You are on page 1of 6

Energy-Efficient Scheduling Algorithms for

Real-Time Data Reporting in 4G LTE


Machine-to-Machine Communication Networks

Po-Chun Shen and Shun-Ren Yang Phone Lin


Dept. of Comp. Sci. & Inst. of Commun. Eng. Dept. of Comp. Sci. & Info. Eng.
National Tsing Hua University National Taiwan University
Hsinchu, Taiwan Taipei, Taiwan
sryang@cs.nthu.edu.tw plin@csie.ntu.edu.tw

Abstract—Machine-to-machine (M2M) communications pro- for human-to-human applications. Only a small portion of
tocols, e.g., Machine Type Communications (MTC) in 3GPP LTE, LTE uplink scheduling algorithms were designed for M2M
support a variety of Machine to Machine (M2M) applications, applications [1], [2], [3], and they typically were based on
which requires real-time data reporting. Many uplink scheduling the fact that the machines are battery-powered (i.e., power
algorithms are primarily designed for human-to-human applica- is limited), and aimed to reduce the energy consumption of
tions rather than for M2M applications. These scheduling algo-
rithms typically are based on the fact that the machines in MTC
machines for M2M communications.
are battery-powered and aim to reduce the energy consumption. The aforementioned studies of M2M uplink scheduling can
We notice that the current literature considers either the problem be divided into two categories. The studies [1], [2] aimed
of sleep-time maximization or transmission power minimization, to minimize the active slots of the machines, so that the
but not both. In this paper, we systematically consider all energy- sleep mode can be as long as possible. However, they did
consumption factors of M2M real-time reporting. Using MTC in
LTE as an example for M2M communications, in this paper, we
not consider the difference in transmission power when dif-
propose two energy-efficient scheduling algorithms to minimize ferent resource blocks (RBs) are utilized to transmit data. The
the total energy consumption of machines for M2M communica- study [3] aimed at the power allocation problem for a machine
tions. The two proposed algorithms utilize the distance between to transmit data over physical channels. Specifically, the RBs
a machine and the serving eNB (i.e., base station in LTE). Our allocated for the data transmission affect the energy con-
performance study shows that our algorithms have performance sumption of signal transmission, so that the data transmission
enhancement in terms of energy consumption compared to other should be allocated with the RBs that have optimal channel
algorithms in the previous works. Furthermore, our algorithms quality. We notice that no previous study has considered the
can maintain high scheduling success ratio and fairness. problem of sleep-time maximization and transmission power
Keywords—Energy efficiency, LTE, machine-to-machine minimization at the same time. Therefore, energy efficiency
(M2M), real-time data reporting may not be accurately reflected. In other words, when the
distance between a machine and the serving eNB is long,
I. I NTRODUCTION the transmission energy consumption accounts for a larger
proportion of total energy consumption. In contrast, sleep-
Machine-to-machine (M2M) communications enables net- time maximization is more important for reducing energy
worked machines to exchange information and exercise with- consumption when the distance is short. Therefore, the two
out human intervention. Examples of M2M applications in- issues should be considered simultaneously.
clude the remote monitoring, e.g., environment monitoring, e-
In this paper, we systematically consider the above energy-
health, smart grid, public surveillance, and positive train con-
consumption factors (including signal transmissions and ma-
trol. Real-time data reporting is one of the major functionalities
chine operations in different modes/states) for M2M real-time
in M2M applications, where more than one type of sensing
reporting over 4G LTE as shown in Fig. 1. The corresponding
data (e.g., temperature, humidity, atmospheric pressure, wind
discussion will be elaborated in subsequent sections. We pro-
speed, direction, etc.) are reported during a predefined time
pose two energy-efficient scheduling algorithms to minimize
period. The sensing data is usually time-sensitive, i.e., if the
total energy consumption of the machines to transmit the
data cannot be received by the M2M application server on
sensing reports to the eNB via LTE uplink channels. We design
time, it becomes useless.
the two algorithms by considering the distance between the
One of the M2M communication protocols is the 4G long- machines and the eNB. The first algorithm is for use when
term evolution (LTE) Machine Type Communications (MTC) the distance is long, and it aims to reduce energy consumption
that satisfies the requirements for M2M real-time reporting when the machines to transmit their data. When the distance
applications. In LTE MTC, machines transmit sensing data to a is longer, a larger proportion of overall energy consumption is
base station (i.e., eNB) via LTE uplink channels. The previous for transmitting data. The first proposed algorithm is proposed
works proposed a variety of uplink scheduling algorithms to reduce total energy consumption for this case. The second
for 4G LTE; however, the majority of these were designed proposed algorithm is applied when the distance is short, and it

