This document analyzes the performance of LTE networks using random linear network coding (RLNC) at the application layer. The study uses a simulation setup with a remote host, eNodeB, and user equipment to send coded and uncoded packets. The results show that applying RLNC at the application layer improves throughput but increases packet delay compared to not using network coding. MIMO antenna techniques are also found to provide higher throughput than SISO.

1. Performance Analysis of LTE Networks with
Random Linear Network Coding
Tewelde Degefa Assefa, Katina Kralevska and Yuming Jiang
Department of Telematics
Norwegian University of Science and Technology, Trondheim, Norway,
Email: tewetgrt@gmail.com, katinak@item.ntnu.no, jiang@item.ntnu.no
Abstract—Random Linear Network coding (RLNC) has
emerged as a promising solution for reliable multimedia delivery
over mobile cellular networks. In this paper, we deploy Applica-
tion Layer-RLNC (AL-RLNC) on the top of the existing Hybrid
Automatic Repeat Request (HARQ) in 4G Long Term Evolution
(LTE) networks. A simple implementation scenario composed
of a user equipment, an eNB and a remote host is considered.
Our results show that AL-RLNC improves the throughput and
the coverage at a cost of a higher packet delay. In addition,
we compare the performance when AL-RLNC is used to the
performance with the advanced LTE system antenna technique
Multiple-Input Multiple-Output (MIMO).
Keywords: RLNC, AL-RLNC, LTE, MIMO, HARQ
I. INTRODUCTION
The number of mobile users and advanced multimedia
delivery services offered over mobile cellular networks have
increased rapidly. It is expected that the total worldwide
mobile traffic will reach more than 127 Exabytes in 2020,
representing 33 times increase compared to 2010 [3]. Thus,
there is an inevitable need for high capacity networks.
Long Term Evolution (LTE) is a standard that provides high-
speed data for mobile phones and data terminals. The key
features of LTE that enable to meet the strict Quality of Service
(QoS) demands are: scalable carrier bandwidths (from 1.4
MHz to 20 MHz), support both Frequency Division Duplexing
(FDD) and Time Division Duplexing (TDD), Orthogonal Fre-
quency Division Multiplexing (OFDM) in combination with
higher order modulation techniques and Adaptive Modulation
and Coding (AMC).
Still novel solutions for reliable and fast delivery of mutime-
dia are needed. Random Linear Network Coding (RLNC) has
become a promising approach for improvement of the network
throughput, efficiency and scalability [4], [7]. In RLNC [6], a
coded packet yj is a linear combination from input packets xi,
multiplied by coefficients cj,i from a Galois Field of size q,
GF(q), where i = 1, . . . , k and j = 1, . . . , n. As the number
of xi packets or the size q of the Galois Field increases, the
probability that each newly generated packet yj is linearly
independent is higher. Given the unique flexibility of RLNC
to efficiently bridge the upper layer content packetization
and the lower layer packet transmission, RLNC is considered
as a powerful cross layer solution for a reliable mutimedia
delivery over the LTE/LTE Advanced networks. Recently,
there have been theoretical and practical evaluations on RLNC
integration in LTE networks such as the work presented in
[5], [9], [12], [14], [15]. The authors in [14], [15] proposed a
modified LTE protocol stack by using RLNC on the Media
Access Control (MAC) layer as a replacement for Hybrid
Automatic Retransmission reQuest (HARQ) in LTE/LTE-A
networks. The performance of RLNC by implementing an
inter-flow and intra-flow MAC-RLNC for a single-user and
multi-users scenarios is evaluated in [5]. Performance analysis
and energy efficiency for integration of RLNC in the MAC
layer is presented in [8]. Practical implementations of RLNC
at the application layer for multimedia streaming on Apple
iPhone platform have been presented in [11], [13].
