Content uploaded by Irina Kochetkova
Author content
All content in this area was uploaded by Irina Kochetkova on Dec 03, 2015
Content may be subject to copyright.
Analyzing Impacts of Coexistence between M2M
and H2H Communication on 3GPP LTE System
Irina Gudkova1, Konstantin Samouylov1, Ivan Buturlin1, Vladimir Borodakiy2,
Mikhail Gerasimenko3, Olga Galinina3, and Sergey Andreev3
1Peoples’ Friendship University of Russia (PFUR), Russia
{igudkova,ksam}@sci.pfu.edu.ru, ivan buturlin@mail.ru
2JSC “Concern Sistemprom”, Russia
bvu@systemprom.ru
3Tampere University of Technology (TUT), Finland
{mikhail.gerasimenko,olga.galinina,sergey.andreev}@tut.fi
Abstract. In this paper, we consider 3GPP LTE cellular system where
machine-to-machine (M2M) devices and human-to-human (H2H) users
transmit their data into the network. By contrast to previous studies
which primarily focused on M2M overload protection and respective con-
trol mechanisms, this work concentrates on system operation when M2M
and H2H data flows coexist in the network. In particular, we propose an
integrated simulation-analytical framework to evaluate relevant perfor-
mance characteristics (data transmission delays, blocking probabilities,
etc.) with both Markov process based analysis and system-level simula-
tions. Our results indicate that the proposed methodology demonstrates
acceptable levels of convergence between analytical and simulations com-
ponents, as well as becomes useful to characterize impacts of M2M/H2H
coexistence on radio resource allocation in 3GPP LTE across a number
of important M2M-centric scenarios.
1 Introduction and Background
Machine-to-machine (M2M) communication is believed to reshape the Internet
as we know it today, as billions of unattended devices (sensors, actuators, smart
meters, etc.) become connected and send their data into the network [1]. Such
massive connectivity offers novel attractive services, but also raises significant
challenges to manage large number of devices, typically transmitting only small
data fragments, across a wide range of emerging applications [2]. This is espe-
cially true for current cellular technology (e.g., 3GPP LTE [3]), which has been
historically optimized for human-to-human (H2H) traffic and therefore creates
inefficiency at every step of M2M communication, from initial network entry to
actual data transmission [4].
Cellular industry, and in particular 3GPP standards community, has recently
been very active with several study and work items identified on M2M communi-
cation [5]. These primarily focused on overload protection, when a large number
of M2M devices attempt to connect to the network in a correlated manner [6].
A. Mellouk et al. (Eds.): WWIC 2014, LNCS 8458, pp. 162–174, 2014.
c
Springer International Publishing Switzerland 2014
Analyzing Impacts of Coexistence between M2M and H2H Communication 163
Such scenarios may be characteristic for modern smart grid deployments, where
a high density of metering devices transmit their ”last gasp” signaling in case
of a massive power outage event. In some situations, this excessive messaging
quickly deteriorates available capacity of LTE signaling channels (i.e., PRACH:
physical random access channel and PDCCH: physical downlink control channel)
and results in significant outage when meters cannot access the network with
their data [7], [8]. Furthermore, at these periods, conventional H2H users suffer
from denial of service by the network as well, as they share the same signaling
channels with M2M devices.
The above overload protection research resulted in respective control mech-
anisms (e.g., EAB: extended access barring) standardized for LTE Release-11
and designed to mitigate initial network entry peaks by barring some of the
(delay-tolerant) M2M devices from accessing the network for predefined periods
of time [9]. These simple mechanisms, whereas offer an immediate solution to the
problem, do not help control regular system operation when both M2M devices
and H2H users already coexist in the network. Little is known about such co-
existence with only a few research works mainly addressing improved scheduler
design by taking into account the typical properties of M2M traffic [10]. These
single-issues papers are primarily build on computer simulations and do not offer
comprehensive understanding of M2M/H2H coexistence.
