66 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 16, NO.

1, FEBRUARY 2015

Information Dissemination Delay in

Vehicle-to-Vehicle Communication
Networks in a Traffic Stream
Lili Du and Hoang Dao

AbstractVehicle-to-vehicle (V2V) communication networks, cars speed, acceleration, or deceleration, in order to keep
as one of the core components of connected vehicle systems, have away from accidents such as road departure and collisions.
been granted many promising applications to address traffic mo- Thus, V2V communication networks can be used to develop
bility, safety, and sustainability. However, only a limited amount of
work has been completed to understand the fundamental proper- driver-assistance systems, which avoid traffic crash and im-
ties of information propagation in such systems, while comprehen- prove driving safety. Alsabaan et al. [4] propose employing
sively considering traffic and communication reality. Motivated V2V as well as traffic-light-signal-to-vehicle communication
by this view, this proposed research develops analytical formula- technologies to enable drivers to adaptively adjust their driving
tions to estimate information propagation time delay via a V2V speed, so that the objectives to promote fuel conservation
communication network formed on a one-way or two-way road
segment with multiple lanes. Distinguished to previous efforts, the and emission reduction within transportation systems can be
proposed study carefully involves several critical communication achieved. Yamaha [5] shows that the V2V communication
and traffic flow features in reality, such as wireless communi- network may detect road condition, such as road surface status
cation interference, intermittent information transmission, and in snowing weather, and then informs the traffic management
dynamic traffic flow. Moreover, this study elaborately analyzes center to adjust traffic control strategies and improve traffic
the interactions between information and traffic flow under sparse
and congested traffic flow conditions. The numerical experiments mobility. In the field of traffic routing, V2V informs drivers
based on Next-Generation Simulation field data illustrate that traffic condition information such as work zone, accident ahead,
the proposed analytical formulations are able to provide very closing lane, etc; hence, they can change route to avoid the
good estimation, with the relative error less than 5%, for the waiting time. The community-based online navigation system
information propagation time delay on a one-way or two-way road developed by Waze [6] has indicated a great potential of devel-
segment under various traffic conditions. The proposed work can
be further extended to characterize information propagation time oping online routing guidance based on V2V communication
delay and coverage over local transportation networks. technologies. In the meantime, many national and international
projects, such as California Partners for Advanced Transit and
Index TermsDynamic traffic flow, time delay, vehicle-to-
vehicle (V2V) communication. Highways (PATH) [7], CarTalk [8], and FleetNet [9], have
been dedicated to test the applicability of V2V technologies in
I. I NTRODUCTION various transportation scenarios.
Although many dramatic applications have been proposed, it

C ONNECTED vehicle systems, as a new generation of the

intelligent transportation system, consist of smart vehi-
cles and roadside infrastructure equipped with wireless com-
is noticed that the availability of real-time traffic information
propagating via V2V communication is one of the critical
bottlenecks to limit their implementation in practice. For ex-
munication facilities, which enable vehicle-to-vehicle (V2V) ample, forward collision warning applications based on V2V
and vehicle-to-infrastructure traffic information exchange. As need the information regarding how quickly a warning can be
one of the key components, V2V communication networks propagated to a vehicle; accident warning applications need
[i.e., vehicular ad hoc networks (VANETs) in literature] have the information to reach as many vehicles as possible in the
been granted many promising applications in traffic safety, local transportation network. Without knowing the information
mobility, and sustainability. For example, the authors in [1][3] propagation characteristics, such as connectivity, transmission
demonstrate that V2V will allow drivers to be aware of other distance, time delay, and coverage, it is difficult to successfully
implement those applications based on V2V technologies. This
requirement has spurred plenty of studies in both transporta-
Manuscript received December 22, 2013; accepted May 12, 2014. Date of tion and wireless communication communities. Research from
publication August 7, 2014; date of current version January 30, 2015. The work wireless communication field mainly focuses on developing
in this paper was supported by NSF Award 1436786. The Associate Editor for advanced communication protocols to ensure efficient informa-
this paper was G. Yan.
The authors are with the Department of Civil, Architectural, and Environ- tion transmission among smart vehicles once the physical con-
mental Engineering, Illinois Institute of Technology, Chicago, IL 60616 USA nections between smart vehicles are built, which is not the focus
(e-mail: ldu3@iit.edu; hdao@hawk.iit.edu). on this study. The proposed study shares common interests
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. with previous studies in transportation community, which seeks
Digital Object Identifier 10.1109/TITS.2014.2326331 to capture the information propagation benchmarks associated

1524-9050 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
DU AND DAO: INFORMATION DISSEMINATION DELAY IN V2V COMMUNICATION NETWORKS IN A TRAFFIC STREAM 67

with traffic stream features. Accordingly, the assumptions on as vehicle density and average vehicle speed; both one- and
successful wireless transmission conditions and dynamic traffic two-way vehicle traffic scenarios are considered. Wang et al.
flow along with their methodologies shaped the characteristics [21] used traffic flow theories, such as car-following models,
of most previous studies. Thus, this study reviews the existing to capture the vehicle mobility and applied a Monte Carlo
literature from these two aspects, which also differentiate the simulation model to evaluate the impacts of traffic flow and
proposed study from the previous studies and highlight our transmission range on the throughput of a VANET. Wang [12]
main contributions. modeled information propagation in VANET as a relay process,
Plenty of research explores instantaneous information propa- and provided the mean and variance of information propagation
gation in a traffic stream, considering that information propaga- distance as well as its distribution in VANET, but considering
tion is instantaneous compared to vehicle movement. Usually, that the presence of equipped vehicles follows an independent
a successful communication is simply identified by the con- homogeneous Poisson process, which is usually denied in
dition stating the geometric distance between two devices actual traffic flow condition. Utilizing the information propaga-
(equipped on smart vehicles or roadside sensors) less than a tion model proposed in [12] to evaluate information travel times
predefined transmission range (with 1 km being the maximum). on the individual arcs, Ng and Waller [22] provided the lower
Representative research in this category studied the topics and upper bounds of information propagation delay between
covering connectivity, intervehicle communication system, in- two nodes in a network, where traffic flow characteristics are
formation transmission distance, and probability of success evaluated by a static traffic assignment model. Wang et al. [23]
for information propagation. For example, Yang and Recker proposed an analytical model to estimate the expected in-
[10] built a simulation framework to test the information dis- formation propagation speed in the early stage of deploying
semination efficiency (coverage and speed) over various traf- V2V communication network, which implies very low smart
fic conditions as information propagating through intervehicle vehicle penetration. Du and Ukkusuri [24] modeled information
communication; Jin and Recker [11] computed the probability propagation along a one-way road segment as a time-expanded
of a successful instantaneous information transmission between network and provided a closed-form formulation to capture
two vehicles in uniform and general traffic streams; Wang [12] the network connectivity over a time period (reachability) for
provided the mean and variance of information propagation VANET. The aforementioned review noticed that many stud-
distance, considering equipped vehicle density and transmis- ies explored information propagation distance or speed, but a
sion range. Both Ukkusuri and Du [13] and Jin and Recker limited amount of work studies information delay (which is the
[14] developed analytical formulations to predict the multihop focus of this study); in addition, all the aforementioned study
connectivity of intervehicle communication network assuming applied a simplified successful transmission condition, which
stationary traffic stream, but different mathematical models are will be substituted by a more comprehensive condition relevant
used. Chen et al. [15] evaluated the performance of multihop to communication interference.
broadcast communication (information propagation distance, Overall, the aforementioned brief survey demonstrates that
and throughput of information package to be received for a previous research has significantly promoted the understanding
given distance) with vehicles following shockwave mobility of information propagation in a traffic stream from different
pattern which mixes free flow and congested flow traffic. angles by enriching realistic traffic flow feature in the modeling
Wang et al. [16] and Yin et al. [17] estimated the expectation, process; nevertheless, the consideration of the communication
variance, and probability distribution of instantaneous infor- side is relatively weak, which degrades the value of the research
mation propagation distance, assuming that vehicles headway in practical application. In addition, there is insufficient study
follows Gamma, Poisson, or Log-normal distribution. Clearly, focusing on the interactions between traffic flow movement
the deficiencies of this group of research are in two aspects: and information propagation under various traffic conditions
1) traffic flow dynamics is not fully considered. Since infor- on two-way roads. Motivated by the aforementioned points,
mation transmission time is omitted, and message propagation this research proposes mathematical description models to
is studied at a snapshot, traffic flow is treated as static flow; strengthen previous research from the following aspects: i) The
2) oversimplify the wireless communication constrains, ignor- proposed approach identifies successfully wireless communica-
ing background noise and interference. These two deficiencies tion by Signal-Interference-plus-Noise Ratio (SINR) condition,
will be addressed in the proposed study. rather than only factoring the transmission range and Euclidean
Some researchers recognized that information propagation distance between communication devices. SINR considers
in V2V communication is significantly impacted by traffic dy- multiple impact factors in wireless communication, such as
namics. For example, Schnhof et al. [18] considered dynamic transmission power and interference between concurrent trans-
communication link in a dynamic traffic flow on a two-way missions. This improvement will make our description model
freeway traffic stream, and then investigated how the smart more realistic from a wireless communication perspective.
vehicle density impacts information propagation speed and ii) This study takes account of information communication
efficiency. Agarwal et al. [19] studied delay tolerant message time to provide the applicability of the proposed formulations
propagation in V2V and developed upper and lower bounds for to the research of network level information dissemination.
information propagation rate as the functions of traffic density, It is realized that the communication time is ignorable for
vehicle speed, and transmission range. Wu et al. [20] indicated measuring the delay on a road segment, but its accumulation
that information propagation distance and speed depend on effect is significant for the information delay over the net-
relative vehicle movement and other traffic characteristics such work. iii) This research elaborately considers the interaction
68 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 16, NO. 1, FEBRUARY 2015

