Content uploaded by Amin Azari
Author content
All content in this area was uploaded by Amin Azari on Nov 25, 2017
Content may be subject to copyright.
Fundamental Tradeoffs in Resource Provisioning for
IoT Services over Cellular Networks
Amin Azari and Guowang Miao
KTH Royal Institute of Technology
Email: {aazari, guowang}@kth.se
Abstract—Performance tradeoffs in resource provision-
ing for mixed internet-of-things (IoT) and human-oriented-
communications (HoC) services over cellular networks are in-
vestigated. First, we present a low-complexity model of cellu-
lar connectivity in the uplink direction in which both access
reservation and scheduled data transmission procedures are
included. This model is employed subsequently in deriving
analytical expressions for energy efficiency, spectral efficiency,
and experienced delay in data transmission of connected devices
as well as energy consumption of base stations. The derived
expressions indicate that the choice of uplink resource provi-
sioning strategy introduces tradeoffs between battery lifetime for
IoT communications, quality of service (QoS) for HoC, spectral
efficiency and energy consumption for the access network. Then,
the impacts of system and traffic parameters on the introduced
tradeoffs are investigated. Performance analysis illustrates that
improper resource provisioning for IoT traffic not only degrades
QoS of high-priority services and decreases battery lifetime of
IoT devices, but also increases energy consumption of the access
network. The presented analytical and simulations results figure
out the ways in which spectral/energy efficiency for the access
network and QoS for high-priority services could be traded to
prolong battery lifetimes of connected devices by compromising
on the level of provisioned radio resources.
Index Terms—Internet of things, Machine-type communica-
tions, Resource provisioning, Energy efficiency, Green cellular
network.
I. INTRODUCTION
Mobile network operators are growing targeting internet
of things (IoT) services in order to decrease the revenue
gap. Motivated by the fact that cellular networks have deeply
penetrated everywhere, cellular IoT also named as machine-
type communications, is expected to play an important role
in realization of IoT [1]. Cellular IoT is generally charac-
terized by the massive number of concurrent active devices,
small payload size, and vastly diverse quality-of-service (QoS)
requirements [2]. Also, most IoT devices are battery driven
and once deployed, their batteries will not be replaced. Then,
long battery lifetime is of paramount importance for them
[3]. Serving the coexistence of ordinary cellular services like
mobile broadband for human-oriented communications, and
the IoT services with ultra-long battery lifetime and ultra-
high reliability requirements within one network constitutes
an interesting research problem which has attracted lots of
interests in recent years [4, 5]. The capacity limits of random
access channel (RACH) of LTE-A for serving IoT services
and a survey of improved solutions are studied in [6]. Among
the alternatives, access class barring (ACB), which reduces the
contention among nodes at the cost of introducing delay, and
cluster-based communications are promising solutions [7]. The
impact of massive IoT on the access probability of human-
type users to the random access channel (RACH) of LTE
has been investigated in [8]. In [9], separation of random
access resources between IoT and HoC has been investigated.
A through survey on LTE scheduling algorithms for mixed
IoT/HoC traffic is presented in [10], which indicates that the
existing IoT scheduling schemes are mainly overlay to HoC
scheduling, i.e. in each radio frame the unused resources for
HoC are allocated to IoT [10]. A time-controlled scheduling
framework for IoT traffic has been proposed in [11], which
divides the traffic into classes based on the QoS requirements
in order to preserve QoS for HoC while satisfying delay
requirements of IoT. Energy-efficient uplink scheduler design
for IoT traffic traffic has been investigated in [12, 13].
The literature study reveals that while the impact of IoT
traffic on the access reservation and scheduled data trans-
mission of cellular networks has been partially investigated,
performance tradeoffs in resource provisioning for upcoming
massive IoT services is missing. The present work is devoted
to deriving a low-complexity model for cellular connectiv-
ity, analytically formulating the key performance indicators
(KPIs), and figuring out performance tradeoffs between intro-
duced KPIs. Towards this end, the main contributions of this
paper include:
•Develop a tractable framework to model energy consump-
tions of IoT-type devices deployed over cellular networks,
experienced delay and spectral efficiency of IoT/HoC
traffic in uplink transmissions, and energy consumption
of the access network.
•Introduce battery lifetime, spectral efficiency, energy ef-
ficiency, and delay tradeoffs in serving mixed IoT/HoC
traffic.
•Explore the impact of resource provisioning, traffic, and
medium access parameters on the introduced tradeoffs.
The rest of this paper is organized as follows. System model
is presented in the next section. Performance indicators are for-
mulated in section III. Performance tradeoffs are investigated
in section IV. Numerical results are presented in section V.