In this work, we bridge the indicated gap by proposing an adequate simulation-
analytical framework to capture the main impacts of M2M communication on
the conventional H2H traffic. In particular, we mathematically characterize the
key performance characteristics of M2M and H2H communications, such as data
transmission times and blocking probabilities [11], and confirm our results by
extensive system-level evaluations across a number of important M2M-centric
scenarios. Our framework allows to optimize radio resource allocation proce-
dures in a cellular network and achieve understanding of resulting system per-
formance to reach good balance between M2M and H2H communication. The
rest of the text is organized as follows. Section 2 details our mathematical model
and introduces its core assumptions. Further, in Section 3, we conduct numerical
analysis of representative M2M-centric scenarios and derive the key performance
characteristics. Section 4 introduces our M2M-aware system-level simulator and
offers some initial performance evaluation results, primarily, for the purposes of
verification of the analytical framework.
2 System Model of LTE Cell with H2H and M2M Traffic
Consider a single cell of LTE network (see Figure 1) with the peak capacity of
Cunits of channel resource (UCR), measured in bps. All users employ identical
H2H-service, such as voice telephony or video streaming. Additionally, the cell
supports transmission of M2M data fragments of a particular type from many
M2M devices. The system reserves ChUCR to offer H2H-services to users. Here-
inafter, the indexes ”m” and ”h” in mathematical expressions differentiate if a
specific parameter applies to M2M or to H2H traffic, respectively. Consequently,
164 I. Gudkova et al.
not more than Cm=C−ChUCR are available for M2M devices, while not less
than ChUCR are available for H2H devices.
A minimum of bmUCR is required to transmit M2M data fragments. Cor-
respondingly, in order to transmit the current number of the data fragments,
UCR are grouped into fixed transmission zones comprising cUCR. Then M=
c/bm=max{y∈N: y≤c/bm}is the maximum number of data fragments
which may be transmitted in one such fixed zone. Further, we assume that the
cell might allocate S=Cm/ctransmission zones to serve M2M user traffic.
The arrival flow of requests from M2M devices to transmit their data is as-
sumed to be Poisson with the rate of λm[1/time-unit = 1/s], whereas the length
of each data fragment is exponentially distributed with the mean θ[UCR×time-
unit = bit]. Denote a=λmθ[UCR] as the corresponding offered load rate.
These simplifying assumptions are made for the sake of analytical tractability
and provide a first-order insight into the performance of the considered system.
Further, H2H-services require bhUCR. We consider the arrival flow of requests
from the users demanding H2H-service to be Poisson with the rate of λh[1/time-
unit], while the duration of H2H-service is exponential with the mean of 1/μ
[time-unit]. Denote as ρ=λh/μ[Erlang] the respective offered load rate by H2H
users.
The considered model is a combination of First Come – First Served streaming
model and Egalitarian Processor Sharing (EPS) elastic traffic model.
Fig. 1. Proposed model of resource distribution in LTE cell
Analyzing Impacts of Coexistence between M2M and H2H Communication 165
In our model, three different scenarios are possible when a new data fragment
transmission request is generated by an M2M device:
1. The request is accepted for service and additional resources are not allo-
cated. This scenario corresponds to the situation when at the moment of the
request generation the number of data fragments is such that the decrease
in their transmission rate (but not less than bm)al lows to serve this new
data fragment.
2. The request is accepted for service and a new fixed transmission zone is
allocated for its service. This scenario corresponds to the situation when at
the moment of the request generation the number of data fragments is such
that the decrease in their transmission rate (but not less than bm)does not
allow to serve this new data fragment. At the same time, there are at least
cUCR of free (unallocated) resources available for M2M service out of the
maximum Cmto allocate a new transmission zone.
3. The request is blocked without any impact on the rate of the spawning
Poisson process.
Similarly, two different scenarios are possible when a new service request is
generated by an H2H device:
1. The request is accepted for service when at the moment of its generation
there are at least bhof ChUCR of free resource.
2. The request is blocked without any impact on the rate of the spawning
Poisson process.
Let Nm(t) be the number of M2M data fragments transmitted at the moment
t≥0, and Nh(t) be the number of users which at the moment t≥0 are receiving
H2H-service. Then the operation of the considered LTE cell model featuring
both H2H and M2M traffic can be described by the compound random process
{(Nm(t),N
h(t)) ,t>0}, over the state space
X={(nm,n
h): nhbh≤C−c(nm),n
mbm≤Cm,n
m≥0,n
h≥0},
|X | =(S·c)/bm
nm=0 C−c(nm)
bh+1
,(1)
where c(nm)=c·nm/M=c·min {y∈N,y≥nm/M}is the number of UCR
allocated for the transmission of nmM2M data fragments.