Fig. 1. Information propagation process.

of traffic flow and information propagation. Namely, various

traffic conditions (e.g., free flow, mild congested traffic flow,
congested traffic flow), different moving directions of traffic
flow and information flow (i.e., information and traffic flows
in the same or opposite direction on one-way or two-way
roads), and their combinations are fully covered. Therefore, the
proposed research comprehensively considers the realistic in
both transportation and communication sides. It will improve
our understanding of real-time traffic information delay in V2V
communication and promotes reliable applications in practices.
The whole of this paper is organized by the following struc-
ture: Section I introduces research background and motivations; Fig. 2. Traffic flow and information flow on a segment.
Section II presents the problem formulations, including traffic
flow model, information flow model, and successful wireless j. As shown, vehicles i and j are consecutive; xij represents
communication condition; Section III proposes our methodol- the corresponding space headway.
ogy to develop mathematical estimation for information prop-
agation time delay in V2V communication networks on a B. Information Flow Model
road segment. Various traffic conditions are considered. The
proposed formulations are validated by numerical experiment The information flow in this study is modeled by specifying
tests presented in Section IV; and the conclusion of this study the following aspects: i) Information always propagates from
is given in Section V. one vehicle to its first nearest neighbor (from vehicle 1 to 2, 2
to 3, until it arrives at the end of the road segment), where in-
formation transmission has the highest opportunity to success,
II. P ROBLEM F ORMULATIONS according to the successful transmission condition introduced
The proposed research is dedicated to exploring mathemati- in Part C of this section. ii) Information propagating from one
cal formulations to estimate information propagation time delay end of the road segment to the other along the direction that the
in a V2V communication network, running on a road segment. road extends is studied. The curvature of the road is ignored.
This study first conduct analyses on traffic flow, information iii) Physical dimensions of the smart vehicle are ignored. The
flow, and successful communication condition, which all to- smart vehicle is represented by a small rectangle without con-
gether serviced as the basis to develop rigorous analysis in the sidering its physical dimensions.
proposed study. Information may flow in either the same or the opposite
direction to the traffic flow. Thus, four possible cases only with
smart vehicles illustrated in Fig. 2 are in consideration. More
A. Road Segment and Traffic Flow exactly, Case (a) or Case (b) represents the situation where
Without loss of generality, this study works on a road seg- information flows in the same or opposite direction as the traffic
ment (either one-way or a two-way) only with an exit and an flow on a one-way (possibly multiple-lane) road. Case (c) and
entrance at each end of the road segment. Namely, no vehicles Case (d) are essentially the same, representing a situation where
exit or enter in the middle of the road segment. The road the information flows opposite to one of the traffic flows on
segment is with length L, and its traffic stream is composed of n a two-way road. The proposed study will cover all these four
vehicles, including both smart and nonsmart vehicles, moving cases.
on either same or opposite directions. Nonsmart vehicles do
not have wireless communication capability; they only impact
C. Successful Condition
overall traffic flow features and do not influence information
propagation directly; thus, they are not counted. Throughout This study assumes that the information transmission be-
this paper, unless noted otherwise, vehicle and smart vehicle tween smart vehicles applies a dedicated short-range com-
are equally used. Smart vehicles are numbered from left to munication (DSRC) radio tuned to the 5.9-GHz frequency,
right, according to their position in a traffic stream. As shown in which is allocated by the Federal Communications Commission
Fig. 1, xij represents the distance between smart vehicles i and for transportation safety and mobility applications on vehicle
DU AND DAO: INFORMATION DISSEMINATION DELAY IN V2V COMMUNICATION NETWORKS IN A TRAFFIC STREAM 69

and infrastructure. The successful commination between two

consecutive vehicles is identified by SINR condition, whose
standard formulation is shown in (1), in which vehicle w is the
transmitter and vehicle i is the receiver. It indicates that vehicle
w will successfully transmit information to vehicle i, if SINR
value at the receiver i is greater than a threshold value

Pw (xwi )
SINR = (1)
N +I
Fig. 3. Interference at vehicle i.
where Pw represents the transmission power of node w; xwi
represents the distance between transmitter vehicle w and re- condition. However, i varies with the location of the transmit-
ceiver vehicle i; is the signal power decay, typically 2 6; ter and receiver. Thus, (3) is the formulation of individuality,
N represents the background noise on the frequent channel which implies that microlevel vehicle distribution information
utilized by network; is the threshold which depends on the is needed. It is not a proper formulation to be used. The
designing modulation and code rate (values which indicate proposed study then explores a uniform formulation (a pseudo
the data transmission rate during a wireless connection) of transmission range derived from SINR condition), which can
wireless network.  = 0.15 is recommended for V2V commu- be used to identify successful information transmission by
nication [25]; I = nj=1,j=w ej pj (xij ) represents the sum known aggregated traffic information. We present our method
of interference power from other vehicles except vehicle w to as follows.
receiver vehicle i. ej = 1, if vehicle j is in transmission status;
otherwise, ej = 0. SINR is a physical model to determine D. Pseudo Transmission Range
the successful reception of a transmission over one hop in
wireless network. It considers many environment factors: the To develop the uniform successful transmission condition
distance between two nodes, path loss of signal, and wireless from SINR, this study labels smart vehicles from the left to right
interference. Thus, using SINR will make our formulations by number 0, 1, . . . , n, as shown in Fig. 3. Next, the spacing
capture more communication reality. xij is approximated by xij = (j i)h, where h is the expected
The standard SINR formulation can be further simplified spacing between two adjacent smart vehicles. Accordingly, i
by considering the communication features in V2V wireless can be approximated by (4), where i is the label of the receiver.
communication network. First, smart vehicles in V2V usually That is
 i
apply broadcast protocol; thus, we have ej = 1, j = 1, . . . , n.
1  1 
ni
1
Second, existing literature [26] shows that the background noise i 2 + (4)
h m2 m=1 m2
in V2V communication usually follows normal distribution m=1
with zero mean. Accordingly, this study applies N = 0 and = 2,
Note that i = 1 implies that the information is transmitted
corresponding to free-space information propagation [27]. Last,
from vehicle 0 to vehicle 1; the case i = 0 is not considered
assuming all smart vehicles adapt the same transmission power
since information starts from vehicle 0 and it will only be
(this assumptions have been widely used in literature such as
a transmitter rather than a receiver in this study; case i = n
[28]), we have Pw = p, w. With the previous four features
indicates that information arrives at the last vehicle so there is
holding in V2V communication network, the standard SINR
no interference coming from its right side.
formulation in (1) is transformed to (2), and further processed
This study
 further 2works on (4) and obtains the approxima-
to obtain the relationships in (3). That is
tion for m=1 (1/m ) in (5), according to [29]. That is
(xwi )2 
2
SINR = 
n (2) 1
= . (5)
(xji )2 m1
m2 6
j=1,j=w

