Adaptive Channel Switching for Centralized MAC Protocols in Multi-hop Wireless Networks

8/7/2019 Adaptive Channel Switching for Centralized MAC Protocols in Multi-hop Wireless Networks

1/7

228 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 58, NO. 1, JANUARY 2010

Adaptive Channel Switching for Centralized MACProtocols in Multihop Wireless Networks

Sheng-Tzong Cheng, Ming-Hung Tao, and Chun-Yen Wang

AbstractThe centralized medium access control (centralizedMAC) protocol, which utilizes an Access Point (AP) to coordinatestations in wireless networks, guarantees the quality of service(QoS) for transmission, and provides good throughput in thehighly congested network. However, the centralized MAC proto-col cannot be adopted in ad hoc networks, because it providesonly single-hop transmission. This paper proposes a multihopmechanism for centralized MAC protocols to operate on variousnetwork topologies. The proposed mechanism provides excellentthroughput for both inter-subnet and intra-subnet links, andalleviates the hidden terminal problem. Experimental resultsreveal that the optimal configuration on the proposed mecha-nism and the comparison between our mechanism and other

multihop forwarding mechanisms. The results demonstrate thatthe proposed mechanism outperforms other multihop forwardingmechanisms in terms of throughput.

Index TermsCentralized MAC, quality-of-service, multihopmechanism, network topologies, hidden terminal problem.

I. INTRODUCTION

CONVENTIONAL MAC protocols in wireless networks

are categorized as distributed MAC protocols or central-

ized (infrastructure based) MAC protocols. Distributed MAC

protocols, such as Distributed Coordination Function (DCF) inIEEE 802.11 [1], realize the multihop wireless networks with

good mobility, but do not guarantee the QoS of a time-boundedservice, since they adopt the contention-based access scheme.

Conversely, centralized MAC protocols, such as Point Coordi-

nation Function (PCF) in IEEE 802.11 [1], arrange the trans-mission in single hop wireless networks to achieve the QoS

requirements of the time-bounded service. Centralized MAC

protocols have a higher system throughput than distributed

MAC protocols, but do not supply the multihop transmission;

the origin of this restriction is the hidden terminal problem

resulting from the discordance between subnets.

An efficient method for centralized MAC protocols to

solve the hidden terminal problem between subnets provides

multiple channels or frequencies for transmission. However,most investigations on multi-channel systems focus on eitherthe multi-channel design for distributed ad-hoc networks [2]

[3] [4] [5] [6] [7] [8] or optimizing the system utilization

Paper approved by M. Zorzi, the Editor for Wireless Multiple Access of theIEEE Communications Society. Manuscript received April 19, 2005; revisedJanuary 11, 2007.

S.-T. Cheng is with the Department of Computer Science and Informa-tion Engineering, National Cheng-Kung University, Tainan, Taiwan (e-mail:[email protected]).

M.-H. Tao is with the SoC Technology Center, Industrial TechnologyResearch Institute, Hsinchu, Taiwan (e-mail: [email protected]).

C.-Y. Wang is with the Information and Communications Research Lab,Industrial Technology Research Institute, Hsinchu, Taiwan.

Digital Object Identifier 10.1109/TCOMM.2010.01.070054

by parallel transmission using multiple channels in a subnet

[9] [10] [11]. Only a few investigations on multi-channelsystems concern the multi-channel architecture for multihop

transmission in centralized wireless networks [12] [13]. K.

Mizuno et al. provides a feasible solution to the hidden ter-

minal problem and developed a multihop relaying scheme for

centralized MAC protocols [12]. It also realizes an end-to-end

QoS guarantee for the time-bounded service. However, in their

protocol, initializing a WLAN is complex, and registering a

new wireless terminal (WT) is time-consuming (chain topolo-

gies are preferred in this protocol). Moreover, a large number

of channels is required to achieve good throughput for thesystem, implying that a WT needs many transceivers. Thus,the implementation cost of this protocol, and the power con-

sumption of each WT are both high. The multiple-frequency

forwarding mechanism proposed by J. Peetz [13] eliminates

the restriction of the one hop configuration in HiperLAN/2 by

employing Multiple Frequency Forwarder Wireless Terminals

(MF-WTs). An MF-WT needs to be located in an area with

two or more overlapping subnets, and join these overlapping

subnets asynchronously by switching the frequency between

these subnets. The inter-subnet links are therefore created and

the multihop functionality is achieved through these inter-

subnet links. However, because the MF-WT must be locatedin an area with overlapping subnets, many network topologies

have no MF-WT available. Additionally, since the throughputof the inter-subnet link depends on the number of MF-WTs

and the synchronization between these MF-WTs, the inter-

subnet links cannot easily have high throughput.

