss. 7, pp. 850860
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-com.2009.0153
www.ietdl.org
used to realise communications in ad hoc networks [5]: single-
hop communications and multi-hop communications. In
single-hop communications, network nodes are able to
cover the whole network, and thus consume energy
according to the longest transmission distance. Hence, with
this xed transmission distance, extra energy is wasted even
when nodes are close to each other. In multi-hop
communications, the total energy consumed for end-to-end
communications increases as a function of both the number
of relaying nodes and each individual transmission distance
(d), where nodes consume energy in a non-linear fashion
with transmission distance according to d
n
(n is the power
index of the channel path loss and usually 2 , n , 4).
Therefore if the hop number is small (while the
transmission distance is large), the energy consumed for
one transmission increases non-linearly. Alternately, if the
hop distance is small (for the same overall end-to-end
distance), the energy consumption will be dominated by
the electronic energy cost in the transceivers and therefore
the total energy increases almost linearly as a function
of the hop number (where the hop number is large). There
is a trade off between hop number and the transmission
range of each hop to obtain the optimal energy efcient.
Geographical adaptive delity (GAF) [6], one of the
topology management protocols, is applied in this paper to
manage the states of network nodes and select the multi-
hop relay nodes that can improve energy efciency. As
shown in Fig. 1, based on vehicular nodes positions
information, the GAF model consists of dividing the
network topology into small virtual grids, where any node
in each grid can communicate with any node in its adjacent
grids. Note that the GAF model ensures that there should
always be in each grid one or more (optimal set of) active
nodes forming the network backbone, whereas the
remaining nodes, which do not transmit or relay packets,
enter into a sleeping state to save more energy.
This paper focuses on saving the energy consumed in
vehicular communications. Our idea consists of enabling
vehicles to adjust dynamically and autonomously their
transmission range based on the data trafc to be
transmitted or relayed, where the relationship between the
optimal transmission range (for energy efciency) and the
network data trafc is derived in this paper. Furthermore,
data trafc is linearly proportional to the transport trafc,
which varies over time and road segments. For example
during the rush hours or over a specic road segment the
vehicle density may be high, and the vehicle density may be
low during other times. Accordingly, vehicles adjust their
transmission range as a function of transport trafc in order
to save more energy.
As a result, more power can be saved in the whole vehicular
communication system while increasing the lifetime of
energy-constrained wireless communication devices
involved in this system. Based on analysis which we have
conducted on real collected vehicular data proles at
inductive loops (ILs) on the M4 motorway in the UK, we
have found out that by enabling vehicles to adjust their
transmission range according to rush hours, the total power
consumed by vehicles distributed along a road segment of
2 km length and over the entire M4 motorway (300 km)
can power approximately 106 and 16 000 home bulbs,
respectively.
The remainder of this paper is organised as follows. In
Section 2, we analyse and discuss some related work. In
Section 3, we explain a vehicles distribution model and the
average generated data trafc from these vehicles based on
statistical analysis of transport trafc data observed at an IL
on the M4 motorway in the UK. In Section 4, we
introduce the energy consumption model and derive the
relationship between optimal transmission range and
network trafc. In Section 5, with the peak time data
trafc model of the M4 motorway, UK, we derive and
compute the energy consumption of different energy
efciency models: equal-grid, adjustable-grid and genetic
algorithm models. In Section 6, we discuss the off peak
time data trafc scenario. In Section 7, we discuss some
analytical and simulation numerical results and compare the
energy consumption results of these models. In Section 8,
we draw a conclusion to this paper.
2 Related work
Many mechanisms have been proposed to save energy in
wireless communication systems. Since the power
consumed in the listening state is less than the power
consumed in the transmitting and receiving states, but
signicantly greater than that in the sleeping state, energy
efciency is achieved if as many as possible nodes are put
into the sleeping state. Based on the fundamental work of
IEEE 802.11 power save mechanism (PSM) [7], the work
of [812] analyses the energy efciency issue and related
performance. The work in [8] proposed a routing protocol
based on PSM to reduce the energy consumption and the
transmission delay. In [9, 10] packet-driven mechanisms
are introduced to induce sleep and wake-up modes in the
nodes, which saved about 60% energy compared to the
Figure 1 Equal-grid and adjustable-grid motorway GAF
model
IET Commun., 2010, Vol. 4, Iss. 7, pp. 850860 851
doi: 10.1049/iet-com.2009.0153 & The Institution of Engineering and Technology 2010
www.ietdl.org
normal PSM. A multi-frequency wake-up mechanism is
proposed in [11] where the nodes utilise different sleeping
durations. A trafc shaping protocol to improve the
performance of normal PSM is proposed in [12] where the
sleeping state is prolonged by shaping the unexpected
trafc packets (only few packets during a nearly empty
duration). It is shown in [12] that about 83% energy is
saved by using the proposed trafc shaping protocol.
Transmission range adjustment as another energy saving
approach has been the focus of numerous studies.
Algorithms are proposed either for the general topology
control purpose [13, 14] or for special tasks [15, 16], for
example routing, data gathering and broadcasting. The
objective of energy saving is usually achieved by computing
the best transmission range based on the geographic
information and the energy model. The work in [17]
proposes a quantitative analysis model for the optimal
transmission range problem in this category. They use
throughput and throughput per unit energy as the
optimisation criteria and conclude that the optimal
transmission power is determined by the network load, the
number of nodes and the network size.
The work in [18, 19] studied the optimisation of radio
transmission range in wireless networks. In [18], the
optimal transmission range that maximises the expected
packet progress in the desired direction was determined for
different transmission protocols in a multihop packet radio
network with randomly distributed terminals. The optimal
transmission ranges were expressed in terms of the number
of terminals in range. It was found that the optimal
transmission radius for slotted ALOHA without capture
capability covers on average eight nearest neighbours in the
direction of the packets nal destination. The study
attempted to improve system throughput by limiting the
transmission interference in a wireless network with heavy
trafc load. It was found in [20] that a higher throughput
can be obtained by transmitting packets to the nearest
neighbour in the forward direction. In [21], the authors
evaluated the optimum transmission ranges in a packet
radio network in the presence of signal fading and
shadowing. A distributed position-based self-recongurable
network protocol that minimises energy consumption was
proposed in [19]. It was shown in [19] that the proposed
protocol can stay close to the minimum energy solution
when it is applied to mobile networks.
The optimisation of transmission range as a system design
issue was investigated in [22]. The wireless network was
assumed to have high node density, and to consist of nodes
with relatively low mobility and short transmission range. As
justied by the assumption of high node density, the authors
further assumed that intermediate routing nodes are always