( + 1) (xwi )2 Moreover, we know that the two lower bounds shown in (6)
and (7) exist, according to [30]. That is
 n



(xji )2 + (xwi )2 i
1  1 1  1 1 2 1
2
2
2
dm = 2
=
j=1,j=w
m=1
m m=1
m m m=1
m i 6 i
i
+1 1
(xwi ) 2
(3) (6)
i

ni 

 1 1 1
where i = nj=1 1/(xji )2 . Equation (3) represents the condi- 2
dm
m m2 m2
tion of a successful transmission between two smart vehicles. m=1 m=1 ni
It factors not only the distance between the transmitter and the
 1 1 2 1
receiver but also the distribution of all other vehicles around = = . (7)
them (represented by i ), reflecting the instantaneous traffic m=1
m 2 ni 6 ni
70 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 16, NO. 1, FEBRUARY 2015

Plugging (6) and (7) into (4), the lower bound for i is given
in (8). That is
  
1 i
1  1
ni
1 2 n
i = 2 2
h m=i
m2 m=i m2 h 3 i(n i)

(1 i n 1) (8a)
 
1  1
i
1 2 1
i = , i = n. (8b)
h2 m=1 m2 h2 6 i

According to Tables II and III in Appendix, it is observed that

the lower bound in (8a) is a tight bound to i . For example, (8a)
is with the maximum relative error equal to 14% as n = 10 and Fig. 4. Information propagation on one-way road.
13% as n = 20, which happens at boundary points; as 1 < i <
n 1, the relative error of (8a) is significantly reduced (less A. Time Delay of Intermittent Transmission
than 2%). Equation (8b) (i = n) is with a very small relative Due to the effect of traffic flow dynamics on wireless
error (0.31% as n = 10 and 0.08% as n = 20). In addition, it is communication connection, it has been recognized that inter-
observed that a larger n value leads to a smaller relative error mittent communication represents a general information trans-
and tighter lower bound. Substituting i in (3) by the lower mission fashion, in which wireless connection is intermittently
bounds given previously, the SINR condition is transformed to connected (leading to instantaneous transmission) and broken
(9). That is (leading to ferry transmission) due to relative movement be-
tween vehicles. Intermittent transmission usually happens in a
+1 1 traffic flow with mild congestion. Pure ferry and instantaneous
xwi communication are two extremes of intermittent transmission.
i
 They usually happen in very sparse traffic and highly con-
gested traffic flow, where the wireless connection between two
2 n , 1 i < n
+1 1
ri = h

3 i(ni) vehicles happens rarely or constantly. Thus, the methodology
(9)

proposed below will focus on intermittent communication.
rn = h 2 1 , i = n.
+1 1
6 i Fig. 4 provides an illustration about the information propaga-
tion in Case (a) during a time interval. An intermittent transmis-
Observing 2 /6 1/n 2 /3 n/i(n i), we know that sion is considered to follow a pattern, in which instantaneous
ri < rn in (9). In addition, it is recognized that ri reaches to transmission and ferry transmission alternatively happen until
the minimum value at i = n/2. By applying the tightest bound the information arrives at the end of the road segment. As an
for ri , SINR condition is led to the format in (10). That is instantaneous transmission occurs, several vehicles are well
connected and information is smoothly transmitted, such as
from node i to node i + k in Fig. 4. The corresponding time
+1 1 delay is calculated by t1 = k , where represents the
xji r = min {ri }ni=1 =h (10)
2
3 4
n transmission time and k represents the number of the hops.
Note that the communication time delay is taken into account to
where r is considered as a pseudo transmission range, which make the proposed model applicable to capture the time delay
limits the successful transmission. Note that, here, r is derived of information dissemination over a large-scale network, where
from SINR condition. It reflects the traffic flow influence by the accumulated effects of a huge number of instantaneous
factoring the space headway between vehicles as well as the communication time delay show impact, although it is ignor-
vehicle distribution around the receiver on the road. It is dif- able on a short road segment. As traffic is very sparse, ferry
ferent to the fixed transmission range specified by transmission transmission happens. Namely, the information will be ferried
power and frequency. by a vehicle until it meets another vehicle, such as the informa-
tion propagation at node (i + k) is broken, and then ferried by
node (i + k) until it reaches to node j at another time in Fig. 4.
The corresponding time delay is calculated by t2 = y/vi , where
III. M ETHODOLOGY
y represents the carrying-on distance and vi represents the
This section presents our methodology to capture the time average speed of the vehicle carrying information. Ignoring the
delay of information spreading on a road segment in Case (a). boundary case, in which instantaneous transmission or ferry
Namely, the information delay on a one-way road is first transmission happens one more times than the other, the ex-
studied, considering information flows in the same direction pected time delay of a piece of information traversing on a road
as traffic flow. Furthermore, we demonstrate that the proposed segment is estimated by the total time delay for one intermittent
methodology is applicable to Cases (b), (c), and (d). transmission (i.e., an instantaneous transmission followed by a
DU AND DAO: INFORMATION DISSEMINATION DELAY IN V2V COMMUNICATION NETWORKS IN A TRAFFIC STREAM 71

ferry transmission and vice versa) multiple by expected times

the intermittent transmission happens. Mathematically, this idea
is presented by (11), as
L
E(T ) = (E(t1 ) + E(t2 ))
E(x) + E(y)
 
E(y) L
= k + (11)
vi E(x) + E(y)