For the considered model, we may derive a system of balance equations. The
equation corresponding to the state (nm,n
h)∈X is given as follows:
p(nm,n
h)×[λm·1{(nm,n
h)/∈B
m}+(c(nm)/θ)·1{nm>0}+
+λh·1{(nm,n
h)/∈B
h}+nhμh]=p(nm−1,n
h)·λm·1{nm>0}+
+p(nm+1,n
h)·(c(nm+1)/θ)·1{(nm,n
h)/∈B
m}+
+p(nm,n
h−1) ·λh·1{nh>0}+p(nm,n
h+1)·(nh+1)μh·1{(nm,n
h)/∈Bh},
where the boundaries of the state space may be defined by means of the sets:
Bm={(nm,n
h)∈X :nhbh>C−c(nm+1)∨(nm+1)bm>C
m},(2)
166 I. Gudkova et al.
Bh={(nm,n
h)∈X :(nh+1)bh>C−c(nm)}.(3)
The random process {(Nm(t),N
h(t)) ,t>0}constitutes a reversible Markov
process with the stationary probability distribution:
p(nm,n
h)=G−1(X)a
M·bmnmnm
i=1 i
M−1
×ρnh
nh!,(nm,n
h)∈X ,(4)
where G(X) is the constant obtained from the normalizing condition.
Further, we consider the primary time-probability characteristics of the pro-
posed LTE cell model and introduce analytical expressions to derive these. To
this end, we write the state space Xas follows:
X=
S
s=0
Xs,Xs={(nm,n
h)∈X :c(nm)=s·c}.(5)
Knowing the distribution (4) and using the state space partitioning in (5), we
arrive at the expression for the M2M request blocking probabilities Bmas well
as those for H2H devices Bh, respectively:
Bm=
(nm,nh)∈Bm
p(nm,n
h)=
S−1
s=0
(C−s·c)/bh
nh=(C−(s+1)·c)/bh+1
p(s·M, nh)+
Ch/bh
nh=0
p(S·M, nh),(6)
Bh=
(nm,nh)∈Bh
p(nm,n
h)=p0,C
bh+
S
s=1
s·M
nm=(s−1)·M+1
pnm,C−s·c
bh.(7)
The resulting formula for the mean M2M data fragment transmission time
may be given as:
Tm=Cm/bm
nm=0 (C−c(nm))/bh
nh=0 nm·p(nm,n
h)
λm(1 −Bm),(8)
where the upper part determines the mean number of the transmitted M2M data
fragments Nm.
Further, we continue by numerically analyzing the operational characteristics
of the considered resource distribution model with the fixed transmission zone
for M2M traffic in LTE cell with H2H users.
3 Numerical Analysis of the Proposed Model
As an example, we consider a single cell of LTE with the peak capacity of
C=52.8 Mbps, which is distributed between H2H users and M2M devices. For
the H2H user service, the system reserves Ch=10.56 Mbps of its capacity. Let
every M2M data fragment of θ=0.88 Mbit require a minimum of bm=0.88
Mbps. As a numerical illustration of an H2H-service, we consider streaming
Analyzing Impacts of Coexistence between M2M and H2H Communication 167
video, which has a requirement of bh=2.64 Mbps on the minimum throughput.
Assume the H2H offered load rate to be ρ= 5 Erlang. Let up to S=2fixed
transmission zones can be allocated for M2M data fragments transmission, each
of which comprising c= 20 Mbps.
Figure 2 introduces plots illustrating H2H request blocking probabilities Bh
calculated as given by formula (7), M2M data fragment blocking probabilities
Bm(6), and mean fragment transmission time Tm(8) on increasing M2M offered
load. The figure indicates that the mean fragment transmission time varies sig-
nificantly with the changing offered load. In order to explain the main reasons
behind the observed effects let us consider the plots of other probability-time
characteristics in our model.