available at the desired location whenever they are needed.
Considering nodes without power control capability, the
authors argued that the optimal transmission range can be
set at the system design stage. Specically, they showed that
the optimal one-hop transmission progressive distance is
independent of the physical network topology, the number
of transmission sources and the total transmission distance;
and that it only depends on the propagation environment
and radio transceiver device parameters.
A similar assumption was made in [23], even though the
node density was only considered for the energy
consumption of overhearing nodes. They investigated the
problem of selecting an energy-efcient transmission power
to minimise global energy consumption for ad hoc
networks. They concluded that the average neighbourhood
size is a useful parameter in nding the optimal balance
point. The authors in [24] studied the optimal transmission
radius that minimises the settling time for ooding in
large-scale sensor networks. In the paper, the settling time
was dened as the time when all the nodes in the network
have forwarded the ooded packet. Regional contention
and contention delay were then analysed. A bit-metre-per-
joule metric for energy consumption in wireless ad hoc
sensor networks was investigated in [25]. The paper
presented a system-level characterisation of energy
consumption for sensor networks. The study assumed that
the sensor network has a relay architecture, and all the
trafc is sent from sensor nodes towards a distant BS. Also,
it was assumed that the source always chooses, among all
relay neighbours, the one that has the lowest bit-metre-per-
joule metric to relay its data packets. In the analysis, the
power efciency metric in terms of average watt per metre
for each radio transmission was rst calculated, and was
then extended to determine the global energy consumption.
The analysis revealed how the overall energy consumption
varies with transceiver characteristics, node density, data
trafc distribution and BS location.
PSM and transmission range adjustment mechanisms can
be utilised together to achieve energy efciency. In [26] the
authors proposed a linear network (where the nodes are
distributed along a line) and, based on GAF topology
management, analysed both equal and adjustable grid
models. A relationship between optimal transmission range
and network trafc is derived in an equal-grid model, which
is used to calculate the grid lengths and network energy
consumption in an adjustable-grid model. It is shown that
the adjustable-grid model is more energy efcient than the
equal-grid model, but the work assumes that the energy cost
in the receiving state is only because of electronic
consumption; however, the energy consumed for processing
data in the receiving state should not be neglected. In our
proposed model, we have considered the energy consumed
for processing data in the receiving state by deriving a
function that correlates the optimal transmission range of
vehicles with the minimum total network energy
consumption. This allows the vehicles to adjusts their
transmission range in order to save more energy during all
communication states. Furthermore, the one-dimensional
linear network in [26] cannot represent most realistic ad hoc
networks that have two- or three-dimensional networks.
852 IET Commun., 2010, Vol. 4, Iss. 7, pp. 850860
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-com.2009.0153
www.ietdl.org
Our rectangular ad hoc motorway network model is more
realistic.
The energy consumed for processing data in the receiving
state is considered in [27], in which the work uses clustering
mechanisms to discuss energy saving in rectangular wireless
networks. In [27] a relationship between optimal
transmission range and trafc is derived based on a static
trafc scenario. However, the work in [27] only considered
a static trafc scenario, so the result is not suitable for the
VANET dynamic trafc scenario.
The work in [28, 29] analyse the energy distribution and
propose a routing mechanism based on the location of
nodes in linear wireless sensor networks. Note that, the
network trafc is determined by the location of nodes in
the models in these papers, therefore the nature of the
work of [28, 29] is based on the relationship between the
network trafc and the transmission radio range. The
results show that the energy consumption can be balanced
and that the bottleneck associated with the network
lifetime can be relaxed. In this paper, however, we consider
a motorway scenario where vehicular nodes are distributed
in a rectangular space based on a real transport trafc
distribution on the M4 motorway in the UK. Considering
this scenario, we derive the optimal transmission range in
three GAF models which can be deployed to save the
energy consumed in motorway ad hoc networks.
3 Statistics trafc of motorway
M4 in UK
As a rst step, we have carried out some statistical analysis on
vehicular trafc proles observed at inductive loops on the
M4 motorway in the UK. Fig. 2 shows the distribution of
vehicles as it is observed at IL (ID) 2244, on the M4, on
Tuesday (23/04/2002), Friday (26/04/2002) and on
Saturday (27/04/2002). On Tuesday and Friday, there are
two peak times (78 am and 56 pm), which correspond to
the morning and evening work rush hours. On Saturday,
there are two peak times (1112 am and 78 pm), which
correspond to the equivalent leisure rush hours. Fig. 3
provides a comparison of vehicles distributions around ILs
on Tuesday (23/04/2002). The statistical analysis shows
that the distributions of vehicles at the ILs IDs 2244, 2246
and 6584 are almost similar. These results clearly show that
the transport trafc varies on a specic location on the M4
according to the time of the day as well as varying in
different locations on the M4 during a specic time
duration. Fig. 4 illustrates these trends, where vehicle
distributions are shown within different time durations, peak
and off peak time, as well as, at different locations between
IDs 2244 and 2256 on Tuesday.
During a specic time duration, the bit rates and statistical
probabilities of different types of data (speech, video, data,
RFID) are presented in Table 1 [30]. Therefore the average
trafc data transmitted by each vehicle is assumed to
be 64 10
3
3=35 320 10
3
5=65 10 10
6
15=
75 10 10
6
1=6 2:9 Mbits=s which is achievable
based on current standards and expected near future
evolution. Based on vehicle distribution trends within
different time durations and at different locations, Fig. 5
illustrates the data trafc transmitted by these vehicles.
This result will be taken into consideration in modelling
transport trafc on a typical motorway in order to achieve
energy efciency in VANETs. Note that the average speed
of vehicles is about 25 m/s (90 km/h) and the distance
Figure 2 Distribution of vehicles over inductive ID 2244
along 3 days
Figure 3 Distribution of vehicles along 1 day over three ILs:
2244, 2246, 6584
Figure 4 Distribution of vehicles along the offpeak and
peak times
IET Commun., 2010, Vol. 4, Iss. 7, pp. 850860 853
doi: 10.1049/iet-com.2009.0153 & The Institution of Engineering and Technology 2010
www.ietdl.org
between two IDs is 500 m. Therefore the average number of
nodes distributed between two ILs is about 20, where we
assume that they are uniformly distributed between any two
ILs on the M4.
4 Preliminaries
The energy consumed by ad hoc wireless network nodes is the
sum of energy consumed for transmitting, receiving and
listening. Based on [31] the energy consumed per second
by a node in these three states could be, respectively,
calculated as follows
E
t
(e
e
e
a
R
n
)T
t
E
r
(e
e
e
p
)T
r
E
l
e
l
T
l
e
l
(1 T
t
T
r
)
(1)
where e
e
is the energy/second consumed by the transceivers
electronics, e
a
is the energy/second consumed in the
transmitter RF amplier, e
p
is the energy/second consumed
for processing in the receivers and e
l
is the energy/second
consumed for listening to the radio environment, where
e
l
e
e
. e
e
, e
a
and e
p
are determined by the design
characteristics of the transceivers. Based on [27], the typical
values of these parameters are: e
e
0.083 J/S, e
p

