Index Coding-based Data Exchange in a Vehicular

Network Junction
Karl Cedric U. Obias Elmer R. Magsino
Electronics and Communications Engineering Department Electronics and Communications Engineering Department
Gokongwei College of Engineering, De La Salle University Gokongwei College of Engineering, De La Salle University
Manila, Philippines Manila, Philippines
[email protected]

Abstract—Traffic lights play crucial roles in maintaining the intelligent traffic light can serve as the RSU installed at the
smooth transition of vehicles from one road segment to another. center of the intersection and can directly communicate with
They have evolved from time-based controllers to adaptive and vehicles on each of the road segments [3].
density-based traffic regulators. They can now even provide
more advanced capabilities such as data processing, storage, Without additional and expensive installation in a road
and computation. In this work, we investigate data dissemination network, the traffic lights become the leading choice to
in an intersection with the traffic light acting as the intelligent exchange information coming from vehicles and their built-
roadside unit (RSU) that can receive and broadcast demanded in sensors [4]. The authors in [5] proposed a traffic light-
environment data from/to its nearby vehicles. Information ex- centric architecture solution that will allow nearby vehicles
change is implemented by utilizing the index coding-based trans-
mission scheme able to optimize the number of transmissions to perform appropriate crossing maneuvers based on available
and transmitted data size while satisfying vehicular demands traffic data. The architecture can be extended to vehicle-to-
depending on its availability. Assuming various mobility scenarios everything (V2X) communication. Their work also provided a
and incorporating the presence or absence of stored information, proof of concept implementation displayed in a 2016 summit.
two transmission schemes are compared. We evaluate the single Another work presented an intelligent traffic light management
junction data exchange between vehicles and infrastructure
in terms of data map throughput, message security, number system focusing on the type and density of vehicles passing
of broadcast transmissions, and amount of consumed wireless by [6]. Given the current lane density and the presence
bandwidth. Our simulation results highlight the advantage of the of emergency vehicles, appropriate lane control signals are
index coding-based transmission scheme over the conventional provided for each road segment while giving priority to an
broadcasting technique by inspecting the performance metrics ambulance, firefighting and police vehicles. Lastly, in [7],
used in the study.
Index Terms—Intelligent traffic light controller, index coding, artificial intelligence is combined with the Internet of Things
data exchange, random broadcasting concept known as (AIoT) to utilize real-time surveillance
videos from network cameras to monitor dynamic vehicular
I. I NTRODUCTION movements, pedestrians, and surroundings. Real world map
To facilitate a smart vehicular environment satisfying com- and traffic data have been used to evaluate AIoT in terms
fortable and convenient travels, autonomous and intelligent of reducing queue lengths for pedestrians and minimizing
vehicles need to know in advance the information pertaining vehicular waiting time at the intersection.
to their trips, which include the road segments they will Index coding with prior information, a special case of
traverse and the up-to-date road happenings, e.g., accidents, network coding, [8] is a source coding technique that exploits
detours, and traffic conditions. This can only be done if they the sender’s carried information by being able to reduce the
communicate with other vehicles [1] or via the roadside unit number of broadcast transmissions. The authors in [9] studied
(RSU) [2]. the scenario of a three-receiver case with a common transmitter
One of the important locations in a road network is the between them. They also assumed that these receivers have
intersection where various road segments connect to and many the same demands and are cooperating with the transmitter.
vehicles converge at. For cars found on parallel road segments Because of its properties, index coding can be used for various
but traveling in opposite directions, the line-of-sight vehicle- applications such as satellite communication and video-on-
to-vehicle (V2V) communication may not be possible because demand [10].
of the separation between them. Note that vehicles within a In this work, we investigate how data dissemination can
car’s transmission range are those found on the same road be carried out at an intersection between infrastructures and
segment, thus, may not need any of the road segment data. vehicles. We observe the effects of the vehicle’s target road
Because of the presence of buildings, line-of-sight commu- segment location and the information it currently carries that
nication is unavailable between vehicles on perpendicular or can be uploaded to the RSU. The Manhattan Mobility Model
intersecting road segments. In this regard, an infrastructure or [11] and some case scenarios are employed to characterize
RSU is needed to allow vehicle-to-vehicle (V2V) communi- the vehicular movements from one road segment to another.