Fig. 2. Blocking probabilities and mean data fragment transmission time
Together with the mean number of transmitted M2M data fragments Nm,we
also consider the following characteristics:
1. Mean number of the allocated fixed transmission zones for M2M devices:
¯s=
(nm,nh)∈X
nm
M·p(nm,n
h)=
S·c/bm
nm=0
(C−c(nm))/bh
nh=0 nm
M·p(nm,n
h).(9)
2. Mean number of UCR allocated for the transmission of a single data frag-
ment:
b1=(nm,nh)∈X,nm=0 c(nm)
nm×p(nm,n
h)=
=S·c/bm
nm=1 (C−c(nm))/bh
nh=0 c(nm)
nm×p(nm,n
h).(10)
168 I. Gudkova et al.
3. Probability that at least one data fragment is being transmitted:
P1=P{nm=1}=
(nm,nh)∈X,nm=0
p(nm,n
h)=
S·c/bm
nm=1
(C−c(nm))/bh
nh=0
p(nm,n
h).(11)
4. Probability that two fixed transmission zones have been allocated for serving
M2M devices:
P{s=2}=
(nm,nh)∈Xs=2
p(nm,n
h).(12)
The plots for the aforementioned characteristics of the LTE cell model are
shown in Figures 3 and 4.
Fig. 3. Time-probability characteristics ¯sand Nm
Let us now consider again the primary parameter for the performance eval-
uation of our model operation, which is the mean time Tmof the M2M data
fragment transmission (see Figure 5). We may further identify three intervals
of the M2M offered load, within which the mean number of fixed transmission
zones ¯sbelongs to the following ranges: 0 ≤¯s≤1, 1 ≤¯s≤2, and ¯s→2.
Over the first interval of the offered load for serving M2M devices, one fixed
transmission range is allocated on average and, correspondingly, 0 ≤¯s≤1. It
is important to emphasize that with the growth of the offered load from a=16
UCR, the mean transmission time Tmis showing non-uniform behavior. Over
the second interval, all UCR of the first fixed transmission zone have been used
for the data fragments transmission 1 ≤¯s≤2, and the probability increases
that two fixed zones will be allocated P{s=2}→1. When the offered load
Analyzing Impacts of Coexistence between M2M and H2H Communication 169
Fig. 4. Time-probability characteristics ¯
b1,P{nm=1},andP{s=2}
Fig. 5. Mean data fragment transmission time
reaches the value of a= 55 UCR, all available UCR are used to transmit M2M
data fragments and ¯s→2.
In what follows, we consider variation in the mean data fragment transmission
time over each of the indicated M2M offered load intervals:
1. Over the interval a=[2,16] UCR, the value of Tmgrows insignificantly
as the number of data fragments in the system is small, P1<1, and they
arrive at the low rate of Nm<1. Accounting for the fact that M2M service
follows the EPS discipline, the amount of resources taken by one M2M device
170 I. Gudkova et al.
increases up to ¯
b1≤9 UCR. Therefore, we observe minimal values of Tmin
this interval.
2. Over the interval a=[16,30] UCR, the value of Tmgrows faster and reaches
the value of Tm≈0,8 seconds. Data fragments begin to arrive with higher
rate and their mean number exceeds one, Nm≥1. Therefore, the amount of
resources allocated for the transmission of one fragment decreases down to
¯
b1≤2. Accounting for such decrease together with increase in the offered
load, the value of Tmgrows significantly.
3. Over the interval a=[32,44] UCR, when one more fixed transmission zone
has been allocated to serve M2M traffic, the value of Tmdecreases slightly.
Probability that an additional fixed transmission zone is available tends to
one P{s=2}→1, and the mean number of UCR allocated for the trans-
mission of one data fragment is ¯
b1.
4. Over the interval a=[44,55] UCR, almost all of the available UCR allocated
across two fixed transmission zones are used for data fragments transmission.
Therefore, the amount of UCR allocated for the transmission of one fragment
decreases further and the value of Tmgrows.
5. Over the interval a=[55,98] UCR, the allocated fixed transmission zones
are completely filled with M2M data fragments, and 40 ≤Nm≤48. The
mean amount of resources allocated for the transmission of one data fragment
tends to the allowed minimum of ¯
b1→bm. Within this interval of the offered
load, it is typical to observe the maximum data fragment transmit times Tm
and high loss probabilities Bm.
We proceed with detailing our simulation methodology to extend the above
mathematical analysis.