The concluding remarks are given in section VI.
II. SYSTEM MODEL
Consider a single cell with one base station at the center,
and a massive number of devices, including IoT-type devices
(c)IEEE copyright applies to this document.
߬;ƐĞĐͿ
ܯூ
ܶ
ோை;ƐĞĐͿ
Wh^,ƌĞƐŽƵƌĐĞ;,ŽͿ
Wh^,ƌĞƐŽƵƌĐĞ;/ŽdͿ
WZ,ƌĞƐŽƵƌĐĞ;,ŽͿ
WZ,ƌĞƐŽƵƌĐĞ;/ŽdͿ
ܯு
ƚŝŵĞ
ĨƌĞƋƵĞŶĐLJ
ܤூ;,njͿ
ܤு;,njͿ
߬;ƐĞĐͿ
Fig. 1: Uplink resource provisioning for mixed traffic
and user equipments (UEs), which are uniformly distributed
in the cell. The links between devices and the BS experience
Rayleigh fading. Here, we focus on the coexistence of two
distinct types of traffic, named PIand PH, in the uplink di-
rection. PIstands for generated traffic from a massive number
of low-cost, energy, and radio front-end devices with small
data payload size, while PHstands for generated traffic from
conventional user equipments like smart-phones. Motivations
for considering only the uplink direction can be the proposed
uplink/downlink decoupling architecture for the 5G [14], and
the proposed multi radio access technology (RAT) architecture
[15], in which cellular users can be served with LTE and WiFi
for uplink and downlink transmissions respectively. Splitting
radio resources for IoT and HoC traffic has been frequently
used in literature [9–11] and standards [16] as their characteris-
tics and QoS requirements are fundamentally different. In split
resource provisioning, two sets of RACH and physical uplink
shared channel (PUSCH) resources are reserved to occur
periodically with period TRAO for serving the mixed traffic.
In each period, RACH and PUSCH resources span over τrand
τpseconds respectively, as depicted in Fig. 1. PIand PHare
generated according to two Poisson processes with arrival rates
of λIand λHrespectively. The numbers of available RACH
preambles in each random access opportunity (RAO) for PI
and PHtraffic are denoted as MIand MHrespectively. Also,
the allocated bandwidths for uplink transmission of PIand PH
traffic over PUSCH are denoted as BIand BHrespectively.
III. FORMULATING THE PERFORMANCE METRICS
A. Energy Efficiency for IoT Communications
For most reporting IoT applications, packet generation at
each device can be modeled as a Poisson process [17], and
hence, energy consumption of a device can be seen as a semi-
regenerative process where the regeneration point is at the end
of each successful data transmission. In each period of activity,
IoT devices wake up, gather and process data, establish uplink
connection with the BS, reserve uplink resources, and transmit
data over granted physical uplink resources. The energy effi-
ciency (EE) metric for IoT communications, denoted as UE,I,
is defined as the ratio between amount of useful transmitted
bits per reporting period and the total consumed energy in the
respective period for communications. Then, UE,I for node i
is formulated as:
Ui
E,I =˜
DI
Ei
CS +Ei
DT
,(1)
h ^
Ğůů/ŶĨŽ
WZ,͗ZĂŶĚŽŵĐĐĞƐƐ
ZĞƋƵĞƐƚ;ZE͕^Z͕ĂƵƐĞ͕
W,Ϳ
W,͗hƉůŝŶŬƐƐŝŐŶŵĞŶƚ
;Z,ƌĞĨĞƌĞŶĐĞ͕Wh^,
ĂůůŽĐĂƚŝŽŶ͕^sZсϬ͕Ͳ
ZEd/ĂƐƐŝŐŶŵĞŶƚͿ
Wh^,͗ĂƚĂƚƌĂŶƐĨĞƌ
;d>>/ͬ^ͲdD^/͕D^s^сϬ͕
ůĂƐƚсƚƌƵĞ͕ĚĂƚĂͿ
W,͗hƉůŝŶŬĐŬ
;d>>/ͬ^ͲdD^/͕ͲZEd/
ĐŽŶĨŝƌŵĂƚŝŽŶ͕^sZсϭͿ
ܶ
௦௬
ܶ
௦
ܶ
௧௫ǡ
ܶ
௧௫ǡ
ܶ
;dƵƌŶƌĂĚŝŽŽŶͿ
;^ůĞĞƉͿ
ƵƚLJLJĐůĞ
ZĞƉŽƌƚŝŶŐWĞƌŝŽĚ
;tĂŬĞƵƉͿ
ĂƚĂŐĂƚŚĞƌŝŶŐ
;tĂŬĞƵƉͿ
ܶ
௦
Fig. 2: Connection establishment and data transmission pro-
cedure for cellular IoT
where ˜
DIdenotes the average size of useful data to be
transmitted; Ei
CS the average consumed energy in connection
establishment; and Ei
DT the average consumed energy in
scheduled data transmission. The presented energy efficiency
model can be used along with any medium access control
(MAC) solution for IoT, including cellular-based solutions. Let
us focus on LTE Release 13 for narrow band IoT [18, chapter
7] as an example, where its connection establishment and data
