-
DBASE : A Distributed Bandwidth
Allocation/Sharing/ExtensionProtocol for Multimedia over IEEE
802.11 Ad Hoc Wireless LAN
Shiann-Tsong Sheu” and Tim-Fang Sheu+
*Department of Electrical Engineering, Tamkang University,
Tamsui, Taipei Hsien, Taiwan‘Institute of Communication
Engineering, National Tsing-Hua University, Hsing Chu, Taiwan
[email protected]. tw, [email protected]
Abstract—In ad hoc networks, carrier sense multiple access(CSMA)
is one of the most pervasive medium access control(MAC) schemes for
asynchronous data traffics. However,CSMA could not guarantee the
quality of real-time traffics. Inthis paper, we will propose a
distributed bandwidthallocation/sharing/extension (DBASE) protocol
to supportmultimedia traffics with the characteristics of variable
bit rate(VBR) and constant bit rate (CBR) over ad hoc WLAN.
Overallquality of service (QoS) will be guaranteed in DBASE.
Suchbandwidth allocation procedure is based on a contention
processthat only occurs before the first successful access and
areservation process after the successful contention. If any
real-time station leaves, the reserved bandwidth will be released
byDBASE immediately. The designed DBASE protocol will notonly
allocate sufficient bandwidth for real-time stations but alsopermit
them to extend bandwidth requirements on demand ifthere is any
excess bandwidtb left. Moreover, the proposedDBASE is still
compliant with the IEEE 802.11 standard. In thispaper, the system
capacity of DBASE is analyzed and theperformance of DBASE is
evaluated by simulations. Simulationsshow that the DBASE is able to
provide high channel utilization,low access delay and small delay
variation for real-time services.
1. INTRODUCTION
Since the beginning of the 1990s, wireless local areanetworks
(WLANS) for the 900 MHz, 2.4, and 5 GHzindustrial, scientific, and
medical (ISM) bands have beenavailable based on a range of
proprietary products [1]. As thespeed and capacity of wireless
networks increase, so does thedemand for improved quality of
service (QoS) for real-timemultimedia applications. The IEEE 802.11
WLAN standard[2] includes a basic Distributed Coordination Function
(DCF)and an optional Point Coordination Function (PCF). The DCFuses
Carrier Sense Multiple Access with Collision Avoidance(CSMA/CA) as
the basic channel access protocol to transmitasynchronous data in a
contention period. An attractivefeature of CSMA/CA protocol is that
it is simple toimplement however, the contention-based MAC
protocolcannot provide delay guarantees to real-time traffic due
toeach packet must contend for transmission. To provideperformance
guarantees, some packet-radio networks haveopted for collision-free
schemes based on polling, fixedassignments (using time or frequency
division multiplexing),or reservation. The PCF in the IEEE 802.11
standard is apolling-based protocol. Real-time stations access
channel in around-robin manner during a contention free period.
However,the use of centralized scheme in PCF constrains the
operation
of wireless LANs. Furthermore, several researchers [3-4]pointed
out that such centralized protocol results in a
poorperformance.
Due to these reasons, a distributed multiple accessprotocol for
voice over WLAN was proposed [5-6]. In thisproposed scheme, voice
stations sort their access rights byjamming the channel with pluses
of energy, which is namedBB contention, before sending their voice
packets. Sincevoice packets must be transmitted repeatedly in a
constantinterval, sending bursts of energy for each packet will
wasteconsiderable bandwidth. Moreover, the BB contention is nota
regular scheme defined in IEEE 802.11 standard, and thus itis
difficult to be overlaid on current CSMA implementations.Although
the chaining scheme proposed in [6] is used toenhance the
efficiency of i3B contention, the splitting ofchain, which occurs
while a node ends a session or its packetis corrupted, will also
reduce the efficiency. Furthermore, theasynchronous data might
transmit its packet in the hole ofchain and make the access
instants of real-time packetsstretched. These real-time packets
might be dropped becausethe packet delay exceeds the
delay-bound.
Another protocol supporting real-time multimedia trafficin a
WLAN based on Group Allocation Multiple Access(GAMA) was proposed
[7-9]. In GAMA, owing to at mostone new session could access the
channel and reserve thebandwidth in a cycle, the medium access
delay mightincrease sharply when the load becomes high. Besides,
thereis no differentiation between the priorities of real-time
andasynchronous data traffics. For the prior reserved members,they
have the higher priorities to occupy more bandwidth,which might
make the QoS of others decrease. Thus GAMAis not a fair protocol
and doesn’t consider the overall QoS.Therefore, we propose a new
distributed protocol, which willnot suffer the potential problems
of antecedents, to supportmultimedia traffics over IEEE 802.11 ad
hoc WLAN.
For simplicity, a station with non-real-time (real-time)traffic
is denoted as rev-station (rt-station) throughout thispaper. To
make sure the rt-traffic meets the delay
restrictions,time-sensitive rt-packets always have higher
priorities thanordinary nrt-packets. In the proposed protocol, the
nrt-stations regulate their accesses to the channel according to
thestandard CSMA/CA protocol. On the contrary, the rt-stationswill
reserve the bandwidth and transmit their rt-packetswithout
contention after the first access. At the first access
0-7803-7018-8/01/$10.00 (C) 2001 IEEE IEEE INFOCOM 2001
-
time, a modified CSMA/CA protocol with RTS/CTS controliiames is
used for compatible with IEEE 802.11 standard. Tosupport both
constant bit rate (CBR) service and variable bitrate (VBR) services
over WLAN, the scare bandwidth will beefficiently allocated,
shared, and released by our proposedDBASE protocol. Moreover, the
DBASE also guarantees thequality of service (QoS) by limiting the
maximum size of thereservation repetition interval.