This work proposes a multihop mechanism named adaptive

channel switching (ACS) for centralized MAC protocols. The

ACS mechanism efficiently utilizes the bandwidth by avoiding

channel divisions between subnets for centralized protocols. It

enables multihop transmission across subnets, and alleviates

the hidden terminal problem using a 3-channel architecture.

Furthermore, ACS can be adapted to various network topolo-gies without complicated initialization procedures or synchro-

nization between subnets.

I I . RELATED WORKS

This section briefly summarizes a representative centralized

MAC protocol, the IEEE 802.11 PCF, and also introduces two

multi-channel mechanisms that provide the centralized MAC

protocol the ability of multihop transmission.

A. IEEE 802.11 PCF

To support time-bounded services, the IEEE 802.11 stan-

dard defines the Point Coordination Function to give stations

0090-6778/10$25.00 c 2010 IEEE

Authorized licensed use limited to: Amal Jyothi College of Engineering. Downloaded on July 07,2010 at 12:58:37 UTC from IEEE Xplore. Restrictions apply.


2/7

CHENG et al.: ADAPTIVE C HANNEL SWITC HING FOR C ENTR ALIZED M AC PROTOC OLS IN M ULTIHOP W IR ELESS NETWORKS 229

priority access to the wireless medium, coordinated by a

station called the Point Coordinator (PC). The time in the

PCF is divided into repeated periods, called superframes. WithPCF, a Contention Free Period (CFP) and a Contention Period

(CP) alternate over time, in which a CFP and the following

CP form a superframe. During the CFP, no contention occurs

between stations, since the stations are polled by the PC.

If the PC receives no response from a polled station after

waiting for PCF Interframe Space (PIFS), then it either pollsthe next station, or ends the CFP. The PC continues polling

other stations until the CFP expires. A specific control frame,

called CF-End, is sent by the PC as the last frame within the

CFP to signal the end of the CFP.

B. The KMH Mechanism

The KMH mechanism was proposed by K. Mizuno et al.[12], and is named after the authors. The KMH mechanism

adopts a PCF based polling scheme in a multihop wireless

network with multiple channels, where the PC and WT each

utilizes two or more transceivers. The station in the KMH

mechanism has three modes for each channel: master mode,

slave mode, and silent mode. In master mode, the station

acts as a PC in the channel. In slave mode, the station

acts as a WT. In silent mode, the station sets NAV, and is

not permitted to send packets during the CFP. After NAV

resetting, a station in silent mode can send packets until

the next beacon frame is sent. In this manner, three modesfor each channel enable communication by all wireless links

using PCF, and offer QoS guarantees from end-to-end. KMH

addresses some issues, such as associating a new station to a

network and guaranteeing the QoS, which are not described

for considerations of space.

C. The Multi-frequency Forwarding Mechanism

The multi-frequency forwarding mechanism was proposed

by J. Peetz [13]. This mechanism enables inter-subnet links,

and extends the one-hop connectivity to a multihop ad hoc

connectivity for the HiperLAN/2 standards [14]. Each subnet

in HiperLAN/2 determines its operation frequency channel

based on interference minimization based on the Dynamic

Frequency Selection (DFS). Figure1 shows an example fora corresponding multihop network configuration consisting

of two interconnected subnets. Both MF-WT1 and MF-WT2

are within the coverage range of Central Controller (CC) 1and 2, where the MF-WT is the WT with the forwarding

functionality. Therefore, increasing the number of MF-WT

capable terminals increases the number of stable inter-subnet

links.