4 Simulation Methodology, Results, and Conclusions
In our past M2M work [12], [13], we were mostly concentrated on the partic-
ular features of IEEE 802.16 and 3GPP LTE technology related to signaling
channel simulations and analysis. For those purposes we employed a simplified
Protocol Level Simulator (PLS) to abstract away many realistic system features
for the sake of simulation speed. By contrast, in this paper we are considering a
more detailed simulation methodology incorporating most of the practical 3GPP
LTE features. Our approach is based on detailed System Level Simulator (SLS)
which has been developed and applied successfully in our recent publications on
next-generation wireless networks [14] focusing H2H traffic. However, this work
extends our SLS tool to enable characteristic M2M scenarios.
The core capabilities of the considered simulator are: detailed LTE MAC-layer
features (according to 3GPP LTE Release-10 specifications, fully calibrated),
dynamic channel modeling, different traffic types, user and eNodeB directivity
and location modeling, as well as many others (see Figure 6). In particular, the
basic features of the LTE implementation inside our SLS tool are: realistic 10
ms FDD frame structure, inter-cell interference, support for several scheduling
Analyzing Impacts of Coexistence between M2M and H2H Communication 171
schemes (round-robin, proportional-fair, etc.). Instead of modeling the control
channels explicitly, the respective control signaling overhead is taken into account
to speed-up the simulations. However, necessary channel procedures could be
easily integrated into the SLS, if required.
Fig. 6. High-level structure of our system-level simulator
Regarding channel models implementation, the most challenging aspects are
interference and pathloss characterization [15]. Basic ITU models (Urban Macro,
Urban Micro, ITU-R M.2135) have been realized and used in the SLS. Interfer-
ence calculation has been somewhat simplified to speed-up simulations further.
Instead of per-RB (resource block) calculations, only the percentage of intersec-
tions between the same time-frequency domain user requests (of different cell in
a sector) is accounted for. Large-scale and small-scale parameters are modeled
employing random variables with a certain deviation and mean; the numbers are
taken from ITU-R M.2135 document.
More advanced Spatial Channel Model (SCM, 3GPP TR25.996), which is
based on multiple ray clusters is currently under implementation. As a conclu-
sion, we emphasize that the methodology behind our SLS tool simplifies physical-
layer implementation to enable better support for MAC-layer features and pro-
cedures across a large-scale system deployment. Furthermore, our abstractions
result in a profound decrease in simulation complexity, which, in combination
with efficient code structures written in Python and C++, delivers attractively
short simulation times: one second of the real-time in a typical 19-cell (3 sector)
deployments with 30 users per cell could be simulated with only around 100
seconds of simulation time.
As a first step in this paper, we calibrate the simulation results with the
above analysis. Along these lines, we choose to disregard realistic interference,
172 I. Gudkova et al.
Fig. 7. Mean data fragment transmission time: analysis and simulation
pathloss, and other complex channel effects. However, we account for the actual
LTE frame structure to verify that simulation results fall well near our analytical
expectations. We further assume that the resource allocated to the H2H and
M2M devices is employing all available frequencies, so that the scheduler is
working in a time-division manner. Additionally, to account for some channel
degradation factors, we enable a simple physical-layer pathloss model described
in [14]. For the purposes of initial calibration and testing, we focus on a tagged
sector of our one-cell scenario. Users are deployed in a 288m-area around eNodeB
(typical for Urban macro model, ISD = 500).
User arrivals and departures are modeled according to the above analysis in
Section 3. At this stage, the interference between the users is considered insignif-
icant, due to the absence of other cells (which may be also the consequence of ap-
propriate network planning). More advanced interference and channel modeling
will be given in our future publications. In Figure 7, we overlay our simulation re-
sults on top of the previously obtained analysis (see Figure 2). Hence, we observe
that simulated mean data fragment transmission times are reasonably close to the
analytical prediction, but they also remain slightly higher due to the increasing in-
fluence of the realistic LTE performance factors not captured by the current anal-
ysis. Our ongoing work is to extend the reported analytical framework towards
the inclusion of practical performance degradation factors explicitly [16], as well
as to build a number of more insightful simulation scenarios mindful of upper-
layer protocols [17]. However, already now we can conclude that the constructed
simulation-analytical framework is a very useful tool to characterize M2M/H2H
coexistence and understand the resulting LTE system behavior.