978-1-5090-0304-4/16/$31.00 ©2016 IEEE 1057


energy consumption for B. Channel Model
signal transmissions
resource The present study implemented algorithms in M2M com-
block
munication networks that will have heavy traffic load. In order
allocated RB to maximum transmission efficiency, we assumed that each
sleep mode node transmitted its data with the best MCS. In other words, we
active mode: idle state
chose the modulation coding scheme that mapped to a channel
quality indicator value of 15. Therefore, we could transmit
active mode: transmit state more data when the frequency was limited. Using this MCS,
we could find the number of the RB needed to transmit for
subframe
a node in a reporting period. Node n is allocated a transport
block (TB), which consists of several RBs, of sn bits. For a
energy consumption for given sn and the MCS, the number of RBs can be found as
machine operations at different modes/states  
sn
x= , (1)
Fig. 1. Energy-Consumption Factors of M2M Real-Time Reporting over 4G ε×b
LTE
where ε is equal to the number of symbols in a subframe
multiplied by the number of the subcarriers in an RB, b is the
M2M node n
subframes spectral efficiency of the MCS, and 1 ≤ n ≤ N .
Type 1 sensor N M2M nodes 1 1 1
煑 The block error rate (BLER) is the average failure rate
PUSCH1 2 2 2
Type m sensor of TBs. To our knowledge, there are no analytical solutions
PUSCH2 3 3 3
煑 to compute the BLER. Therefore, a computationally efficient
PUSCH3 RBs
Type Mn sensor 4 4 4 method is called least-squares approximation, which is in
煑 煑 煑 煑 [3]. [3] also proves least-squares approximation which closely
Type m M2M data
PUSCHN K-1 K-1 K-1
eNB approximates the signal-to-noise ratio (SNR) as
Size: sn,m
K K K
Reporting period: rn, m γn,ef f ≈ ax exp(bx sn ) − γ0,x . (2)

Fig. 2. SC-FDMA Shared Channel The ax , bx and γ0,x are given in [3] for BLER = 10%. Using
the above function, along with channel estimates, it is easy to
obtain the required applied power to achieve a target block
error rate.
aims to reduce the number of the active slots for the machines.
Note that the second algorithm also determines whether it is
worth for the machine to go into the sleep mode, or just stay in C. Power Consumption
the idle state of the active mode. The experimental results show
When a node transmits the data, it needs to consume power
that our algorithms have better performance (i.e., less energy
for the signal transmission. For giving BLER = 10% and [3],
consumption) than the previous algorithms. Furthermore, our
we can calculate now the node’s effective SNR (γn,ef f ) from
algorithms can also maintain high scheduling success ratio and
Eq. (2). According to Eq. (2), we can translate the unit of
fairness, whose definitions will be elaborated later.
SNR from db to SNR rate as γ. In [4], the power of the signal
transmission would be considered as
II. S YSTEM M ODEL
N × Lp (D)
P0 = γ × , (3)
A. Simplified SC-FDMA MAC Model |q|2
Fig. 2 shows the simplified MAC model of a multiuser where P0 is transmission power, q is the channel gain, N is
single-carrier frequency-division multiple access (SC-FDMA) noise power and Lp (D) is the path loss between the eNB and
system that was used for uplink in 3GPPs LTE. Each M2M the node.
node has access to a single physical uplink shared channel for
transmission of its uplink data. There are N M2M nodes and In addition to the nodes signal transmission power, a node
an eNB within a single cell. The cell spectrum consists of K also needs extra power to run. The node can be in awake mode
RBs, each with 12 subcarriers, and 14 symbols in a subframe. or sleep mode. Awake mode is divided into two states. The first
It was assumed that the system operated in frequency division of these is transmit state, in which the node transmits data, and
duplexed mode. For the purpose of the present study, we the second is idle state in which the node does not transmit
also assumed that inter-cell interference was negligible. SC- data. When the node is in sleep state, it will turn off its radio
FDMA is employed in LTE uplink and in our system model. frequency (RF) and consume less power. We define Pt , Pi ,
Unlike in orthogonal frequency-division multiple access, in Ps to represent a node’s power consumption of transmit state,
SC-FDMA only adjacent sub-carrier blocks (RBs) can be idle state and sleep state. In addition, if a node needs to wake
allocated to a given station at one time. In addition, the data up to transmit the data when it is sleeping, it should power on
relating to individual nodes are further limited, such that the all relative modules. Thus, the node will consume more power.
use a common modulation coding scheme (MCS) mode per We considered power consumption Pw of a node that transfers
subframe. its state from sleep state to transmit state.

