You are on page 1of 13

IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING, VOL. 13, NO.

1, JANUARY/FEBRUARY 2016 5

V2V QoS Guaranteed Channel Access


in IEEE 802.11p VANETs
Che-Yu Chang, Hsu-Chun Yen, and Der-Jiunn Deng, Member, IEEE

Abstract—IEEE 802.11p Wireless Access in the Vehicular Environment (WAVE) has been serving as the de facto wireless protocol
for a Vehicular Ad hoc Network (VANET) with the explosive growth of vehicular applications. These applications require guaranteed
Quality-of-Service (QoS). However, the fundamental channel access mechanism of WAVE, Enhanced Distributed Channel Access
(EDCA), is not able to provide guaranteed QoS due to the unpredictable random access. To remedy this problem, we propose a novel
channel access scheme, called Earliest Deadline First based Carrier Sense Multiple Access (EDF-CSMA). EDF-CSMA based on
EDCA dynamically adjusts the priority of real-time streaming to avoid collision and introduces an admission control policy according to
time constraints to provide guaranteed QoS in multi-channel environments. An analytical model is carried out to study and compare
the channel utilization of EDF-CSMA and QoS-aware Hybrid Coordination Function (HCF) controlled channel access (HCCA) method.
The result shows that 60 percent of channel utilization is improved by EDF-CSMA. Additionally, real video-based simulations are
conducted to evaluate the performance of EDF-CSMA and the existing EDCA method. The results show that EDF-CSMA reaches
better QoS support than EDCA while maintaining efficient channel utilization.

Index Terms—VANET, WAVE, 802.11p, QoS, EDF-CSMA

Ç
1 INTRODUCTION

S INCE the technically approved standard of the IEEE


802.11p in 2010 [1], WAVE have become the primary
vehicular communication technology in VANETs. VANETs
started defining the WAVE [8], [9], [10], [11] communication
protocol for ITS applications to seek fundamental impro-
vement for VANETs. Fig. 1 shows the structure of WAVE,
use Dedicated Short-Range Communications (DSRC) technol- wherein the 802.11p working group is responsible for
ogy to provide a safe and reliable transmission on specified medium access control (MAC) and physical (PHY) layer
channels. In US, the Federal Communications Commission standardization. The PHY layer introduces the orthogonal
(FCC) allocated 75 MHz of spectrum in the 5.9 GHz band frequency-division multiplexing technique, with a maxi-
for Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure mum transmission rate of 27 MB/s that can support a trans-
(V2I) communications. And the allocated spectrum is further mission range of 300 to 1000 meters. The MAC layer adjusts
divided into seven sub-channels as shown in Table 1. the authentication mechanism to reduce the access time
Within all the sub-channels, the Control Channel (CCH), between the vehicle and the access point. To improve com-
CH178, is mostly for transmitting control and safety packet, munication efficiency, IEEE 802.11p defines an Outside the
while the other six channels are Service Channels (SCHs) for Context of a Basic service set (OCB) operation model, which
transmitting data packets, providing non-security-related allows data transmission between the nodes in different
services. With the maturation of VANETs technology, the Basic Service Set (BSS). The 1609 working group is responsi-
demand for Intelligent Transportation System (ITS) related ble for standardizing the technique stack and general DSRC
applications such as collision warning [2], traffic manage- applications above the link layer, which includes resource
ment [3], [4], [5], [6], overtaking assistance [7], and voice/ manager (1609.1), security services (1609.2), network layer
video service between vehicles has increased. In general, services (1609.3), multi-channel operation (1609.4), and
ITS applications require not only reliable data transmission electronic payment (1609.11).
but also short transmission latency, and usually these appli- EDCA and HCCA are two mechanisms in WAVE to sup-
cations become much richer in content and more diverse port QoS. EDCA is a contention-based channel access proto-
in traffic patterns, which make the development of ITS appli- col based on CSMA with Collision Avoidance (CSMA/CA).
cations more difficult than traditional wireless applications. In EDCA, high-priority traffic has a higher probability to
In view of this, the IEEE 802.11p and 1609 working groups access channel, but unpredictable delay could occur due
to the characteristics of random access. HCCA is an access
protocol based on polling architecture, wherein the Hybrid
 C.Y. Chang and H.C. Yen are with the Department of Electrical Engineering, Coordinator (HC) controls data transmission between all
National Taiwan University, Taiwan.
nodes. After the HC compiles demands of data streams, it
E-mail: {D99921030, hcyen}@ntu.edu.tw.
 D. J. Deng is with the Department of Computer Science and Information generates a fixed polling table that allows data transmission
Engineering, National Changhua University of Education, Taiwan. to specific nodes at a specific time to reduce packet waiting
E-mail: djdeng@cc.ncue.edu.tw. time and avoid collision. It thus has a great effect for data
Manuscript received 30 June 2014; revised 18 Dec. 2014; accepted 12 Jan. streams that require less latency. However, according to
2015. Date of publication 3 Feb. 2015; date of current version 20 Jan. 2016. related studies [12], [13], [14], [15], [16], [17], [18], [19], [20],
For information on obtaining reprints of this article, please send e-mail to:
reprints@ieee.org, and reference the Digital Object Identifier below. [21], the use of a great number of control packets such
Digital Object Identifier no. 10.1109/TDSC.2015.2399912 as CF-POLL and CF-ACK significantly reduces channel
1545-5971 ß 2015 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.
6 IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING, VOL. 13, NO. 1, JANUARY/FEBRUARY 2016

TABLE 1
DSRC Spectrum

utilization. With a fixed polling table, HCCA is only suitable


for real-time data streams with the characteristics of Con-
stant Bit Rate (CBR), such as Voice Over Internet Protocol Fig. 1. WAVE structure.
(VOIP). However, it is not suitable for real-time data streams
with the characteristics of Variable Bit Rate (VBR), such system has been designed by WAVE to support sequential
as video applications. Furthermore, when implementing operation between CCH and SCH. Data transmission in
HCCA in VANETs, due to the characteristics of infrastruc- WAVE can be divided into three types namely control
ture networks, HCCA can be applied to V2I scenarios, but frames, management frames, and data frames. Control
cannot be applied to V2V scenarios. And in addition, both frames are used by the IEEE 802.11p MAC to send channel
EDCA and HCCA consider neither multi-channel status in access coordinating messages. Management frames contain
network resource allocation. Overall, it still lack an efficiency the timing advertisement frame and WAVE Service Adver-
QoS support in VANETs; nonetheless, QoS support is an tisements (WSAs), with the former used to send time-based
important issue in VANETs since it affects the safety of driv- synchronized messages, and the latter used to send mes-
ing. For example, collision warning, traffic management, sages related to channel service. Data frames are upper-
and overtaking assistance are all useful services for driver. layer application data, mostly WAVE Short Message (WSM)
However, if the data of those services transmit in a longer or and IP-based data. The CCH is mostly retained for manage-
unpredictable latency, driver would be confused and thus ment frames and WSMs with a high priority while the SCH
makes the wrong decision. Hence, in order to remedy the is mostly used for WSMs with a low priority and IP-based
above-mentioned problems, in this paper, we propose a col- data for users. To coordinate CCH and SCH, WAVE uses
lision-free QoS-guaranteed solution in the WAVE multi- a preset alternating channel switch mode, as shown in
channel structure, called EDF-CSMA. The innovation of Fig. 2b. WAVE divides the wireless channel access area into
EDF-CSMA is that it fusions the distributed and centralized alternating CCH and SCH intervals and vehicle nodes have
channel access schemes, i.e. EDCA and HCCA, to reduce the to switch to the corresponding channel within an individual
latency of transmission and increase the channel utilization. interval. According to WAVE protocol, a pair of CCH
EDF-CSMA provides a collision-free QoS-guaranteed envi- and SCH interval is defined as a sync interval with a preset
ronment for nodes by introducing the concept of EDF duration of 100 ms. It is similar to 802.11’s superframe con-
scheduling and reduces controlled packet transmission by structed by two different periods (in a superframe, which
integrating the properties of CSMA/CA distributed access. are Contention Period (CP) and Contention Free Period
EDF-CSMA is applicable to the V2I scenario and can also (CFP) respectively). However, for flexibility, WAVE also
apply in V2V scenario by constructing a dynamic service defines three extra modes of channel coordination.
group according to vehicle characteristics with its clustering As shown in Fig. 2a, 2c, 2d, they are as follows: continuous
technique. To guarantee QoS transmission and to achieve CCH/SCH access, allowing the nodes to monitor incoming
high network usage, EDF-CSMA introduces a delay based data from the CCH/SCH; immediate SCH access, allowing
admission control module which takes multi-channel status
into account to reduce the probability of delayed packet and
increase network capacity. In other word, the contribution of
this paper is that it achieves offering high-efficiency trans-
mission in VANET environment and maintains an accept-
able implementation complexity.
The rest of the paper is organized as follows: Section 2
describes the WAVE channel access structure, and its
related research and problems. Section 3 introduces the
EDF-CSMA’s principle of operation and the design of
admission control. Section 4 presents EDF-CSMA’s theoreti-
cal performance analysis and network utilization evalua-
tion. Section 5 describes the simulation results and index
evaluation. Section 6 concludes the paper.