cation via vehicle-to-infrastructure (V2I) communication. An We then adopt the Data Availability Probability (DAP) [12] to
assume the presence or absence of prior (or side) information
of any vehicle. Road data exchange between vehicles via RSU
takes place by utilizing the optimal index coding transmission
(1J-IdxCd) scheme [13]. 1J-IdxCd exploits the linear XOR
operation between two input road map data. Not only it
compresses and encodes two or more map data, it also offers
various advantages such as throughput and security.
The outline of the paper is as follows. Section II talks
about the experimental setup of a single junction in a road
network and the vehicles and infrastructure located on it. We
also talk about the index coding scheme utilizing the demand
graph of vehicular demands and road map data availability. We
then present various simulation results arising from various
mobility models and data availability in Section III. Lastly,
the research study is concluded in Section IV.
II. E XPERIMENTAL S ETUP
In this section, we discuss the single junction scenario in
a vehicular network and the index coding-based information
exchange between vehicles via RSU, implemented as a traffic
light.
Fig. 1. Scenario at a single junction intersection with vehicles on all road
segments. The traffic light controller at the center acts as the RSU receiving
A. Single Junction Scenario data uploads and vehicular requests and broadcasting data demands.
Fig. 1 shows a junction with four road segments leading to
it, labeled as Road Segment 1 to Road Segment 4. Directional
arrows show where these vehicles want to go, therefore, need- 4. mx denotes the road map data at road segment x. As the
ing the current information about it for decision-making. This vehicles in Road Segment 1 and Road Segment 3 receive the
request is defined as the data demand. Given the road segment encoded message m1 ⊕m3 , they will decode it accordingly. Ve-
destination of each vehicle in Fig. 1, the corresponding data hicle at Road Segment 1 will perform m1 ⊕m3 ⊕m1 = m3 and
demand graph is illustrated in Fig. 2. R1 represents Road vehicle at Road Segment 3 will perform m1 ⊕m3 ⊕m3 = m1 .
Segment 1 and so on. We also emphasize that these vehicles However, cars from Road Segment 2 and 4 will not be able to
on each of the road segment carries a side information of their decode m1 ⊕ m3 successfully. They have to wait for the next
corresponding road segment only. transmission for their desired road segment map data.
Vehicles within the RSU transmission range (represented
by the yellow circles) can upload their road segment data
and then request the RSU for other available data from other
vehicles. The uploaded data are the data availability of the road
segment. To satisfy all vehicular requests, the RSU must send
all necessary road map data. To facilitate data dissemination,
the random Rand and optimal index coding 1J-IdxCd [13]
broadcasting schemes will be utilized. In Rand, all requests
are transmitted individually based on demand. In the given
example, any of the road map data can be sent first. Thus,
Rand will always require four transmissions to satisfy all
vehicular demands since there are four distinct road segment
data. In [14], it has been shown that as the number of n road Fig. 2. Road segment data demand and availability based from Fig.1.
segments increases, the number of transmissions per time slot
by Rand also increases. In fact, the number of transmissions The demand graph in Fig. 2 is mathematically represented
is always equal to the number of road segments, n attached by D in (1), where δij = γij αij . When γij = 1, vehicle on
to the junction. road segment i wants to proceed to road segment j, otherwise,
However, in 1J-IdxCd, road maps are encoded first accord- γij = 0. On the other hand, when αij = 1, the vehicle does
ing to demands before being broadcasted. By following the not have the road map data of segment j, else, αij = 0. Thus,
algorithm of 1J-IdxCd, two transmissions are only needed to δij = 1 when a vehicle on road segment i wants to proceed
satisfy all vehicular demands. These encoded and broadcasted to road segment j and has no data on it. Also, δij = 0, i = j
messages are: m1 ⊕m3 , to satisfy vehicles on road segments 1 to signify that U-turning vehicles do not need the data of the
and 3 and m2 ⊕m4 , to satisfy vehicles on road segments 2 and road segment they are on.