1058
III. E NERGY-E FFICIENCY S CHEDULING P ROBLEM not allocated to a given node and 1 if it is for transmission.
Based on equation (4) and (5), the objective function can be
In this section, we describe the energy consumption prob-
formulated as
lem when the eNB schedules the data of the M2M nodes, and XT XN
we made two assumptions. First, as the size of sensory data is min E n (f ) (6)
extremely small and the data must be transmitted immediately, f =1 n=1
we ignored the time it takes for the M2M nodes to collect
the data. Second, we assumed that each piece of sensory
1) Allocation Constraint: As the frequency is limited, the
data cannot be segmented into smaller segments to avoid data
number of the RB is stationary in a subframe. Therefore, the
reassembling overhead.
amount of RBs that are allocated to the nodes cannot exceed
the amount of RBs in a subframe. Thus, this constraint can be
A. M2M Communication Network Architecture formulated as
The M2M communication network architecture is shown in K
N X
Fig. 2. N M2M nodes that connect to an eNB. Node n has Mn
X
βn,k (f ) ≤ K, ∀f ∈ {1, 2, ..., T }. (7)
data types and data type m has size sn,m and reporting period n=1 k=1
rn,m . We define an observation cycle T which represents the
scheduling cycle of the eNB. We set the scheduling observation 2) Contiguity Constraint: Additionally, SC-FDMA is lim-
cycle as the least common multiple of all of the data’s reporting ited to RB allocations in contiguity. As a result, we can
periods. In an observation cycle T , the data will be transmitted formulate the contiguity constraint as
several times. We use dn,m,p to represent the pth data of the
data type m collected in the node n. βn,k (f )−βn,k+1 (f )+βn,x (f ) ≤ 1, ∀f ∈ {1, 2, ..., T }, x = k+2, ..., K.
(8)
B. Problem Formulation
3) Real-Time Packet Constraint: The sensory data must be
The present study proposed a scheduling algorithm to collected before their deadline, otherwise they will be dropped.
minimize energy consumption when M2M nodes transmit their Therefore, we formulated the real-time packet constraint as
sensory data to eNB. For node n, we defined that E n (f )
represents the energy consumption within subframe f and let (p−1)·rn,m < tn,m,p ≤ p·rn,m , ∀n, m, p ∈ {1, 2, ..., (T /rn,m )}.
Etn (f ), Ei and Es be the energy consumption when the node (9)
n is in the transmit state, idle state, and sleep state, and let Ew
be the energy consumption of a transition from sleep state to IV. S CHEDULING A LGORITHMS
transmit state. Note that the value of Ei , Es and Ew are fixed
for a node, but the value of Etn (f ) is dynamic. Accordingly, the In order to extend the life time of the nodes, we supposed
energy consumption for node n in subframe f can be expressed minimization Eq. (6). Eq. (6) consists of two kinds of energy
as consumption:
PK 1) signal transmission energy consumption (i.e.,
P n,k (f )×t f ×βn,k (f )) and 2) state energy consumption
E n (f ) = Etn (f )×αnt (f )+Ei ×αni (f )+Es ×αns (f )+Ew ×αnw (f ), (i.e.,
k=1
Pt × tf + Ei + Es + Ew ). We proposed two scheduling
(4) algorithms to reduce the total energy consumption of the
where αnt (f ), αni (f ), αns (f ), and αnw (f ) are the indication nodes. The first was designed for reducing signal transmission
functions. If node n is in transmit state within subframe f , energy consumption. Therefore, Algorithm 1 selects the RB
n n n
αt (f ) = 1, αi (f ) = 0 and αs (f ) = 0. If node n is in with the better channel quality. The second algorithm was
n n
idle state within subframe f , αi (f ) = 1, αt (f ) = 0 and designed for reducing the active subframes. Thus, Algorithm 2
αns (f ) = 0. If node n is in sleep state within subframe f , allocates the data of the same node to the same subframe, and
n n n
αs (f ) = 1 and αt (f ) = 0 and αi (f ) = 0. In addition, more frequently makes the nodes go into sleep or idle state.
if node n experiences a state transition (from sleep state to Therefore, a node will consume less energy.
n
transmit state) from subframe f −1 to subframe f , αw (f ) = 1,
n
otherwise αw (f ) = 0.
A. The First Energy-Efficiency Scheduling Algorithm
Furthermore, Ei = Pi × tf , Es = Ps × tf and Ew =
Pw × tf , where tf is the time of a subframe. In particular, 1) Concept of Algorithm 1: In order to minimize the energy
Etn (f ) consists of the energy consumption of the transmission PK of the M2M nodes, we propose Algorithm 1 to
consumption
state and the energy consumption of the signal transmission. reduce k=1 Pn,k (f ) × tf × βn,k (f ). First, this algorithm
Therefore, Etn (f ) can be expressed as decides the scheduling order, according to the sensory data’s
K
reporting period and size. Algorithm 1 schedules the data with
n
X short transmission deadline first. When the data has the same
Et (f ) = Pt × tf + Pn,k (f ) × tf × βn,k (f ), (5)
transmission deadline, the order of the large data are ordered
k=1
before the small data. Therefore, this can not only satisfy
where Pn,k (f ) is the power consumption of node n for signal the datas transmission deadline, but also increase the nodes
transmission when the k th resource block is allocated to node throughput. Next, Algorithm 1 must choose the specific RB
n within subframe f and its value is according to equation that consumes less energy for allocating the data. Algorithm
(3), and βn,k (f ) is a binary indicator to represent if the data 1 will thoroughly search every possible combination of RBs.
of the node n is allocated in resource block k within subframe When allocating the data, Algorithm 1 should follow some
f . The indicator βn,k (f ) is equal to 0 if the resource block is principles.