The rest of the paper is organized as following. Section
2describes the DBASE MAC protocol consisting of DCF fornrt-traftlcs
and the distributed bandwidth allocation/sharing/extension protocol
for rt-traffics over WLAN. The systemcapacity of DBASE in WLAN is
analyzed in Section 3.Simulation models and results of the proposed
DBASE arediscussed in Section 4. Finally, Section 5 summarizes
theconclusions.
2. DBASE PROTOCOL
In this section, we will describe the access procedures
fortransmitting nrt-packets and rt-packets separately. In DBASE,we
divide the frames into three priorities as the standard does.The
frames of different priorities have to wait different inter-fiame
spaces (IFSS) before they are transmitted. The shortIFS (SIFS) is
used by immediate control flames, whichalways have the highest
priority, such as clear to send (CTS)and acknowledge (ACK). The
priority IFS (PIFS) is used bythe rt-frames, such as reservation
frame (RF) and the requestto send (RTS) of voice and video packets.
The DCF IFS(DIFS) is the longest IFS and used by the nrt-frames,
whichalways have the lowest priority, such as the data packets.
Theaccess procedure for transmitting asynchronous data packetsis
based on the CSMA/CA protocol and is briefly describedin section
2.1. We control the rt-stations to access themedium by a modified
CSMA/CA protocol, which still canbe overlaid on the IEEE 802.11
standard implementation. Theaccess procedure for rt-sources is
described in section 2.2.
2.I The Access Procedure for Asynchronous Data Stations
The basic access method for nrt-stations is based onconventional
DCF [2]. A nrt-station with a data packetwaiting for transmitting
needs to monitor the channel untilthe medium is determined to be
idle for the duration of aDIFS period and the backoff procedure is
finished. After then,the nrt-station generates a random backoff
time for the datacontention window. The data backoff time (DBT) is
derivedby
DBT = rand(a,b) x Slot_time,
where rand(a, b) returns a pseudo random integer withininterval
[a, b], which b grows exponentially for eachretransmission attempt
and the range of b is ilom predefineb~,. to b~u. By the IEEE 802.11
standard, a is set to O; b~,nand b.m are set as 32 and 1024 for
simplicity. That is
b= b~,. X 2’ ~ brew.
where r is the number of retransmission times. The
Slot_time,which is defined as the time needed for a station to
detect apacket, to accumulate the time needs for the
propagationdelay, the time needed to switch from the receiving
state tothe transmitting state (~_TX_Turnaround_Time), and thetime
to signal to the MAC layer the state of the channel (busydetect
time) [10]. The backoff time counter is decreased aslong as the
channel is idle, and froze while the mediumbecomes busy. The
nrt-station transmits its packet (or RTS ifany) only when its
backoff time counter becomes zero. Whenthe destination receives the
packet correctly, it will transmitan ACK to the source within a
SIFS.
2.2 The Access Procedure for Real-time Stations
In DBASE, the rt-stations contend for the medium by therequest
packets (RTSS) to join the Reservation Table (RSVT)and reserve the
needed bandwidth. The RSVT is a virtualtable built and maintained
by each rt-station respectively, andit records the information of
all rt-stations that have finishedthe reservation procedure
successfully. The informationincludes the access sequence, the MAC
address, the packetlength, the service type and the required
bandwidth of eachrt-station. After a rt-station (STA~~) joins into
the RSVTsuccessfidly, it will not need to contend the medium
anymore during this whole session. In order to maintain thecorrect
access sequence, each rt-station needs to be equippedwith a
sequence ID (SID) register and an active counter (AC).The SID is
used to record the access order among all activert-stations and the
AC is used to record the total number ofactive rt-stations at this
moment.
Now, we will describe how the rt-stations reserve,allocate,
share and extend the bandwidth in the following
foursubsections.
2.2.1 Reservation Procedure
If STA~~ intends to start a session at time t,itwill monitorthe
channel for detecting the Reservation Frame (RF) in theinterval
(t,t+DmJ, where D.u means the smallest maximaltolerance delay among
all active rt-connections. D~a stronglydepends on the
characteristic of rt-service. Parameter D~m isusually a constant
value and predefine in system. The RF isa broadcast ti-ame and used
to announce the beginning ofcontention free period (CFP). The RF is
sent by the first rt-station (SID= 1) and this rt-station has the
right to send its rt-packet first in the CFP. This rt-station which
has theresponsibility to initiate the CFP (i.e. by issuing a RF
flame)periodically is named as contention fkee period
generator(CFPG). We consider that the repetition interval is equal
toD.a because the real-time information delayed for more thanD.a
will cause unacceptable quality and must be discarded.The RF flame
mainly carries the information of the numberof active rt-stations
(AN) in the basic service set (BSS) andthe information of all
rt-stations recorded in the RSVT ofCFPG.