0.017 J/S, e
a
2 10
25
J/S/m
2
; n is the power index of
the channel path loss, which is between 2 and 4. R is the
node transmission range; T
t
and T
r
denote, respectively, the
time durations for transmitting and receiving the trafc
data, which can be expressed as follows
T
t
D
t
=d
R
T
r
D
r
=d
R
(2)
where D
t
andD
r
are the transmittedandreceivedtrafc data bits,
respectively, and d
R
is the transmitted or received data rate (bits/
s) of each network node. Thus, during 1 s, T
l
, which denotes the
time for listening radio environment, is: T
l
1 T
t
T
r
,
where 0 T
l
1, thus 0 (1 D
t
=d
R
D
r
=d
R
) 1.
For the static trafc data scenario, where D
t
D
r
D, the
value range of D can be described as follows
0 D
1
2
d
R
(3)
In this paper, we focus on the uplink communication from
vehicles to a BS. Based on GAF mechanism, our motorway
model is divided into virtual grids, and the data trafc is
ultimately forwarded to the BS. Since the distance between
two BSs is about 4 km, in our model we consider a 2 km
(L 2 km) motorway network section with a BS located at
the right end of the network. As we mentioned before, about
20 vehicles are uniformly distributed in 500 m (distance
between two IDs), the average number of nodes in the 2 km is
80 which corresponds to a similar trend to that shown in
Fig. 5 (observe the variation with time of day/rush hour).
Vehicles are classied as normal nodes (only transmits data
out) and head nodes (relay trafc to next grid).
Assume that a constant trafc (D) ows along the
motorway network (from left end to the BS) and the nodes
in the network do not transmit data, the relationship
between the optimal transmission range and network trafc
can be derived as follows. In the equal-grid motorway GAF
model, where all the virtual grids are the same size, L and
W represent the length and width of the network,
respectively, whereas r and R represent the grid length and
the transmission range of nodes in each grid, respectively.
Therefore the number of grids, m, is
m
L
r
(4)
and the transmission range (R
i
) of nodes in the ith grid is
derived based on GAF mechanism as follows
R
i