where x and y represent the information propagation distance

following instantaneous and ferry transmissions, respectively;
E(x) and E(y) represent their expected values; k represents the
expected number of hops in instantaneous transmission; L is the
length of the road segment. Equation (11) covers various traffic
flow conditions. E(x) dominates the time delay in congested
traffic condition, but E(y) mainly accounts for the time delay
in sparse traffic condition; E(x) and E(y) together capture
the feature in the intermediate congested traffic condition. The Fig. 5. One transmitter has one, two, or three receivers. (a) Scenario I:
follows of this paper further present our approaches to develop Connected to the nearest receiver. (b) Scenario II: Connected to the nearest
two receiver. (c) Scenario I: Connected to the nearest three receiver.
the formulations for E(x), E(y), and k, incorporating traffic
flow features and communication limits. in this study for three reasons. i) It is difficult to decide how
many vehicles one transmission will cover. It depends on the
vehicle distribution around a transmitter. As the traffic distribu-
B. Expected Information Propagation Distance in an
tion is not uniform, this number is uncertain and becomes very
Instantaneous Transmission
difficult to decide. ii) This study observed that the information
As the space headway between a transmitter and a re- propagates the same distance along the road segment under
ceiver satisfies S < r, information propagates by instantaneous these two scenarios, given we ignore the distance between
transmission. Considering information always propagates to its vehicles vertical to direction that the road extends. Fig. 5
nearest neighbor, the conditional random variable S|S r rep- provides examples to illustrate this observation, where a piece
resents the space headway in instantaneous transmission. Given of information propagates on a road segment with same vehicle
an instantaneous transmission includes k hops on average, we distribution, but the instantaneous transmission Fig. 5(a) fol-
obtain the formulation for E(x) as lows scenario I, i.e., a transmitter only reaches to the nearest
neighbor; the instantaneous transmission in Fig. 5(b) or (c)
E(x) = k E(S|S < r) (12) follows scenario II, i.e., a transmitter reaches to the nearest
two or three neighbors, respectively. Then, if vehicle 5 fails
where E(S|S < r) is the expected space headway in instan- to reach its nearest neighbor, vehicle 6, then we know that
taneous transmission; E(S|S < r) can be calculated by (13), vehicle 5 cannot reach other vehicle further in Fig. 5(b) and (c),
which is derived by the mathematical process given in (14) and according to our successful transmission condition given in
(15), given the spacing distribution f (s) is known. That is (10). In this context, vehicle 4 will fail to reach vehicle 6
r in both Fig. 5(b) and (c) since it is more difficult to build up
sf (s)ds connection between vehicles 4 and 6 than vehicles 5 and 6. Fol-
0
E(S|S < r) = r (13) lowing the same thought, vehicle 3 cannot connect to vehicle 6
f (u)du either in Fig. 5(c). As a result, the information propagates the
0 same distances from vehicle 1 to vehicle 5 under the three
F (S|S < r) = P (0 < S < s|0 < S < r) examples in Fig. 5, although one transmitter only connects the
P (0 < S < s, 0 < S < r) nearest neighbor in scenario I and two or three receivers in
=
P (0 < S < r) scenario II. This is a good quality for the proposed methodology
s in (11). iii) The instantaneous transmission time is very small
f (u)du
= 0r (14) (in microseconds); hence, the time delay difference between
0 f (u)du scenario I and scenario II is small and negligible. Hence, this
dF (S|S < r) f (s) study focuses on scenario I to study the information propaga-
f (S|S < r) = = r . (15)
ds tion time delay on road segment.
f (u)du
0
C. Expected Information Propagation Distance in a
This study noticed that, in reality, one transmitter may suc-
Ferry Transmission
cessfully transmit one piece of information to multiple receivers
(referred to as scenario II) rather than only to the nearest neigh- As the space headway between a transmitter and a receiver
bor (referred to as scenario I). However, we focus on scenario I (two consecutive vehicles) satisfies S > r, information will be
72 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 16, NO. 1, FEBRUARY 2015

2) Event Ac , 1 c k represents the cth hop in an in-

stantaneous communication. For instance: if the instan-
taneous transmission starts at the second vehicle, A1
represents the transmission between second and third
vehicle, A2 represents the transmission between third
and fourth vehicle, etc. With the pseudo transmission
Fig. 6. Instantaneous transmission on a segment. range r (derived from Section II-D), we calculate P (Ac ),
i.e., the probability of a successful information transmis-
sion from any vehicles to its nearest neighbor by (20).
spread by a ferry transmission. A ferry transmission will stop as That is
the spacing between this transmitter and a receiver satisfies S <
r. Therefore, the expected information propagation distance
r
by a ferry transmission, i.e., E(y), can be calculated by (16). P (Ac ) = P (0 S r) = f (s)ds = P. (20)
That is 0

E(S|S > r) r 3) Event Pk represents k hops of successive transmission,

E(y) = vi (16)
vij k [0, n] Using the same notation for the event and its
probability, we have Pk = P (A1 A2 Ak ).
where E(S|S > r) represents the expected spacing given a 4) Event Pk Bi represents an instantaneous transmission
ferry transmission happens; vij is the average relative speed starts at vehicle i and only successively propagates k
between two vehicles i and j. Considering S|S > r as a hops. Then, we have P (Pk Bi ) = P (Pk |Bi )P (Bi ). It
conditional random variable, E(S|S > r) can be calculated is noticed that for a given i, k [0, n i]. Namely, if an
by (17), which is derived by the cumulative distribution and instantaneous transmission starts from the ith vehicle, its
probability density formulations for S|S > r given by (18) and maximum number of successive hops is (n 1). Table I
(19), respectively. That is provides the calculations for all possible P (Pk |Bi ).
5) Event g(k) represents an instantaneous transmission with

+ 
+
only k hops. g(k) = ni=0 Pk Bi .
sf (s)ds sf (s)ds
E(S|S > r) = r
= r
(17) Table I below demonstrates the calculations of P (Pk |Bi )

+ r and g(k) as k = 0, . . . , n, and i = 0, . . . , n. For example,
f (u)du 1 f (u)du
r 0 P (Pk |Bi ) = P2 (1 P ) when k = 2 and i = 2. It indicates that
the instantaneous transmission starts from vehicle 2, propagates
F (S|S > r) = P (0 < S < s|r < S < +) two hops, and then the connection is broken. Each row in
s Table I provides the probabilities that an instantaneous trans-
P (r < S < s) r f (u)du mission spreads k hops, given this instantaneous transmission
= =  + (18)
P (r < S < +) f (u)du r starts at any vehicle i, i = 0, . . . , n. By summing P (Pk |Bi )
in each row, we obtain the general formulation: g(k) = ((n
f (S|S > r) =
dF (S|S > r)
= +
f (s)
. (19) k)Pk (1 P ) + Pk )/(n + 1) (note that the same notation is
db  used for event g(k) and its probability).
f (u)du
r
With the solution given in Table I, k can be calculated as

n
k= kg(k) = g(1) + 2g(2) + + ng(n)
1
D. Expected Hops in an Instantaneous Transmission
A piece of information may propagate multiple hops in an (n k)Pk (1 P ) + Pk
g(k) = . (21)
instantaneous transmission along the well-connected vehicle n+1
network on a road segment, until the communication link is bro-
Clearly, Pk is the key component to calculate k. The follow-
ken and the information propagation turns to ferry transmission.
ing study investigates the formulation for Pk as well as k under
To develop the formulation for k (the expected number of hops
different traffic conditions.
in an instantaneous transmission), we consider there are n + 1
1) Free Flow Traffic Condition: Under free flow condi-
number of vehicles running on the road segment, and label the
tion, the large spacing between vehicles guarantees vehicles
vehicles from left to right with the number from 0 to n as we did
movement freedom without concerning safety. This implies
before. Fig. 6 shows an example. Based on that, we consider k
the independence of the spacing and successful information
as a random variable and explore its expectation by the events
transmission between any two consecutive vehicles so that
defined as follows.
Pk = P (A1 A2 Ak ) = P k , and then, g(k) can be
1) Event Bi represents an instantaneous transmission start- calculated by
ing at the ith vehicle, i [0, n]. Considering an instanta-
neous communication may start at any individual vehicles (n k)P k (1 P ) + P k
g(k) = . (22)
evenly, we have P (Bi ) = 1/n + 1. n+1
DU AND DAO: INFORMATION DISSEMINATION DELAY IN V2V COMMUNICATION NETWORKS IN A TRAFFIC STREAM 73