0-7803-7018-8/01/$10.00 (C) 2001 IEEE IEEE INFOCOM 2001
-
A new active rt-station STA~~ will detect the RF fi-ame inthe
interval (t, t+D~J if there is any active rt-station
alreadyreserved the bandwidth. If a RF flame is not received
duringthe interval (t, t+D~J, it means that there is no active
rt-station. Thus, when the channel still perceives idle in the
interval (t+D.w+&, t+D.a+c +PIFS] and none RF frame
isdetected, STA~~ will execute the backo~jscheme. The symbolc
denotes the remaining transmitting time of the current PDU(protocol
data unit) at the time instance t+Dmm. Thecontention process is
still based on the CSMA/CA protocol.The real-time backoff time
(RBT) of a rt-station is defined asfollowing:
RBT = rand(c, d) x Slot_time,
where the function rand(c, o returns a pseudo randomnumber in
the range from c to d. In our scheme, c and d areset to O and 3,
respectively. The RBT time counter isclecreased as long as the
channel is idle, and suspended whilethe medium is sensed busy. If
RBT reaches zero, STA~~willtransmit its RTS to the destination
station to setup aconnection. If no collision occurs, the
rt-station STA~~,whichsends RTS tlame, will receive the CTS ffame
within SIFS,and then STA~~ will play the role of the CFPG in
network.Meanwhile, it sets both its SID and AC to one and
transmitsthe RF frame and its rt-packet right away. In the
proposedDBASE, the RTS/CTS control frames for rt-traffics will
begenerated only before the first successful access. The lengthclf
the first rt-packet of each session is limited as the averagelength
decided by the traffic type of this session.
To prevend nrt-station’s transmissions from disturbing
rt-station’s transmissions, the relationships among three spacesand
two contention windows are shown in Figure 1 and mustsatis& the
following two constrains:
SIFS + Slot_time s PIFS,
PIFS + Slot_time + max{RBT} < DIFS + min{DBT}.
In this paper, the propagation delay can be ignored becausethe
diameter of a Basic Service Area (BSA) is only on thecinderof 100
feet [10].
Contrarily, if collision occurs when transmitting RTS,
theI’-persistent scheme is used to decide whether the
collidedstations insist on accessing channel in the next
Slot_time.Such collision is detected by a rt-station not receiving
its CTSwithin the following duration of SIFS after transmitting
theRTS frame. The collided rt-station will retransmit the RTS inthe
following Slot_time with a probability p. With aprobability q= l-p,
it will defer at least one Slot_time andcontend at the next
real-time contention window. That is, itwill recalculate the RBT by
the following equation
RBTP = rand(c + 1, d) x Slot_time.
This scheme will efficiently reduce the contentionresolving
period. As soon as the CFPG is being generated,
non-real-tlmc
. DIFs___ contention windowH
real-t]me contention (DBT) ‘+::34 g~ b
,/”’ ,,,‘ ,,, ,/L-—E!!??__ Bwkoff ,/ ‘ Busyv b*”
SlOi_tlme
Figure 1. The relations between three interframe spaces
(SIFS,PIFS and DIFS) and two contention windows (DBT and RBT).
other rt-stations will detect the RF and the rt-packet of
theCFPG and then content for the second access position in theRSVT.
If a new active rt-station (STAK~)detects the RF framein the
interval (t,t+D~J, itknows that at least one active rt-station
already reserved the bandwidth. To avoid disturbingthe rt-stations
access channel in the CFP, a new rt-station thatwants to join into
the RSVT by contention must wait until theCFP finishes. The length
of CFP can be calculated by theinformation in RF. During the
waiting period, STA~~monitors the activity of channel. If the
channel idles aSlot time during CFP, it implies a rt-station
disconnectssess~n and the CFP should be decreased. Afler the
CFPfinishes, STA~~ follows the backoff scheme to contend for
itsreservation by sending a RTS. While in the backoff periodRBT,
STA~~ should keep monitoring the channel to checkwhether any
rt-station joins into the RSVT successfidly. Untilthe RBT becomes
zero, STA~~ will send its RTS. If nocollision occurs, the content
of AC will be increased by oneand the SID of STA~~ is set as the
content of AC. At thismoment, it transmits its first rt-packet
right away. Based onthis access procedure, each rt-station will
increase their ACby one as soon as it listens a CTS during the
real-timecontention period. On the contrary, if collision occurs,
thecontention resolution also follows the P-persistent scheme
asmentioned above. To make sure the rt-packets can betransmitted
periodically and the repetition cycle will not belonger than D.a,
we define a parameter RP~,, to limit thecontention period.
Figure 2 demonstrates how to add a real-time session intoa RSVT
by contending the medium for its reservation. Figure2(a) is the
case that no collision occurs and Figure 2(b) showsa case that a
collision occurs. In Figure 2(a), we assume thereare 5 stations
where STA 1 and STA3 want to transmit rt-packets to STA2 and STA4,
respectively. Moreover, STA5has a nrt-packet that is waiting for
transmission. In this case,if no RF is received by STA1 and STA3
after listening to thechannel for D.m, they believes that no
rt-station exists. Then,if STA1 and STA3 detect an idle period of
PIFS after adetecting period D.a, each of them generates a backoff
time(RBT) and starts to count down. We assume that RBTs~Al
issmaller than RBTs~A3.When RBTs~Al counts down to zero,STA1 sends
out a RTS and waits for its CTS. If STA1receives a CTS within SIFS,
there is no collision occurredand STA1 adds into the RSVT
successfully. Because STA1 isthe first active station in the RSVT,
ithas the responsibility to
0-7803-7018-8/01/$10.00 (C) 2001 IEEE IEEE INFOCOM 2001
-
11Ieal.tim.cml!c”tmr, .On.real.thlle
Wi”dnw :;’,’,’Cmtmtio”widow
?)7s SIRS DBS ~w~ I%d%.1.,-payload
I
1A 1 1 ~ ““./.,””’”TS RF .I-pa,kcSTA1 RF Wack t
CY$ ‘laSTA2 @
STA3 r,.pack , “.pmkelT
STA4 TC w @ c
STA5 T nrt.data,,,
(a) without collisionPIPS SIFS PIFS
4 #“’”$\ 4w ,,.pckct RF .I-pmket tb
c-iSTA2 c cl
,,.pack.t ..packtb
STA4 T c
. bm-dehFDm=
(b) with a collision
1%
Figure 2. An example of a new rt-station joining into the
RSVT.
transmit a RF before ita rt-packet. After STA1
finishestransmitting its first rt-packet, STA3 continues to count
down.When RBTs~N counter becomes zero, STA3 sends out itsFLTS and
waits for receiving its CTS. When its CTS isdletected, it means
STA3 has already reserved its bandwidthand joined into the RSVT.