(2r)
2
W
2
_
(5)
The total energy consumed in the ith grid, E
i
, is the sum of
the energy consumed in the listening, transmitting and
Figure 5 Distribution of data trafc along the offpeak and
peak times
Table 1 Parameters for computing data trafc
speech bit
rate
64 kB/s speech ON/OFF
probability
5/35
video bit rate 320 kB/s video ON/OFF
probability
5/65
data bit rate 10 MB/s data ON/OFF
probability
15/75
RFID bit rate 10 MB/s data ON/OFF
probability
1/12
854 IET Commun., 2010, Vol. 4, Iss. 7, pp. 850860
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-com.2009.0153
www.ietdl.org
receiving states of the nodes in the grid (1), thus
E
i
E
t
E
r
E
l
e
p
T
r
e
l
e
a
R
n
i
T
t
(6)
Therefore based on (2), (4) and (5), the energy consumed in
the entire network can be expressed as follows
E
total
mE
i

L
r
e
p
D
r
e
l
=d
R
e
a

(2r)
2
W
2
_ _ _
n
D
t
_ _
(7)
As mentioned, data trafc is constant through the network,
thus D
t
D
r
D. To compute the minimum energy
consumption of the network nodes, we take the rst
derivative of E
total
in terms of grid-length, r, and let
@E
total
=@r 0. For n 2 (free space propagation scenario)
r
2

e
p
D e
l
=d
R
W
2
e
a
D
4e
a
D
r

e
p
D e
l
=d
R
W
2
e
a
D
4e
a
D
_ (8)
where r

is the optimal grid length. Based on (5), the optimal

transmission range, R

, can be described as follows

R

e
p
D e
l
=d
R
2W
2
e
a
D
e
a
D
_
(9)
Equation (9) shows that the optimal radio transmission
range, R

, relates to the static network trafc data, D, and

width W, where W 10 m in our motorway model.
5 Energy consumption in peak
time trafc model
In what follows, we discuss energy-efciency models where
our analysis and simulations are based on 100 s duration.
5.1 Equal-grid motorway GAF model
In the equal-grid model, as shown in Fig. 1, the entire
network is divided into equal grids and the trafc is
forwarded grid by grid to the BS. To calculate the energy
consumption of the motorway network, the vehicular nodes
are classied as normal nodes (N
n
) and head nodes (N
h
).
The energy consumed by the normal nodes is only for
transmitting, which can be derived based on (1) and (2)
E
N
n
(e
e
e
a
R
n
i
)D
t
=d
R
(10)
where R
i

r
2
i
W
2
_
(normal nodes are able to transmit
data to the head node, no matter where their location is), r
i
is the grid length of the ith grid. In the equal-grid model,
r
i
r.
The energy consumed by the head nodes of the ith grid
can be derived as (7)
E
N
h
i
e
p
D
ri
e
l
=d
R
e
a