Analyzing Impacts of Coexistence between M2M and H2H Communication 173
Acknowledgment. The reported study was partially supported by RFBR, re-
search projects No. 13-07-00953 a and No. 14-07-00090, GETA, and the Internet
of Things program of Digile, funded by Tekes.
References
1. David, K., Vinodrai, V., Yao, J.: WWRF Introduction and Vision (2010)
2. Wu, G., Talwar, S., Johnsson, K., Himayat, N., Johnson, K.: M2M: From mobile
to embedded Internet. IEEE Communications Magazine 49(4), 36–43 (2011)
3. 3GPP LTE Release 10 & beyond (LTE-Advanced)
4. Gotsis, A., Lioumpas, A., Alexiou, A.: M2M scheduling over LTE: Challenges and
new perspectives. IEEE Vehicular Technology Magazine 7(3), 34–39 (2012)
5. Study on RAN Improvements for Machine-Type Communications. 3GPP Technical
Report (TR) 37.868 (2011)
6. Cheng, M.-Y., Lin, G.-Y., Wei, H.-Y., Hsu, A.: Overload control for machine-type-
communications in LTE-Advanced system. IEEE Communications Magazine 50(6),
38–45 (2012)
7. Andreev, S., Larmo, A., Gerasimenko, M., Petrov, V., Galinina, O., Tirronen, T.,
Torsner, J., Koucheryavy, Y.: Efficient small data access for machine-type commu-
nications in LTE. In: Proc. of the IEEE International Conference on Communica-
tions, pp. 3569–3574 (2013)
8. Dementev, O., Galinina, O., Gerasimenko, M., Tirronen, T., Torsner, J., Andreev,
S., Koucheryavy, Y.: Analyzing the overload of 3GPP LTE system by diverse classes
of connected-mode MTC devices. In: Proc. of the IEEE World Forum on Internet
of Things (2014)
9. Hasan, M., Hossain, E., Niyato, D.: Random access for machine-to-machine com-
munication in LTE-Advanced networks: Issues and approaches. IEEE Communi-
cations Magazine 51(6), 86–93 (2013)
10. Zheng, K., Hu, F., Wang, W., Xiang, W., Dohler, M.: Radio resource allocation in
LTE-Advanced cellular networks with M2M communications. IEEE Communica-
tions Magazine 50(7), 184–192 (2012)
11. Borodakiy, V.Y., Buturlin, I.A., Gudkova, I.A., Samouylov, K.E.: Modelling and
analysing a dynamic resource allocation scheme for M2M traffic in LTE networks.
In: Balandin, S., Andreev, S., Koucheryavy, Y. (eds.) NEW2AN 2013 and ruS-
MART 2013. LNCS, vol. 8121, pp. 420–426. Springer, Heidelberg (2013)
12. Andreev, S., Galinina, O., Koucheryavy, Y.: Energy-efficient client relay scheme
for machine-to-machine communication. In: Proc. of the IEEE Global Telecommu-
nications Conference (2011)
13. Gerasimenko, M., Petrov, V., Galinina, O., Andreev, S., Koucheryavy, Y.: Im-
pact of MTC on energy and delay performance of random-access channel in LTE-
Advanced. Transactions on Emerging Telecommunications Technologies 24(4),
366–377 (2013)
14. Andreev, S., Pyattaev, A., Johnsson, K., Galinina, O., Koucheryavy, Y.: Cellular
traffic offloading onto network-assisted device-to-device connections. IEEE Com-
munications Magazine 52(4), 20–31 (2014)
15. Andreev, S., Koucheryavy, Y., Himayat, N., Gonchukov, P., Turlikov, A.: Active-
mode power optimization in OFDMA-based wireless networks. In: Proc. of the
IEEE Global Telecommunications Conference Workshops (2010)
174 I. Gudkova et al.
16. Moltchanov, D., Koucheryavy, Y., Harju, J.: Loss performance model for wireless
channels with autocorrelated arrivals and losses. Computer Communications 29,
2646–2660 (2006)
17. Dunaytsev, R., Koucheryavy, Y., Harju, J.: TCP NewReno throughput in the pres-
ence of correlated losses: The slow-but-steady variant. In: Proc. of the IEEE Inter-
national Conference on Computer Communications, INFOCOM (2006)