When a nrt-station (STA5) detectschannel being idle for DIFS, it is
clear that no rt-station wantstransmit and STA5 can try to send out
its RTS a soon asDBTsTA5 counter becomes zero.
In Figure 2(b), we only assume STA1 and STA3 want totransmit
rt-packets to STA2 and STA4 at the same timerespectively. In this
case, if RBT~~Alis equal to RBTs~N, acollision will occur because
STA1 and STA3 send out theiramm RTS at the same time. Then
P-persistent scheme will beapplied. When STA1 and STA3 don’t
receive their CTSwithin SIFS, they have the probability p to
persist ontransmitting their RTS in the next Slot_time. Suppose
thatSTA1 gets the right to persist on transmitting but STA3
doesnot, STA1 will retransmit its RTS right away and STA3
willgenerate a new backoff time RBTPsT~. If STA1
successfullyreceive its CTS, it will send out a RF and its
rt-packet. Assoon as STA3 counts down RBTPsAT~to zero, it begins
tosend out its RTS flame just as the previous case. If anycollision
occurs, the rt-stations will repeat the P-persistentscheme under
the constrain that the following contentionprocess will not cross
the RPmboundary.
After art-station STA~~ has joined into the RSVT and thepassing
time from the last access is over the RPm., it willstart to monitor
the radio channel for its contention-freeaccess. If the STA~~ is
the CFPG (SID= 1), it will issue theFWframe as soon as the channel
is detected idle for a PIFS.
3 4 4
Owation control(101) ~
1 ‘ 1 Aibi’)NOEF
..~........ . ---------...............
MACHeader ‘m yg~ A&bess 1 Adress 2 Adress 3 s:”#;
Adress40m301 1
Figure 3. The duration field formatofanIEEE802.11 MPDU.
Also it will be the first one to transmit rt-packet. On the
otherhand, as soon as the other STA~~ receives the RF,
STARTwillupdate its RSVT by the broadcast information in the RF
andthe access instant of each session will be decided. When
thechannel is idle for a Slot_time, we assume that a rt-stationwith
AID (i.e. SID = AID), which should deliver its packet atthis
moment, has stopped transmitting. Each following stationwith larger
SID will shift forward its access sequence in theRSVT. Due to the
characteristic of the multimedia traffic, it isreasonable to
release the reserved bandwidth when a stationtears down a
rt-session. If the channel is still idle for the nextSlot_time, the
release step will be repeated. After each stationfinishes sending
its packet in the current cycle, it still keepsmonitoring the
channel to check whether any session behindit tears down or any new
session succeeds to add into theRSVT.
2.2.2 Allocation Procedure
To filly utilize the resource in a distributed system,
theinformation of RSVT can be got and updated by the RF frameand by
checking the duration field of MAC header in eachMAC PDU (MPDU).
The duration format in a MAC headeris shown in Figure 3. This
duration field consists of five sub-fields: control field, type
field, next degree (ND) field,extension flag (EF) and raise degree
(RD) field. The firstthree bits in the duration field of a packet
is control field. Weassume that the control field with ‘101‘
indicates that thispacket is a rt-packet. The value ‘101’ has been
reserved bystandard The 4-bit type field indicates what type of the
rt-packet is. (e.g. voice, video, MPEG bit stream, etc.)
Thesetraffic types are predefine in system. Each session
utilizesthe ND field to inform other stations the demandedbandwidth
at the next cycle. The request is decided by theamount of buffered
packets in each station. Since ND fieldonly occupies 4 bits, the
required bandwidth is represented as16 degrees for each service
type. To reduce the overhead ofdescribing the length of required
bandwidth, let u(i) denotethe unit bandwidth of the multimedia type
i of a session. Thenthe required bandwidth can be described by
identifying howmany bandwidth unit u(i) a station needs. The u(i)
can beobtained by considering two characteristic parameters:
PBR(i)(peak bit rate) and MBR(i) (minimal bit rate) of
multimediatype i. That is, we have
[
PBR(i) – MBR(i) D~=u(i) = x
16 1UST x CDR “
0-7803-7018-8/01/$10.00 (C) 2001 IEEE IEEE INFOCOM 2001
-
where UST is the unit slot time and CDR is the channel datarate.
In this equation, a minimal amount of bandwidth isreserved for each
session in each cycle.
Let MS(i) and ND(i) respectively denote the minimumnumber of UST
needed for an active session in every cycleamd the required
bandwidth degree (excluding the minimumguaranteed bandwidth MY(i))
of a session at thewith service type i. Thus, we have
next cycle
‘s(i)=MBR(i)x[us:::DR1il
QLAD(i) = —
u(i) ‘
where QL denotes the queue length of each rt-session whichis
measured in UST. The derived ND will be no mare than 16.As long as
art-station has the right to access the channel, theND is
calculated immediately according to the queue lengthin its buffer.
Based on this concept, in proposed DBASE,after rt-stations update
their RSVTS, each station will beaware of the reserved bandwidth
for each session in everyCFP. Now we will describe how to
dynamicallyaJlocate/share/extend bandwidth for each session
withclifferent bandwidth requirements in WLAN.