TABLE I
P ROBABILITY T HAT AN I NSTANTANEOUS T RANSMISSION P ROPAGATES k H OPS G IVEN IT S TARTS AT THE iTH V EHICLE

k
Based on that, it can be proved that the sum of the last column By observing b=1 P (Ac Ab ) k, (25) provides a new
in Table I equals 1; thus, the correctness of the probability upper bound for Pk as
distribution in Table I is verified. In addition, we obtain the
1
k
closed-form formulation to predict the expected hops of an
Pk = P (A1 Ak ) 1 P (Ac )2 = 2P P 2 .
instantaneous transmission, i.e., k c=1
(26)

n
k = kg(k) = g(1) + 2g(2) + + ng(n) Accordingly, we obtain the lower and upper bounds for k in the
1 following:


n 
n
(n k) (kP (k 1)) (1 P ) + kP (k 1)
(n k)P k (1 P ) + P k k k
= k . (23) n+1
1
n+1 1
(27)

n
(n k)(2P P 2 )(1 P ) + 2P P 2
2) Congested Flow: Under congested traffic condition, the k k . (28)
spacing between two vehicles is relatively small. The move- 1
n+1
ment of a following vehicle needs to consider the movement
of the leading vehicle in front to keep safety; therefore, the At this point, by combining (11), (12), (16), (23), (27), and (28),
spacing between any two consecutive vehicles is dependent. we are ready to calculate the expected time delay of information
Accordingly, it brings the difficulty to accurately calculate propagation along a road segment in Case (a).
Pk = P (A1 A2 Ak ) and the associated k in (21). As
a compromise, this study develops the lower and upper bounds E. Extension to Other Cases
of Pk , which further lead to lower and upper bounds of k.
The following presents our methods. According to Bonferroni This study next demonstrates the applicability of the pro-
bound [31] and Caen bound [32], the lower and upper bounds posed approach to Cases (b), (c), and (d) in Fig. 2.
for Pk are given as 1) Case (b)One-Way Road Segment With Traffic and Infor-
mation Flowing in the Opposite Direction: Case (b) represents
a situation that information flow spreads in the opposite direc-
Pk = P (A1 A2 Ak ) tion to traffic flow. It is observed that information propagation
in Case (b) also follows the same pattern, alternatively present-

k
1 P (Ac ) = kP (k 1) (24) ing instantaneous transmission and ferry transmission. More
c=1 specifically, as information spreads by instantaneous transmis-
 sion, we may ignore the movement of smart vehicles since it

k is much slower than wireless information spread. Accordingly,
Pk = P (A1 A2 Ak ) = 1 P Ac the time delay resulting from instantaneous transmission in
c=1 Case (b) can be measured by t1 = k. However, as a ferry
transmission happens in Case (b), the vehicles conducting

k
P (Ac )2
1 k . (25) ferry transmission may carry information backward to the
c=1 b=1 P (Ac Ab ) information propagation direction for time E(y)/vi . Hence,
74 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 16, NO. 1, FEBRUARY 2015

transmission between the ferry vehicle and the other vehicle in

the same way (such as vehicle 4) and implies (SEB r)/vEB
(S  r)/vEW , or in the other way (such as vehicle 3) and
implies (SEB r)/vEB (S  r)/vEW . These two possible
ferry transmissions are denoted by f1 and f2 with probability
p1 and p2 ; they lead to the expected forward ferry transmission
distance E(yf1 ) and E(yf2 ), respectively.
Scenario (2): The example shown in Fig. 8 indicates that a
previous multihop instantaneous transmission stops at vehicle 2
Fig. 7. Ferry transmission on vehicle 1 in EB direction on two-way segment.
with the last hop from vehicle 1 in EB direction. Thus, the
ferry transmission starts on a ferry vehicle (such as vehicle 2)
carrying information backward (i.e., given SWB > r, and S  >
r, vehicle 2 carries information in WB direction). This ferry
may stop at a vehicle behind itself on the same way (such as
vehicle 3), but not on the other way (such as vehicle 4) due
to the opposite moving direction. However, as this backward
carrying happens, information is still possible to move forward
since it is noticed that vehicle 1 will carry information and move
forward. This study next performs more elaborate discussions
for this scenario, which may include two other situations.
Fig. 8. Ferry transmission on vehicle 2 in WB direction on a two-way
segment. (a) If vehicle 3 meets vehicle 1 (x13 r) before it meets
vehicle 2 (i.e., x23 r), then information is carried forward
by vehicle 1. The ferry transmission conducted by vehicle 1 is
the distance that a piece of information being spread forward
similar to the forward ferry f2 we discussed in Scenario (1).
in one cycle of the transmission pattern (an instantaneous
The only difference is that the forward ferry here will cancel
transmission followed by a ferry transmission or vice versa)
the propagation of information resulting from previous instan-
equals to E(x) E(y). Accordingly, the time delay of the
taneous information transmission from vehicle 1 to vehicle 2.
information propagation along a road segment under Case (b)
Considering it is difficult to measure the information propaga-
will be calculated by (29). That is
tion distance from vehicle 1 to vehicle 2 under this situation,
 
E(y) L this study ignores these details and considers it as the forward
E(T ) = + k . (29) ferry f2 . Note that the backward ferry conducted by vehicle 1
vi E(x) E(y)
under this situation does not have an effect on information
Note that (i) as E(x) E(y) 0, the information will never propagating forward. (b) If vehicle 3 meets vehicle 2 before
reach to the end of the road segment since it is always carried it meets vehicle 1, then information is carried backward
back by ferry transmission; (ii) (29) is a variant of (11); all by vehicle 2 before the instantaneous transmission happens
the elements such as E(x), E(y) and k can be measured by between vehicle 2 and vehicle 3. We denote this backward ferry
the formulations proposed in previous sections. Therefore, our as fb happening with probability pb . Accordingly, it results
approach also works for Case (b). in the expected backward ferry transmission distance E(yb ).
2) Cases (c) and (d)Two-Way Road: Case (c) and Case (d) Clearly, to identify this backward ferry, we need to recognize
are essentially the same. They both illustrate a situation that the distance between vehicle 1 and vehicle 3, and it is very
information spreads in the same direction to one of the traffic difficult to get. To address this issue, we observed that, if there
flows [such as in East Bound (EB)] but opposite to the other are many other vehicles between vehicle 1 and vehicle 2, then
way [such as in West Bound (WB)]. The time delay formulation vehicle 1 is very likely relatively far away, and the chance
developed for instantaneous transmission in Case (a) still works that vehicle 1 meets vehicle 3 earlier than vehicle 2 is low;
for these two cases. This study next provides more discussions then we are sure about the backward ferry when SSB > r;
about ferry transmission based on the examples in Figs. 7 and 8, on the other hand, if vehicle is almost adjacent to vehicle 2,
where S  represents the spacing distance between vehicles on then the average distance between vehicle 1 and vehicle 3
EB direction and vehicles on WB direction; SEB and SWB can be approximated by S  . In this context, we think that the
represent the spacing distance between vehicles on EB and WB backward ferry happens as (S  r)/vEW (SWB r)/vWB
direction, respectively. It is observed that there are two possible and SWB > r, and S  > r.
scenarios for a ferry transmission. Clearly, E(yf1 ), E(yf2 ), and E(yb ) can be calculated by the
Scenario (1): The example shown in Fig. 7 indicates that formulations given in (16) and (17). Note that the spacing and
a previous multihop instantaneous transmission stops at ve- relative speed used to calculate E(yf2 ) should be measured for
hicle 1, with the last hop from vehicle 2 in WB direction vehicles moving in opposite directions.
occurring. Thus, a ferry transmission starts on a ferry vehicle This study next develops the formulations for p1 , p2 , and pb
(vehicle 1) carrying information forward (i.e., given SEB > r, based on the examples in Figs. 7 and 8. It is observed that the
and S  > r, vehicle 1 carries information in EB direction). probability that a ferry transmission starts (the last hop of the
This ferry transmission will be stopped by an instantaneous previous instantaneous transmission stops) on the EB direction
DU AND DAO: INFORMATION DISSEMINATION DELAY IN V2V COMMUNICATION NETWORKS IN A TRAFFIC STREAM 75

or the WB direction depends on the spacing of vehicles on these

two directions. Namely, if the vehicles on the EB directions
are sparser than on the WB direction, then the chance that a
ferry transmission starts from EB direction is higher than from
the WB direction. With this observation, this study estimates the
probability that a ferry transmission happens on the WB or the
EB direction by SWB /(SWB + SEB ) and SEB /(SWB + SEB ).
Furthermore, we develop the formulations for p1 , p2 , and pb
based on the aforementioned discussions for f1 , f2 , and fb .
That is
Fig. 9. Testbeds. (a) US Highway 101. (b) Peachtree St. (c) Lankershim Blvd.
P (f1 )
p1 = P (f1 |f ) =
P (f ) instantaneous transmission and forward ferry) and backward
 transmission (through backward ferry). The expected ferry
SEB SEB r
= P S  r SEB r transmission distance is calculated by the weighted average
SWB + SEB vEB
 of forward ferry transmission and backward ferry transmis-
S r 1 sion. Mathematically, this difference is calculated by E(x) +
(30)
vEW P (f ) p1 E(yf1 ) + p2 E(yf2 ) pb E(yb ).
Clearly, (33) presents the same underline logic as (11) and
P (f2 )
p2 = P (f2 |f ) = (29). So far, we claim that the proposed approaches cover all the
P (f ) four cases in Fig. 1. To validate the proposed formulations, this