With the MT Absence function, the H/2 RLC standard

enables WT to withdraw from communication. The WT

transmits the message RLC MT ABSENCE to inform the

CC that it is unavailable for a time interval of 0 63 MAC frames. When the CC re-sponds with RLC MT ABSENCE ACK, the WT changes

to the absent state, and the absence timer is started. The

communication between WT and CC is continued immediatelyas soon as the absence timer expires. MT Absence is applied

for the novel interconnection concept to facilitate WTs to hold

(a) (b) (c)

(c) (a)(b)

(a) RLC_MT_ABSENCE (b) RLC_MT_ABSENCE_ACK (c) RLC_MT_ALIVE

CC1

CC2

MF-WT1

MF-WT2

Fig. 1. MF-WT operation in subnet 1 and 2.

connection to more than one CC. The aim of the MT Alive

procedure is to check whether a CC and WT can communicatewith each other. The MF Alive function may be used to

indicate the presence of an MF-WT to the CC by sending anRLC MT ALIVE message after switching and synchronizing

to the new frequency channel.

The multi-frequency forwarding mechanism is founded on

an intermittent presence of forwarding WTs at each subnet

to be interconnected. Therefore, the MF-WT periodically

withdraws from a current transmission for a certain number of

MAC frames by using the RLC functionsMT Absence and MT Alive. Figure 1 shows the operationof two MF-WTs successfully associated with the CCs of

two subnets. Assume that an MF-WT is alternating between

CC1 and CC2. To leave the current CC, for example CC1,

it sends the RLC MT ABSENCE message containing the

parameter. When the MF-WT receives theacknowledgement from CC1, the radio connection to CC1 is

intermitted, and the absence period counter is started from

the following MAC frame. The Broadcast Channel (BCH)

transmitted by CC2 then has to be detected and decoded by

the MF-WT for synchronization.

III. THE ADAPTIVE CHANNEL SWITCHING MECHANISM

This section proposes a multihop mechanism called adaptive

channel switching for centralized MAC protocols. The ACS

mechanism has the following features: 1) it avoids channel

divisions between subnets, allowing the system to use the

bandwidth efficiently; 2) it enables the multihop transmissionacross subnets, and alleviates the hidden terminal problem;

3) it eliminates the need for complex initialization and syn-chronization between subnets; 4) it can be adapted to various

network topologies, and 5) it uses the smallest possible number

of transceivers to realize these goals.

Since this work describes the ACS mechanism based on

enhancing the IEEE 802.11 PCF, the following description

uses the term PC instead of AP, and uses the term WT

to indicate a non-PC station. The ACS mechanism divides

the total bandwidth into three channels, namely the Control

Channel (C-channel), the Data Channel (D-channel), and the

Relay Channel (R-channel). C-channel is adopted for the



3/7


PC1 PC2WT1 WT2 WT3

CPF CP

D1

poll

D1

poll

U1

ack

U1

ack

D2+ack

+poll

D2+ack

+poll

ack

ack

Beacon

frame

Jamming

packets CF-End

RTS

RTS

CTS

CTS

Data

Data

ack

ack

R-mode

Time

Time

C-channel packet D-channel packet R-channel packet

WT1

PC1

WT2

WT3

PC2

Fig. 2. An example illustrates the packet exchange in the ACS mechanism.

exchange of control signals such as beacon frames and CF-

End; D-channel is adopted for the transmission of data packets,

which occupies the most system bandwidth, and R-channel is

mainly used by the boundary stations in a subnet to relay

packets to adjacent subnets. D-channel can be accessed incontention or contention-free ways, while R-channel can only

be accessed with contention. Notably, each station can access

only one channel through a particular transceiver in it, so three

transceivers are required for each station. A station can operatein either the Free-Mode (F-mode) or the Restricted Mode

(R-Mode). When operating in F-mode, stations send control

signals via the C-channel and transmit data packets via the

D-channel, they also use the R-channel to communicate with

other stations in R-mode. When operating in R-mode, stations

are restricted to using only the R-channel to transmit data

packets based on the CSMA/CA and RTS/CTS mechanisms.

Although the R-mode stations can send packets through only

the R-channel, they can hear the data packets sent in other

channels. Therefore, the boundary WT of a subnet can enter

R-mode to participate in the CFP of its subnet, and relaythe outgoing packets at the same time, where the outgoing

packets are the packets belonging to other subnets. Besidesthe data packets, the R-mode station sends RTS, CTS, ACK

and polling response via R-channel; the station that receives

the packets from the R-channel should respond to the sending

station through R-channel if needed. The example in Fig. 2

provides a good understanding of the channels and modes

defined in the ACS mechanism.