transmission procedures have been depicted in Fig. 2. In this
case, Ei
CS consists of: (i) energy consumption in receiving
primary system information (PSI) and performing synchro-
nization: (Tsyn +Tpsi)Pc,where Pcdenotes the consumed
energy in electronic circuits and Tsyn +Tpsi has been specified
in [18]; and (ii) energy consumption in random access:
1
Fr,I Ttx,r (Pc+ξPi,r )+PcTas,(2)
where Ttx,r and Tas denote the average spent times in sending
a preamble and receiving response as specified in [18], ξ
the inverse power amplifier efficiency, Fr,I is the average
probability of successful resource reservation over RACH. The
transmit power over RACH can be modeled as
Pi,r =Po,r/gi,
where Po,r is the preamble received target power, which is
broadcast by the BS [18]. Also, the consumed energy in data
transmission over physical uplink shared channel (PUSCH) is
modeled as
Ei
DT =1
Fp,I PcTack +(Pc+ξPi,p)τp
¯
SI,
where the average probability of successful transmission over
PUSCH is denoted as Fp,I , and the average spent time in
receiving acknowledgment is denoted as Tack. Denote channel
gain between device iand the BS as gi=hd−σ
i, where h∼
exp(1),didenotes the communications distance, and σthe
pathloss exponent. Then, the transmit power over PUSCH is
modeled as:
Pi,p =[2
DI
τpBI/[¯
SI+¯
WI]−1]ΦIγ0diσ/h, (3)
in which, noise power spectral density (PSD) at the receiver
is modeled as N0, size of data plus overhead to be transmitted
as DI, numbers of present and newly arriving devices to be
scheduled as ¯
WIand ¯
SIrespectively, SINR gap between chan-
nel capacity and a practical coding and modulation scheme as
γ0,ΦI=(ΘI+N0)BI, and the upperbound on the PSD of out-
of-cell interference as ΘI. Regarding the uniform distribution
of devices in the cell, one can find the average required
transmit power over PUSCH and RACH as:
¯
Pp=R
0
Pi,p
2y
R2dy =σ+1
0.5R2−σ[2
DI
τpBI/[SI+WI]−1]ΦIγ0,
(4)
¯
Pr=R
0
Pi,r
2y
R2dy =Rσ−2Po,r.(5)
One sees that all parameters that contribute in formulating
UE,I have been specified, unless ¯
SI,¯
WI,Fp,I , and Fr,I ,
which are derived using steady state analysis in the following
subsections.
1) Markov Chain for Modeling Uplink IoT Communica-
tions: The number of deployed IoT devices in future cellular
networks is expected to become much higher than the existing
UEs. Due to the surge in the number of IoT devices, access
class barring has been standardized to prevent network con-
gestion [6]. Based on the ACB, in case collision occurs in
preambles allocated to IoT, the colliding nodes will contend
after QT
RAO seconds, where Qis geometrically distributed
with parameter q. Denote the total number of PIdevices in
the cell as NI, total number of available preambles for PIas
MI, number of devices attempting for RACH access in the
kth RAO as Uk, and number of connected devices which have
requested uplink service before kth RAO but have not been
scheduled yet as Wk. By considering (Uk,W
k)as the state of
the system, one can form a Markov chain in order to evaluate
the performance. We have:
Uk+1 =Uk−Sk+Ak+1,
Wk+1 =Wk+Sk−Zk,
where Skdenotes number of devices which successfully pass
the RACH procedure in the kth RAO, Ak+1 number of newly
arriving devices to the system between kth and k+1
th RAO,
and Zknumber of nodes which successfully receive uplink
service and leave the system. The probability of transition from
state (i1,j
1)to (i2,j
2)is derived as:
p(i1,j1);(i2,j2)=
min{M,i1}
s=0
j1+s
z=0
pr(Uk+1=i2,W
k+1=j2|Uk=
i1,W
k=j1,S
k=s, Zk=z)pr(Sk=s, Zk=z|Uk=i1,W
k=j1),(6)
where,
pr(Uk+1=i2,W
k+1=j2|Uk=i1,W
k=j1,S
k=s, Zk=z)=
pr(Uk+1=i2|Uk=i1,W
k=j1,S
k=s, Zk=z)×
pr(Wk+1=j2|Uk=i1,W
k=j1,S
k=s, Zk=z,Uk+1=i2),
pr(Wk+1=j2|Uk=i1,W
k=j1,S
k=s, Zk=z, Uk+1=i2)
=1if j2=j1+s−z,
0O.W.