1059
Algorithm 1 Algorithm 2 to illustrate Phase 2 of the Algorithm 1. We assume that
(a) (e) there are only seven RBs in a subframe and allocate
node data type reporting period data size RB node data type reporting period data size RB d1,1,1 to the RBs first. We found the possible options
1 2 subframes 4000 bits 5 RB
1
2 6 subframes 1600 bits 2 RB
1
1
2
2 subframes
6 subframes
6400 bits
1600 bits
8 RB
2 RB
for d1,1,1 , as shown in Fig. 3(c) and allocate d1,1,1 to
2
1 3 subframes 1600 bits 2 RB 1 2 subframes 2400 bits 3 RB an option with less energy. Note that 1) each allocation
2 6 subframes 2400 bits 3 RB 2 2 3 subframes 3200 bits 4 RB
3 1 12 subframes 3200 bits 4 RB 3 6 subframes 2400 bits 3 RB
of the nodes should follow the contiguity constraint, and
2) energy consumption will increase when the allocation
(b) observation cycle=lcm(2,3,6,12)=12 (f) observation cycle=lcm(2,3,6)=6
causes the node to consume the energy of transmit state
(Pt × tf ). Finally, all of the data were allocated, as shown
d1,1,1 d1,1,2 d1,1,3
d2,2,1
d1,1,4 ... d1,1,1 d1,1,2 d1,1,3
in Fig. 3(d).
d2,1,1 d1,2,1 d2,1,2 d2,1,3
d2,1,1 d2,2,1 d2,1,2 d2,2,2 d2,1,3 d2,3,1
2 3 4 6 8 9 2 3 4 6
d1,2,1 Phase 3 “State Deciding”: In this phase, after allocation of
(c) (g) (1) resource block
all of the sensory data, we determined the state of the
resource block
nodes. In Fig. 3(d), for node 2, subframe 2, 6, 7, 9 and 12
d1,1,1 d2,2,1 d1,1,2 d2,2,2 d1,1,3
must be in transmit state because node 2 has to transmit
d2,1,1 d2,1,2 d2,1,3 subframe
data. Node 2 will be in sleep state within the subframes
d2,3,1
that do not transmit the data, and it is worth consuming
(2) d1,2,1 (3)
1 2 1 2 1 2 1 2 1 2 1 2 the switching power (Pw ). In addition, node 2 will be in
subframe d2,3,1 idle state within other subframes. Therefore, subframe 1,
(d)
resource block
d1,1,1 d2,2,1 d1,1,2d2,2,2 d1,1,3 d1,1,1d2,2,1 d1,1,2d2,2,2 d1,1,3
d1,2,1 7, 10, 11 are in idle state and subframe 3, 4, 5 are in
d2,1,3
d2,1,1 d2,1,2 d2,1,3 d2,1,1 d2,1,2 d2,1,3 sleep state.
d2,2,1
d1,1,2 d3,1,1 d1,1,5d1,1,6 (h) resource block
d1,1,4
d2,1,1
d1,1,1
d1,2,1
d1,1,3 d2,2,1
d2,1,3 B. The Second Energy-Efficiency Scheduling Algorithm
d2,1,2 d1,2,1
d1,1,1 d2,2,1 d1,1,2 d2,2,2 d1,1,3