2 PRELIMINARY
As aforementioned, the DSRC spectrum comprises one Fig. 2. Channel access options: (a) continuous, (b) alternating, (c) imme-
CCH and six SCHs; therefore, a channel access coordinating diate, and (d) extended.
CHANG ET AL.: V2V QOS GUARANTEED CHANNEL ACCESS IN IEEE 802.11P VANETS 7

the nodes to switch to SCH from CCH without waiting Minimum PHY rate (R). The minimum physical bit rate
for the next SCH interval; and extended SCH access, allow- assumed by the scheduler for calculating transmission time
ing the nodes to continue transmission on SCH without (in bits per second)
pausing and switching to CCH. The definitions of other parameters can be found in the
As for Media Access Control (MAC) implementation, IEEE 802.11e standard [12]. Meanwhile, the HC internally
WAVE specifies a communication stack for DSRC spec- implements an admission control unit (ACU). After receiv-
trum optimization. The internal structure contains two ing ADD-TS request, the ACU will evaluate if nodes can be
IEEE 802.11p MAC entities which operate on the CCH and added to its schedule. This request would be denied if the
SCH, respectively. The two entities have their own MAC channel status did not fulfill the node requirement and vice
configuration, and individually use the IEEE 802.11e HCF versa. It is clear that HCCA has a better QoS achievement
as their channel access mechanism. HCF provides two than EDCA, but in fact, HCCA still has a scope for improve-
channel access modes: the EDCA mode, mainly used by ment, in which a scheduling algorithm is of the highest pri-
WAVE, and the optional HCCA mode. EDCA optimizes ority. The HCCA scheduler only considers a few TSPEC
the IEEE 802.11 Distributed Coordination Function (DCF) parameters to construct a fixed schedule; hence, it cannot
to allow support for differentiated services. It provides meet the requirement of traffics that have VBR property.
four Access Categories (ACs), namely, AC4, AC3, AC2, In the relevant literature, many papers have proposed
and AC1. Each category has its access parameters such as improvements to the scheduling algorithm. In [13], a TXOP
Arbitration Inter-Frame Spacing (AIFS), Contention Win- timer has been used to replace the original service cycle
dow (CW), and Transmit Opportunity (TXOP). In EDCA, time for each data stream, and EDF strategy has been
the node will randomly select one value from the CW[AC] applied to identify nodes available for data transmission.
interval and set it as the initial value of its backoff timer. To optimize the allocated TXOP duration, [14] proposes a
When a node detects the channel status changes from busy mechanism that uses the queue size information based on
to idle, it will defer an AIFS[AC] time, and then starts the the IEEE 802.11e QoS header. In [15], an adaptive scheduler,
backoff countdown procedure. In particular, during the which can provide different service cycles for different data
procedure, if a node detects the channel status changes streams to optimize the VBR effect has been proposed. In
from idle to busy, it will freeze the backoff timer and wait [16], data transmission time and service cycle is calculated
for the idle status. A node can start data transmission only based on packet loss rate and traffic characteristics. Linear
when the backoff timer decreases to zero. The duration of programming has also been implemented to optimize chan-
AIFS[AC] and CW[AC] decides the transmission priority nel schedule of scalable video streaming [17].
of nodes. A higher AC level means lower deferring and Another problem that needs to be addressed in HCCA is
backoff time. In such cases that the data has a high proba- low efficiency. Since HC follows a fixed polling order to
bility to be sent. Though EDCA can provide differentiated serve nodes, the channel will be reserved for each node no
services, there is still a probability that nodes select the matter it has data or not. It thus created a large number of
same backoff value, and data collision will occur. Hence, it unnecessary polling frames. Against to this, state-based
is not enough to provide QoS for these highly time-sensi- polling, simultaneous polling, and multistep polling had
tive data such as collision warning, intersection warning, been proposed to save polling time when a node has no
and real-time voice/video. data [18], [19], [20]. However, the hardware implementation
On the contrary, HCCA manages the channel resource by for these polling/scheduling algorithms is difficult owing
applying a centralized channel access method. All the trans- to their high complexity. Hence, the Deterministic Backoff
missions in HCCA are scheduled by the HC, and as Time (DEB) polling algorithm proposed in [21] uses the assigned
Division Multiple Access (TDMA), nodes will transmit data backoff value based on CSMA/CA to achieve polling
in a designated period to avoid data collision. To achieve scheme and to save extra time of the polling frame. How-
this, HCCA exploits polling packet such as QoS CF-POLL ever, [21] does not consider node’s QoS requirement in its
and QoS CF-ACK to organize the transmission of nodes. schedule. Packet loss and delay expired flags might occur
Nodes can only initiate a transmission after receiving QoS due to ignoring the real-time properties.
CF-POLL. Besides, nodes that intend to transmit data have In addition to these issues, the main reason that HCCA
to send an ADD-TS request which contains a Traffic Specifi- cannot be the main access mode of WAVE is that it lacks
cation (TSPEC) information that records the traffic proper- a complete strategy to apply in a wireless ad hoc network
ties and QoS requirement to the Hybrid Coordinator (HC) (i.e. V2V communication). Besides, although applying
first. The common fields of TSPEC are shown as follows: EDCA has mentioned advantages, the network usage will
Peak data rate (P). The maximum bit rate allowed for be low because the feature of WAVE multi-channel has still
packet transfer (in bits per second); not been considered.
Mean data rate (r). The average bit rate for packet trans-
mission (in bits per second); 3 PROPOSED WAVE/VANET SOLUTION
Maximum burst size (M). The maximum size of a data Based on the above-mentioned requirements, this paper
burst that can be transmitted at the peak data rate (in bytes); proposes an EDF-CSMA channel access solution in WAVE.
Delay bound (D). The maximum delay allowed in trans- EDF-CSMA has two main projects namely WAVE service
porting a packet across the wireless interface (including group configuration and multi-channel access mechanism.
queuing delay, in milliseconds); The former focuses on constructing a complete environment
Nominal MAC Service Data Unit (MSDU) size (L). The for data transmission, while the latter focuses on satisfying
nominal size of a packet (in bytes); QoS transmission and maximizing network usage.
8 IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING, VOL. 13, NO. 1, JANUARY/FEBRUARY 2016

Fig. 4. The format of MAL packet.