To fairly utilize the resources, if the demand of each
rt-station for bandwidth in the next cycle does not excess
itsawerage bandwidth requirement (AVD), the demandedbandwidth will
be allocated. Let AVD(i) denote the averageclegree of a multimedia
type i. The A VD(i) can be obtainedfrom the ABR(i) (average bit
rate) of multimedia type i. Thatis,
~ ~(j)= (ABR(i) - A4BR(i)) x ~m
u(i) x UST x CDR
If the requested bandwidth is larger than A VD(i), only
theaverage bandwidth quota of its multimedia type i will be
firstallocated. Therefore, the maximal bandwidth reserved for
allactive rt-sessions in every CFP is the sum of A VDS of allactive
rt-sessions, and simply described as the CFPmX.If theCFP~U is
larger than the actually required bandwidth, there isresidual
bandwidth that can be shared by some overloaded rt-stations. In the
next subsection, we will describe how residualbandwidth is
shared.
Because the interference exists in the wirelessenvironment and
the delay bound for the rt-traffics is limited,an efficient
retransmission scheme should be designed toimprove the quality of
rt-traffics and the stability of aclistributed system. In DBASE
protocol, after passing CFP,the CFPG will broadcast the
retransmission mapping (RTM)frame. The RTM is a bit mapping to
inform all rt-stationswhich stations can retransmit their packets.
For example,RTM is “001”, it means that the rt-station whose SID =
3 canretransmit its r-t-packet after RTM frame when AC = 3. The
length of the retransmitted packet is limited as the
negotiatedaverage packet length (AVD). If the retransmission
succeeds,every rt-stations monitoring the channel can detect the
NDfield of the retransmitted packet and this information will
berecorded into each RSVT, Otherwise, the ND field for
thisretransmitted station in the RSVT will be set as AVD(j) if
themultimedia type of the retransmitted station is j. During
theretransmission scheme, the rt-stations still need to
checkwhether the time instant is over the boundary of RPmXor not.If
the time exceeds the boundary, the retransmission schemewill be
terminated immediately.
2.2.3 Sharing Scheme
Before reallocating residual bandwidth for the
overloadedrt-stations, each active rt-station will first accumulate
thespare bandwidth from those sessions whose requests are lessthan
A VD. Let CD be the required bandwidth degree of thecurrent cycle,
which is copied horn the ND that has beeninformed in the previous
cycle. For simplicity to demonstratethe following equations, we
omit the service type parameterin notations A VD, CD, MS and u. Let
SS denote the numberof USTS that can be shared by those rt-stations
whosedemanded degree excesses its A VD. (The SS can be treated
asthe residual bandwidth in a contention free cycle.) Therefore,we
have
SS = f(A VDk – min(A VDk, CDk )) x ui ,k=l
To fairly share the residual bandwidth among
over-loadedsessions, the proportional approach is used. That is,
the exactnumber of reserved unit slot (RS) for sessionj is
RSJ =
1(CD] - A VDj ) x U] 1 XSS+AVDIXU1~(CDk -A VDk)
xukCDk>AVDk~MS + 2x SIFS+ TACK
UST, where CD] > A VD1
CD, X1.t, +MS+ 2X S1:;TACK
1 , where CD] < A VD1where TACKis the time needed for
transmitting the ACK frame.According to the above equation and the
RSVT in each activert-station, the packet length of each session
can be calculatedby each rt-station individually. Consequently, we
note that theaccess instant of each reserved real-time session and
thelength of reserved period can be got easily.
2.2.4 Extension Procedure
In the case that the burst traffic arrives just after a
rt-station issued the last bandwidth reservation information,
thedelay bound for the excess data may be violated. To solve
0-7803-7018-8/01/$10.00 (C) 2001 IEEE IEEE INFOCOM 2001
-
[~ undetectaipacket SIFS
gzEEEg smF& ~F,pp \ .’ 1,:’ ;,/-’
? c RT B TTD
ND-s ND+ I-ID+ ND=5
RPmCCYC16Dm‘4$ JJAsp_E~:- -~’+-’
Figure 5. Superfiame structure.(a) Interferenceoccurs in
CFP.
Becauseofcolhsiono- m rcbanwnissm7 proposed DBASE.
SIFSy +>,
Duicd,themkseta$ -~IFS PIFS AVE
‘Due to the characteristic of time-sensitive for
rt-services,!1
1 R ;:;,%A ~~. C R ?\#,;, T T D w~hno~ ~ ~ ~ ~ ~. . ..
ND-5 ND-5 ND=5 I ‘AVDCD=5 CD=5RP_
-
bit rate (Mbps) bit rate (Mb}s) .MaximumBit Rate
‘/c 3 m
44~,time(ms)
(exponentialdistributed)
stateholdongtime(exponentialdistributed)
Figure 6. The source model for VBR.
Table 1. Numerical values for the CBR model.
Table 2. Numerical value for the VBR model.
Parameter ~Maximum Bit Rate 420 KbpsMinimum Bit Rate 120
KbpsMean Bit Rate 240 KbpsMean State Holding Time 160 msecMean
Video Transmission Time 180 secMaximum Packet Delay 75 ms
Table 3. The characteristics of two trat%c types for
simulations.
mDEEllma3EKlmmmmlmCBR I 64 Kbps ] 64 Kbps I 64 Kbps I 15 0 0VBR
1420 Kbps1240 Kbpsl 120 Kbpsl 28 6 I 5
Parameter L.-X&-JConversation Length 180 sec
Table 4. System parameters of simulations.