d2,1,1 d2,1,2
d1,2,1
d2,1,3 subframe
Algorithm 1 was dedicated to allocating the data to RBs
for
node 2 transmit state idle state sleep state
for node 1
that consume less energy, and it searches all the possible com-
binations of RBs. Therefore, Algorithm 1 is highly complex. In
Fig. 3. Scheduling Algorithm addition, the distance between the eNB and the node will affect
the energy consumption of the RB. We proposed a second
algorithm to overcome the issues that are mentioned above
1) As the sensory data are renewed after their reporting 1) Concept of Algorithm 2: From our knowledge and
period, they should be scheduled before their deadline. equation (3), if the distance between the eNB and the node is
2) The data will be allocated in a subframe that has sufficient long, the node needs more power to transmit the sensory data
continuous RBs (follow the SC-FDMA requirement that to eNB than if the distance is short and vice versa. Therefore,
is used in LTE uplink), otherwise the data cannot be when the distance between the eNB and the node is short, the
allocated. sum of Pt × tf , Ei , Es and Ew accounts for a large proportion
3) Allocate the data in an RB combination that consumes of the total energy consumption in a node. We may accordingly
less energy. The energy consumption includes 1) signal allocate the data of the same node to the same subframe when
transmission energy and 2) the transmit state energy the distance is short. Thus, the node can avoid waking up in
(Pt × Tf ) when the node’s data are allocated to this a new subframe and consuming Pt × tf .
combination’s subframe.
We proposed Algorithm 2 to reduce the total state energy
Finally, after all of data is allocated, Algorithm 1 decides the consumption, when the distance between the nodes and the
state (transmit state, idle state and sleep state) in each subframe eNB is short. First, this algorithm sorts the data and decides the
of the nodes. scheduling order. The data with a short transmission deadline
will be scheduled first. When the transmission deadline is
2) The Proposed Algorithm 1: This is composed of three
the same, the data of bigger size will be sorted before the
phases. An example from phase 1 to 3 in Fig. 3(a)-(d). data of smaller size. Second, Algorithm 2 attempts to allocate
Phase 1 “Data Sorting”: In this phase, the observation cy- the data to a suitable subframe. In order to minimize the
cle is determined, the sensory data are sorted, and the energy consumption of the nodes, Algorithm 2 reduces the total
scheduling order of the data is decided. The sorting rules amount of active subframes. The allocation has the following
are: two possible scenarios:
1) Sort the sensory data in increasing order of their Case 1: To reduce the active subframes, Algorithm 2 allocates
transmission deadline. dn,m,p to the subframes which node n has been scheduled.
2) Sort the sensory data with the same transmission In addition, if there are several subframes which the
deadline in decreasing order of size. node has been scheduled, Algorithm 2 will select the
On the basis of Fig. 3(a), the observation cycle T = subframe that has minimum remaining RBs. Thus, we
lcm(2, 3, 6, 12) and the sensory data are sorted as more efficiently used the RBs.
Fig. 3(b). Case 2: If the subframes in which node n has been sched-
uled do not have sufficient RBs for dn,m,p to allocate,
Phase 2 “Resource Block Choosing”: We take Fig. 3(b), Algorithm 2 can only select the subframes which node n
which has already been sorted by Phase 1, as an example has not been scheduled and those subframes are before