Fig. 3. The concept of WSG. with a being a parameter that can be configured from 0 to 1.
After calculation, the node with the greatest weight will be
3.1 WAVE Service Group Configuration selected as the GH. Subsequently, the selected GH will
At the initial state, we define a group of nodes that can broadcast a JOIN-INVITE packet to one-hop nodes. And
transmit data between each other as a WAVE Service nodes that receive the packet will reply a JOIN-REPLY
Group (WSG), as shown in Fig. 3, with two different roles: packet to the GH and list itself as a GM of the WSG, thereby
Group Head (GH), for coordinating data transmission; completing a V2V WSG. Note that, in V2I scenario, the
and Group Member (GM), for data transmission. A WSG Road Site Unit (RSU) is as it should be the group head due
may contain one GH and multiple GMs. In the process of to its infrastructure property, and the other processes are
WSG construction, there are existing two different scenar- the same as V2V.
ios namely V2V and V2I, depending on the situation. In When the GH has received all one-hop node’s JOIN-
V2V scenario, we introduce the clustering technique of REPLY or after a preset timeout, the GH will broadcast a
Mobile Ad hoc Networks (MANETs) and apply cluster MAC Addresses List (MAL) packet to all GMs. As shown in
head election to nominate specific nodes as the GH of Fig. 4, MAL packet contains the MAC address of all GMs.
WSG. In MANET-related research, many election algo- After GMs receive the MAL, they will check the sorting
rithms have been proposed, such as Lowest-ID [22] and position of their MAC address in the MAL and set the posi-
Max-Min D-hop clustering Algorithm [23]. However, these tion information as its Member ID (MID). For instance, sup-
algorithms are not applicable for driving situations that pose there are three MAC addresses contained in the MAL
have the characteristic of high mobility and variability. For namely aa:aa:aa:aa, bb:bb:bb:bb, and cc:cc:cc:cc. When the
any random VANET cases, the GH must be the node that GM of bb:bb:bb:bb received the MAL, it will find out that
is closest to the center of WSG to have the shortest trans- the position of its MAC address is No. 2, and then configures
mission distance to nearby nodes. And it must have a rela- its MID as 2. After finishing the process, the WSG will start
tively stable speed to reduce the times for GH re-selection. the proposed EDF-CSMA multi-channel access mechanism.
Against to this, we follow the concept of [24] to propose the
relative position and mobility election criterion, and apply 3.2 Multi-Channel Access Mechanism
it to the WSG construction. We configure that every node
As mentioned earlier, the main purpose of the EDF-CSMA
periodically broadcasts a HELLO packet to other adjacent
is to satisfy the QoS transmission of real-time traffic stream-
(one-hop) nodes together with its Global Positioning System
ing. For driving safety, QoS support is to ensure the real-
(GPS) location and speed information. And each node keeps
time safety information, such as collision warning and inter-
a copy of the neighbor table to record the broadcasting
section warning, can be transmitted within the time limit.
information of all one-hop nodes. Assuming that there are n For real-time multimedia, the data beyond the delay bound
one-hop nodes recorded in the neighbor table, and then we have to be discarded to achieve high-quality user experien-
have two sets as V :fV1 ; . . . ; Vn g and L :fL1 ; . . . ; Ln g, with Vi ces, such as a clear voice call and smooth video streaming.
and Li being the speed and location of ith nodes. We Obviously, packet delay is the most important factor that
define a function SðVi Þ as Vi ’s sorting position in set V, and affects the QoS performance. On the other hand, to achieve
MedianðVÞ as the median of set V after sorting. Then, we
high network usage, GMs in a WSG are allowed to switch
can express the speed weight of the ith node as
between different SCHs. In other word, EDF-CSMA consid-
WV ðVi Þ ¼ 1=jSðVi Þ  MedianðVÞj : (1) ers not only sending a packet within an acceptable delay
Next, we calculate the central position Lc by range but also organizing the resource of each wireless
! channel. For this purpose, we divide the whole process into
1 X n Xn
two phases: QoS requirement collection and channel access
Lc ¼ Longi ; Lati ; (2)
n i¼1 i¼1 coordination.

where Long and Lat are the information of Longitude and


Latitude respectively and express position weight of the ith 3.2.1 Phase I: QoS Requirement Collection
node as In this phase, all of the GMs that intends to start a QoS
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi transmission will issue a request to the GH, and the GH has
WL ðLi Þ ¼ 1 ðLongi  Longc Þ2 þðLati  Latc Þ2 ; (3) responsibility for organizing these requests. Without loss of
generality, we consider that all GMs are distributed around
which is the distance between ith node and Lc . Finally, we the CCH and six SCHs. In order to receive requests from
define the total weight of the node as CCH and SCH, the GH will switch between CCH and SCH-
Ws ðiÞ ¼ aWV ðVi Þ þ ð1  aÞWL ðLi Þ (4) 1 (by default) with a period of 50 ms defined in the WAVE
standard. As Fig. 5 shown, the interval from t ¼ 0 to 50 ms
CHANG ET AL.: V2V QOS GUARANTEED CHANNEL ACCESS IN IEEE 802.11P VANETS 9

Fig. 5. The procedure of QoS requirement collection.

is represented as CCH-Interval, it means that the GH will section). After the reserved GMs and assigned-channels
stay at CCH in this interval. Likewise, the SCH-Interval have been determined, as shown in Fig. 6a, the GH classifies
means that the GH will stay at SCH-1 in this interval. GMs these GMs into several categories according to the assigned-
will know the switching frequency of GH, i.e., when the GH channel and GM’s delay bound specified in TSPEC. GMs in
will stay at CCH or SCH-1. GMs will send a so-called a particular category have the same assigned-channel and
Request For Service (RFS) packet to the GH at the corre- delay bound.
sponding interval. For instance, the GMA in Fig. 5 intends to The classified result will be encapsulated into an Aggre-
start a QoS data stream at CCH. In such cases, GMA will gated Category Table (ACT) packet and broadcasted to the
switch to CCH during the CCH-Interval and send the CCH- WSG at the begin of next same interval. More specifically,
RFS packet to GH. Similarly, GMB intends to issue a QoS when the CCH/SCH-RFSs have been received by GH dur-
voice call in SCH, so it will switch to SCH-1 during the ing CCH/SCH-Interval, a CCH/SCH-ACT will be sent at
SCH-Interval and send the SCH-RFS packet to GH. The RFS the begin of next CCH/SCH-Interval. Note that to ensure
packet which is like HCCA’s ADD-TS contains the destina- that the ACT packet has the highest transmission priority,
tion MAC address and the QoS requirement of the traffic the defer time of transmission is set as PCF Inter-Frame
stream, i.e., TSPEC. Once the GH has received the RFS, an Space (PIFS); moreover, to increase the channel efficiency,
acknowledge frame will be sent back to the GM. Further- the GH only broadcasts a ACT packet since the entities of
more, when the current SCH-Interval or CCH-Interval has reserved GMs is changed. The detailed structure of ACT is
terminated, the GH will decides that whether the requested shown in Fig. 7. As can be seen, the ACT packet declares
GM can be reserved and which channel can be assigned to three kinds of field namely category_num, category_info,
it by adapting the Admission Control Unit (ACU) (the and station_info. The field of category_num contains 4 bits,
behavior of the proposed ACU will be described at later which is used to represent the number of categories, i.e.,
the maximum number of categories is 15. And based on
the number of categories, the field of category_info contains
one or more subfield, and each subfield uses 15 bits to
represent a category information in which 3 bits are used to
record the assigned-channel (000 is represented as CCH,
001 to 110 are represented as SCH-1 to 6 respectively), 4 bits
to record the delay bound with a unit of 10 ms, and 8 bits to
record the number of GMs in the category, i.e., the maxi-
mum delay bound is 150 ms, and the maximum number of
GMs in a category is 255. The total length of category_info
will be the number of categories multiplied by 15 bits. And