All stations are initially in F-mode. The PC of a subnetbroadcasts the beacon frame or CF-End message through the

C-channel to announce the beginning or the end of a CFP. The

Operates in F-mode

Receives a jamming

packet from other

subnets

Records the CFP

length indicated in

the jamming packet

Operates in R-mode

Receives

a jamming packet

from other

subnets?

Does the

remaining time in

R-mode expire?

Records the CFP

length indicated in

the jamming packet

The number of outgoing

packets exceed Pout?

Records the CFP

length of its ownsubnet

Yes

Yes

Yes

No

No

No

Operates in F-mode

Receives a beacon

frame from its PC

The number of outgoing

packets exceed Pout?

Operates in R-mode

Receives a CF-End

frame from its PC

Yes

No

Fig. 3. The flowchart illustrating the passive restriction and self-restriction(dotted portion) procedures.

WTs that receive the beacon frame sequentially send jamming

packets, indicating the length of CFP to their neighbors

through the C-channel as depicted in Fig. 2. The sequence of

the transmitting jamming packets is included in the beacon

frame. The stations that receive the jamming packets then

begin the passive restriction procedure, and switch to R-mode.

An R-mode station entering the passive restriction procedure

continues recording all incoming jamming packets (comingfrom other subnets), and returns to F-mode after the duration

of the latest recorded CFP expires. If a PC intends to start a

CFP, but is not allowed to send the beacon frame, i.e., it is in

R-mode, then it immediately transmits the beacon frame to its

WTs after switching to F-mode. If a WT returning from R-

mode to F-mode finds that its subnet is current in CFP, then it

immediately transmits a jamming packet to its neighbors. The

flowchart of the passive restriction procedure is presented in

Fig. 3 (please ignore the dotted portion of the figure at this

stage).

The other condition (denoted as self-restriction procedure

in the following) for a WT to operate in R-mode is shown asthe dotted portion in Fig. 3. An F-mode WT with more than

packets that are destined to other subnets in its next-hopswitches to R-mode if it receives the beacon frame from its

PC, where denotes the threshold of the outgoing packets(PC is not allowed to switch to R-mode by the self-restriction

procedure). This R-mode WT automatically returns to F-mode

if it receives the CF-End frame from its PC. Notably, after the

R-mode duration of a WT restricted by the passive restriction

procedure expires, there is still a chance for the WT to remain

in the R-mode. The chance is when the WT has more than

outgoing packets to send and finds that the subnet iscurrently in CFP (please refer to the dotted portion mixedwith the solid portion in Fig. 3). The R-mode duration is then

extended to the end of the CFP.



4/7


A B

Case 1

A B

Case 2

Fig. 4. The illustration of the topologies presented in the analytical model.

The self-restriction procedure complementing the passive

restriction procedure can increase the throughput of the inter-

subnet link. The improvement of throughput due to the self-

restriction procedure can be revealed by considering an ex-

ample in Fig. 2, where a data flow is sent from PC1 to PC2.

This data flow is blocked in halfway when PC1 starts the CFP,

and WT2 operates in F-mode. This block occurs frequently,

and reduces the throughput of the inter-subnet link if the self-

restriction procedure is not introduced. The self-restriction

procedure can force WT2 to operate in R-mode (WT3 operates

in F-mode since it does not receive the jamming packet fromthe R-mode WT2), and facilitate the use of R-channel between

WT2 and WT3.

It should be noticed that the negotiated QoS may be violated

in the situation where a WT operates in R-mode during CFP.

It is because the traffic delivered from the PC to this WT

(through D-channel) may collide with the traffic transmittedin adjacent subnets. Moreover, if there are too many adjacent

R-mode stations transmitting their packets synchronously,

the contention-based R-channel will suffer serious collisions

which result in the QoS deterioration to inter-subnet traffic.

A. Analytical model

This subsection presents an analytical model for evaluating

the relay performance of the ACS mechanism under different

configurations and topologies. The analytical model applies

the ACS to two topologies shown in Fig. 4: the first topology,denoted as case 1, consists of two subnets in which the

transmission range of the PCs do not overlap, while thesecond topology denoted as case 2 comprises two subnets in

which the transmission range of the PCs do overlap. Since the

performance of relay transmission is the most important factor

in evaluating a multihop mechanism, the proposed model

focuses on analyzing the queue length of relay WTs (the gray

nodes in Fig. 4).