1060
the deadline of dn,m,p . dn,m,p will be allocated to the Algorithm has the same goal as our algorithms, and minimizes
subframe that has maximum remaining resource blocks. energy consumption in the M2M network, but it does not
Thus, node n has a higher probability of allocating its consider signal energy consumption and total state energy
other data to that subframe and avoiding transmission of consumption together. In the following subsection, we compare
data in another subframe. these algorithms in 1) the impact of the number of nodes and
2) the impact of the distance between the nodes and the eNB.
In addition to deciding the subframe to which the data are
allocated, the algorithm must also allocate the data to specific A. The Impact of Number of Nodes
RBs. Nevertheless, when the distance is short, the gap of the
RB’s energy consumption between great channel quality and We discuss the simulation result in four aspects: (1) energy
poor channel quality is low. We may allocate the data to the consumption, (2) scheduling success ratio, and (3) fairness.
RBs randomly, not thoroughly. According to the strategies With these metrics, we describe the impacts of the number of
above, Algorithm 2 is far less complex than Algorithm 1. M2M nodes transmitting its sensory data to eNB.
Finally, Algorithm 2 decides the state of the subframe for each 1) Energy Consumption: To measure the energy saving
node. capability, we use Eq. (6) as the total energy consumption of
2) The Proposed Algorithm 2: This is composed of three all nodes. We increase the number of nodes when the distance
phases. An example from phase 1 to 3 for Algorithm 2 is between the eNB and the nodes is 800 meters. The result
shown in Fig. 3(e)-(h). is shown in Fig. 4(a). Our algorithms and Chen’s Algorithm
aim to reduce energy consumption of the nodes. Thus, the
Phase 1 “Data Sorting”: In this phase, Algorithm 2 is the result shows that our algorithms and Chen’s Algorithm perform
same as Algorithm 1. better than EDF, and WF2 Q in terms of energy efficiency.
Phase 2 “Subframe Choosing”: Fig. 3(f), which has already Moreover, our algorithms consider signal energy consumption
been sorted by Phase 1, can be taken as an example and total state energy consumption together. Therefore, the
to illustrate Phase 2 of Algorithm 2. According to the performances of our two algorithms are better than that of
strategies above, Algorithm 2 allocates the data to the Chens Algorithm.
same node within the same subframe. Therefore, we can We also observed that when the number of nodes increased,
see the result before d2,3,1 is as shown in Fig. 3(g)-(1). the energy consumption of all algorithms increased; in particu-
When deciding the transmission time of d2,3,1 , d2,3,1 may lar, Algorithm 1 was better than Algorithm 2. The reason is that
be allocated within subfames 2, 4, and 6. Furthermore, to total state energy consumption accounts for a large proportion
more efficiently use the bandwidth, d2,3,1 was allocated of the total energy consumption in a node when the distance is
to the subframe with the minimum remaining RBs, which 800 meters. Therefore, the energy consumption of Algorithm
was subframe 2 as shown in Fig. 3(g)-(2). 2, which is dedicated to reducing state energy consumption, is
In addition, if the sensory data can not be scheduled less than Algorithm 1.
within the subframe which the sensory data’s node has
been scheduled, like d1,2,1 . Algorithm 2 allocates d1,2,1 to 2) Scheduling Success Ratio: It is important to transmit
the subframe with the maximum remaining RBs. Finally, the sensory data before their deadline for real-time reporting,
the algorithm allocate d1,2,1 to subframe 6 as shown in otherwise these data will lose their effectiveness and will be
Fig. 3(g)-(3). dropped. To measure the freshness of sensory data, we consider
Phase 3 “State Deciding”: In this phase, Algorithm 2 is scheduling success ratio Rs defined as
again the same as Algorithm 1. N
1 X
Rs = 1 − zn ,
V. P ERFORMANCE E VALUATION N n=1
In this section, we discuss the performance evaluation of where zn is the percentage of the sensory data collected in
our algorithms. Initially, we introduce the scenario of our M2M node n within T that are dropped.
algorithms. The simulations were conducted using C. We The result is shown in Fig. 4(b). We know that EDF trans-
consider that 20 M2M nodes (default) transmit their sensory mits the data that have the earliest deadline first. Therefore,
data to an eNB, via LTE uplink. The average number of M2M EDF must have a higher scheduling success ratio. Neverthe-
data types collected in each M2M nodes is set as 6. There are less, we can see that our proposed algorithms have almost the
25 kinds of data type for nodes to collect and their size (bits) same scheduling success ratio as EDF in this scenario, indi-
can be 800, 1600, 2400, 3200, 4000, and their reporting period cating that the scheduling success ratio will not be influenced
(tf ) can be 6, 7, 9, 14, 18. The other parameters to consider by our algorithms, which are used for power-saving. However,
in the simulation are according to [2] and the parameter of we noted that with the limited 10MHz channel bandwidth in
Equation (3) is according to [4] and [5]. In addition, we this experiment, our algorithms can support up to 25 nodes
assumed that the channel bandwidth is 10MHz. and each node senses six kinds of data type by consuming
We simulate 1) earliest deadline first (EDF) [6] and less energy while maintaining high scheduling success ratio,
2) worst-case fair weighted fair queueing (WF2 Q) [7] 3) that is, more than 0.95.
Chen’s Algorithm [1] to compare with our algorithms. EDF 3) Fairness: To measure fairness, we use Jain’s Fairness
is designed for real-time environment and the data that have Index F , which is defined as
earliest deadline will be scheduled first. Thus, data will be PN
promised a low delay. WF2 Q is a packet data scheduling ( n=1 zn )2
F = ,
algorithm that provides data with fairness guarantee. Chen’s N· N
P 2
n=1 zn