Fig. 6. Classifying sample: (a) GMs classifying. (b) ACT with two catego-
ries and three GMs. Fig. 7. The format of ACT packet.
10 IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING, VOL. 13, NO. 1, JANUARY/FEBRUARY 2016

based on the number of GMs, the field of station_info con- at C i can be found by:
tains one or more subfield, and each subfield uses 12 bits to 20 1 3
record the classifying result of a GM in which 4 bits are Chi
1
6B XSi
Ch C 7
used to record the category that a GM belongs to and 8 bits bij ¼ 1 þ r  4@ Nk i A þ oij  15 (6)
to record the GM’s position in that category. The subfield of k¼1
station_info are sorted by MID; therefore, according to
GM’s MID, each GM can grab the classified result by check- with oij being the position of GM ij in C i , and r being the
ing its subfield. Once the ACT is received by GMs, they number of tolerable re-transmissions in which r ¼ 1 means
will parse the category_info length according to the value of one time transmission, i.e., no re-transmission when a fail-
category_num, and then search for the corresponding ure occurs. More specifically, the bij of GM ij is obtained
subfield in station_info by using their own MID, and to get by the total number of GMs in categories whose assigned-
their own category and position. Note that, the category channel is the same to C i and deadline sorting position is in
and position information would be both 0 if the GM is not front of C i plus GM ij ’s position in C i . For example, the
reserved. Fig. 6b shows an example with two categories and backoff value of GMs 1 to 3 shown in Fig. 6(b) is 1, 1, and
three GMs. Suppose that GM3 has received ACT packet, it 2 respectively.
can find out the number of categories by checking bit 1 to 4, In particular, the delay bound dij that the GM ij transmits
and it will know there are three categories. Then based on to GH in the QoS requirement collection phase is con-
the number, it will parse the information of each category structed by
by checking bit 5 to 34, and using the categories informa-
tion, GM3 can re-construct the category table. Moreover, dij ¼ gij þ d þ O; (7)
GM3 will know that the start bit of station_info is bit 35, and
according to its MID, it will get the third subfield by shifting where gij is the Packet Generation Interval (PGI) of GM ij , d
24 bits. Using the third station_info, the GM3 will eventually is a configurable parameter which reflects the real time level
know that it belongs to category 1, and the position is 2. of a packet, and O is the time of overhead.1 In order to
Through the above process, all GMs can store the same cate- reduce packet delay and jitter, we set d as zero in EDF-
gory table individually and know their category number CSMA, i.e., we try to make the delay bound close to PGI.
and position. In addition, once a GM’s QoS transmission With this approach, delay bound might be considered as a
requirement has completed, an End Of Service (EOS) packet GM’s service cycle while deadline ensures the execution of
will be issued to the GH. Therefore, the GH can remove the each service during this cycle. Once a GM completes the
specified GM from the reserved GMs. transmission of current cycle, it may selectively switch to
CCH to monitor safety packets or change to sleep mode until
3.2.2 Phase II: Channel Access Coordination the next PGI (after the deadline), and then switch back to
After storing the category table, each reserved GM uses SCH to start a new transmission. Due to the different service
EDF-CSMA for channel access in the rest of the CCH/SCH- cycles of GMs, the backoff value has to be refreshed dynami-
Interval. EDF-CSMA is a channel access policy based on cally. As mentioned above, each GM stores the category
CSMA/CA, which is the listen first, speak later policy. table; therefore, it can calculate the deadline of each cate-
Same as the DCF, all the reserved GMs will kick off the gory. Once the time reaches or exceeds one of these catego-
backoff procedure when the channel is sensed idle, and ries’ deadline, each GM must refresh the deadline
start transmission after backoff timer decreasing to zero. information using Equation (5) and re-calculate backoff
Different from DCF, in this paper, the reserved GM uses value using Equation (6). Through these operations, each
EDF-Backoff selection to get the backoff value rather than GM has a unique backoff value at each point of time, i.e.,
DCF backoff selection. With EDF-Backoff selection, a GM there will be no more than one GMs counting down to 0 at
calculates the latest service time (i.e. deadline) of each cate- the same time. Therefore, it achieves collision-free channel
gory based on the delay bound information recorded in the access based on CSMA/CA. Moreover, since the order of
category table, and then decides the most suitable backoff each category’s deadline is considered in the backoff calcula-
value for itself. To describe the detailed process, at the first tion, the earliest expired GM has the shortest backoff value.
we define the category table which contains n categories as Therefore, it achieves the concept of earliest deadline first.
C ¼ fC1 ; C2 ; ; Cn g and ith category as Ci ¼ fChi ; di ; Ni g, in Except the QoS transmission service provided for
which Chi , di , and Ni represent the ith category’s assigned- reserved GMs, for sending RFS and other non-QoS packets,
channel, delay bound and number of GMs, respectively. we propose another access method to avoid collision. As
Let t be the current time, the deadline DLi of category C i above mentioned, an ACT packet is broadcasted to GMs at
can be calculated by the beginning of the CCH/SCH-Interval; therefore, except
the category information, each GM knows that how many
DLi ¼ di  ðt mod di Þ : (5) GMs are reserved. Non-reserved GMs that intend to send
RFS or non-QoS packets will follow Algorithm 1 to select a
We then define Sm i as Ci ’s deadline sorting position in cate- unique backoff value. To describe the algorithm, consider a
gories whose assigned-channel is m and Nkm as the number GM x which is a non-reserved GM with MID x is performing
of GMs in a category whose assigned-channel is m and this algorithm, then GM x will check 1) the biggest backoff
deadline sorting position is k. Note that, the deadline sort-
ing is an ascending sort in which a lowest position means 1. The overhead includes channel switching time, physical transmis-
earliest deadline. Then the backoff value bij of GM ij located sion delay, and so on.
CHANG ET AL.: V2V QOS GUARANTEED CHANNEL ACCESS IN IEEE 802.11P VANETS 11