A. Our Contribution
This paper analyzes the performance of LTE networks with
intra-flow RLNC over unicast flows. Only packets within
one flow are coded when intra-flow RLNC is used [2]. We
implement RLNC at the application layer as a complement of
the HARQ mechanism. To the best of our knowledge, this is
the first work that uses the Kodo network coding library [10] to
examine the performance of LTE networks with AL-RLNC. At
the time of conducting this work, Kodo library functionalities
are implemented in the transport layer. In comparison to MAC-
RLNC, AL-RLNC provides a simple integration of RLNC
into LTE networks without any modification to the LTE
protocol stack. In addition, it turns out that deployment of
network coding is less complex with AL-RLNC in terms of
compatibility with current devices and applications. Moreover,
MAC-RLNC requires packets to be coded at the MAC layer
which Kodo library does not support.
We analyze the throughput and delay performance which are
the key design parameters of any system subject to strict
QoS requirements. The results show that RLNC improves the
throughput at the cost of a higher packet delay. In addition,
we compare the performance when Single-Input Single-Output
(SISO) and Multiple-Input Multiple-Output (MIMO) antenna
techniques are used.
The paper is organized as follows. In Section II, we present
the background of AL-RLNC. Section III describes the simu-
lation setup and the system flow. In Section IV we discuss the
results obtained from the simulations. Conclusions and future
work are given in Section V.
MIPRO 2016/CTI 673

2. II. BACKGROUND
When LTE networks support AL-RLNC, the standard LTE
protocol layer stack remains unchanged as shown in figure
1. AL-RLNC is deployed on top of the existing MAC layer
HARQ based packet transmission process. In this solution,
RLNC encoded IP packets enters the eNB Packet Data Con-
version Protocol (PDCP) layer. PDCP layer performs header
compression and ciphering then the PDCP encapsulated IP
packets are delivered to the RLC layer. The RLC layer
performs segmentation/ concatenation of IP packets into RLC
packets to fit the MAC frame size requirements. Each MAC
frame is allocated a single physical layer transport block for
transmission over the eNB/UE interface. The physical layer
carries all the information from the MAC transport channels
over the air interface. In addition, it has link adaptation
functionality that provides a matching of the modulation and
the coding techniques to the radio link interface condition [8].
In this work packets are coded at the application layer by
using the RLNC funcionalites from the Kodo network coding
library.EEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 11, NO. 12, DECEMBER 2012
g used to
ical and
on delays
d MAC-
gurations
paper.
ning in
/network
(RNC)1
message
h l bits.
e.g., an
arbitrary
is of the
nts their
efficients
ode x as
encoded
mination
encoded
obal en-
ork, due
a header
iver pair
s (RNG)
changed
) which
K of the
symbols.
the UE
......
PHY TB
MAC PDU
... ......TB
PDU
RLC PDU
MAC PDU
PHY TB
PDU
1
1 8
TB10
TTI TTI
PDCP PDU
RLC Segmentation ARQ Process
H-ARQ Process
(8 MAC PDUs inParallel)
RLC Layer
IP/PDCP Layer
MAC Layer
PHY Layer
RLC PDU
... .........
RLC SDU
1 Radio Frame 10ms
Fig. 1. eNB E-UTRAN protocols: MAC-HARQ solution.
so that the resulting RLC PDUs fit the MAC PDU size
requirements, which in turn depend on the PHY transport
block (TB) sizes to be used on the upcoming transmission
time intervals (TTIs). The PHY TB size depends on the
adaptive modulation/coding (AMC) scheme selected by the
MAC layer scheduler based on the channel quality indicators
(CQI) continuously reported by the UE. In other words,
adaptation to dynamic PHY TB sizes propagates up to the
RLC layer where RLC PDUs are created to match these
requirements.
The appropriate size RLC PDUs are forwarded to the
MAC layer which transmits groups of 8 MAC PDUs using 8
parallel HARQ processes within blocks of 8 consecutive TTIs
(Fig. 1). If any of the MAC PDUs is not received correctly,
its new incremental redundancy (IR-HARQ)-based version
is transmitted within the same slot of the following TTI
octet. The TTI octet period (8 ms) between retransmissions is
sufficient for ACK/NACK feedback reception and generation
of a new IR-HARQ-based MAC PDU description, if needed.