1) Analysis of the ACS mechanism in case 1 without self-

restriction procedure: To evaluate the queue length of the

relay WTs (the gray nodes in subnet A) in case 1, the mean

arrival and service rates of the relay WTs are derived according

to several assumptions mentioned below, by using the //1Markovian Birth-Death Queueing Model. The CP length is

assumed to be much shorter than the CFP length, allowing

all behaviors in CP to be negligible. Additionally, the relay

WTs in subnet A (or subnet B) are assumed to spend half

of their lifetime in R-mode, and the other half in F-mode.

Let and denote the number of WTs in subnets A andB; and denote the number of relay WTs in subnetsA and B; denote the system bandwidth (Mbps), and

denote the ratio of the bandwidth reserved for D-channel to

the total bandwidth. Additionally, each WT in subnet A is

assumed to send data streams with data rate to PC B (thepacket inter-arrival time follows the exponential distribution).

The relay WTs in subnet A are regarded as one aggregated

node (denoted as in the following), which is analyzed forthe mean queue length in different modes.

The mean arrival rate of in F-mode equals , and

should be bounded by + 12 , where denotes the channel utilization of CFP, and is formulated as

follows.

=(1 )

( + )(1) + + , (1)

where denotes the payload size of a packet; denotesthe packet header, and denotes the probability that astation has no packets to transmit when polled. The mean

service rate of equals 0, since only communicateswith the stations within subnet A. Therefore, [ ], themean queue length of measured at each end of F-mode

can be calculated as:

[ ] =

{

, if( )

2

( +

2 )

, otherwise

,

(2)

where denotes the duration of a superframe.The mean arrival rate when is in R-mode is the same

as that when it is in F-mode; the service time follows the

exponential distribution with the mean equal to 1/(1 )(), where () denotes the channel utilizationof CP with contending stations. () is discussed byBianchi [15], and is formulated as follows.

= ()() / ((1 ())+()() +()(1 ())),(3)

Here, without RTS/CTS, denotes the average time that thechannel is sensed busy due to a successful transmission; denotes the average time that the channel is sensed busy byeach station during a collision; denotes the duration of anempty slot time; () denotes the probability that at leastone transmission occurs among stations in the consideredtime slot, () is the conditional probability that exact onestation transmits on the channel given that there is at least

one transmission among stations in the considered time slot.These Bianchis parameters are given in (4) (where denotes

the propagation delay, and denotes the minimumcontention window).

= + + + + + + , = + + + ,() = 1 (1 )

,

() =(1)1

()= (1)

1

1(1) ,

= 2/( + 1).

(4)

For simplicity, is calculated under the assumption that noexponential backoff is considered. According to the //1queueing model, the mean queue length of measured ateach end of R-mode, given by [ ], can be obtained as:

[ ] =2

( ). (5)



5/7


Based on the assumption that stays equally in F-mode andin R-mode, [], the mean queue length of measuredat each end of superframe, can be calculated as:

[] = ([ ] + [ ])/2. (6)

2) Analysis of the ACS mechanism in case 1 with self-

restriction procedure: Because of the two assumptions: stays equally in F-mode and R-mode, CP can be neglected

compared to the duration of CFP, we can say that issure to enter F-mode after its R-mode duration expires, and

vice versa. However, when the self-restriction procedure isapplied, probably remains in R-mode after its current R-mode duration expires. That is, the time that stays in R-mode is longer than the time it stays in F-mode. The following

equation expresses the mean queue length of .

[] = 1 [ ] + 2 [ ]. (7)

where 1 and 2 denote the limiting state probabilities of stateA1 ( stays in F-mode) and A2 ( stays in R-mode) ina two-state Markov chain. The values of

1and

2can be

calculated by solving the following equation,{1 + 2 = 1

112 = 221, (8)

where 11, 12, 21 and 22 denote the transition probabilitiesof 1 1, 1 2, 2 1, and 2 2,respectively. Since is sure to enter F-mode after its R-modeduration expires, 12 = 1 and 11 = 0. 22 is the probabilitythat remains in R-mode after its current R-mode durationexpires. In other words, 22 is the probability that the numberof packets in s queue exceeds , which denotes the

threshold of the outgoing packets for the aggregated relaynode. If = /, then the probability that there are packetsin s queue (denoted as ) has the following expression,

= (1 ). (9)

Therefore, 22 can be calculated as follows.