value of reserved GMs and 2) its order in non-reserved


GMs, and the order will be right-rotated according to the
deadline refresh times. Finally, the selected backoff value
will be the summation of these two number. For instance,
suppose that GM 3 , GM 4 , GM 6 which are performing this
algorithm are non-reserved GM and GM 1 , GM 2 , GM 5 are
reserved GMs whose backoff value are 1, 2, and 3 respec-
tively. And since applying EDF-CSMA, there are two times
refreshing for deadline synchronization. After calculating,
GM 3 , GM 4 , GM 6 will get the backoff value 6, 4, and 5
respectively. Note that, the deadline refresh times should be Fig. 8. The structure of flow shaper.
calculated by each GM in each time to make Algorithm 1
works. Also, once the current time reaches or exceeds one of the GH evaluates the QoS requests of GMs using the admis-
these categories’ deadline, non-reserved GMs still need to sion control unit and then broadcasts an ACT packet
re-perform Algorithm 1 since the reserved GMs will update appended with the result of the evaluation to each GM at
their backoff value at this moment. next corresponding interval. 4) After receiving the ACT
packet, the reserved GMs will execute EDF-Backoff selection
Algorithm 1. Unique backoff value selection for non- to obtain an unique backoff value and start counting down.
reserved GM The QoS packet will be sent after the backoff timer decreas-
ing to zero. 5) After completion of the QoS request, the GM
1: ref: deadline refresh times
will send an EOS packet at next transmission opportunity to
2: n, m: the number of total and non-reserved GMs
3: res: the biggest backoff value of reserved GMs
inform the GH to remove it from the reserved GMs. 6) After
4: o: the order of target GMs in non-reserved GMs the GH removes the GM, it will broadcast an ACT packet to
5: bi : GMi ’s backoff value calculated by (6) the GMs to update their category table at next corresponding
6: res 0, o 1, i 1 cycle. 7) Returning to step 2.
7: while i <¼ n do
8: if GMi is in the list of reserved GMs then 3.3 Admission Control
9: res ¼ max (res, bi ) In this section, we will start to introduce the proposed admis-
10: else sion control unit, which aims at reducing packet delay and
11: increase m increasing network usage. At the first, to ensure that each
12: if i < target GM’s MID then GM’s data transmission matches the proposed TSPEC speci-
13: increase o fication and facilitates ACU design, this paper introduces a
14: end if flow shaper to each GM’s data stream, referring to the con-
15: end if cept proposed in [25]. As shown in Fig. 8, the flow shaper has
16: increase i two token buckets in which the first is used to restrict flow
17: end while peak data rate Pi while the second is used to restrict flow
18: output ¼ res þ ððo þ ðref mod mÞ  1Þ mod mÞ þ 1 mean data rate ri . Both parameters are configured as the cor-
responding values proposed in TSPEC. When a GM requires
Besides, there is other different between legacy CSMA/ data transmission, it can only be allowed after that the sec-
CA and EDF-CSMA. In front, a node kicks off backoff proce- ond bucket has the token number as the transmitted data.
dure once it has packets queued in the buffer while in later When the transmission is done, the second bucket will
only the non-reserved GMs follow this rule. The reserved remove the corresponding number of the token. We assume
GMs have to kick off backoff procedure once the time that ai ðt; t þ tÞ is the data generated by traffic flowi , in
reaches or exceeds one of these categories’ deadline; there- t; t þ t, then after flow shaping, ai ðt; t þ tÞ will be shaped as
fore, it can ensure that the backoff value of each GM is
unique and synchronized. And as mentioned, after the 
Pi t; 0 < t < hi
backoff timer decreasing to zero and the current TXOP is bi ðt Þ ¼ (8)
Bi þ ri t; t  hi :
finished, GM can switch to CCH or change to sleep mode
until the next service cycle. However, there is still a proba- Here, Bi ¼ Mi ð1  ðri =Pi ÞÞ, and hi ¼ Mi =Pi , Mi being the
bility that packets are generated after current TXOP and maximum packet length declared within TSPEC. Through
before next service cycle. In this case, the same to non- this process, any random program ai ðt; t þ tÞ can be shaped
reserved GMs, this GM has to choose an unique backoff to the data stream that matches TSPEC. Subsequently, we
value by using Algorithm 1 and start backoff procedure. construct a Network Dependent Graph (NDG) which repre-
Finally, for the EDF-CSMA operational procedure we can sents the relation of traffic source and sink between each
conclude the following: 1) In any VANET environment, a GM. As shown in Fig. 9, given an NDG: G(V, E) being a
group of nodes that can communicate with each other choo- directed graph in which the vertex is represented as GM
ses a node as GH using cluster head election, and the rest while the directed edge is represented as data transmission.
nodes act as GM. 2) When wireless channel becomes idle, the For example, E12 represents that GM 1 is wanting to have a
GM with a QoS transmission request selects a unique backoff data stream with the destination GM 2 . For allocating net-
value via Algorithm 1 and then transmits RFS packets after work resource in multi-channel, all of the GMs related to a
the backoff timer decreasing to zero. 3) After receiving RFS, sender have to be assigned in the same channel. In an NDG,
12 IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING, VOL. 13, NO. 1, JANUARY/FEBRUARY 2016

Fig. 10. The structure of aggregated data transmission.


Fig. 9. The concept of NDG.

we can easily figure out the related GMs by finding all the rate fj of stream j can be expressed as
connected sub-graph. Following that, we aggregate the
TSPECs of related GMs into an Aggregated TSPEC (A- qj ðtÞ
fj ¼ lim ¼ r aggrj
rj : (13)
TSPEC). For example, in Fig. 9, subgraph A contains three t!1 t
GMs in which there are two TSPECs provided by GM 3 and
The data rate variance of stream j can be expressed as
GM 5 , respectively. Then these two TSPECs will be aggre-
gated into an A-TSPEC. After that, the ACU can be formed !
a aggr ½0; 0 þ t
RVj ðt Þ  Var
j
as a function which has the input A-TSPEC, and it outputs a
decision to decide whether the related GMs can be added to t
h i h i2
services and which channel should be assigned to the
E aaggr j ðt Þ2  E aaggr j ðt Þ
related GMs. we consider a system with multiple aggre- ¼
gated data streams as shown in Fig. 10. We define D aggri , t2 (14)
Z b aggrj ðtÞ
L aggri , P aggri , B aggri , r aggri , b aggri ðt Þ, and a aggri ðt Þ 1
¼ 2 x2 dFt;j ðxÞ
as ith aggregated parameters, and for each stream entering t 0
queue, we set the accumulated data amount within time t Z b aggrj ðtÞ !2
as a random variable Ej ðtÞ. Then the probability for a ran- 1
 2 xdFt;j ðxÞ
dom packet that has a delay greater than minimum delay t 0
bound (D) for all data streams is b aggrj ðtÞfj t  fj t
with data distribution ft;j ðxÞ¼ b aggrj ðtÞ dðxÞ þb aggrj ðtÞ dðx
( )
X b aggrj ðtÞÞ. Taking the upper limit for above equation,
probfd>Dg  max prob Ej ðt ÞCt  CD ; (9)
0tb
j2C we can obtain

where C is the linked system capacity, and in particular, we fj qj ðt Þ


RVj ðt Þ ¼  f2j
calculate link capacity C as 8t  
 < r aggrj P aggrj  r aggrj
r2j ; 0 < t < hj
R  number of SCH; SCH request ¼
C¼ (10) :r aggrj B aggrj r2j
R; CCH request; t ; hj  t  by :

where R is the physical bandwidth of wireless channel. The (15)


central limit theorem is then applied. If a random variable Busy time b is as shown below:
Yj has a mean of mj and variance s 2j , and Y1  Yn are mutu- ( !)
ally independent, Y1 þ Y2 þ    þ Yn is a Gaussian distribu- X
P P
b¼ min t>0 q ðtÞ  Ct ; (16)
tion with a mean nj¼1 mj and a variance nj¼1 s 2j , and we j2C j
derive the formula
  where qj ðtÞ is the upper data limit that the accumulated
probfd>Dg max prob E^j ðtÞCt  CD ; (11) data flow reaches to consider data usage efficiency, and is
0tb
expressed by the following equation:
P P
wherein E^j ðtÞ  Nð j2C tfj ; j2C t RV j ðtÞÞ is a Gaussian
2

distribution, fj is the mean data rate of data stream j, a aggrj ½t; t þ t  qj ðt Þ