Also,
pr(Uk+1=i2|Uk=i1,W
k=j1,S
k=s, Zk=z)
=pr(Ak+1=i2-(i1-s)|Uk=i1,W
k=j1,S
k=s, Zk=z),
=B(Nh−i1−j1+z,i2−i1+s, 1-e−λhTRAO ),
for i1-s≤i2≤Nh-s-j1+z,
and 0otherwise. In this expression,
B(x, y, z)=x
yzy(1 −z)x−y.
Furthermore,
pr(Sk=s, Zk=z|Uk=i1,W
k=j1)=
pr(Sk=s|Uk=i1,W
k=j1)pr(Zk=z|Sk=s, Uk=i1,W
k=j1),
pr(Sk=s|Uk=i1,W
k=j1)=pr(Sk=s|Uk=i1)=
i1
v=1 pr(Sk=s|Vk=v, Uk=i1)pr(Vk=v|Uk=i1),
where Vkdenotes number of nodes that decide to contend for
channel access out of Ukattempting nodes due to the ACB
scheme, and we have:
pr(Vk=v|Uk=i1)=B(i1,v,q).(7)
In the following subsections, we investigate probabili-
ties of failure in RACH and PUSCH transmissions, i.e.
pr(Sk=s|Vk=v, Uk=i1)and pr(Zk=z|Sk=s, Uk=i1,W
k=j1).
2) Probability of Failure in RACH Transmission: If several
devices select a preamble simultaneously in a RAO, collision
in RACH occurs. Considering the problem of distributing v
distinct objects in MIdistinct boxes,
pr(Sk=s|Vk=v, Uk=i1)=pr(Sk=s|Vk=v)
is the probability of having boxes occupied with one object,
and is derived as [19]:
[−1]sMI!v!
[MI]vs!
min(MI,v)
l=s
[−1]l[MI−v]v−l
[l−s]![MI−l]![v−l]!.(8)
3) Probability of Failure in PUSCH Transmission: If the
required transmit power for successful transmission of DI
bits over the assigned PUSCH resources is higher than the
maximum allowed transmit power, uplink transmission will be
unsuccessful. Then, one can derive the successful transmission
probability over PUSCH as:
pr(Pi,p <P
max,I )(a)
=exp −dσ
iγ0ΦγI
Pmax,I ,(9)
where Pmax,I is the maximum allowed transmit power, and (a)
is due to the fact that his exponentially distributed. Regarding
the uniform distribution of devices in the cell, the probability
distribution function (PDF) of the distance between a device
and the BS is f(y)= 2y
R2, where Ris the cell radius and y
is the communications distance. Then, the average probability
of successful transmission over PUSCH is derived as:
Fp,I =R
0exp(−yσγ0ΦIγI
Pmax,I
)2y
R2dy. (10)
The integral in (10) can be evaluated using the integral tables
in [20]. For example, when σ=4:
Fp,I =√πErfc(AR2)/2AR2,
where A=γ0ΦIγI
Pmax,I , and Erfc(·) is the error function [20].
Now, we have:
pr(Zk=z|Sk=s, Uk=i1,W
k=j1)=B(s+j1,z,F
p,I ).
4) Deriving Steady-State Probabilities: To finalize our
analysis, the steady state probabilities are derived here. As
Ukand Wkchoose their values from {0,1,···,N
I}, the total
number of states is (NI+1)2. In order to simplify the analysis,
one can reduce the number of states to NuNw, where:
Nu=argmin
i2{p(i1,j1);(i2,j2)≤ε, ∀i1,j
1,j
2},
Nw=argmin
j2{p(i1,j1);(i2,j2)≤ε, ∀i1,i
2,j
1},
and εis a very small value. Now, the steady state probabilities
of the system, i.e. π=[π(0,0),···,π
(Nu,Nw)], are calculated
by solving:
π=πP,
Nu
i=0 Nw
j=0 π(i,j)=1,(11)
where Pis the transition matrix, and its entries have been
derived in (6). Also, the average probability of successful
RACH transmission can be derived as:
Fr,I =Nu
i=0 Nw
j=0 π(i,j)i
l=2 B(i, l, q)fr,I (l, MI),
(12)
where fr,I (l, MI)is the success probability when lde-
vices contend for MIpreambles, and is derived as
T[(MI−1)/MI]l−1.Furthermore, the average number of PI
devices that successfully pass the RACH procedure in each
RAO is derived as:
¯
SI=Nu
i=0 Nw
j=0 π(i,j)
i
l=1 B(i, l, q)l
x=1 xB(l, x, fr,I (j)),(13)
while the average number of devices that are to be scheduled
per RAO is:
¯
SI+¯
WI=¯
SI+Nu
i=0 Nw
j=0 π(i,j)j.