(2r)
2
W
2
_ _ _
n
D
ti
(11)
where
D
ti
(L (i 1)r)Wl
D
ri
(L ir)Wl
(12)
l denotes the trafc data (bits) intensity per metres
2
(m
2
),
which can be calculated from the our data trafc model
(l 1:16 10
4
bits=m
2
=s).
By using (10) (12), the energy consumption of the entire
network is
E
total

m
i1
E
N
h
i
(N m) E
N
n
(13)
where N is the number in vehicular nodes in our motorway
network (N 80).
5.2 Analysis of network connectivity
Based on [32], the probability of network connectivity (P)
can be calculated by
P (1 e
d2R
)
N
(14)
where d is the node density, R is the transmission range of
nodes and N is the total number of nodes in the network.
To meet P . 99%, and using the parameters stated earlier
in this section, R is solved to be R . 113 m. Using the
mechanism of equal-grid GAF model, the number of grids
is about 36 (m 36), where the energy consumption of the
network is 3841 J, which is larger than the minimum energy
consumption of the equal-grid model (E
min
3567 J).
5.3 Adjustable-grid motorway GAF model
Obviously, nodes near the sink will relay more trafc than nodes
at the other end of the network. Therefore to optimise their
energy consumption according to the relationship between
optimal radio range (optimal grid length) and network trafc
from (9) [(8)], the nodes near a sink should have a smaller
radio range. Apparently, an adjustable-grid motorway GAF
model can lead to lower energy consumption compared to
equal-grid GAF model, if properly designed. As illustrated in
Fig. 1, the adjustable-grid motorway GAF model divides the
network into m variable-length grids, where the length of
each grid is determined according to (9).
IET Commun., 2010, Vol. 4, Iss. 7, pp. 850860 855
doi: 10.1049/iet-com.2009.0153 & The Institution of Engineering and Technology 2010
www.ietdl.org
The energy consumed by the normal nodes is only for
transmitting, which can be derived based on (1) and (2)
E
N
n
(e
e
e
a
R
n
i
)D
t
=d
R
(15)
where R
i

r
2
i
W
2
_
(normal nodes are able to transmit
data to the head node, no matter where their location is), r
i
is the grid length of the ith grid.
The energy consumed by the head nodes of the ith grid
can be derived as (7)
E
N
h
i
e
p
D
ri
e
l
=d
R
e
a

(2r)
2
W
2
_ _ _
n
D
ti
(16)
where
D
ti
(L r
1
r
2
; . . . ; r
i1
)Wl
D
ri
(L r
1
r
2
; . . . ; r
i
)Wl
(17)
l denotes the trafc data (bits) intensity per metre
2
(m
2
),
which is the same as the equal-grid model
(l 1:16 10
4
bits=m
2
=s).
By using (15) (17), the energy consumption of the entire
network is
E
total

m
i1
E
N
h
i
(N m) E
N
n
(18)
where N is the number of vehicular nodes of our motorway
network (N 80). The energy consumption of the
adjustable-grid motorway GAF model is 3012 J, when the
number of grids m 21.
5.4 Genetic algorithm motorway
GAF model
Genetic algorithms (GA) [33, 34] are optimisation and
search techniques based on the principles of genetics and
natural selections. GA is applied to search the minimum
value of the tness function, and the variables are processed
by three main GA operators, which are dened as follows:
Selection: This operator selects chromosomes in the
population for reproduction, and the tter chromosome has
a higher probability of being selected to reproduce.
Crossover: This operator chooses a locus and exchanges or
crosses over two chromosomes to create two offspring.
Mutation: This operator randomly ips some bits in a
chromosome.
Each variable is considered as a chromosome that evolves
through iterations (generations). The chromosomes of the
next generation are created by merging or modifying the
chromosomes together from the current population, where
the crossover and mutation operators are applied in the
process. During each population, the chromosomes are
selected according to their selection probabilities, which are
determined by the selection operators. This has been
explained in the owchart shown in Fig. 6.
In our GA model, the tness function is
E
t

m
i1
E
i
(19)
where
E
i
e
p
D
ri

e
l
d
R
e
a

(r
i
r
i1
)
2
W
2
_
_ _
n
D
ti
(20)
which is a constraint by the set
_
{(r
1
, r
2
, . . . , r
m
):0 , r
i
, L,