RV j ðtÞ is variance of mean data rate of data stream j, and b    
¼ min P aggri
t
rj ; B aggrj þ r aggrj t
rj ;
is the busy time. Knightly [26], [27] uses approximated tail
distribution and experiment to express the above formula as (17)
!
1 ðC ðtþDÞmt Þ2 where rj is the transmission time after considering network
probfd>Dg max pffiffiffiffiffiffi exp (12)
0tb 2p 2s 2t resource utilization ratio under the coordinating system,
P which is the time with transmission time of overhead data
with the total average data flow mt ¼ j2C tfj , the total divided by the time without transmission time of overhead
P
data traffic variance s 2t ¼ j2C t 2 RVj ðtÞ. The average data data, as shown below:
CHANG ET AL.: V2V QOS GUARANTEED CHANNEL ACCESS IN IEEE 802.11P VANETS 13

Fig. 11. The procedure of channel access in (a) HCCA scheduler and (b) EDF-CSMA.

Tj and
rj ¼
; (18)
8
L aggrj

TSYM n  ni
bpSjD P ðy ¼ iÞ¼ 1  Pempty Pempty i ; (21)
ni
where TSYM is the symbol time in seconds, bpSjD
is the data
modulation rate in bits per symbol, and L aggrj is the aggre- respectively. According to the properties of PGF transfor-
gated average packet size in bytes. The transmission time Tj mation, the PGF of (20) and (21) will be
with overhead is shown in the equation below:
GX ðZ Þ ¼ ðPempty þ ð1  Pempty Þ  ZÞn (22)
Tj ¼ TDIFS þ TSlot þ TDATA þ TSIFS þ TACK : (19)
and
Through the above derivation, we can easily calculate
acceptable QoS, i.e., P fd > Dg, and the main task of channel GY ðZ Þ ¼ ðð1  Pempty Þ þ Pempty  ZÞn ; (23)
access coordination becomes the simple comparison of
P fd > Dg with threshold . A requesting GM is reserved respectively. From x and y, the total channel access time of
if the calculated probability is less than  and vice versa. EDF-CSMA and HCCA scheduler can be shown as:
Therefore, this probability can be deemed as blocking
probability to a certain degree.
THCCA ¼ TBeacon þ TPIFS þ TCF END
X
y
4 CHANNEL UTILIZATION ANALYSIS þ n  ðTSIFS þ TCF POLL Þ þ fTPIFS g
EDF-CSMA achieves collision-free by applying a distrib- j¼1 (24)
uted access mechanism rather than centralized access mech- X
x

anism; therefore, in this section we will analyze channel þ fTSIFS þ Di þ TSIFS þ TACK g
i¼1
utilization in detail and emphasize that EDF-CSMA has a
greater channel utilization than formal scheduler. We con- and
struct a statistical probability analysis model and analyze
EDF-CSMA and the HCCA scheduler channel utilization
using a Probability Generating Function (PGF) [28], [29], TEDF CSMA ¼ TPIFS þ TACT þ n  ðTDIFS þ TSlot Þ
[30]. For HCCA scheduler, we assume that the channel is X
x
applied to CFP operation completely and without consider- þ fDi þ TSIFS þ TACK g; (25)
ing any virtual carrier sensing technique such as Ready i¼1

To Send (RTS) and Clear To Send (CTS). Fig. 11 shows the


procedure of channel access in HCCA scheduler and EDF- where TBeacon , Di , TCF POLL , TCF END , TACK , and, TACT
CSMA. We consider that there are n nodes needed to be represent time of beacon frame, the ith transmitted data, CF-
served in a WSG. Then, two random variables x and POLL frame, CF-END frame, acknowledgement frame, and
y ¼ n  x are given to represent the number of nodes with ACT frame, respectively. Let ui ¼TSIFS þ Di þ TSIFS þ TACK
data transmission and the number of nodes without data and vi ¼Di þ TSIFS þ TACK , the PGF of above equation can
transmission respectively. Let Pempty be the probability that be shown as
a node’s queue is empty, the probability distribution for x
 Px   Py 
and y can be formed as the binomial distribution THCCA ðZ Þ ¼ E Z i¼1 ui  E Z i¼1 TPIFS
 
n  i  E Z TBeacon þTPIFS þTCF END þnðTSIFS þTCF POLL Þ
P ðx ¼ iÞ¼ 1  Pempty Pempty ni (20)
i (26)
14 IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING, VOL. 13, NO. 1, JANUARY/FEBRUARY 2016

TABLE 2 and
Network Parameters
TEDF CSMA ¼ TPIFS þ TACT þ nðTDIFS þ TSlot Þ
  (31)
þ n  1  Pempty ðD þ TSIFS þ TACK Þ:

From the above derivations, we can express channel utiliza-


tion as time cost of data transmission within unit time,
which is
 
n  1  Pempty  D
Utilization ¼ : (32)
total length of CFP

Then we substitute time parameter with Table 2 and simu-


late channel status by adjusting P empty . Pempty ¼ 0.01 repre-
sents that the channel is saturated, and each node always
and has packets queued in its buffer, while Pempty ¼ 0.5 repre-
sents that channel is not saturated, and each node has a
 Px  probability of 50 percent to transmit packet. We set the
TEDF CSMA ðZ Þ ¼ E Z i¼1 vi 
  (27) experimental period as 200 second; the maximum data size
E Z TPIFS þTACT þnðTDIFS þTSlot Þ : that a node can transmit is configured as 2000 to 5000 bytes,
and the maximum physical rate is set as 11 Mbps. Fig. 12
Let the PGF of x, u, and v are represented as XðzÞ, UðzÞ, and shows the analysis results. The channel utilization of HCCA
V ðzÞ. From (22) and (23), we can obtain scheduler in both saturated and non-saturated states are
    around 20 to 45 percent. When the maximum data size
THCCA ðZ Þ ¼ Pempty þ 1  Pempty  U ðzÞ increasing, the wasted bandwidth of HCCA becomes larger.
  