 TABLE I
  S IZES OF COMPRESSED 3D POINT CLOUD DATA FOR THE STATIC MAP
0 δ12 ··· δ1j δ1n DATA SHOWN IN F IG . 9 ( A ).
 δ21 0 ··· δij δ2n 
Map Data Data Size (MB)
 
D =  ... .. .. .. .. (1)
 
 . . . . 
 m1 5.838
δ(n−1)1 δ(n−1)2 ··· 0 δ(n−1)n  m2 5.254
m3 2.763
δn1 δn2 ··· δn(n−1) 0
m4 3.184
m1 ⊕m2 2.126
For example, the demand matrix D of Fig. 2, assuming m1 ⊕m3 2.812
αij = 1, is given below in (2). By employing graph theory m1 ⊕m4 2.630
given the adjacency matrix D, the encoded messages needed m2 ⊕m3 2.631
m2 ⊕m4 2.363
to satisfy all vehicular demands can be deduced. m3 ⊕m4 2.126
 
0 0 1 0
0 0 0 1
D=
1
 (2) A. 1J-IdxCd vs. Rand Broadcast Evaluation
0 0 0
0 1 0 0
Fig. 3 illustrates the instantaneous number of transmissions
B. Advantages of Index coding-based Transmission and Per- in a span of 24 hours when the number of vehicles and their
formance Metrics desired road segment destinations are randomly generated. In
this scenario, the average numbers of transmissions are 3.09
When implementing index coding in these scenarios, the
and 2.36 for the Rand and 1J-IdxCd schemes, respectively,
following advantages are demonstrated. First is data map
resulting in a corresponding power saving. At best, the Rand
throughput. By using index coding, the RSU broadcasts more
transmission can only equal the number of transmissions
map information in an encoded packet. The higher the en-
completed by the optimal index coding technique.
coding level is, the more information is sent to requesting
vehicles. Second, security in message transmission is accom-
plished. When the RSU broadcasts an encoded message, a
vehicle cannot decode the message unless the vehicle has side
information, therefore, implying security in the broadcast.
From these two advantages, the performance metrics that
will be used to evaluate the Rand and 1JIdx-Cd broadcast-
ing schemes are: number of transmissions and transmitted
data size, data map throughput, T , and message security, S.
Throughput T is measured by the number of road map data
that has been sent per transmission. If the message is encoded,
T = 2, else, T = 1. Security S, on the other hand, is
measured by how many coded messages have been received
and successfully decoded by the surrounding vehicles.
In this work, we concentrate only on level-2 encoding. Since
we employ at most binary coded packets in our scheme, a low-
complexity solution to the dissemination of road map data in
the vehicular network can happen in polynomial time. We call
such encoding as level 2, since two maps are combined. A
more complex encoding, i.e., level n, involves n road maps to
be encoded.
Fig. 3. Instantaneous number of transmissions using the Rand and 1J-IdxCd
broadcasting schemes.
III. R ESULTS AND D ISCUSSION
We present various simulation scenarios and results in this On the other hand, Fig. 4 shows the instantaneous transmit-
section and provide discussions on our findings. Table I, from ted data size during transmission. For the Rand transmission,
[13], contains the file sizes of the static environment data a total of 1.254 GB (4.48 MB on average) road map data are
assumed for each of the road segments shown in Fig. 1. These broadcasted to the surrounding vehicles, while 0.977 GB (3.49
file sizes are already compressed by at least 60% when using MB on average) of environment data are sent when the optimal
Octree compression [12]. Rows 6 – 11 also show the file index coding transmission is utilized. We note that having a
size when the road maps have been encoded using the XOR larger transmitted data size does not automatically mean that
operation, i.e., index coding technique. more road map data have been sent, as depicted in Table I.