m
i1
r
i
L} (21)
where r
i
is represented by a chromosome in our GA
algorithm, based on which new chromosomes are produced
in each generation by crossover and mutation operations on
parent population, as shown in Fig. 6. Given enough
generations (about 20 000 generations), the minimum of
the tness function (minimum energy consumption) is
computed: E
t
2946 J, where m 20. This result will be
analysed in Section 7.
Figure 6 Flowchart of the GA
856 IET Commun., 2010, Vol. 4, Iss. 7, pp. 850860
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-com.2009.0153
www.ietdl.org
6 Energy consumption in off peak
time trafc model
In the off peak data trafc scenario, the average number of
vehicles in the motorway section GAF model is only about
4. Since the maximum number of grids is 4, with the same
mechanism of the peak time trafc model, the minimum
energy consumption of the equal-grid model is achieved at
this number of grids, where m 4. Since the data trafc
density is too low and the number of vehicles is very small,
the energy consumption results of the adjustable-grid,
equal-grid and genetic algorithm models show comparable
results in terms of energy consumption.
Fig. 5 shows that the average data trafc transmitted by
vehicles during the peak time is much larger than that during
off peak time and varies at different locations during specic
time durations. The huge difference in data trafc calls for
enabling vehicles to transmit at different transmission ranges
according to time duration and location on a motorway in
order to save more energy. Therefore vehicles adjust their
transmission range based on transport trafc density in order
to maintain vehicular network connectivity as well as to save
more energy. When the number of vehicles along a specic
time duration or over a specic location is small, then, each
adjusts its transmission range to a large value in order to
maintain network connectivity. Whereas, when the number
of vehicles increases then each vehicle should adjust its
transmission range based on the optimal transmission range
and total energy relationship functions derived in this paper
so that the total energy consumption can be minimised. As a
result, vehicles will be able to automatically tune their wireless
communication cards to a transmission range, which
corresponds to a specic driving time or a specic location on
a motorway in order to save more energy.
7 Numerical results
We have conducted simulations to evaluate the energy efciency
of the equal-grid, adjustable-grid and GA models. Eighty
vehicular nodes (determined by the M4 motorway trafc
model) were randomly distributed in the motorway network
which is divided into virtual grids following the GAF models
introduced. In each grid, the head node to which all the other
nodes transmit was randomly selected. The data are relayed by
the head node of each grid until it reaches the sink node.
Each simulation was repeated 1000 times, and the total
average energy consumption was computed. In what follows,
we discuss the results based on the relationships derived and
the energy-efciency GAF models considered.
Based on the relationship between the optimal transmission
range and data trafc transmitted by vehicular nodes, which
we have derived in Section 5, Fig. 7 depicts the relationship
between optimal grid length and network trafc. Three
important points can be observed in the results shown in
Fig. 7. First, the optimal grid length (or optimal radio
transmission range) increases sharply when the network trafc
is low, where the number of transmissions is small and
therefore the global network energy can be minimised by
minimising the number of nodes involved in the transmission,
which corresponds to a large node transmission range. The
analysis agrees with this observation and predicts this trend,
where a large number of nodes are in the sleeping state and
therefore the transmission range is large. The majority of the
network is in the sleeping state when the network trafc is less
than 2 10
6
bits in 1 s. Second, when the network trafc is
large, the optimal transmission range, R

, takes smaller values

and therefore the optimal grid length becomes smaller, and
hence, there will be more relaying nodes (more grids).
7.1 Comparison of GAF models for energy
efciency
To evaluate the total network energy consumption with
dynamic trafc in the equal-grid motorway GAF model, we
evaluated the total energy consumption in terms of the
number of network grids, m, based on (13). The relationship
between E
total
and m is shown as in Fig. 8. The minimum
Figure 8 Energy consumption comparison
Figure 7 Relationship between optimal grid length and
network trafc (n 2)
IET Commun., 2010, Vol. 4, Iss. 7, pp. 850860 857
doi: 10.1049/iet-com.2009.0153 & The Institution of Engineering and Technology 2010
www.ietdl.org
energy consumption of the equal-grid model is achieved when
m 27. Fig. 8 shows the optimal transmission range in terms
of the number of grids covering a network in the equal-grid
model. When the number of grids is small, the length of the
grid is large. Therefore the d
n
(n 2) propagation loss
component dominates the energy consumption, thus the
energy consumption decreases sharply for m , 27 with
increase in the number of grids. However, when the number
of grids is large, the length of each grid is small. Therefore
the d
n
propagation loss component becomes smaller, where
the transmitter and receiver electronics energy consumption
per bit become large. This linearly increases in turn the
energy consumption with the number of grids, as shown in
Fig. 8, for m . 27.
As presented in Section 5, the energy consumption of the
adjustable-grid and GA models has been computed.
Accordingly, the network energy consumption of the equal-
grid, adjustable-grid and GA models are compared in Fig. 8.
This shows that the network energy consumption of the
equal-grid model reaches the minimum energy consumption
(3567 J) at m 27, and the network energy consumption of
GA model reaches the minimum energy consumption
(2946 J) at m 20. Compared to the minimum network
energy consumption of the equal-grid model, the adjustable-
grid model saves 19.3% energy and the GA model saves
21.1% energy (by comparing the minimum energy
consumption of both models). Note that, there is only one
single value for the network energy consumption in the
adjustable-grid model because in the adjustable-grid model,
the length of each grid is computed according to the network
trafc, and the number (m) of grids is determined once the
sum of the length of these grids exceeds the length of the
network (2 km).
7.2 GAF models with variable
transmission ranges
In the numerical results obtained so far, the transmission
range in each grid is considered to be static based on GAF
models. In these any node in one grid is able to
communicate with any node of its neighbouring grid,
which means that the transmission range in one grid is
determined by the size of its neighbour and itself. Here,
however, we assume that the transmission range, R
i
, from
one head node to its neighbour head node is only
determined by their positions on the motorway:
R
i