 1  Pempty þ Pempty  Z TPIFS The main reason is that HCCA scheduler uses a huge
amount of controlling packets to coordinate node packet
 E½Z TBeacon þTPIFS þTCF END þnðTSIFS þTCF POLL Þ transmission and distributes a fixed TXOP which considers
(28) the maximum data size to each node whether nodes have
data to transmit. Although it satisfies the transmission
and requirement of each node, the scheduler causes extra band-
    width waste. Conversely, the channel utilization of EDF-
TEDF CSMA ðZ Þ ¼ Pempty þ 1  Pempty  V ðzÞ  CSMA may reach up to 80 percent no matter what maxi-
(29)
E½Z TPIFS þTACT þnðTDIFS þTSlot Þ : mum data size is. The main reason is that EDF-CSMA uses
distributed transmission strategy to save extra controlling
According to Equations (28) and (29), by taking the deriva- packet transmission time. Also, with the CSMA/CA tech-
tive and setting Z to unity, we can obtain the average time nique, if a node has no data to transmit, other nodes can
length of EDF-CSMA and HCCA scheduler: continue their operation after TDIFS þ TSlot to save extra
transmission time.
THCCA ¼ TBeacon þ TPIFS þ TCF END
þ nðTSIFS þ TCF POLL Þ 5 SIMULATION
(30)
þ n  Pempty  TPIFS In this section, extensive simulations to validate the EDF-
 
þ n  1  Pempty ðD þ 2TSIFS þ TACK Þ CSMA solution are performed, and the results are com-
pared and discussed in terms of throughput, mean packet

Fig. 12. Channel utilization with (a) Pempty ¼ 0:01, (b) Pempty ¼ 0:5.
CHANG ET AL.: V2V QOS GUARANTEED CHANNEL ACCESS IN IEEE 802.11P VANETS 15

TABLE 3 to the property of random access, and it leads that the chan-
TSPEC Parameters nel is occupied with fewer nodes. In contrast, the data trans-
mission time in EDF-CSMA will be shorter cause of the
organized transmission order; therefore, EDF-CSMA can
accommodate a large number of nodes. Fig. 13b shows the
packet loss rate of video transmission. For the same reason,
after 10 nodes with 12 Mbps and 17 nodes with 27 Mbps,
there is no more bandwidth can be assigned to nodes
in EDCA, and the extra data transmission will be dropped
due to expiring the deadline. In addition, the probability
delay, and loss rate for EDF-CSMA and the main scheme of probfd > 40msg calculated by admission control unit is
WAVE, i.e., EDCA. Furthermore, to make the simulation also presented in this figure. As it shown, the probability is
close to reality, the evaluation of real video is introduced, slightly higher than the simulation. Fig. 13c shows a compar-
and the result is compared in terms of PSNR. The simula- ison of mean packet delay. Before reaching the limit of band-
tion tool we used is network simulator 3 [31], in which the width, both of EDF-CSMA and EDCA have lower packet
wave model is adopted. The simulations are performed in a delay. However, it should be note that, the packet delay of
V2V scenario with increasing vehicular nodes, where each EDF-CSMA is still lower than EDCA’s because the node in
node moves with velocities 120 km/h and transmits a uni- EDCA which has data to transmit has to defer the access
cast VBR video stream to another node. The Nakagami time for a random period while the defer time in EDF-
propagation loss model is also applied in channel fading. CSMA is determined.
The other related parameters such as SIFS, DIFS, slot time, At the second, the real video akiyo [33] which is encoded
etc. are the same as mentioned in Table 2. Besides, for facili- by MPEG-4 and formed as a video stream with 256 Kbps is
tating comparison, only one CCH/SCH is adopted for data adopted as the video source. To apply real video in NS-3,
transmission. we introduce the EvalVid [34], which is a validation tool to
At the first, the trace file Jurassic Park I [32] is adopted as validate the performance of video streaming. Then the
the video source, and the related parameters are listed in received videos are evaluated by using the index of PSNR,
Table 3. Fig. 13a compares the throughput of EDF-CSMA which is calculated by
and EDCA, and in which two kinds of physical rate namely
12 Mbps and 27 Mbps are adopted. As can be seen, the 2552
capacity of EDF-CSMA is higher than EDCA either with PSNR¼10log P M1 PN1 0ði;jÞ 2
: (33)
j¼0 ½xði; jÞ  x
1
12 Mbps or 27 Mbps. The main reason is that a node trans- M
N i¼0
mits data in EDCA will cause longer transmission time due

Fig. 13. The comparison of (a) Throughput, (b) packet loss, (c) packet delay, and (d) PSNR.
16 IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING, VOL. 13, NO. 1, JANUARY/FEBRUARY 2016

