Sunteți pe pagina 1din 5

MI-VANET: A New Mobile Infrastructure Based

VANET Architecture for Urban Environment


Jie Luo Xinxing Gu Tong Zhao Wei Yan
School of Electronics Engineering and Computer Science, PKU, Beijing, China
{luojie,gxx,zt,yanwei}@net.pku.edu.cn

Abstract—The vehicular ad hoc network (VANET) has at- also presented. MIRT estimates the bus density of each road
tracted a lot of interest recently. However, traditional VANET segment based on the bus line information for road segment
is just an instantiation of MANET in which nodes are treated selection.
equally for data delivery. We first analyze the unique features
of urban VANET that vehicles have different types, and move At last we choose a downtown area in Beijing as the simu-
as clusters due to the influence of traffic lights. Then a two tier lation area. Simulation results show significant performance
architecture called Mobile Infrastructure Based VANET (MI- improvement in packet delivery ratio and throughput. Our
VANET) is proposed. In this architecture, the buses constitute a contribution is summarized as:
mobile backbone for data delivery while the low tier is composed
• A two-tier VANET architecture MI-VANET which is
of ordinary cars and passengers. MI-VANET will not only bring
the benefit that ordinary cars do not have to forward packets more suitable for urban environment than traditional
for other nodes, but also improve the network connectivity. VANET is proposed. In MI-VANET, the buses constitute
The corresponding Mobile Infrastructure Registering (MIRG) a mobile backbone for data delivery.
and Mobile Infrastructure Routing (MIRT) algorithms are also • The corresponding register method and routing algorithm
presented. The bus line information is made full use in MIRT.
Simulation results show that there is a 40-55% improvement in for MI-VANET are also developed. The bus line infor-
delivery ratio while the throughput is even doubled compared to mation is used for transmission quality estimation in the
GPSR in traditional VANET. routing algorithm.
I. I NTRODUCTION The rest of this paper is organized as follows: The related
work is given in Section II. In section III, we analyze the
Vehicular ad hoc networks have recently received consider- unique features of urban VANET. In section IV, the new
able attention. The VANETs provides both roadside-to-vehicle architecture MI-VANET is proposed and analyzed. Then cor-
communication (RVC) and inter-vehicle communication (IVC) responding register method and routing algorithm for MI-
capability. IVC works like a multi-hop mobile ad hoc network VANET are described in Section V. Section VI presents
(MANET) with its own unique characteristics. simulation settings and results. Finally, conclusions and future
In traditional VANET architectures, nodes are all treated works are summarized in section VII.
equally and may participate in packet forwarding. However,
vehicles are different in reality. Some vehicles are not willing II. R ELATED W ORK
to forward packets for others, and their wireless devices might Wireless network architecture has attracted a lot of interest.
be shut down by the driver at any time. While the buses are In [1], CF Huang proposed a two-tier architecture for MANET.
much easier to manage by the bus company to provide contin- The goal of this architecture is to support Internet access,
uous service. Another unique feature of urban VANETs is that and the high tier nodes are no different from the others
traffic lights have great influence on the vehicle movement, so except they have extra network interface connecting to the
vehicles are moving like clusters. These features should be cellular/infrastructure network.
considered in the VANET architecture design. Our proposed architecture is somehow like a Wireless Mesh
Considering the unique features of urban VANETs, a two- Networks (WMNs) [2]. A WMNs consist of mesh routers and
tier architecture called Mobile Infrastructure Based VANET mesh clients, where mesh routers have minimal mobility and
(MI-VANET) is proposed in this paper. In MI-VANET ar- form the backbone of WMNs. Different from WMS, the high
chitecture, the buses constitute a mobile backbone for data tier nodes in MI-VANET can also move with a high speed.
delivery, and the low tier is composed by ordinary cars and To deal with the rapidly changing network topology, many
passengers. MI-VANET will not only bring the benefit that routing techniques for VANET are based on location informa-
ordinary cars do not have to forward packets for other nodes, tion. One famous strategy is GPSR [3]. GPSR selects the node
but also improve the network connectivity. The correspond- that is the closest to the destination among the neighboring
ing Mobile Infrastructure Registering (MIRG) and Mobile nodes one-hop away. When local maximum occurs, the algo-
Infrastructure Routing (MIRT) algorithms for MI-VANET are rithm recovers by routing around the perimeter of the region.
Due to local maximums are common in urban VANETs, GPSR
This work is co-supported by the National Key Basic Research Program of
China (No. 2009CB320504) and the National Key Technology R&D Program seems not the best choice. Some other routing protocols for
of China (No. 2008BAH37B09) VANETs have also been proposed. One example is GyTAR