One sees solving (11) requires knowledge on ¯
SI,¯
WI, and
Fr,I , which are in turn functions of π(i,j). Then, this problem
can be solved in an iterative way. Now, one can assemble the
derived expressions in the previous subsection, and formulate
the energy efficiency metric as:
UE,I =˜
DIPc(Tsyn +Tpsi)+ PcTtx,r +ξRσ−2Po,rTas
Fr,I
+
1
Fp,I PcTack +[Pc+ξ2(σ+1)
R2−σ[2
DI
τpBI/[SI+WI]–1]ΦIγ0]τp
¯
SI.
(14)
The couplings between energy efficiency, individual and net-
work battery lifetime for PItraffic have been investigated in
[21], where the interested reader may refer to see how network
lifetime can be derived from the EE expression.
B. Experienced Delay in Data Transmission
The second KPI which is investigated in this work is the
experienced delay (ED) in uplink communications, i.e. from
first access reservation transmission to service completion
epoch. The average experienced delay by PHtraffic in uplink
communications can be modeled as:
UD,H =1
Fr,H −1+TRAO +1
Fp,H
TRAO,(15)
where [x]+=max{0,x},Fr,H is the probability of successful
reservation over allocated RACH resources to PHtraffic, and
Fp,H =Dtx
DHis defined as the ratio between average size of
data that can be transmitted over allocated PUSCH resources
per device per RAO and the average size of queued data to
be transmitted per PHdevice denoted by DH.Dtx is derived
as:
Dtx =R
0
BHτp
¯
SH+¯
WH
log(1 + Pmax,H
ΦHγ0
r−σ)dr, (16)
σ=4
=BHτp/ln(2)
¯
SH+¯
WH√aR2
2Atan(R2
√a)+log(1+ a
R4),
where a=Pmax,H
ΦHγ0.Also, ¯
SH,¯
WH, and Fr,H are derived
following the same procedure as we presented in subsections
III-A1 - III-A4 for ¯
SI,¯
WI, and Fr,I . Furthermore, the average
experienced delay by PItraffic can be modeled as:
UD,I =1
Fr,I −1+TRAO +1
Fp,I
TRAO,(17)
where Fp,I and Fr,I have been found in (10) and (12)
respectively.
C. Uplink Spectral Efficiency
Spectral efficiency (SE) in uplink communications is an-
other important KPI for cellular networks which is investigated
in this section. Spectral efficiency, in terms of bit/sec/Hz, can
be modeled as:
US=MI
MI+MH
τr
τr+τp
+BI
BI+BH
τp
τr+τp
US,I
+MH
MI+MH
τr
τr+τp
+BH
BI+BH
τp
τr+τp
US,H (18)
where,
US,I =Fp,I (¯
SI+¯
WI)DI
MI
MI+MH(BI+BH)τr
τr+τp+BI
τp
τr+τp
,
US,H =(¯
SH+¯
WH)Dtx
MI
MI+MH(BI+BH)τr
τr+τp+BH
τp
τr+τp
.
D. Energy Consumption Modeling for the Uplink Module
Energy consumption (EC) of the uplink module of the BS
per unit of time can be modeled as:
UBS =TsPs+1−TsPes,(19)
where Psand Pes denote power consumption in service and
energy-saving modes respectively. Also, Tsindicate percent-
age of time the BS spends in the service mode. As PIand PH
traffic are served separately, Tsfor PItraffic can be modeled
using an M/D/1 queuing model as:
Ts,I =τr
τr+τp
+λI
1
Fp,I
τp
¯
SI+¯
WI
.
Also, the Tsfor PHtraffic is modeled as:
Ts,H =τr
τr+τp
+λH
1
Fp,H
τp
¯
SH+¯
WH
.
Then, one may derive the average percentage of time the BS
spends in the service mode as:
Ts=τr
τr+τp
+max{λI
1
Fp,I
τp
¯
SI+¯
WI
;λH
1
Fp,H
τp
¯
SH+¯
WH}.