The maximum number of retransmissions is three.
Fig. 1. eNB RAN protocol, MAC-HARQ solution; AL-RLNC is deployed
on top of the standard MAC-HARQ solution [8]
We have decided to implement AL-RLNC solution due to
two reasons. First, the Kodo library is applicable only to the
upper layers and it is not designed to use the encoding and
decoding functionalities of the library in the lower layers.
Second, from an implementation point of view, deploying the
RLNC in the application layer allows to simply integrate the
Kodo library coding scheme on top of the current stack and
without affecting the functionality of the LTE protocol stack
and/or compatibility of UEs and eNBs.
III. AL-RLNC
In order to deploy RLNC in LTE networks, we use ns-31
and Kodo2
library.
1A C++ based open source simulation library for networking research. We
used ns-3 version 3.22
2A C++ library for implementing random linear coding. We used Kodo
version 19.0.0
Fig. 2. LTE-EPC simulation topology. The remote host sends coded packets
to UE through the SGW/PDN-GW
A. Topology and Architecture
In ns-3, only FDD for LTE is supported and all the obtained
results in this work are based on FDD operation mode.
The QoS aware scheduler is a key component in LTE for
the achievement of a fast adjusted and efficiently utilized
radio resource. In this experiment we have used a QoS aware
scheduler called Priority Set Scheduler (PSS) from the ns-
3 module. During the TTI duration, i.e. 1 ms, the UEs
report their perceived radio quality as an input to the eNB
scheduler to decide which AMC should be used. The scheduler
prioritizes the QoS requirements amongst the UEs. Then it
informs the UEs of the allocated radio resources both in
downlink and uplink direction.
Figure 2 shows the topology of the LTE network used for
the performance analysis. The topology consists of a remote
host located in an external network, SGW/PDN-GW as one
entity, eNB and a UE. Two applications are implemented at
the remote host and the UE for encoding and decoding packets,
respectively. On each of these applications a UDP socket is
created to transmit and receive the encoded IP packets over
the LTE network. The coded packets are generated from a
large file with a specified generation size (number of source
packets), packet size and coding over a specific Galois field
using the RLNC functionality in the Kodo library. We use this
simple topology because the LTE-EPC simulation module in
ns-3 supports only a point-to-point connection between remote
host and UEs/eNBs located in different networks, i.e., it does
not support broadcasting.
The architecture in Figure 3 shows how the encoded packets
are transmitted from the remote host and decoded at the re-
ceiver, UE. As shown in Figure 3, a packet to the SGW/PDN-
GW node which is connected to the Internet by the SGi
interface is forwarded by Internet routing. SGW/PDN-GW
determines the eNB to which the UE is connected to by look-
ing at the UE IP destination address. It classifies the packet
by using traffic flow templates to identify to which Evolved
packet system (EPS) bearer it belongs. Each EPS bearer has
a one-to-one mapping to the S1-U interface. Following the
mapping it sends out the coded packet to the intended eNB
via the S1-U interface. Upon a reception of the coded packet,
the eNB forwards it to the UE over the LTE-Uu interface
based on the bearer ID. Finally the UE receives the encoded
packet. The application installed on the UE checks whether the
received packet is linearly independent. It drops the linearly
dependent packets and stores the linearly independent packets
in a matrix form for decoding. The remote host (encoder)
674 MIPRO 2016/CTI

3. Fig. 3. Architecture showing how coded packets are sent down to the UE
using RLNC
is transmitting encoded packets until it gets a notification
from the UE (decoder) that signals the reception of sufficient
linearly independent packets to recover the whole file. In this
process the HARQ performs its functions independently from
the application layer, i.e., it retransmites a coded packet after
unsuccessful reception.
Figure 4 gives an overview of the steps for creating the
simulation architecture.