978-1-4244-3574-6/10/$25.00 ©2010 IEEE


[4], in which the real time road traffic variation is taken into the communication range is 200m, their connection time is
account. GyTAR dynamically selects the junctions through about 250s. If they only have a 150m communication range,
which a packet should pass through to its destination. The the connection time is still about 200s.
carry-and-forward mechanism is used for network disconnec- As we mentioned above, in urban VANETs, there are
tion. However, these routing techniques require the forwarding at least two types of vehicles which are ordinary cars and
ability of each node. Vehicles differences are not considered buses. The number of buses is much less than ordinary cars
in these algorithms. under normal conditions. For example, according to Beijing
To evaluate the proposed architecture and routing protocol Traffic Management Bureau [6], about 80 percent of all motor
by simulation, various traffic mobility models have been vehicles in Beijing are ordinary cars, and buses only compose
studied. VanetMobiSim [5] is a well known and validated 20 percent. In addition, the buses are larger and more powerful,
traffic generator, which is developed by Eurecom. We use this therefore they can carry better wireless equipment with a larger
traffic generator in our simulation studies. transmission range than ordinary cars. We hope to design a
new network architecture in which the buses constitute the
III. U NIQUE F EATURES O F U RBAN V EHICULAR
M OVEMENTS A NALYSIS backbone network. This will not only bring the benefit that
cars do not have to forward packets for other nodes, but also
Vehicular communication is a new technology including can improve the network connectivity.
roadside-to-vehicle communication (RVC) and inter-vehicle We also studied the bus line character from google map. The
communication (IVC). The roadside-to-vehicle communica- main road segments in the real area showed in figure 2 have 7
tion is greatly influenced by the limited connection time. And bus lines on average. And the departure time interval of each
most of the previous research on inter-vehicle communication bus line is about 15 minutes. As there are two directions in
is just an instantiation of MANETs in which nodes are treated each road, the average time between two adjacent buses is 7∗215
equally. In the urban environment, vehicular ad hoc network minutes. Another simple experiment is also taken in both rash
has many unique features. Firstly, vehicle movements are hour and normal time. In this experiment we have randomly
constrained by roads. Secondly, in urban VANETs, traffic taken 10 buses. With the help of the GPS record, we found
lights have great influence on the vehicle movement. Another that the highest speed of buses in Beijing is about 50km/h,
important feature of urban VANETs is that vehicles have at and the average speed is only about 15km/h, because buses
least two different types which are ordinary cars and buses as take a large proportion of time on bus stops and traffic lights.
analyzed in RBM [7]. All these features are not considered in So the average distance between buses is not very long that
the traditional VANETs architecture, so we believe that a new buses can form a mobile backbone.
network architecture is needed for urban VANETs.

Fig. 2. A real map from Southern Beijing


Fig. 1. Distance variation between a car and a bus in the experiment

In urban VANET, traffic lights have great influence on the IV. A RCHITECTURE
vehicle movement. Red traffic light at junction stops vehicles Traditional VANET is an instantiation of MANETs where
from approaching, which is an important factor to network nodes are treated equally. On the contrary, in MI-VANET,
disconnection. When the traffic light turns green, stopped the mobile infrastructure including buses and other public
vehicles will continue to move and those moving in the same transports constitute the backbone of the networks to provide
direction will be close to each other like a cluster. The nodes services for other vehicles and passengers. MI-VANET is
in the same cluster often have a much longer connection time somehow like a mesh network, but the mobile infrastructures
than road side units, because they have a high probability to are moving much more rapidly than mesh routers.
move with the similar patterns. A simple experiment has been
taken in a normal street in Beijing to know the connection time
A. Traditional Vehicular Ad Hoc Networks
feature. In this experiment, two vehicles moving in the same
direction are used, one is an ordinary car and the other is a bus. In the traditional VANETs, when a node wants to send a
To know their connection time, the car is 300m behind the bus message, all the nodes may participate in the data forwarding
at the beginning. After about 350 seconds, the car is 250m in according to the routing algorithm. Figure 3 shows a possible
front of the bus. Figure 1 shows their distance variation. If route example of message delivery.
Fig. 5. (a) Uniformly distributed vehicles (b) Vehicle moves like clusters