IV. TRADEOFFS IN RESOURCE PROVISIONING
From the derived expressions in the previous section, one
sees that the energy/spectral efficiency, delay, and energy
consumption performance metrics can be controlled by: (i)
frequency of occurrence of RAOs, i.e. 1/TRAO; (ii) RACH
resource partitioning, i.e. MIand MH; and (iii) PUSCH
resource partitioning, i.e. BIand BH.
From (2), one sees that the average energy efficiency for PI
traffic decreases when the success probability in RACH reser-
vation decreases, i.e. when the allocated RACH resources to
PIdecreases, which in turn increases the collision probability
in respective resources. Also, one sees in (3) that an increase
in the amount of allocated PUSCH resources to PItraffic
increases energy efficiency as it decreases the required transmit
power for reliable data transmission to the BS. In other words,
with more radio resources, each device can decrease the data
transmission rate and send data with a lower transmit power.
Furthermore, a decrease in TRAO or τp, results in a lower
collision probability because in turn it decreases the number
TABLE I: Parameters for numerical analysis
Parameters Value
Cell inner and outer radius 50 m, 500 m
Pathloss 128.1+37.6 log( d
1000 )
Bandwidth 10 MHz
NI,λ
I,D
I24000,1/450,62 Bytes
NH,λ
H,D
H6000,1/150,30 KBytes
RACH configuration Config. 0 in LTE-A
Number of preambles per RAO 54
τr,τ
p2, 16 ms
Pc,Ps,Psl 0.05, 130, 10 W
Pmax,I ,P
max,H 0.2,2W
of devices that contend at each RAO, and hence, decreases the
collision probability.
From (16), one sees the average experienced delay by PH
traffic decreases when the success probability over RACH
increases, i.e. when the allocated RACH resources to PH
traffic increases. Also, as the average data payload size for PH
traffic is much larger than the PItraffic, the experienced delay
for PHtraffic decreases when Dtx increases. Dtx increases
when either the allocated PUSCH resources to PHtraffic, i.e.
BH, increases or the allocated uplink resources to PUSCH, i.e.
τp, increases. From (19), one sees a non-optimal allocation
of RACH and PUSCH resources to PIand PHtraffic not
only degrades QoS of connected devices by decreasing the
energy efficiency and increasing the experienced delay, but
also increases the energy consumption of the BS because
in this case BS will be in the service mode for a longer
period of time. Also, from (19) one sees that change in
the frequency of occurrence of RACH resources can either
increase or decrease energy consumption of the BS. The
optimal τp, which minimizes BS’ energy consumption for a
given set of system and traffic parameters can be derived from
(19). Finally, comparing (14) with (18) indicates that while
further RACH and PUSCH resource allocation to PItraffic
can improve their energy efficiency and battery lifetimes, it
may reduce spectral efficiency of the network. This is due to
the fact that PItraffic usually consists of a large number of
short-lived sessions, where each of them delivers only a small
amount of data to the network.
From an overall system perspective, we aim at minimizing
the consumed energy and radio resources in the access net-
work, maximizing the energy efficiency of communications,
and minimizing the experienced delay in data transmission.
Unfortunately, as discussed in the above, these objectives
cannot be treated separately because they are coupled in
conflicting ways such that improvements in one objective may
lead to deterioration of the other objectives. In the next section,
we investigate these tradeoffs by simulations.
Energy Efficiency (bits/joule)
α
: RACH alloc. to IoT
β
: PUSCH alloc. to IoT
0
10
0
1
5
×10
5
0.8
0.6
10
-2
10
0.4
0.2
0
Max energy efficiency
(a) Energy efficiency for IoT traffic
Exprienced Delay (sec)
β: PUSCH alloc. to IoT
0
100
10-1
10-2
α
: RACH alloc. to IoT
1
0.8
0.6
10-3 0.4
0.2
0
1
2
Min delay
(b) Experienced delay for HoC traffic
β
: PUSCH alloc. to IoT
α
: RACH alloc. to IoT
0
100
0.5
1
Spectral Efficiency (bits/sec/Hz)
1
0.8
10-2 0.6
1.5
0.4
0.2
10-4 0
Maximum
spectral efficiency
X: 0.92
Y: 0.004
Z: 1.311
(c) Uplink spectral efficiency
β
: PUSCH alloc. to IoT
α
: RACH alloc. to IoT
110
1
0.5
10-3
10-2
10-1
120
0100
Avg. cons. energy
per unit time (J)
130
X: 0.582
Y: 0.07
Z: 112
Min Energy
(d) Energy consumption for the BS
Fig. 3: Performance analysis
10
-3
10
-2
10
-1
10
0
β
: PUSCH alloc. to IoT
110
120
130
Ave. cons. energ
y
per unit time (J)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
α
: RACH alloc. to IoT
110
120
130
Ave. cons. energy
per unit time (J)
Min Energy
Min Energy
β
=0.07
α
=0.582
(a) Energy consumption for the BS versus αand β
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
x 10
5
0
0.2
0.4
0.6
0.8
1
Time (× reporting period)
CDF of individual lifetimes
α=28%, β=0.5%
α=28%, β=5%
α=37%, β=0.5%
α=37%, β=5%
α=74%, β=0.5%
α=74%, β=5%
(b) CDF of individual battery lifetimes. Battery capacity: 500 J, static
energy consumption per reporting period: 50 μJ, and λI=1/150.