(x
i
x
i1
)
2
(y
i
y
i1
)
2
_
, where x
i
and y
i
are the
coordinates of the active node in the ith grid.
As a result, Fig. 8 also shows the energy consumption of
the GAF models which considers variable transmission
ranges, and it shows that the model based on the variable
transmission range can save more energy than those which
are based on a static transmission range. Based on our
computation, it has been found that the GAF models
based on variable transmission ranges can save energy by
35.3% in (equal-grid model), 34.5% in (adjustable-grid
model) and 36.1% in (GA model), compared to the GAF
models which are based on static transmission ranges.
Furthermore, the comparison is more clearly shown in Fig. 9.
7.3 Comparison of power consumption
The results above are all based on 100 s simulations, in which
the power consumptions (J/s) of these models are calculated
and shown in Fig. 9. Without GAF mechanisms, we assume
that every vehicle can transmit data directly to the BS. Thus
the transmission range of each vehicle is considered to be
xed, and is set to be the network length (L 2 km). Based
on (7), the energy consumed by the entire network is
computed to be about 6:4 10
5
J, thus the power
consumption is 6:4 10
3
W, which is shown in Fig. 9.
Compared to the power consumption without using GAF
mechanism, the GAF models proposed in this paper could
save power consumption by 99:4 99:7% (about 6364
6381 W), and the comparison is given in Fig. 9. Note that, in
Fig. 9, we only consider the minimum energy consumption
of both equal-grid and GAs models, which occur respectively
at m 21 and 27. As an example, if we consider a 60 W
bulb, then, the power saved in the 2 km section of the
motorway network can power 106 bulbs and over the entire
motorway (300 km approximately) 16 000 bulbs.
8 Conclusion
Based on transport trafc proles observed at different ILs on
the M4 motorway in the UK, this paper introduced, analysed
and evaluated three GAF models to optimise energy
consumption in vehicular ad hoc networks. We have derived
the relationship between the number of grids and energy
consumption, where the minimum number of network grids
can be determined in the equal-grid GAF model,
determining the minimum energy consumption in this
model. The relationship between the network energy
consumption and nodes transmission range has been derived
Figure 9 Power consumption in three models
858 IET Commun., 2010, Vol. 4, Iss. 7, pp. 850860
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-com.2009.0153
www.ietdl.org
in the adjustable-grid GAF model. This enabled us to nd the
optimal transmission range, which corresponds to the
minimum energy consumption in adjustable-grid model. The
performance of these two GAF models is compared to the
performance of a genetic algorithm introduced to optimise
energy consumption in VANETs. The comparison results
show that the GA model is the most energy efcient one and
saves 21.1% compared to the equal-grid model, whereas the
adjustable-grid model saves 19.3% compared to the equal-
grid model. It has been found out that a GAF model with
variable transmission ranges can save about 35% energy
compared to the GAF models based on a static transmission
range. This paper draws the rst step towards building a
software utility to control vehicle energy consumed in
vehicular communication. Consequently, vehicles will be able
to tune their communication transmission range based on the
driving time or specic locations along their trajectory.
Furthermore, the proposed transmission range adjustment
mechanism can be deployed in other ad hoc network scenarios
where dynamically varying node densities are expected, for
example in social networking.
9 References
[1] BUNDGEN R.: White paper green ict. Technical Report,
T-Systems Enterprise Service GmbH, 2008
[2] HODGES R., WHITE W.: Go green in ict. Technical Report,
GreenTech News, 2008
[3] GROUP S.: ICT system and network for reducing
environmental impacts. Technical Report, Ministry of
Internal Affairs and Communications of Japan, March 2007
[4] NGUYEN T.D., BERDER O., SENTIEYS O.: Cooperative mimo
schemes optimal selection for wireless sensor networks.
IEEE Vehicular Technology Conf. (VTC) Spring, May 2007,
pp. 8597
[5] JURDAK R.: Wireless ad hoc and sensor networks:
a cross-layer design perspective (Springer, 2007)
[6] XU Y., HEIDEMANN J., ESTRIN D.: Geography-informed energy