 Fig. 5. Results for varying instances of data availability probability among
vehicles approaching the junction while following Manhattan Model.
Fig. 4. Instantaneous transmitted data size using the Rand and 1J-IdxCd
broadcasting schemes.

B. Effects of Mobility Patterns

In another experiment, we assume that the randomly gener-
ated vehicles follow the Manhattan Model [11] when selecting
which road segment to proceed, i.e., γij = 0.25 of turning left
or right and γij = 0.5 moving forward. We introduce in this
setup the Data Availability Probability (DAP) to take on the
values of αij . It pertains to the absence or presence of side
information carried by a vehicle. When DAP approaches zero,
the requesting vehicle does not have any prior information of
any road segments. As DAP increases, the vehicle has already
acquired other road segment data elsewhere, e.g., on-board
sensors or gossiping [15].
Fig. 5 illustrates the performance of Rand and 1J-Idxcd
schemes when 0 ≤ DAP ≤ 1 and if the vehicles are following
the Manhattan Model. We run each simulation scenario for Fig. 6. Comparing the throughput and security performance metrics.
100,000 times to arrive at the results. 1J-IdxCd still performs
better than Rand. Quite noticeable is that when using the 1J-
in vehicles approaching the junction. These are enumerated
IdxCd scheme, the transmitted data size is consistent even at
below.
low values of DAP. This means that 1J-IdxCd encodes road
map data appropriately. 1) All vehicles want to proceed to a specific road segment.
Fig. 6 displays the amount of road map data file throughput For example, vehicles on road segments 2 to 4 want to
and the number of unsecured data map transfers. Note that move to road segment 1, therefore, δ21 = δ31 = δ41 = 1.
the advantages of employing 1J-IdxCd over Rand are easily 2) All vehicles have the mapping data of a specific road
noticeable. These plots mean that the prevalence of encoded segment, e.g. road segment 1, therefore, δi1 = 0.
messages is always experienced in the junction. For each 3) All vehicles on a specific road segment, e.g. road
transmission slot, even though 1J-IdxCd transmits less than segment 4, have the mapping data of the other road
Rand, 1J-IdxCd sends out more road maps at a lesser segments, thus, δ4j = 0.
transmitted file size. This encoding technique, is a compression Fig. 7 exemplifies the effects when vehicles do not carry or
procedure, too, thus, further reducing bandwidth consumption carry side information. Cases 1 and 2 represent the absence
and presence of available side information of a certain road
C. Effects of Road Segment Data Availability segment from approaching vehicles, respectively, setting the
Three cases of index coding are performed to determine performance limits of 1J-IdxCd. On the other hand, Case 3
the effects of the absence or presence of prior knowledge of is closer to Case 1 when DAP ≤ 0.5 and approximates the
other road segments’ mapping data, called side information, Case 2 behavior as DAP approaches 1. These results verify
the non-optimal trends earlier presented in [12] and emphasize [6] S. Jahnavi, G. Prasanth, D. Priyanka, A. Sneheth, and M. Navya, “Intel-
the importance of having prior road segment knowledge that ligent traffic light management system,” in Proceedings of International
Conference on Advances in Computer Engineering and Communication
can be accomplished by efficient information dissemination Systems. Springer, 2021, pp. 489–498.
through communications between vehicles and RSUs and [7] Y. Liu, L. Liu, and W.-P. Chen, “Intelligent traffic light control using
having on-board sensors for sensing the environment. distributed multi-agent q learning,” in 2017 IEEE 20th international
conference on intelligent transportation systems (ITSC). IEEE, 2017,
We note that the Rand performance will be much worse pp. 1–8.
than any of these three cases, as manifested by Fig. 3 and [8] Z. Bar-Yossef, Y. Birk, T. Jayram, and T. Kol, “Index coding with side
Fig. 4. information,” IEEE Transactions on Information Theory, vol. 57, no. 3,
pp. 1479–1494, 2011.