0.6
0.8
1
β: PU
SC
H alloc. to IoT
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
α
: RACH alloc. to IoT
10-2
Max energy efficiency for IoT
EE, EC, and ED
increase; SE decrease
EC increases;
EE and ED decrease
Min BS energy consumption
EC and SE
increase
Min Delay for HoC
Max uplink spectral
efficiency (HoC+IoT)
(c) Performance tradeoffs
Fig. 4: Detailed performance analysis
V. P ERFORMANCE EVALUATIONS
The implemented system model in this section is based on
the uplink of a single cell with coexistence of IoT and HoC
traffic, in which devices are randomly distributed according
to a spatial Poisson point process. The simulation parameters
can be found in Table I. Fig. 3a illustrates energy efficiency
in uplink communications for IoT traffic versus αand β, the
fraction of allocated RACH and PUSCH resources to IoT and
HoC traffic respectively. It is evident that by increasing the
amount of provisioned resources, the energy efficiency in data
transmission, and hence the battery lifetime, can be improved
significantly. Fig. 3b represents the experienced delay in data
transmission for HoC Traffic. One sees that increase in the
amount of allocated RACH and PUSCH resources to IoT
traffic increases the experienced delay by HoC traffic. Fig.
3c depicts network spectral efficiency (SE) versus αand β.
We see that the maximum SE is achieved when 0.92% of
RACH resources and 0.4% of PUSCH resources are allocated
to IoT traffic. This is due to the fact that the payload size for
IoT traffic is much less than the payload size for HoC traffic,
while the expected number of concurrent access requests from
IoT traffic is higher than the one of HoC traffic. Fig. 3d
illustrates average energy consumption of the BS in serving
mixed IoT/HoC traffic. It is interesting to see that EC is
jointly quasi-convex over αand β. One sees that the minimum
EC for the BS is achieved when 58% of RACH and 0.7%
of PUSCH resources are allocated to IoT traffic. Detailed
analysis of EC as a function of αand βis depicted in
Fig. 4a, in which, one sees how uplink resource improvising
affects energy consumption of the BS, and where are the
optimized operation points. Fig. 4b represents the cumulative
distribution function (CDF) of individual lifetimes of IoT-type
devices. From this figure, it is evident that increase in the
amount of allocated RACH/PUSCH resources to IoT traffic
can substantially prolong the battery lifetime. For example,
by considering the first energy drain (FED) as the network
lifetime, increasing βfrom 0.5% to 5% can prolong the
FED network lifetime by 30% when α= 37%. To ease
understanding of coupling among battery lifetime for IoT, ED
for HoC, CE for the BS, and network SE, in Fig. 4c the
above derived individually optimized operation points have
been depicted together. The background of this figure is a 2D
view of Fig. 3d, in which different energy consumption levels
are depicted by different colors. Considering the optimized
EC operation point as a reference, one sees in Fig. 4c that the
average energy consumption of the BS, energy efficiency of
IoT communications, and experienced delay by HoC increase
when extra radio resources are allocated to IoT traffic. The
increase in energy efficiency of IoT communications is due to
the fact that the access probability over RACH and success
probability over PUSCH increase in the amount of allocated
resources, which in turn results in decreasing QoS for HoC.
VI. CONCLUSION
In this work, we have investigated battery lifetime, spectral
efficiency, energy efficiency, and delay tradeoffs in green
cellular network design with coexistence of IoT and HoC
traffic. For a cellular network in which two types of distinct
traffic are served, analytical expressions for energy consump-
tion of the BS in serving the mixed traffic, as well as
energy/spectral efficiency and experienced delay of connected
devices in uplink data transmission have been derived as a
function of system, traffic, and resource provisioning param-
eters. Then, the performance impacts of control parameters
on the introduced tradeoffs have been studied. Significant
impacts of uplink resource provisioning on the battery lifetime
of energy-limited devices, energy/spectral efficiency of the
network, and experienced delay in uplink communications
have been presented using analytical and simulation results.