But vehicles are not uniformly distributed. Moving vehicles


Fig. 3. An example of message delivery in traditional VANETs
are affected by traffic lights, and when the light turns green,
those moving in the same direction will be close to each other
B. Mobile Infrastructure Based VANET like clusters as figure 5(b). In traditional VANET, vehicles
Our MI-VANET is a two-tier vehicular ad hoc architecture. in different cluster may not able to communicate, but things
In urban environment, the mobile infrastructure including are different in MI-VANET. Suppose the road segment length
buses and other public transports constitute the high tier of the is l, the period of red traffic light is t, and the average
MI-VANET. Each high tier node is equipped with two wireless velocity is v, then the expectation distance between two
interfaces. One interface has the same communication range clusters is min(t ∗ v, l). In this case, without having a five
with the low tier nodes to provide service to them. The other times transmission range, connectivity of MI-VANET can be
has a relatively larger transmission range on another channel, improved compared to traditional VANET, if the high tier com-
and the high tier nodes are communicated with each other in munication range is wider than the distance between adjacent
ad hoc mode by this interface. clusters. Additionally, buses are dispatched periodically, so the
Other vehicles and passengers constitute the low tier of distribution of buses in the network is relatively dispersive.
the MI-VANET. Each low tier node has only one wireless So our proposed MI-VANET architecture will not only bring
interface, which is used to register and communicate with the benefit that ordinary cars do not have to forward packets
its registered bus. In urban environment such as Beijing, we for other nodes, but also improves the network connectivity.
believe the number of buses is large enough that common cars
V. T HE R EGISTER A ND T HE ROUTING
can reach a bus with only one hop in the main roads. This can
be expanded to multi-hop if the bus density is low. In MI-VANET, only the buses have to periodically send bea-
con packets containing its location and velocity information.
The beacon packets are used to inform the low tier nodes that
services can be provided and tell the neighbor buses to renew
the neighbor table for packet forwarding. The basic problems
for MI-VANET are which bus the low tier node chooses to
register and how the packets are routed on the backbone.
A. Register
Each low tier node in our proposed MI-VANET has to
Fig. 4. (a) MI-VANET architecture example (b) An example of message
delivery in MI-VANET register on a bus. This method is called MIRG (Mobile
Infrastructure Registering). When the wireless device of a
Figure 4(a) shows one example of the proposed architecture. low tier node is opened, it will wait a beacon period time to
The high tier and low tier are revealed by solid and dash lines, know the nearby buses. Then the node computes the expected
respectively. When a common car or a passenger wants to connection time of each nearby bus, and choose the bus with
send a message, it first sends the message to its registered bus a longer expected connection time and shorter distance to
(which is called source bus for short). Once the source bus register. Suppose dist is the distance between the bus and the
receives the message from a low tier node, the high tier nodes car, vc is the velocity of the car, and vb is the velocity of the
will forward the message to the bus which the destination bus. If the two vehicles are moving in the same or opposite
direction, the expected connection time T is about R− dist
registered on (called destination bus). At last, the destination v c −v b .
bus will send the message to the destination node. Figure 4(b) Otherwise the two vehicles are not in a same straight road,
shows an example of message delivery in MI-VANET. and their communication quality will be bad because obstacles
are common in this situation. Therefore, we do not prefer to
C. Architecture Analysis choose these buses and T = 0 in this situation.
Assume that vehicles are uniformly distributed on the road Then a score S of each nearby bus will be calculated by:
as figure 5(a). If the average distance between two vehicles is