[5] A. Vinel, D. Staehle, and A. Turlikov, “Study of beaconing for


car-to-car communication in vehicular ad-hoc networks,” in
Proc. ICC Workshops Commun. Workshops, 2009, pp. 1–5.
[6] A. Vinel, “3gpp lte versus ieee 802.11 p/wave: which technol-
ogy is able to support cooperative vehicular safety
applications?” Wireless Commun. Lett., vol. 1, no. 2, pp. 125–128,
Apr. 2012.
[7] A. Vinel, E. Belyaev, K. Egiazarian, and Y. Koucheryavy, “An
overtaking assistance system based on joint beaconing and real-
time video transmission,” IEEE Trans. Vehicular Technology,
vol. 61, no. 5, pp. 2319–2329, June. 2012.
[8] IEEE Std. 1609.1-2006, “Ieee trial-use standard for wireless access
in vehicular environments (wave)—resource manager,” IEEE,
Sept. 2006.
[9] IEEE Std. 1609.2-2006, “Ieee trial-use standard for wireless access
in vehicular environments—security services for applications and
management messages,” IEEE, Jul. 2006.
[10] IEEE P1609.3/D1.0, “Draft standard for wireless access in
vehicular environments (wave)—networking services,” IEEE,
Dec. 2008.
Fig. 14. The captures of received video: (a) EDF-CSMA with five nodes, [11] IEEE P1609.4/D1.0, “Draft standard for wireless access in vehicu-
(b) EDCA with five nodes, (c) EDF-CSMA with 15 nodes, and (d) EDCA lar environments (wave)—multi-channel operation,” IEEE, Dec.
with 15 nodes. 2008.
[12] IEEE Std. 802.11e, “Part 11: Wireless lan medium access control
Fig. 13d shows the comparison of PSNR and Fig. 14 shows (mac) and physical layer (phy) specifications amendment:
the captured image. As can be seen, EDF-CSMA has a Medium access control (mac) quality of service enhancements,”
IEEE, Nov. 2005.
higher index than EDCA in each case. In MPEG-4, each [13] A. Grilo, M. Macedo, and M. Nunes, “A scheduling algorithm for
frame is classified as I-frame or P-frame, and I-frame has qos support in ieee 802.11e networks,” IEEE Wireless Comm. Mag.,
the original data while P-frame records the different vol. 10, no. 3, pp. 36–43, Jun. 2003.
between I-frame and next frame. It is obvious that missing [14] D. Skyrianoglou, N. Passas, and A. Salkintzis, “Arrow: An
efficient traffic scheduling algorithm for ieee 802.11e hcca,”
I-frame will cause undesirable video quality, and it is also IEEE Trans. Wireless Comm., vol. 5, no. 12, pp. 3558–3567,
the main reason causing the lower performance in EDCA. Dec. 2006.
[15] I. Inan, F. Keceli, and E. Ayanoglu, “An adaptive multimedia qos
scheduler for 802.11e wireless lans,” in Proc. IEEE Int’,l Conf. Com-
6 CONCLUSION mun., vol. 11, Jun. 2006, pp. 5263–5270.
This paper has proposed an EDF-CSMA channel access [16] G. Boggia, P. Camarda, L. Grieco, and S. Mascolo, “Feedback
based control for providing real-time services with the 802.11e
solution for IEEE 802.11p WAVE to improve the QoS data mac,” IEEE/ACM Trans. Netw., vol. 15, no. 2, pp. 323–333, Apr.
transmission and increase the network usage in VANETs. 2007.
An admission control unit which considers the multi-chan- [17] M. van der Schaar, Y. Andreopoulos, and Z. Hu, “Optimized scal-
able video streaming over ieee 802.11 a/e hcca wireless networks
nel resource has been constructed to optimum the channel under delay constraints,” IEEE Trans. Mobile Comput., vol. 5, no. 6,
access for the reserved nodes. A channel utilization analysis pp. 755–768, Jun. 2006.
has been also conducted to highlight that EDF-CSMA has [18] O. Sharon and E. Altman, “An efficient polling mac for wireless
higher channel utilization than other well-known mecha- lans,” IEEE/ACM Trans. Networking, vol. 9, no. 4, pp. 439–451,
Aug. 2001.
nisms. Simulation experiments indicated that the proposed [19] S. C. Lo, G. Lee, and W.-T. Chen, “An efficient multipolling mech-
EDF-CSMA solution is able to provide high network capac- anism for ieee 802.11 wireless lans,” IEEE Trans. Comput., vol. 52,
ity and guarantee QoS support. As future work, we will no. 6, pp. 764–778, Jun. 2003.
[20] B. Kim, S. Kim, Y. Fang, and T. Wong, “Two-step multi-polling
extend the proposed scheme to a multi-hop VANETs and mac protocol for wireless lans,” IEEE J. Sel. Areas Commun.,
heterogeneous networks. Der-Jiunn Deng is the correspond- vol. 23, no. 6, pp. 1276–1286, Jun. 2005.
ing author of this paper. [21] Y. He and X. Ma, “Deterministic backoff: Toward efficient
polling for ieee 802.11e hcca in wireless home networks,”
IEEE Trans. Mobile Comput., vol. 10, no. 12, pp. 1726–1740,
ACKNOWLEDGMENTS Dec. 2011.
This work was supported by ROC NSC under Contract 102- [22] M. Gerla and J. T.-C. Tsai, “Multicluster, mobile, multimedia radio
network,” ACM-Baltzer J. Wireless Netw., vol. 1, no. 3, pp. 255–265,
2221-E-018-012-MY3. Der-Jiunn Deng is the corresponding Sep. 1995.
author of the article. [23] A. D. Amis, R. Prakash, T. H. P. Vuong, and D. T. Huynh, “Max-
min d-cluster formation in wireless ad hoc networks,” in Proc.
REFERENCES IEEE INFOCOM Conf., vol. 1, March. 2000, pp. 32–41.
[24] S. C. Lo, Y. J. Lin, and J. S. Gao, “A multi-head clustering algo-
[1] IEEE P802.11p/D6.01, “Part 11: Wireless lan medium access con- rithm in vehicular ad hoc networks,” Int. J. Comput. Theory Eng.,
trol (mac) and physical layer (phy) specifications—amendment 7: vol. 5, no. 2, pp. 242–247, Apr. 2013.
Wireless access in vehicular environments,” IEEE, Apr. 2009. [25] C. T. Chou, S. N. Shankar, and K. G. Shin, “Achieving per-stream
[2] X. Xiang, W. Qin, and B. Xiang, “Research on a dsrc-based qos with distributed airtime allocation and admission control in
rear-end collision warning model,” IEEE Trans. Intell. ieee 802.11e wireless lans,” in Proc. IEEE INFOCOM, vol. 3, Apr.
Transportation Syst., vol. 15, no. 3, pp. 1054–1065, June. 2014. 2005, pp. 1584–1595.
[3] A. Vinel, V. Vishnevsky, and Y. Koucheryavy, “A simple analyti- [26] E. W. Knightly, “Second moment resource allocation in multi-
cal model for the periodic broadcasting in vehicular ad-hoc service networks,” in Proc. ACM SIGMETRICS, vol. 25, no. 1, Jun.
networks,” in Proc. IEEE GLOBECOM Workshops, 2008, pp. 1–5. 1997, pp. 181–191.
[4] C. Campolo, A. Vinel, A. Molinaro, and Y. Koucheryavy, [27] E. W. Knightly, “Enforceable quality of service guarantees for
“Modeling broadcasting in ieee 802.11 p/wave vehicular bursty traffic streams,” in Proc. IEEE INFOCOM, vol. 2, 1998,
networks,” Commun. Lett., IEEE, vol. 15, no. 2, pp. 199–201, pp. 635–642.
Feb. 2010.
CHANG ET AL.: V2V QOS GUARANTEED CHANNEL ACCESS IN IEEE 802.11P VANETS 17

[28] S. V. Krishnamurthy, A. S. Acampora, and M. Zorzi, “Polling- Hsu-Chun Yen received the BS degree in electri-
based media access protocols for use with smart adaptive array cal engineering from National Taiwan University,
antennas,” IEEE/ACM Trans. Networking, vol. 9, no. 2, pp. 148–161, Taiwan, in 1980, the MS degree in computer
Apr. 2001. engineering from National Chiao-Tung Univer-
[29] A. Papoulis and S. U. Pillai, Probability, Random Variables and Sto- sity, Taiwan, in 1982, and the PhD degree in
chastic Processes, 4th ed, McGraw Hill, Jan. 2002. computer science from the University of Texas at
[30] J. M. Chung, M. Kim, Y. S. Park, M. Choi, S. Lee, and H. S. Oh, Austin, Austin, TX, in 1986. He is currently a pro-
“Time coordinated v2i communications and handover for wave fessor of electrical engineering at National Tai-
networks,” ieee journal on selected areas in communications,” wan University, Taiwan, where he initially joined
IEEE J. Sel. Areas Commun., vol. 29, no. 3, pp. 545–558, Mar. 2011. in August 1990. From August 1986 to July 1990,
[31] Ns-3. [Online]. Available: http://www.nsnam.org/, Jun. 2014. he was an assistant professor of computer sci-
[32] Video traces for network performance evaluation. [Online]. ence at Iowa State University, Ames, IA. His current research interests
Available: http://trace.eas.asu.edu/, Jun. 2014. include automata theory and formal languages, Petri net theory, graph
[33] Yuv video sequences. [Online]. Available: http://trace.eas.asu. drawing, design and analysis of algorithms, and formal methods.
edu/yuv/, Jun. 2014.
[34] Evalvid—a video quality evaluation tool-set. [Online]. Available:
http://www2.tkn.tu-berlin.de/research/evalvid/fw.html, Jun. Der-Jiunn Deng received the PhD degree in
2014. electrical engineering from the National Taiwan
University, Taiwan, in 2005. He joined the
Che-Yu Chang received the BS degree in indus- National Changhua University of Education,
trial education, and the MS degree in applied Taiwan, as an assistant professor in the Depart-
electronics technology from National Taiwan ment of Computer Science and Information Engi-
Normal University, Taiwan, in 2006 and 2008, neering in August 2005 and then became a full
respectively. He is currently working toward the professor in February 2012. His research inter-
PhD degree in computer science from National ests include multimedia communication, quality-
Taiwan University. He is currently a software of-service, and wireless networks. In 2010, 2011,
engineer at Institute for Information Industry, and 2012, he received the Research Excellency
Taiwan. His research interests include vehicular Award of National Changhua University of Education. In 2012, he also
ad hoc networks and wireless sensor networks. received the Outstanding Faculty Research Award of National Chan-
ghua University of Education. He served or is serving as an editor and
guest editor for several technical journals. He also served or is serving
on several symposium chairs and technical program committees for
IEEE and other international conferences. He is a member of the IEEE.

" For more information on this or any other computing topic,


please visit our Digital Library at www.computer.org/publications/dlib.

You might also like