conservation for ad hoc routing. Proc. ACMMobil01, July
2001, pp. 7084
[7] I. 802.11: Wireless lan mediumaccess control(MAC) and
physical layer(phy) specications, IEEE Standard 802.11, 1999
[8] ZHOU Y., TAN Y.E., LAURENSON D.I., MCLAUGHLIN S.: Impact of
power saving mac scheme on ad hoc network routing
protocol. Vehicular Technology Conf. 2005. VTC 2005-
Spring, 2005 IEEE 61st, 2005
[9] LI N., WANG H., ZHENG S.: An energy-saving scheme for
wireless lans. Int. Conf. on Communication Technology,
April 2003
[10] LI N., XU Y., XIE S.: A power-saving protocol for ad hoc
networks. IEEE Int. Conf. on Wireless Communications,
Network and Mobile Computing, September 2005
[11] MILLER M.J., VAIDYA N.H.: Ad hoc routing for multilevel
power save protocols. April 2008, vol. 6, pp. 210225
(www.elsevier.com/locate/adhoc)
[12] CHANDRA A.V.S.: Application-specic network management
for energy-aware streaming of popular multimedia formats.
Usenix Annual Technical Conf., 2002
[13] RODOPLU V., MENG T.H.: Minimum energy mobile wireless
networks. IEEE Int. Conf. on Communications, ICC98, June
1998, vol. 3, pp. 16331639
[14] LI N., HOU J.C., SHA L.: Design and analysis of an mst-
based topology control algorithm. IEEE INFOCOM 2003,
March 2003, pp. 17021712
[15] CHANGJ.H., TASSIULASL.: Energy conserving routing in wireless
ad-hoc networks. IEEE INFOCOM 2000, 2000, pp. 2231
[16] BANERJEE S., MISRA A.: Minimum energy paths for reliable
communication in multi-hop wireless networks. MobiHoc,
2002, pp. 146156
[17] PARK S.J., SIVAKUMAR R.: Quantitative analysis of
transmission power control in wireless ad-hoc networks.
Int. Workshop on Ad Hoc Networking (IWAHN), August 2002
[18] TAKAGI H., KLEINROCK L.: Optimal transmission ranges for
randomly distributed packet radio terminals, IEEE Trans.
Commun., 1984, COM-32, pp. 246257
[19] RODOPLU V., MENG T.H.: Minimum energy mobile wireless
networks, IEEE J. Sel. Areas Commun., 1999, 17, pp. 13331344
[20] HOU T.C., LI V.O.K.: Transmission range control inmultihop
packet radio networks, IEEE Trans. Commun., 1986,
COM-34, pp. 3844
[21] ZORZI M., PUPOLIN S.: Optimum transmission ranges in
multihop packet radio networks in the presence of
fading, IEEE Trans. Commun., 1995, 43, pp. 22012205
[22] CHEN P., ODEA B., CALLAWAY E.: Energy efcient system
design with optimum transmission range for wireless ad
hoc networks. IEEE ICC, 2002, pp. 945952
[23] CHEN Y., SIRER E.G., WICKER S.B.: On selection of optimal
transmission power for ad hoc networks. 36th Hawaii Int.
Conf. on Syst. Sci., 2003, pp. 300309
[24] ZUNIGA M., KRISHNAMACHARI B.: Optimal transmission
radius for ooding in large scale sensor networks. 23rd
Int. Conf. on Distrib. Comput. Syst. Workshops, 2003,
pp. 697702
IET Commun., 2010, Vol. 4, Iss. 7, pp. 850860 859
doi: 10.1049/iet-com.2009.0153 & The Institution of Engineering and Technology 2010
www.ietdl.org
[25] GAO J.L.: Analysis of energy consumption for
ad hoc wireless sensor networks using a bit-meter-
per-joule metric. Technical Report, IPN Progress Report,
August 2002
[26] GAO Q., BLOW K.J., HOLDING D.J., MARSHALL I.W., PENG X.H.:
Radio range adjustment for energy efcient wireless
sensor network, Ad hoc Netw., 2006, 4, pp. 7582
[27] YIN B., HONGCHI S., SHANGY.: Analysis of energy consumption
in clustered wireless sensor networks. Second Int. Symp. on
Wireless Pervasive Computing, February 2007
[28] GOKTURK M.S., ERCETIN O., GURBUZ O.: Energy distribution
control in wireless sensor networks through range
optimization. PIMRC 2008, 2008
[29] ERCETIN O.: Distance-based routing for balanced
energy consumptioninsensor networks. Globecom2008, 2008
[30] QAZI B.R., DE GRADOVIVEROA., ALANI O.Y.K., ELMIRGHANI J.M.H.: Pico
cellular mac protocol for multimedia communication. AINA
2009, 2009
[31] BHARDWAJ M., GARNETT T., CHANDRAKASAN A.P.: Upper bounds
on the lifetime of sensor networks. Proc. ICC01, June
2001, pp. 785790
[32] BETTSTETTER C.: On the minimum node degree and
connectivity of a wireless multihop network. Int. Symp.
on Mobile Ad Hoc Networking & Computing Archive Proc.
Third ACM Int. Symp. on Mobile Ad hoc Networking &
Computing, 2002
[33] HAUPT R.L., HAUPT S.E.: Practical genetic algorithms
(Wiley, New York, 2004)
[34] MELANIE M.: An introduction to genetic algorithms
(A Bradford book, The MIT Press, 1999)
860 IET Commun., 2010, Vol. 4, Iss. 7, pp. 850860
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-com.2009.0153
www.ietdl.org