X, the average distance between two buses will be X/20%, S = a ∗ (R − dist) + b ∗ T, a and b are weighting f actors
because buses only compose 20 percent of all motor vehicles. The bus with the highest score will be chosen to register.
It seems that to improve the connectivity of the network, When the car is leaving the communication range of its
transmission range required between the high tier should be registered bus, another bus will be chosen with the same
over five times those of other vehicles. method.
B. Routing Otherwise, the transmission quality is not good on this road
Once a bus (called source bus) receives a packet from a segment. As the network connectivity on each road segment
low tier node, the high tier nodes will forward the message to is positively related to the bus density, we can estimate the
the bus which the destination registers on (called destination hop count on road segment j by the formula:
bus). The routing algorithm used on the high tier is MIRT
(Mobile Infrastructure Routing). We assume that each bus D ∗ Lj Lj Nj
Cj = ∗ ,( < D)
knows its location through the GPS service, and has a digital Nj R2 Lj
street map including bus line information. We also suppose
the availability of a location service through which a source Nj
Where Lj is the bus density of road segment j, D is a
bus can get the location information about destination bus. D∗L N
constant bus density, so Nj j is the ratio between D and Ljj .
MIRT is a location based reactive routing protocol including
two essential parts: 1) selecting an optimal route which con- This formula is not absolutely accurate, but it is simple and
sists of a sequence of road segments with the best estimated can be used as an estimation of the hop count. A large hop
transmission quality, and 2) efficiently forwarding packets hop- count number will appear if the transmission quality of a road
by-hop through each road segment in the selected route. segment is rather low. Routing algorithm prefers not to choose
Similar to position-based source routing such as GyTAR them for the optimal route.
[4], MIRT adopts the road segments based routing approach. Now that we know the expected hop count for each road
Therefore, data packets will be routed between buses, fol- segment, the total hop count for a certain route can be
lowing the street map topology. In MIRT, the road segments calculated. Afterwards, the Dijkstra algorithm can be used to
are chosen one by one, considering the transmission quality select a shortest route with the minimal expected hop count.
of each road segment. When selecting the next forwarding Two implementations can be used. One is that the best route
road segment, the bus (source bus or an intermediate bus consisting of a sequence of road segments is computed by the
near a junction) checks the route table and chooses the best source bus and is put into the packet header. The disadvantage
neighboring road segment with a min estimated hop count to of this method is the increase of packet size and network
the destination. bandwidth cost. The other is each bus calculates the route
To select an optimal route with the min estimated hop count, of each junction pairs and puts these information in a route
we have first to estimate the hop count of each road segment. table. And the next road segment is chosen by checking the
As the packets are only routed on the buses and each bus route table when packet is near a junction . This scheme
has a fixed path, we can estimate the hop count of each road requires less bandwidth. Our protocol MIRT uses the second
segment by the expectation bus density and road length. implementation.
For convenience, we first define the following notations:
• R1 : the low tier transmission range on one channel.
• R2 : the high tier transmission range on another channel.
• Xi : the total route length of bus i
• Lj : the length of road segment j
• Nj : the expected number of buses on road segment j
• Cj : the estimated hop count on road segment j
Suppose the route length of bus i is Xi , the length of road
segment j is Lj , and the route of path i contains road segment
L
j. So the probability that bus i is on road segment j is Xji
and the expected number of buses Nj on road segment j is: Fig. 6. Route selection in MI-VANET