I Parameter l~mChannel data rate CDR ] 11 MbpsMaximum delay D
.QI 25 msSlot time Slot time 20 usUnit slot time UST [0 us
In the first subsection, the traffic models used in
thesimulations are introduced. In the second subsection,
theperformance measurements are defined. In the last subsection,the
simulation results are presented and some discussions aremade. in
order to investigate the multimedia over WLAN, weconsider the WLAN
with 1lMbps, which WLAN adapter hasbeen announced recently. Each
simulation run sustains 3x106Slot_times.
4.1 Traffic Models
In order to evaluate the performance of the proposedprotocol,
three different traffic models are considered:
Voice Trafjc Model (CBR): The voice traffic is usuallyconsidered
as a service with constant bit rate (CBR) traffic. Inthe
simulations, the voice traffic is modeled as a two-stateMarkov
process with talkspurt and silent-gap states. Eachvoice source is
assumed to equip a slow speech activitydetector (SAD) [11-12]. In
the talkspurt state, we considerthat the voice source generates a
continuous bit-stream. Onthe other hand, in the silent state, there
is no packet to begenerated. The duration of talkspurt and
silent-gap bothfollow exponential distribution with the mean
duration equalto 1 and 1.35 seconds respectively. A voice packet is
assumedto be dropped if it suffers a delay longer than D.a
(=25ms).We assume that the data rate of the voice traffic is
64Kbps.The parameters of voice model are summarized in Table 1.
Variable-bit-rate J4deo Traf@ Model (VBR): This modelis a
multiple-state model (shown in Figure 6) where a stategenerates a
continuous bit stream for a certain holdingduration [13-14]. The
bit rate values for different states areobtained from a truncated
exponential distribution. Thisdistribution is defined with a
minimum and a maximum bit
Short interfkame space SIFS 10 usPriority interframe space PIFS
30 usDCF interframe space DIFS 110 usMAC header HMAC 272 bitsPHY
header H 128 bitsThe packet length of RTS T,,~::DR 288 bitsThe
packet length of CTS TC,,XCDR 240 bitsThe packet length of ACK
T.C.XCDR 240 bitsBasic backoff window size of nrt-sessions ~Lnb
32Maximum backoff window size of nrt- b
1024sessionsPersistent probability P. 0.8
rate values. The holding times of the states are assumed to
bestatistically independent and exponential distributed. Weassume
that each state has the same mean holding time. Inour simulations,
we considered 16 levels to approach the bitrates. Table 2
summarizes the numerical values used for theVBR model. Table 3
shows the characteristics of two traffictypes for simulations.
Data Traffic Model: We assume that data packets arrive ateach
station following the Poisson process with the meanvalue
Z=(W’/iV’~, where L means the arrival rate per nrt-station. The
total data load p can be estimated as~ = anrl x EIT~~~lOa~], where
A “r~ is the mean arrival rate of
total asynchronous packets . Through the simulations, the
EITJ~-lOaJ] is set as 744 ~s (=8184 bits/CDR), A is 0.1,
buffer size is fixed at 100 packets and the length
ofasynchronous packet is fixed at 8184 bits.
According to the traffic models defined above, the
trafficparameters of DBASE can be summarized in Table 4.
0-7803-7018-8/01/$10.00 (C) 2001 IEEE IEEE INFOCOM 2001
-
1 I I
0.8 —. .— —
‘-’’’-’n
+ VBR=IOla VBR=20
g 0,4 a VBR=3 o
x VBR=400.2
o~0,1 0.3 0.5 0.7 0.9 1.1 1.3 1.5
Dataload
(a) VBR real-timeservices
0.s r— ,
0.6
Pfl ‘
, .y~- ‘ ‘ ~*-
m
+ CBR=50
El E CBR=l 00
$ 0.4 ,*’ a CBR=150
c? + CBR=200- 0.2 y II - CBR=250 ]o~ - CBR=300
0.1 0.3 0.5 0.7 0,9 1.1 1.3 1.5Data load
(b) CBR real-timeservices
Figure 7. A comparison of the goodput, as the number of
real-time sessions varies. Show two traffic types respectively.
4.2 Performance Measurements
The performance measurements used in our analysis aredefined as
follows:l Goodput: the goodput is defined as the percentage of
the
time used by both rt- and nrt-stations to successfullytransmit
their pure payload data. The control andmanagement signals, the
idle time, the collided packets,the error packet caused by the
interference and theheader bits are excluded from the goodput.
l Packet delay dropped probability (PDDP): the packetdelay
dropped probability is defined as the flaction ofdiscarded packets
due to the delay-sensitivity of the rt-traffics.
4.3 Simulation Results
To obvious the performance of DBASE, we change thenumber of
rt-stations and the traffic types. Figures 7 and 8plot the goodput
and PDDP of CBR and VBR trafficsrespectively. The number of
nrt-stations N“” is fixed at 10 andthe asynchronous data load (p)
increases fi-om 0.1 to 1.5. Thenumber of VBR sessions ranges from
10 to 40, and thenumber of CBR sessions ranges from 50 to 300.
Figure 7 shows the relations of the goodput and the dataload
under different number of rt-stations. In Figure 7, wecan find that
the goodput can be up to about 80% for VBRtraftlcs and 67’%for CBR
traffics. We know that the overheadcaused by headers of packets
will be quite obvious when thepayloads of packets are small.