Fig. 4. Overall system flow for setting up the simulation
B. Simulation Setup and System Flow
We use the simulation parameters shown in Table I in
order to simulate and collect data for the performance analysis
and evaluation of the defined network topology. As shown
in the table an EPS bearer is established to classify the
packets according to TFTs and the defined QoS class, as
they cross the EPC core towards the UE in the LTE access
network. The simulation time is 20 seconds. The run-times
TABLE I
LTE-EPC PERFORMANCE SIMULATION SETUP PARAMETERS
Parameter Value
Simulation time (s) 20
Application run-time (s) 20
Downlink Bandwidth (MHz) 5
Resource Block 25
Distance dependent pathloss
model
COST231 propagation loss
model
Antenna type 2x2 MIMO, SISO
eNB Tx power (dBm) 30, 46 and 60
Fading model Fast fading
Shadowing deviation 7
UE speed of interest (Kmph) 0
Operating frequency band 2GHz
Downlink EARFCN 100
Traffic model video
Video packet generation inter-
val (ms)
100
EPS bearer type NGBR
MAC Scheduler Type PSS
Encoding scheme Systematic RLNC
Galois field Binary8
Generation Size 100
Packet Size (byte) 1024
of the applications on the remote host and the UE are the
same as the simulation duration so that enough packets are
generated and encoded within the given duration at a packet
generation interval of 100ms. Furthermore, the used distance
dependent propagation model is COST231 propagation loss
model3
. In addition, a fast fading model generated with a
Rayleigh channel is included in order to consider the impact
of the fading on the radio signals. Some of the values for the
parameters are taken from [1] so that we simulate a topology
that is as close as possible to a real urban area LTE network
topology. The table also shows the coding parameters. A
systematic coding is performed over the Galois Field GF(256).
The generation size is 100 packets where each packet has size
of 1024 bytes.
IV. SIMULATION RESULTS AND ANALYSIS
We evaluate the performance when encoded and non-
encoded packets are sent. Accordingly a comparison between
the throughput and the delay performance for the both schemes
is made. The throughput is defined as the rate of successful
data delivery, while the delay is defined as the end-to-end delay
for all the packets sent from the remote host to the UE are
successfully decoded.
A. Throughput performance analysis
To analyse the throughput performance with RLNC, 100
packets each of size 1024 bytes, are encoded and sent to the
UE from the remote host. The simulation is performed for
three different eNB transmission powers, UE-eNB distance in
the range between 0.3km and 4km and bandwidth of 5MHz.
We run the simulation with these parameters so that we can
3COST231 propagation loss model is applicable to urban areas to evaluate
path loss of radio signals in frequency range 1500 MHz to 2000 MHz and
link distance of up to 20 km.
MIPRO 2016/CTI 675

4. 0.3 0.35 0.4 0.45 0.5
101
102
Distance(km)
Throughput(Kbps)
With-NC
Without-NC
Fig. 5. Throughput performance for 30dBm eNB transmission power: SISO,
5MHz bandwidth
0.8 0.9 1 1.1 1.2 1.3 1.4 1.5
100
101
102
Distance(km)
Throughput(Kbps)
With-NC
Without-NC
Fig. 6. Throughput performance for 46dBm eNB transmission power: SISO,
5MHz bandwidth
observe the effect of the distance on the number of successfully
received packets. If the UE is close enough to the eNB there
will be no packet loss since there is only one UE for the given
system bandwidth.
Figures 5, 6 and 7 show the difference in the throughput
performance obtained with and without RLNC, for 30, 46
and 60dBm eNB transmission power, respectively. We can
conclude from these figures the throughput is improved with
RLNC. When RLNC is used, the only requirement is enough
linearly independant packets to be received at the UE, i.e., an
exact packet should not be received in a specific order. Note
that RLNC is performed over HARQ which by itself is an
efficient way of recovering lost packets with retransmission
requests. The lowest value of the distance on the x-axis in
Figures 5, 6 and 7 indicates that with the given power level
all the sent packets are fully recovered up until that distance.
However, as we increase the distance, the number of lost
packets is increasing. For instance, for 30dBm power the
throughput is the same with and without RLNC up until
0.3km. When the distance is more than 0.3km, the throughput
performance is better with RLNC.