SEB SEB r study conducts the numerical experiments in the next section.
= P S  r SEB r
SWB + SEB vEB

S r 1 IV. N UMERICAL E XPERIMENTS
(31)
vEW P (f ) This section presents the numerical experiments to validate
P (fb ) the proposed mathematical formulations.
pb = P (fb |f ) =
P (f )

SWB SWB r A. Testbed and Input Data
= P S  r SWB r
SWB + SEB vWB The field data collected by Next-Generation Simulation


S r 1 (NGSIM) is used to validate the proposed methodology and
(32) formulations. The data set provides vehicle trajectory data, in-
vEW P (f )
cluding the attributes: vehicle ID, frame ID, total frames, global
where P (f1 ), P (f2 ), and P (fb ) represent the corresponding time, local X, local Y, global X, global Y, vehicle length, vehicle
probabilities of forward ferry transmission f1 , f2 and backward width, vehicle class, vehicle velocity, vehicle acceleration, lane
ferry transmission fb ; P (f ) represents the probability of ferry ID, preceding vehicle ID, following vehicle ID, space headway
transmission; vEB represents the relative speed on the lanes in (in same lane), and time headway (also in same lane). Three
EB direction. vEW represents the relative speed between the testbeds are selected, so that the experiments cover one-way or
lane on EB direction and the lane on WB direction. two-way road segment as well as free flow and congested traffic
Overall, a piece of information may spread on a two-way conditions. Below provides details for the experiments which
road segment through instantaneous and ferry transmissions. are setup on a one-way and two-way road segment.
By following the ideas to develop the time delay formulation 1) One-Way Traffic Flow: The validation experiments for
for Case (a) and Case (b) [i.e., (11) and (29)], we develop (33) one-way traffic flow was conducted on a road segment of US
to estimate the expected information propagation time delay for Highway 101 in Los Angeles, CA [see Fig. 9(a)]. The study
Case (c) and Case (d). That is area was a 2100-ft-long one-way segment with five lanes,
  on which vehicles move from North to South throughout the
E(yf 1 ) E(yf 2 ) E(yb )
E(T ) = k + p1 + p2 + pb section. Traffic data were collected during two 15-min periods
vf 1 vf 2 vb on June 15th, 2005. Given the speed limit of the road segment is
L 55 mi/h, the data set collected during (7:508:05 A . M .) rep-
(33) resents free flow traffic condition with a flow rate equal to
E(x) + p1 E(yf 1 ) + p2 E(yf 2 ) pb E(yb )
9500 vph and average speed equal to 48 mi/h; the data set
where vf and vb are the average vehicle speed in a traffic collected during (8:208:35 A . M .) represents an intermediate
flow with the same and opposite direction to information flow, congested traffic flow with average flow rate equal to 7800 vph
respectively. Note that (33) implies that the distance that a and average speed equal to 25 mi/h.
piece of information moving forward in one cycle including an 2) Two-Way Traffic Flow: The validation experiments for
instantaneous transmission and a ferry transmission is calcu- two-way traffic flow were conducted on two road segments:
lated by the difference between forward transmission (through Peachtree Street, Atlanta, GA [see Fig. 9(b)] for free flow and
76 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 16, NO. 1, FEBRUARY 2015

Lankershim Blvd, Los Angeles, CA [see Fig. 9(c)] for con- Step 3) Check if the information reaches to the end of the
gested flow. The segment of Peachtree Street is 650 ft long with road segment.
five lanes. Traffic data were collected during (12:501:00 P. M .) 1) Yes, record current time te ; T g = te t0 , kg =
on November 8th, 2006. Given the speed limit of 55 mi/h, average (k), calculated T M and k M ; go to Step 4
the data indicate free flow traffic condition with the average 2) No, change transmission scenario to ferry (in-
speed equal to 50 mi/h. The segment of Lankershim Blvd is stantaneous), go back to step 2
600 ft long with six lanes. Traffic data were collected during Step 4) If all data examined, stop; otherwise, t0 = t0 + t,
(8:509:00 A . M .) on June 16th, 2005. Given speed limit of go to Step 1.
55 mi/h, the collected data indicate a congested flow with The accuracy of T M is evaluated by RMSE and relative error
average speed equal to 23 mi/h. (e) given in (34) and (35) as follows. RMSE demonstrates
the average difference between T M and T g over all tracked
information. Relative error (e) measures the percentage of the
B. Experiment Design error between T M and T g to T g , thus giving us the idea how
This section designs the experiments to evaluate the perfor- significant the error is. That is

mance of the proposed mathematical estimation for the time N  g M 2

delay that a piece of information propagates through a road i=1 Ti Ti
RMSE = (34)
segment. Considering that the expected number of hops in an N
instantaneous transmission (k) is one of the key components of
1  TiM Tig
N
those mathematical estimation formulations, we also check the e= 100% (35)
accuracy of its formulation. N i=1 Tig
The overall ideas of the experiment are presented first. Based
on the field data collected from the selected testbeds, this study where N represents the number of scenarios, Tig represents the
first measures field information propagation time delay (T g ) as field time delay in a scenario i, TiM represents the correspond-
well as the expected number of hops in an instantaneous trans- ing mathematical estimation in free flow (the average of the
mission (k g ), and then, we calculate the corresponding mathe- upper and lower bounds in congested flow). A negative e value
matical estimations for the time delay (T M ) and the expected indicates an underestimation over all experimental scenarios,
number of hops (k M ), respectively. After that, we compare (k g ) and a positive e value means the other way around. The same
to (k M ) and (T g ) to (T M ), and demonstrate the accuracy by evaluation work will be conducted for k M .
root-mean-square error (RMSE) and relative error (e).
The field information propagation time delay T g is defined as C. Experimental Results and Insights
the time interval that a piece of information propagates through
a road segment, given a successful transmission between any 1) One-Way Segment: This section presents our numerical
every two vehicles is identified by SINR condition in (1). T g experimental results and the insights we obtained for one-way
is considered as the ground truth in this study. Mathematical road segment. The results given in Fig. 10 indicate that k g is
estimation T M is calculated by the proposed formulations, well bounded by our mathematical lower and upper bounds
given the needed distribution and parameters are obtained from [calculated by (27) and (28)] in congested flow, and accurately
the field data. These experiments select Log-normal distribution estimated by the mathematical model (23) in free flow. More
to represent the space headway distribution vehicles on one- exactly, RMSE values for k g under congested and free flow are
way or two-way road segment after it was calibrated by the field 1.64 and 1.37, respectively, and the relative errors are 3.71%
data. However, the applicability of the proposed approaches and 4.22%, respectively. In addition, we see that, on average,
does not depend on the distribution selection. Along the process k g = 8.68 and k M = 9.04 in the tested congested flow, and
to calculated T g and T M , k g and k M are also calculated k g = 6.36 and k M = 6.58 in the tested free flow; they are very
through field counts and proposed mathematical estimation close. Thus, our mathematical formulations provide reliable
formulations, respectively. k g represents the ground truth. estimations for k under both congested and free flow on one-
Next, we provide the experimental procedure. For every t way road segments.
(= 6 or 5) seconds, the experiment starts to track a piece of The results for evaluating the mathematical formulation (11)
information just launching on the start of the road segment to estimate the information propagation time delay on one-
until it reaches to the end of the road segment. Accordingly, way road segment are given in Fig. 11. Fig. 11(a) shows that
T g (and k g ) and T M (and k M ) are checked every t seconds. field time delay T g in congested flow is well bounded by our
According to the field data set, 151 (or 121) pieces of informa- mathematical bounds. The relative error 4.79% and RMSE
tion in total are tracked for one-way (or two-way) testbeds. The equal to 5 s; Fig. 11(b) also indicates that the mathematical
detailed experiment steps to track information propagation are formulation can estimate the field time delay very well in the
given as follows. free flow case; the relative error is 4.34% and RMSE equals
to 5.65 s. The negative sign of e indicates that, on the average,
Step 1) At time t0 , a piece of information launches on the the proposed mathematical formulation underestimates the time
start of the road segment. delay. In addition, the results show that T g = 42.07 s and
Step 2) Track instantaneous (or ferry) information propaga- T M = 39.84 s in the tested congested flow; T g = 46.40 s and
tion until it is broken; record k. T M = 44.03 s. Clearly, the average of the field time delay T g
DU AND DAO: INFORMATION DISSEMINATION DELAY IN V2V COMMUNICATION NETWORKS IN A TRAFFIC STREAM 77