Therefore, the saturatedgoodput of VBR is higher than that of CBR
since the packetlength of VBR (average length= 58 UST) is longer
than that
0.01tOa%assssaas S@ @as
0.1 0.3 0.5 0.7 0.9 1.1 1.3 1.5Data load
(a) VBR real-time services
‘:G
0.1 0.3 0.5 0.7 0.9 1,1 1.3 1.5DataIoad
(b) CBR real-time services
.+ CBR=50
w CBR=I 00
A
a cBR=150
.Y CBR=200
+ CBR=250
l CBR-300
Figure 8. A comparison of the packet delay dropped
probability(PDDP) of real-time stations, as the number of real-time
sessionsvaries.
of CBR (average length = 15 UST). Because the number ofdata
stations is fixed at 10, the increasing data load will onlyincrease
the queue length of the data buffer but not thecollision
probability when the traffic load is saturated.Accordingly, the
goodput will not decrease after the goodputis saturated. In Figure
7(b), because the payload length of nrt-session is much larger than
that of CBR sessions, the curveswith different number of CBR
sessions cross each other.While the number of CBR stations
decreases, the number ofnrt-packets transmitted successfully will
become more.Therefore the saturated goodput with 50 CBR sessions
ishigher than others when the asynchronous data load is heavy.
Figure 8 plots the PDDP of CBR and VBR trafficsrespectively.
According to the Section 3, we can estimate thesaturated capacity
of the system by those simulationparameters. If the traffic type is
CBR, the saturated capacityis about 112 CBR rt-sessions. Similarly,
the saturatedcapacity for the VBR traffic model is about 39 VBR
rt-sessions. In Figure 8(a), since 40 VBR sessions are just alittle
over the saturated number of stations, the PDDP is notzero but it
is still very small. The derived PDDP is about zerofor other cases.
Because the reservation process is applied,packets will be dropped
only when the rt-sessions are still inthe contention procedure.
Therefore, the PDDP will notincrease rapidly even though data load
is still increasing andthe rt-sessions are over the saturated
capacity (39). In Figure8(b), we find that the PDDP is not over 3%
even when thenumber of CBR sessions is up to 200, which is over
thesaturated active sessions of CBR traffics. This is because
theON-OFF model is used in the CBR traffic model.
0-7803-7018-8/01/$10.00 (C) 2001 IEEE IEEE INFOCOM 2001
-
1 I I
()~20 60 100 140 180 220 260 300
Numberof red-time sessions
Figure 9. A comparison of the goodput of DBASE and DCF fora
variety of traffic types in the clear radio environment.
In papers [15-16], authors concluded that the packetizedvoice
communications can tolerate only a small amount(1%-3%) of dropped
packets before suffering a large qualitydegradation. Based on these
criteria, in Figure 8(b), themaximum number of CBR stations can be
up to 200. We findthat the asynchronous data load does not affect
the PDDPbecause the rt-traffic has a higher priority than the
nrt-traffic.The low PDDP of DBASE implies that DBASE can performa
high quality of service even when the traffic load is heavy.
To compare our proposed protocol DBASE with the IEEE802.11 MAC
protocol based on pure DCF, the followingsimulations are made. For
simplicity, the conventionalprotocol based on IEEE 802.11 DCF is
named as DCF. InFigure 9, we assume that the number of nrt-sessions
is stillfixed at 10 and the A is 0.1. The curves DCF(CBR)
andDBASE(CBR) indicate the performance of DCF and DBASEfollowing
the CBR rt-traflic model respectively and so as theDCF(VBR) and
DBASE(VBR) following VBR rt-trafficmodel. For the CBR model, when
the active number of rt-sessions is low, the DBASE performs similar
to DCF.However, as the number of CBR sessions is larger than
100,the DBASE performs much better than DCF. This is becauseDBASE
will reserve the bandwidth and the increasingcontentions caused by
the increasing number of rt-sessionswill not dominate the channel
resource. For the VBR model,the DBASE also performs much better
than DCF because ofthe same reasons mentioned above. We emphasize
that thederived goodput of DBASE(VBR) can get up to 90%.
Thisimplies that almost network resources are fblly utilized
byDBASE protocol.
To show the effect of the interference in the
wirelessenvironment on the DBASE and DCF, we set the packet
errorrate (PER) as 0.1. Meanwhile, we assume that the number
ofnrt-session is zero. In Figure 10(a), the curves show that
thegoodput of DBASE is higher than that of DCF under thesame
traffic types. We know that a higher PER will result in alower
goodput. However, the goodput of DBASE(VBR) canalso hold on 80’%
even though the PER is up to 0.1 but thegoodput of DCF(VBR)
decreases to about 30%. In Figure10(b), we find that the PDDP of
DCF is much higher than
o~ I
20 60 100 140 180 220 260 300
Number of real-timesessions
(a) The goodput of DBASE and DCF
b DBASE(CBR)
M DBASE(VBR)
a DCF(CBR)
x DCF(VBR)
20 60 100 140 180 220 260 300
Numberof real-timesessions
(b) The PDDP of DBASE and DCF
Figure 10. A comparison of goodput and PDDP of DBASE andDCF in
the interfering environment (PER= 0.1).
that of DBASE under the same traffic types. Figure 10(b)show
that the PDDP of DBASE(CBR) will begin to increasewhen the number
of CBR sessions is larger than 220, whichis the saturated capacity
under the consideration of ON-OFFmodel and PER=O. 1. However the
PDDP of DCF(CBR) willincrease rapidly as long as the number of CBR
sessions islarger than 100. Similarly, the PDDP of
DBASE(VBR)increases at about 40 VBR sessions and PDDP of DCF(VBR)is
up to about 20’%while the number of VBR sessions is only20. This is
because the rt-trafics are delay-sensitive but thePER and
contentions make the packet delay probability ofDCF increase
extremely. Since the bandwidth is reserved forthe rt-packets to
retransmit in DBASE, the packet error ratedoes not influence the
PDDP of DBASE seriously when thenumber of rt-sessions is small.