The reason why we depict the throughput performance when
different eNB transmission power levels are used is to check
the effect of network coding at different distances and to show
2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4
101
102
Distance(km)
Throughput(Kbps)
With-NC
Without-NC
Fig. 7. Throughput performance for 60dBm eNB transmission power : SISO,
5MHz bandwidth
how the performance can be improved not only with network
coding, but also by increasing the transmission power. As we
can see better throughput performance in LTE networks is
achieved with increasing the transmission power. For example
the throughput and the coverage that can be obtained with
30dBm and RLNC can easily be achieved by increasing the
transmission power to 46dB and not using RLNC. However
this is only for the downlink communication where eNB power
transmission is not a big concern, but for UEs increasing the
transmission power could have its own effect on the battery
life of the UE. Thus, we can say that without having to
deal with the complexity of deploying new RLNC scheme
on the existing HARQ scheme the required performance can
be achieved with increasing the transmission power.
Note that the throughput is in Kbps, because only 100
packets each of size 1024 bytes are sent during 20s simulation
time.
B. Delay performance analysis
Figures 8, 9 and 10 show the delay performance at 30,
46 and 60 dBm eNB transmission power levels. As shown
the delay is higher with RLNC. This is because in RLNC
in addition to the transmission delay, the delay includes the
time to encode a packet, receive enough linearly independent
packets and decode them. We also see how the delay reduces
for higher distance between the UE and the eNB. This is in
line with the decreasing throughput which indicates the total
number of received packets are less so the delay as well.
The way how we calculate the delay is imposed by ns3. The
flow monitor tracks the received packets and calculates the
delay for these packets. When the distance is big, the receiver
does not receive packets during the simulation time. That is
the reason why the delay goes towards zero. In real-world
scenarios, the delay actually goes towards infinity.
C. AL-RLNC vs. MIMO
In this part, it is shown that MIMO could be an alternative
to deploying RLNC in LTE networks. MIMO techiques are
one of the major enablers for LTE. They allows higher data
rate transmission through the use of multiple antennas at the
676 MIPRO 2016/CTI

5. 0.3 0.35 0.4 0.45 0.5
100
101
Distance(km)
Delay(s)
With-NC
Without-NC
Fig. 8. Delay performance for 30dBm eNB transmission power: SISO, 5MHz
bandwidth
0.8 0.9 1 1.1 1.2 1.3 1.4 1.5
100
101
Distance(km)
Delay(s)
With-NC
Without-NC
Fig. 9. Delay performance for 46dBm eNB transmission power: SISO, 5MHz
bandwidth
transmitter and the receiver. Figure 11 shows the throughput
obtained for 60dBm eNB transmission power for MIMO and
Single-input Single-output (SISO) systems with and without
network coding. As shown in the figure the throughput ob-
tained with MIMO system without using RLNC is higher at
any given distance as compared to SISO with network coding.
We can also see that the MIMO system increases the coverage
of the network up until 5km. Thus, we can expect a better
2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4
100
101
Distance(km)
Delay(s)
With-NC
Without-NC
Fig. 10. Delay performance for 60dBm eNB transmission power: SISO,
5MHz bandwidth
2.5 3 3.5 4 4.5 5
101
102
Distance (km)
Throughput(Kbps)
Without-NC-MIMO
With-NC-SISO
Without-NC-SISO
Fig. 11. MIMO, SISO and RLNC performance comparison for 60dBm eNB
transmission power
throughput performance than the graph shows if we combine
MIMO with network coding.
V. CONCLUSIONS
This paper presented an AL-RLNC integration in LTE
networks deployed on top of HARQ with a goal of achieving
an efficient and flexible multimedia delivery mechanism. The
presented results indicate that the throughput performance of
LTE network is improved by using RLNC but at the cost of a
higher packet delay. We have also shown that the performance
improvement with RLNC can also be obtained either with
increasing the eNB transmission power or by using MIMO
techique. The results presented in this paper are obtained for
a simple topology and the performance can further be analysed
and evaluated with multiple UEs and remote hosts.