1061
Energy consumption v.s. Number of nodes and WF2 Q. The reason is the same as that which explains the
550
impact of the number of M2M nodes on energy consumption.
450

Energy(J)
WF²Q
350
EDF
Most importantly, we can observe that Algorithm 2 is
250
Chen's an improvement on Algorithm 1 when the distance is below
150 Algo. 1 900 meters and vice versa. This illustrates why we chose
50 Algo. 2
15 20 25 30 35 40 Algorithm 2 to schedule the M2M nodes data when the
Number of nodes
distance was below 900 meters in this scenario, because state
(a) energy consumption accounts for a large proportion of the total
energy consumption in a node. In contrast, in this scenario, we
Scheduling success ratio v.s. Number of nodes chose Algorithm 1 to schedule the M2M nodes data when the
1
distance was beyond 900 meters, because signal transmission
Scheduling success ratio

0.95
0.9
0.85
energy consumption accounts for a large proportion of the total
WF²Q
0.8
0.75 EDF
energy consumption in a node. The energy consumption of
0.7
Chen's Algorithm 2 increases more quickly than that of Algorithm 1,
0.65
0.6 Algo. 1 and the latter will approach other algorithms (except Algorithm
0.55 Algo. 2
0.5 1) when the distance is long.
15 20 25 30 35 40
Number of nodes
VI. C ONCLUSION
(b)
This paper investigated the energy-consumption scheduling
Fairness v.s. Number of nodes
1
problem for M2M real-time reporting over 4G LTE. We con-
0.98 sidered all the energy consumption factors of such reporting,
0.96
Fairness