[9] S. Unal and A. B. Wagner, “General index coding with side informa-
tion: Three decoder case,” in 2013 IEEE International Symposium on
Information Theory. IEEE, 2013, pp. 1137–1141.
[10] E. Byrne and M. Calderini, “Index coding, network coding and broadcast
with side-information,” in Network Coding and Subspace Designs.
Springer, 2018, pp. 247–293.
[11] F. Bai, N. Sadagopan, and A. Helmy, “Important: A framework to
systematically analyze the impact of mobility on performance of routing
protocols for adhoc networks,” in IEEE INFOCOM 2003. Twenty-second
Annual Joint Conference of the IEEE Computer and Communications
Societies (IEEE Cat. No. 03CH37428), vol. 2. IEEE, 2003, pp. 825–
835.
[12] K. F. Chu, E. R. Magsino, I. W.-H. Ho, and C.-K. Chau, “Index coding of
point cloud-based road map data for autonomous driving,” in 2017 IEEE
85th Vehicular Technology Conference (VTC Spring). IEEE, 2017, pp.
1–7.
[13] I. W.-H. Ho, S. C.-K. Chau, E. R. Magsino, and K. Jia, “Efficient 3d road
map data exchange for intelligent vehicles in vehicular fog networks,”
IEEE Transactions on Vehicular Technology, vol. 69, no. 3, pp. 3151–
3165, 2019.
Fig. 7. Results for varying instances of data availability among vehicles
[14] E. R. Magsino, G. P. Arada, and C. M. L. Ramos, “Investigating
approaching the junction following the three scenarios.
data dissemination in urban cities by employing empirical mobility
traces,” in 2020 IEEE 12th International Conference on Humanoid,
Nanotechnology, Information Technology, Communication and Control,
IV. C ONCLUSION Environment, and Management (HNICEM). IEEE, 2020, pp. 1–5.
[15] P. Mashwama, S. G. Fashoto, E. Mbunge, and S. Gwebu, “Development
The data exchange between vehicles and infrastructure in of a mobile inter-vehicular communication system based on gossip
a single junction with four road segments employing index algorithm.” International Journal of Interactive Mobile Technologies,
coding has been presented and evaluated in this work. Various vol. 14, no. 11, 2020.
vehicular mobility and data availability scenarios have been
evaluated to determine the effects of data demand and avail-
ability. Compared with the conventional random broadcasting
scheme, the index coding-based transmission technique has
been able to reduce the number of transmissions needed in
order to satisfy vehicular demands while conserving wireless
bandwidth with more data file throughput. It is also guaranteed
that when using the index coding transmission method, the
transmitted messages are secured and can only be decoded by
vehicles having the correct side information.
R EFERENCES
[1] R. Wang, Z. Xu, X. Zhao, and J. Hu, “V2v-based method for the
detection of road traffic congestion,” IET Intelligent Transport Systems,
vol. 13, no. 5, pp. 880–885, 2019.
[2] E. R. Magsino and I. W.-H. Ho, “Roadside unit allocation for fog-based
information sharing in vehicular networks,” in Proceedings of the 1st
ACM International Workshop on Smart Cities and Fog Computing, 2018,
pp. 7–12.
[3] H. Wei, G. Zheng, H. Yao, and Z. Li, “Intellilight: A reinforcement
learning approach for intelligent traffic light control,” in Proceedings
of the 24th ACM SIGKDD International Conference on Knowledge
Discovery & Data Mining, 2018, pp. 2496–2505.
[4] L. F. P. de Oliveira, L. T. Manera, and P. D. G. da Luz, “Development
of a smart traffic light control system with real-time monitoring,” IEEE
Internet of Things Journal, 2020.
[5] N. Varga, L. Bokor, A. Takács, J. Kovács, and L. Virág, “An architecture
proposal for v2x communication-centric traffic light controller systems,”
in 2017 15th International Conference on ITS Telecommunications
(ITST). IEEE, 2017, pp. 1–7.