Fig. 12. Comparison of number of hops between field and mathematical

estimations on two-way segment. (a) Congested flow. (b) Free flow.
Fig. 10. Comparison of number of hops between field and mathematical
estimation on one-way segment.

Fig. 11. Comparison of time delay between field and mathematical estima- Fig. 13. Comparison of time delay between field and mathematical estima-
tions on one-way segment. (a) Congested flow. (b) Free flow. tions on two-way segment. (a) Congested flow. (b) Free flow.

is very close to the average of the estimated time delay T M . 2) Two-Way Segment: The performances of the proposed
Moreover, in both cases, the relative error is around 4%; hence, approaches [i.e., (33) and all the related equations] on two-way
our mathematical formulations work well. road are also evaluated by the same way that we did for the
78 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 16, NO. 1, FEBRUARY 2015

TABLE II
R ELATIVE E RROR OF THE L OWER B OUNDS FOR i (W ITH n = 10)

one-way segment. The results in Figs. 12 and 13 show that the view, the proposed research aims to develop more reliable for-
field values (time delay or the expected number of hops) under mulations to estimate the time delay of a piece of information
congested flow are well bounded by the mathematical bounds, propagating through a traffic stream, considering more traffic
and they are accurately estimated by the mathematical formula- flow scenarios, such as on one-way or two-way road segments
tions for free flow. More exactly, the relative error for k g under under either free or congested traffic flow, and realistic wireless
free flow (or congested flow) is 4% (or 4.5%) with RMSE equal communication constraints, such as interference, information
to 1.26 (or 1.4), implying that the average difference between flow direction, instantaneous, and ferry transmission.
our estimation and the field value for k is about 1. In addition, Stochastic and probabilistic models combined with analytical
the results show that, on average, k g = 6.28 and k M = 6.49 approaches are adopted to develop the pseudo transmission
in the tested congested flow, and k g = 5.84 and k M = 6.03 in range from SINR condition, to estimate the expected num-
free flow. Clearly, they are very close. The relative error for T g ber of hops in instantaneous transmission, and to estimate
under free flow (or congested flow) is 4.36% (or 4.15%) with the expected transmission distance under instantaneous and
RMSE equal to 3.11 s (or 3.24 s). Moreover, the results show ferry transmission individually. Based upon these analytical
that, on average, T g = 25.4 s and T M = 26.02 s in the tested formulations and elaborate analyses, this study further proposes
congested flow, and T g = 17.8 s and T M = 18.48 s in the closed-form analytical formulations to estimate exact time de-
tested free flow. Again, they are very close. The experimental lay values for information propagating through free flow, and
results indicate small estimation errors. Thus, we claim that our provides analytical solutions to estimate the upper and lower
mathematical estimation formulations for two-way road also bound time delay for information propagating through con-
perform well. gested flow. Numerical experiments are conducted to validate
Overall, the numerical experiments indicate that the pro- our approaches, based on the field data collected by NGSIM
posed approaches perform pretty well under both free flow for one-way segment on US101 (congested and free flow),
and congested traffic flow on a one-way segment and two-way and two-way segments on Peachtree Street (free flow) and
segment. In more details, our mathematical estimations perform Lankershim Blvd (congested flow). The experimental results
a bit better for free flow traffic condition than for congested indicate that the proposed mathematical formulations provide
traffic flow condition on a one-way road segment and two-way reliable estimation to information propagation time delay, with
road segment, due to the fact that probabilistic bounds rather the relative error about 4% under various traffic conditions on a
than closed-form formulations are developed to estimate k in one-way or two-way road segment. The ground true time delays
congested flow. in congested flow on either one-way or two-way road segment
are well bounded inside the analytical upper and lower bounds.
There is some potential future research stemmed from this
V. C ONCLUSION AND F UTURE W ORK
study. First, the presented study can be extended to network
V2V communication holds promising future applications to level since it counts the time delay resulting from instantaneous
improve traffic safety, sustainability, and mobility. However, to transmission, which is negligible in a single road segment but
successfully implement these applications and cash these ben- significant as information spreads over a large scale of network.
efits, practitioners are still lacking reliable formulations to es- Information propagation at intersection is another interested
timate the information propagation time delay, based on traffic related research issue in this context. Second, the proposed
flow characteristics and communication limitations. Although methodology can be further extended to establish informa-
plenty of research worked on this issue in the literature, over- tion propagation dynamics overtopping traffic flow dynamics.
simplified communication or traffic flow assumptions weaken Information flow throughput over both temporal and spatial
their applicability in practice. Motivated by the aforementioned dimensions can be explored. Clearly, the proposed analytical
DU AND DAO: INFORMATION DISSEMINATION DELAY IN V2V COMMUNICATION NETWORKS IN A TRAFFIC STREAM 79

TABLE III
R ELATIVE E RROR OF THE L OWER B OUNDS FOR i (W ITH n = 20)