Although DBASE has theretransmission scheme to reduce PDDP of the
reserved rt-sessions, the new rt-sessions using CSMAICA to
reservebandwidth still makes the PDDP increase as the rt-traffic
loadincreases.
5. CONCLUSIONS
In this paper, we proposed a distributed
bandwidthallocation/sharing/extension (DBASE) protocol to
supportmultimedia traffic over ad hoc WLAN. In DBASE, rt-stations
can reserve and free channel resources dynamically.The system
capacity of proposed DBASE was analyzed.Simulation results show
that the proposed protocol cansupport both constant bit rate and
multimedia services of
0-7803-7018-8/01/$10.00 (C) 2001 IEEE IEEE INFOCOM 2001
-
variable bit rate in ad hoc WLAN. Simulation results also
[12]demonstrate that the DBASE performs very well and muchbetter
than the conventional IEEE 802.11 standard. Thechannel efficiency
of DBASE is up to about 90°/0 for [13]
supporting variable bit rate (VBR) traffics. Besides, the
delaydropped probability of rt-packets, caused by the
delaylimitation and noise interference, is very low even though
the
total traffic load is heavy. [14]
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
Richard van Nee, Masabiro Morikura, Hitoshi Takanashi,Mark
Webster, Karen W. Halford, “New High-Rate
[15]
Wireless LAN Standards”, IEEE CommunicationsMagazine, Volume:
3712, pp. 82-88, Dec. 1999.IEEE 802.11, “Part 11: Wireless LAN
Medium AccessControl (MAC) and Physical Layer (PHY)
Specifications”,IEEE 802.11 Std, Aug. 1999.
[16]
W.K. Kuo, C.Y. Chan and K.C. Chen, “Time BoundedServices and
Mobility Management in IEEE 802.11Wireless LANs”, Proceedings of
the 1997 IEEEInternational Conference on Personal
WirelessCommunications, pp. 157-161, 1997.Jaime Sanchez, Ralph
Martinez, and Michael W. Marcellin,“A Survey of MAC Protocols
Proposed for Wireless ATM”,IEEE Network, vol.11 6, pp. 52-62,
Nov.-Dee. 1997.J. L. Sobrinho and A. S. Krishnakumar,
“DistributedMultiple Access Procedures to Provide
VoiceCommunications over IEEE 802.11 Wireless Networks”,Proceedings
of the IEEE GLOBECOM’96, vol. 3, pp. 1689-1694, 1996.Joao L.
Sobrinho and A. S. Krishnakumar, “Quality-of-Service in Ad Hoc
Carrier Sense Multiple Access WirelessNetworks”, IEEE Journal on
Selected Areas inCommunications, vol. 178, pp. 1353-1368, Aug.
1999.Andrew Muir and J.J Garcia-Luna-Aceves, “SupportingReal-Time
Multimedia Traffic in a Wireless LAN”,Proceedings of the SPIE
Multimedia Computing andNetworking, 1997.Andrew Muir and J.J
Garcia-Luna-Aceves, “GroupAllocation Multiple Access with Collision
Detection”,Proceedings of the INFOCOM’97, Kobe, Japan, pp.
1182-1190, April 1997.Andrew Muir and J.J Garcia-Luna-Aceves,
“GroupAllocation Multipke Access in Single-Channel WirelessLANs”,
Proceedings of the Communication Networks andDistributed Systems
Modeling and Simulation Conference,Phoenix, Arizona, 1997.F. Cali,
M. Conti, E. Gregon, “IEEE 802.11 Wireless LAN:Capacity Analysis
and Protocol Enhancement”,Proceedings of the IEEE INFOCOM’98, San
Francisco,USA, pp. 142-149, APril 1998.Antonio Iera, Antonino
Modafferi and Antonella Molinaro,“Access Control and Handoff
Management in Multi-tierMultimedia Wireless Systems”, IEEE
WirelessCommunications and Networking Conference WCNC’99,vol. 3,
pp. 1518-1522, 1999.
Chunhung Richard Lin and Mario Gerla, “AsynchronousMultimedia
Multihop Wireless Networks”, Proceedings ofthe IEEE IiVFOCOM ’97,
vol. 1, pp. 118-125, 19997.Ian F. Akyildiz, David A. Levine, and
Inwhee Joe, “ASlotted CDMA Protocol with BER Scheduling for
WirelessMultimedia Networks”, IEEE/A CM Transactions onNetworking,
vol. 7, no. 2, pp. 146-158, April 1999.M. Decina and T. Toniatti,
“Bandwidth Allocation andSelective Discarding for a Variable Bit
Rate Video andBurst Data Cells in ATM Networks”, ht. J. Digital
AnalogCommunication System, vol. 5, no. 2, pp. 85-96,
Apr.-June1992.H. Heffes and D. M. Lucantoni, “A Markov
ModulatedCharacterization of Packetized Voice and Data Traffic
andRelated Statistical Multiplexer Performance,” IEEEJournal on
Selected Areas in Communications, vol. SAC-4,no. 6, pp. 856-868,
Sep. 1986.K. Snrarn and W. Whitt, “Characterizing
SuperpositionArrival Processes in Packet Multiplexer for Voice
anddata,” IEEE Journal on Selected Areas in Communications,vol.
SAC-4, no. 6, pp. 833-846, Sep. 1986.
0-7803-7018-8/01/$10.00 (C) 2001 IEEE IEEE INFOCOM 2001