REFERENCES
[1] Vodafone Alcatel-Lucent, picoChip Designs. Simulation assumptions
and parameters for fdd henb rf requirements. 3GPP TSG RAN WG4
(Radio) Meeting, May 2009.
[2] Szymon Chachulski, Michael Jennings, Sachin Katti, and Dina Katabi.
Trading structure for randomness in wireless opportunistic routing. In
Jun Murai and Kenjiro Cho, editors, SIGCOMM, pages 169–180. ACM,
2007.
[3] UMTS Forum. Mobile traffic forecasts 2010-2020 report. UMTS Forum
Report 44, January 2011.
[4] Christina Fragouli, Jean-Yves Le Boudec, and J¨org Widmer. Network
coding: an instant primer. Computer Communication Review, 36(1):63–
68, 2006.
[5] Pavel Loskot Hassan Hamdoun. Implementing network coding
in LTE and LTE-A. International Workshop on Smart Wireless
Communications-SWICOM2012, 2012.
[6] Tracey Ho, M. Medard, R. Koetter, D.R. Karger, M. Effros, Jun Shi,
and B. Leong. A random linear network coding approach to multicast.
Information Theory, IEEE Transactions on, 52(10):4413–4430, Oct
2006.
[7] Sachin Katti, Hariharan Rahul, Wenjun Hu, Dina Katabi, Muriel M´edard,
and Jon Crowcroft. XORs in the air: practical wireless network coding.
IEEE/ACM Trans. Netw, 16(3):497–510, 2008.
[8] Chadi Khirallah, Dejan Vukobratovic, and John S. Thompson. Per-
formance analysis and energy efficiency of random network coding
in LTE-advanced. IEEE Transactions on Wireless Communications,
11(12):4275–4285, 2012.
MIPRO 2016/CTI 677

6. [9] Ashutosh Kulkarni, Michael Heindlmaier, Danail Traskov, Marie-Jos´e
Montpetit, and Muriel M´edard. An implementation of network coding
with association policies in heterogeneous networks. In Vicente Casares
Giner, Pietro Manzoni, and Ana Pont, editors, Networking Workshops,
volume 6827 of Lecture Notes in Computer Science, pages 110–118.
Springer, 2011.
[10] Morten Videbæk Pedersen, Janus Heide, and Frank H. P. Fitzek.
Kodo: An open and research oriented network coding library. In Vi-
cente Casares Giner, Pietro Manzoni, and Ana Pont, editors, Networking
Workshops, volume 6827 of Lecture Notes in Computer Science, pages
145–152. Springer, 2011.
[11] Hassan Shojania and Baochun Li. Random network coding on the
iphone: fact or fiction? In Wei Tsang Ooi and Dongyan Xu, editors,
NOSSDAV, pages 37–42. ACM, 2009.
[12] Surat Teerapittayanon, Kerim Fouli, Muriel M´edard, Marie-Jos´e Mont-
petit, Xiaomeng Shi, Ivan Seskar, and Abhimanyu Gosain. Network
coding as a wiMAX link reliability mechanism. CoRR, abs/1208.4766,
2012.
[13] P. Vingelmann, F.H.P. Fitzek, M.V. Pedersen, J. Heide, and H. Charaf.
Synchronized multimedia streaming on the iphone platform with net-
work coding. Communications Magazine, IEEE, 49(6):126–132, June
2011.
[14] Dejan Vukobratovic, Chadi Khirallah, Vladimir Stankovic, and John S.
Thompson. Random network coding for multimedia delivery over LTE-
advanced. In ICME, pages 200–205. IEEE Computer Society, 2012.
[15] Dejan Vukobratovic, Chadi Khirallah, Vladimir Stankovic, and John S.
Thompson. Random network coding for multimedia delivery services in
LTE/LTE-advanced. IEEE Transactions on Multimedia, 16(1):277–282,
2014.
678 MIPRO 2016/CTI