formulations, capturing the time delay of information propagat- [6] Google, Free GPS Navigation Turn by Turn-Waze. 2013. [Online]. Avail-
ing through a road segment, provide a firm base to investigate able: http://www.waze.com/
[7] California Partners for Advanced Transit and Highways (PATH),
these advanced topics. This research team will work on them in Berkeley. [Online]. Available: http://www.path.berkeley.edu
the near future. [8] D. Reichardt, M. Miglietta, L. Moretti, P. Morsink, and W. Schulz,
CarTALK 2000Safe and comfortable driving based upon inter-
vehicle-communication, in IEEE Intelligent Vehicle Symp., Versailles,
A PPENDIX France, Jun. 2002, pp. 545550.
[9] Johannes Steiner Fleetnet (Internet on the road), Germany. [Online].
See Tables II and III. Available: http://www.fleetnet.de/, Johannes Steiner
[10] X. Yang and W. Recker, Simulation studies of information propagation
in a self-organizing distributed traffic information system, Transp. Res.
ACKNOWLEDGMENT Part C, Emerging Technol., vol. 13, no. 5, pp. 370390, Dec. 2005.
[11] W. Jin and W. Recker, Instantaneous information propagation in a traf-
The authors would like to thank Dr. Xiang-Yang Li at Illinois fic stream through inter-vehicle communication, Transp. Res. Part B,
Institute of Technology, Dr. Stephen Boyles at The University Methodol., vol. 40, no. 3, pp. 230250, Mar. 2006.
[12] X. Wang, Modeling the process of information relay through inter-
of Texas at Austin, and Dr. Yafeng Yin at University of Florida vehicle communication, Transp. Res. Part B, Methodol., vol. 41, no. 6,
for their valuable comments on the early version of the work, pp. 684700, Jul. 2007.
and also the reviewers, whose constructive comments help us to [13] S. Ukkusuri and D. Lili, Geometric connectivity of vehicular ad hoc
networks: Analytical characterization, Transp. Res. Part C, Emerging
improve the articulations and presentation of this paper in the Technol., vol. 16, no. 5, pp. 615634, Oct. 2008.
final version. [14] W. Jin and W. Recker, An analytical model of multihop connectivity of
inter-vehicle communication systems, IEEE Trans. Wireless Commun.,
vol. 9, no. 1, pp. 106112, Jan. 2010.
R EFERENCES [15] R. Chen, W. Jin, and A. Regan, Multi-hop broadcasting in vehicular
[1] L. Strandn, E. G. Strm, and E. Uhlemann, Wireless communications ad hoc networks with shockwave traffic, in Proc. 7th IEEE CCNC,
vehicle-to-vehicle and vehicle-to-infrastructure, Vehicle Traffic Safety Jan. 2010, pp. 15.
Centre at Chalmers, Gothenburg, Sweden, Apr. 2008. [16] X. Wang, K. Yin, and Z. Yunlong, An exact Markov process for multihop
[2] J. Santa and A. F. Gomez-Skarmeta, Sharing context-aware road connectivity via intervehicle communication on parallel roads, IEEE
safety information, IEEE Pervasive Comput., vol. 8, no. 3, pp. 5865, Trans. Wireless Commun., vol. 11, no. 3, pp. 865868, Mar. 2012.
Sep. 2009. [17] K. Yin, X. Wang, and Z. Yunlong, Vehicle-to-vehicle connectivity on two
[3] Virginia Smart Road Project, Virginia Department of Transportation, parallel roadways with a general headway distribution, Transp. Res. Part
Jan. 2013. [Online]. Available: http://www.virginiadot.org/projects/ C, Emerging Technol., vol. 29, pp. 8496, Apr. 2013.
constsal-smartrd.asp Virginia Smart Road Project, [18] M. Schnhof, A. Kesting, M. Treiber, and D. Helbing, Coupled ve-
[4] M. Alsabaan, K. Naik, and T. Khalifa, Optimization of fuel cost and hicle and information flows: Message transport on a dynamic vehicle
emissions Using V2V communications, IEEE Trans. Intell. Transp. Syst., network, Phys. A, Statist. Mech. Appl., vol. 363, no. 1, pp. 7381,
vol. 14, no. 3, pp. 14491461, Sep. 2013. Apr. 2006.
[5] H. Yamaha et al., Development of next-generation road services on [19] A. Agarwal, S. David, and D. C. L. Thomas, Analytical model for mes-
SMARTWAY project, in Proc. 14th World Congr. ITS, Beijing, China, sage propagation in delay tolerant vehicular ad hoc networks, in Proc.
Oct. 2007. IEEE VTC-Spring, May 2008, pp. 30673071.
80 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 16, NO. 1, FEBRUARY 2015

[20] H. Wu et al., Spatial propagation of information in vehicular networks, Lili Du received the B.S. degree in mechanical
IEEE Trans. Veh. Technol., vol. 58, no. 1, pp. 420431, Jan. 2009. engineering from Xian Jiaotong University, Xian,
[21] Z. Wang, L. Wei, and W. Jin, A study on information throughput of inter- China, in 1998; the M.S. degree in industrial
vehicle communications in a unidirectional traffic stream, in Proc. CMC engineering from Tsinghua University, Beijing,
WRI Int. Conf., Jan. 2009, vol. 2, pp. 396401. China, in 2003; and the Ph.D. degree from
[22] M. Ng and T. S. Waller, A static network level model for the information Rensselaer Polytechnic Institute, Troy, NY, USA,
propagation in vehicular ad hoc networks, Transp. Res. Part C, Emerging in 2008.
Technol., vol. 18, no. 3, pp. 393407, Jun. 2010. She is currently an Assistant Professor with the
[23] W. Wang, S. S. Liao, X. Li, and J. S. Ren, The process of informa- Department of Civil, Architectural, and Environ-
tion propagation along a traffic stream through inter-vehicle communi- mental Engineering, Illinois Institute of Technology,
cation, IEEE Trans. Intell. Trans. Syst., vol. 15, no. 1, pp. 345354, Chicago, IL, USA. Her work has appeared in the As-
Feb. 2014. sociation for Computing Machinery Workshop on Vehicular Ad Hoc Networks
[24] D. Lili and S. Ukkusuri, The relative mobility of vehicles improves publication, Transportation Research Part B: Methodological, Transportation
the performance of information flow in vehicle ad hoc networks, Netw. Research Part C: Emerging Technologies, Networks and Spatial Economics,
Spatial Econ., vol. 10, no. 2, pp. 209240, Jun. 2010. Renewable Energy, and International Journal of Production Research, among
[25] M. Andrew and M. Dinitz, Maximizing capacity in arbitrary wireless others. Her research interests include vehicle-to-vehicle communication net-
networks in the SINR model: Complexity and game theory, in Proc. works, connected vehicles, intelligent transportation systems, real-time traffic
IEEE INFOCOM, Apr. 2009, pp. 13321340. sensing, and transportation network design and system analysis.
[26] B. X. Wang et al., The process of information propagation in a traffic Dr. Du is a member of the Transportation Research Board Transportation
stream with a general vehicle headway: A revisit, Trans. Res. Part C: Network Modeling Committee (ADB30).
Emerging Technol., vol. 18, no. 3, pp. 367375, Jun. 2010.
[27] A. Ozgur, R. Johari, D. N. Tse, and O. Lvque, Information-theoretic
operating regimes of large wireless networks, IEEE Trans. Inf. Theory,
vol. 56, no. 1, pp. 427437, Jan. 2010.
[28] P. Gupta and P. R. Kumar, The capacity of wireless networks, IEEE
Trans. Inf. Theory, vol. 46, no. 2, pp. 388404, Mar. 2000.
[29] D. Kalman, Six ways to sum a series, College Math. J., vol. 24, no. 5,
pp. 402421, Nov. 1993.
[30] Estimating the value of a Series. [Online]. Available: http://tutorial.math.
lamar.edu/Classes/CalcII/EstimatingSeries.aspx
[31] K. J. Worsley, An improved Bonferroni inequality and applications,
Biometrika, vol. 69, no. 2, pp. 297302, Aug. 1982. Hoang Dao received the B.S. degree in civil engi-
[32] D. De Caen, A lower bound on the probability of a union, Discrete neering from Hanoi University of Civil Engineering,
Math., vol. 169, no. 13, pp. 217220, May 1997. Hanoi, Vietnam, in 2009 and the M.S. degree in civil
[33] Q. Du and C. I. Chang, A signal-decomposed and interference- engineering from Illinois Institute of Technology,
annihilated approach to hyperspectral target detection, IEEE Trans. Chicago, IL, USA, in 2012, where he is currently
Geosci. Remote Sens., vol. 42, no. 4, pp. 892906, Apr. 2004. working toward the Ph.D. student in the Depart-
[34] G. Zhang, Y. Wang, H. Wei, and Y. Chen, Examining headway distri- ment of Civil, Architectural, and Environmental
bution models with urban freeway loop event data, Trans. Res. Rec., J. Engineering.
Trans. Res. Board, vol. 1999, no. 1, pp. 141149, Jan. 2007. His research interests include intelligent trans-
[35] S. Yin et al., Headway distribution modeling with regard to traffic portation systems and vehicle-to-vehicle communi-
status, in Proc. IEEE Intell. Veh. Symp., Jun. 2009, pp. 10571062. cation networks.