The derived results figure out the ways in which scarce radio
and energy resources for the BS and QoS for human-oriented
communications could be preserved while coping with the
ever increasing number of energy-limited IoT-type devices in
cellular networks.
REFERENCES
[1] S. Andreev et al., “Understanding the IoT connectivity landscape: a
contemporary M2M radio technology roadmap,” IEEE Communications
Magazine, vol. 53, no. 9, pp. 32–40, September 2015.
[2] 3GPP TS 22.368 V13.1.0, “Service requirements for machine-type
communications,” Tech. Rep., 2014.
[3] Nokia Networks, “Looking ahead to 5G: Building a virtual zero latency
gigabit experience,” Tech. Rep., 2014.
[4] H. Tullberg et al., “Towards the METIS 5G concept: First view on
horizontal topics concepts,” in European Conference on Networks and
Communications. IEEE, 2014, pp. 1–5.
[5] M. I. Hossain, A. Azari, and J. Zander, “DERA: Augmented random
access for cellular networks with dense H2H-MTC mixed traffic,” 2016,
to be published.
[6] A. Biral et al., “The challenges of M2M massive access in wireless
cellular networks,” Digital Communications and Networks, vol. 1, no. 1,
pp. 1–19, 2015.
[7] A. Azari and G. Miao, “Energy efficient MAC for cellular-based
M2M communications,” in 2nd IEEE Global Conference on Signal and
Information Processing, 2014.
[8] T. P. de Andrade et al., “The impact of massive machine type commu-
nication devices on the access probability of human-to-human users in
lte networks,” in 2014 IEEE Latin-America Conference on Communica-
tions, 2014, pp. 1–6.
[9] Y.-C. Pang et al., “Network access for M2M/H2H hybrid systems: a
game theoretic approach,” IEEE Communications Letters, vol. 18, no. 5,
pp. 845–848, 2014.
[10] M. Mehaseb, Y. Gadallah, A. Elhamy, and H. El-Hennawy, “Classifica-
tion of LTE uplink scheduling techniques: An M2M perspective,” IEEE
Communications Surveys Tutorials, no. 99, 2015.
[11] S. Y. Lien and K. C. Chen, “Massive access management for QoS
guarantees in 3GPP machine-to-machine communications,” IEEE Com-
munications Letters, vol. 15, no. 3, pp. 311–313, March 2011.
[12] A. Aijaz et al., “Energy-efficient uplink resource allocation in LTE net-
works with M2M/H2H co-existence under statistical QoS guarantees,”
IEEE Transactions on Communications, vol. 62, no. 7, pp. 2353–2365,
July 2014.
[13] A. Azari and G. Miao, “Lifetime-aware scheduling and power control
for cellular-based M2M communications,” in IEEE Vehicular Technology
Conference (VTC Spring), 2015.
[14] H. Elshaer, F. Boccardi, M. Dohler, and R. Irmer, “Downlink and uplink
decoupling: A disruptive architectural design for 5G networks,” in IEEE
Global Communications Conference, Dec 2014, pp. 1798–1803.
[15] O. Galinina et al., “5G Multi-RAT LTE-WiFi ultra-dense small cells:
Performance dynamics, architecture, and trends,” IEEE J. Sel. Areas
Commun, vol. 33, no. 6, pp. 1224–1240, June 2015.
[16] 3GPP TS 45.820, “Cellular system support for ultra-low complexity and
low throughput internet of things (ciot,” Tech. Rep., (Rel. 13).
[17] 3GPP, “USF capacity evaluation for MTC,” Tech. Rep., 2010, TSG
GERAN 46 GP-100894.
[18] 3GPP TS 36.213, “Evolved universal terrestrial radio access , physical
layer procedures,” Tech. Rep., (Rel. 13).
[19] D. Shen and V. O. Li, “Performance analysis for a stabilized multi-
channel slotted ALOHA algorithm,” in IEEE Proceedings onPersonal,
Indoor and Mobile Radio Communications, vol. 1, 2003, pp. 249–253.
[20] D. Zwillinger, Table of integrals, series, and products. Elsevier, 2014.
[21] G. Miao, A. Azari, and T. Hwang, “E2-MAC: energy efficient medium
access for massive M2M communications,” IEEE Transactions on
Communications, no. 99, 2016, to be published.