 Lj
Nj = fi,j ∗
X i Figure 6 shows an example of how the next road segment
i
is selected. Once a bus receives a message from vehicle S, it

1, bus line i contains road j (1) checks the route table. Considering the distance and expected
fi,j = number of buses, the road segment with shadow will be chosen
0, bus line i doesn’t contain road j (2)
as the next road segment.
N
So the density of buses on the road segment j is Ljj . When the next road segment is determined, the improved
Suppose when the density of buses on the road segment is greedy forwarding strategy can be used to forward packets
high enough to reach a constant number D (which we choose on the road segment. Each bus maintains a neighbor table in
D = R22 in our simulation), the transmission quality on this which position, velocity and direction of each neighbor bus
road segment is rather high. And the hop count Cj on this are recorded. This table is updated through beacon messages
road segment can be estimated by: exchanged periodically. Thus, when a packet is received, the
Lj Nj forwarding bus computes the new predicted position of each
Cj = ,( > D)
R2 Lj neighbor, and then greedily selects the next hop neighbor.
VI. P ERFORMANCE E VALUATION
A. Simulation Model
In our experiments we use version 2.32 of the ns-2 simula-
tor. The simulated area is based on a real map from Southern
Beijing with a 1700m*1000m size, as shown in figure 2.
The vehicle movement trace is generated by VanetMobiSim,
which is a well known and validated traffic generator. Other
Fig. 8. Throughput of networks with 200 nodes
simulation parameters are listed in TABLE 1.
TABLE I To compare the throughput, we randomly select 50 com-
SIMULATION PARAMETERS
munication pairs. Each source node generates CBR traffic
for a period of 600 seconds with the sending rate from
Parameter Value 5kb/s to 30kb/s. As shown in figure 8, the throughput of
R1 150m
R2 400m
MI-VANET MIRT outperforms all the other protocols. The
bandwidth for both channel 2Mb throughput of MI-VANET MIRT is about twice that of VANET
beacon inerval 1.0s MIRT, the main reason is that the buses have another channel
Vehicle velocity 0-30m/s for transmission and has a larger communication range. In
Number of nodes 100-250
The bus percent 20% addition, the packet control overhead of MI-VANET MIRT is
TTL 32 small, because only the buses have to send beacon packets.
Packet size 512bytes
CBR data rate 128bytes/s VII. C ONCLUSION
Simulation time 600s In this paper, we have analyzed the unique features of urban
VANETs and proposed a two-tier VANET architecture called
Mobile Infrastructure Based VANET (MI-VANET). In MI-
We compare the performance of MIRT in MI-VANET
VANET architecture, the buses constitute a mobile backbone
architecture with GPSR in traditional VANET in both delivery
for data delivery, and the low tier is constituted by ordinary
ratio and throughput. Additionally, we also simulate MIRT
cars and passengers. MI-VANET will not only bring the
in traditional VANET called VANET MIRT and GPSR. In
benefit that ordinary cars do not have to forward packets
VANET MIRT, the route selection is the same as MI-VANET
for the others, but also improve the network connectivity.
MIRT. But different from MI-VANET, vehicles are equally in
Then the corresponding register method (MIRG) and routing
VANET MIRT, which means bus has only one interface and
algorithm (MIRT) for MI-VANET are also introduced. MIRT
all nodes are participating in packet forwarding.
is a geographical routing using the map topology and the bus
B. Simulation Results line information to facilitate route selection. Additionally, the
algorithmic complexity of MIRT is low.
In the simulation study, we demonstrate that our proposed
architecture with MIRT outperforms much better than GPSR
and VANET MIRT in terms of packet delivery ratio and
throughput.
As the future work, we will incorporate more realistic
factors into the protocol design for MI-VANET, such as the
real time or the history information of vehicle traffic.
R EFERENCES
[1] Chi-Fu Huang,Hung-Wei Lee and Yu-Chee Tseng. A Two-Tier Heteroge-
Fig. 7. The data delivery ratio in diffrent network density neous Mobile Ad Hoc Network Architecture and Its Load-Balance Routing
Problem. Mobile networks and applications, 2004
[2] Ian F.Akyildiz, Xudong Wang, and Weilin Wang. Wireless mesh networks:
As shown in figure 7, MI-VANET MIRT achieves the a survey. Computer Networks, 2005
highest packet delivery ratio. GPSR incurs a highest data loss [3] B. Karp and H.T.Kung. Gpsr: Greedy perimeter stateless routing for
wireless networks. MobiCom 2000: Proceedings of the 6th annual
rate, because simple greedy forwarding strategy without taking international conference on Mobile computing and networking, 2000.
urban environment characteristics into account is not suitable [4] M. Jerbi, S.-M. Senouci, R. Meraihi, and Y. Ghamri-Doudane. An im-
for VANET. After considering some urban environment char- proved vehicular ad hoc routing protocol for city environments. ICC’07.
IEEE International Conference on Communications, June 2007.
acteristics in VANET MIRT, 10-40% more packets can be [5] J. Härri, F. Filali, C. Bonnet, and M. Fiore. Vanetmobisim: generating
received. And our proposed architecture with MIRT achieves realistic mobility patterns for vanets. VANET ’06: Proceedings of the
10-25% gain in delivery ratio than VANET MIRT for the 3rd international workshop on Vehicular ad hoc networks, 2006.
[6] http://www.bjjtgl.gov.cn.
following reasons: as vehicles move like clusters and buses are [7] J. Wang and Wei Yan. Rbm: A role based mobility model for vanet. 2009
more uniformly distributed, increasing the bus communication International Conference on Communications and Mobile Computing, Jan
range can notably improve the network connectivity. 2009.

S-ar putea să vă placă și