0.94
WF²Q
for both signal transmissions and machine operations in differ-
EDF
0.92
0.9 Chen's ent modes/states. We proposed two energy-efficient scheduling
0.88 Algo. 1 algorithms to minimize total energy consumption of machines
0.86
15 20 25 30 35 40
Algo. 2
when machines transmit their data to the eNB via LTE uplink
Number of nodes channels. Algorithm 1 is for use when the distance between
a machine and the eNB is long, and Algorithm 2 is for use
(c)
when the distance is short. Our algorithms were compared with
Energy consumption v.s. Distance EDF, WF2 Q, and Chen’s algorithm, and the results showed
550 that our algorithms consumed less energy and also maintained
450
acceptable scheduling success ratios and fairness.
Energy(J)

WF²Q
350
EDF
250
Chen's
150
Algo. 1
ACKNOWLEDGMENT
50 Algo. 2
600 700 800 900 100011001200130014001500 This work was supported in part by the Ministry of Science
Distance(m)
and Technology of Taiwan (MOST) under Contracts 103-2221-
(d) E-007-076-MY2 and 104-3115-E-007-004.

Fig. 4. Simulation Results R EFERENCES


[1] Yi-Bei Chen, Shun-Ren Yang, Jenq-Neng Hwang, and Ming-Zoo Wu. An
energy-efficient scheduling algorithm for real-time machine-to-machine
where zn is defined in Sec. V-A2. (m2m) data reporting. In 2014 IEEE Global Communications Conference
(GLOBECOM), pages 4442–4447, Dec 2014.
Fig. 4(c) shows that WF2 Q resulted in greater fairness [2] Kuo-Chang Ting, Hwang-Cheng Wang, Chih-Cheng Tseng, and Fang-
than the other algorithms. This makes sense because WF2 Q Chang Kuo. Energy-efficient drx scheduling for qos traffic in lte
is dedicated to increasing fairness. Our algorithms maintained networks. In 2011 IEEE 9th International Symposium on Parallel and
a good fairness value above 0.92 when the number of nodes Distributed Processing with Applications (ISPA), pages 213–218, May
was increased up to 45. In addition, we noted that EDF was 2011.
associated with the worst fairness of all the algorithms. This [3] D.J. Dechene and A. Shami. Energy-aware resource allocation strategies
for lte uplink with synchronous harq constraints. IEEE Transactions on
is because EDF does not sort the data according to their size. Mobile Computing, 13(2):422–433, Feb 2014.
Thus, the data that are larger will have high a probability of [4] Tze-Ping Low, Man-On Pun, Y.P. Hong, and C.-C.J. Kuo. Optimized
being dropped under high traffic load. opportunistic multicast scheduling (oms) over wireless cellular networks.
IEEE Transactions on Wireless Communications, 9(2):791–801, February
2010.
B. The Impact of the Distance [5] 3GPP TR 36.931. LTE;Evolved Universal Terrestrial Radio Access
(E-UTRA);Radio Frequency (RF) requirements for LTE Pico Node B
We have mentioned that the energy consumption relates (Release 9), v9.0.0 edition, May 2011.
to the distance between the nodes and the eNB. Therefore, [6] M. Caccamo, L.-Y. Zhang, S. Lui, and G. Buttazzo. An Implicit
we investigated the impact of the distance between the nodes Prioritized Access Protocol for Wireless Sensor Networks. IEEE Real-
and the eNB on energy. The result can be seen in Fig. 4(d), Time Systems Symposium, pages 39–48, 2002.
which shows that our two algorithms are an improvement on [7] J.C.R. Bennett, and H. Zhang. WF2 Q: Worst-Case Fair Weighted Fair
other algorithms, and that Chens Algorithm is superior to EDF Queueing. Proceedings IEEE INFOCOM, 20:120–128, March 1996.

1062

You might also like