22 =

=

=(1 )

1 = . (10)

Consequently, 21 = 1 22 = 1 . By substituting

12 = 1 and 21 = 1 into (8), the limiting stateprobabilities can be obtained:

1 =1

2 , 2 =

1

2 . (11)

3) Analysis of the ACS mechanism in case 2: In this case,

the common relay nodes of subnet A and subnet B depicted

in the right hand side of Fig. 4 are analyzed for their queue

length. During the analysis, these common relay nodes are

regarded as an aggregated node denoted as . Subnets Aand B operate in CFP in turn (nodes A and B enter R-mode in

turn) according to the ACS mechanism. To evaluate the queue

length of , its mean arrival rate and mean service ratemust be derived based on which subnet is currently in CFP.Each node is assumed to be able to hear from other nodes

within the same subnet.

When subnet A operates in CFP, all nodes in subnet B

except the common relay nodes are sure to be in R-mode. Thus

the mean arrival rate of is equal to , and should bebounded by + /2, where denotes thenumber of the common relay nodes; the mean service rate of

equals 0. Consequently, [ ], which denotes themean queue length of measured at each end of F-mode,can be calculated as:

[ ] =

{

, if( )

2

( +

2 )

, otherwise

.

(12)

All nodes in subnet A except the common relay nodes are

sure to be in R-mode when subnet B operates in CFP. Thus,the mean arrival rate of (denoted as

) equals ,and should be bounded by + (1 )( ); the mean service rate of (denoted as

) equals

/2. According to the //1 queueing model,[ ], which denotes the mean queue length of measured at each end of R-mode, can be calculated as:

[ ] = 2

( ). (13)

The mean queue length of measured at each end ofsuperframe (denoted as []) can then be obtained as:

[] = ([ ] + [ ])/2. (14)

IV. EXPERIMENTAL RESULTS

The ACS mechanism (with = 7/11, = 5) was com-pared with the multi-frequency forwarding mechanism, the

KMH mechanism, and IEEE 802.11 DCF, based on enhancing

IEEE 802.11 PCF. Three simulation scenarios were designed

to evaluate the performance of these mechanisms operatingon different network topologies. In each scenario, the ACSmechanism was compared with other multihop mechanisms

in terms of the throughput of the inter-subnet traffic. These

scenarios were programmed in C++ and configured with

the following settings. The total bandwidth of the system

was 11Mbps, and the simulation time was 60 seconds. Thetransmission and interference ranges of a station were 40m

and 45m, respectively. Each station had at most 50 packets

in its queue, and routed packets according to the DSR [16]

routing protocol. All traffic was generated from CBR sources

generating fixed size packets (1000 bytes).

A. The transmission range of each PC is not overlapped

The upper portion of Fig. 5 shows the network topology

in this scenario. PC1 coordinated the left subnet, and PC2

coordinated the right subnet; the transmission range of each

PC did not overlap. Dataflow 1 transmitted from WT1 to WT8

was simulated as the inter-subnet traffic. Dataflow 2 was sent

from WT2 to WT3, and Dataflow 3 was transmitted from WT6

to WT7; both were simulated as the intra-subnet traffic. The

load of each data flow was set at 1500 Kbps.

In the topology depicted in Fig. 5 (the upper portion), the

ACS mechanism was compared with other multihop mecha-nisms by varying the load of each dataflow from 300Kbps to

1500Kbps. Figure 6 illustrates the variation of the cumulative



6/7


WT2

WT1

WT3

WT4

PC1

WT5

WT6

WT7

WT8

PC2

(40,80)

(10,50)(40,50)

(40,20)

(70,50) (100,50)

(130,80)

(130,50)

(130,20)

(160,50)

WT1 PC1 WT2 PC2 WT3 PC3 WT4

(10,50) (40,50) (70,50) (100,50) (130,50) (160,50) (190,50)

Dataflow 1 Dataflow 2 Dataflow 3

Fig. 5. The topologies in the experiments.

300 400 500 600 700 800 900 1000 1100 1200 1300 1400 15000

1

2

3

4

5

6

7

8

9

10x 10

6

Offered load (Kbps)

Totalthroughput(bytes)

ACS

DCF

MultiFKMH

Fig. 6. The cumulative throughput of the inter-subnet traffic vs. offered loadin the first scenario.

throughput on Dataflow 1. These results indicated that the

ACS mechanism strongly outperformed other mechanisms in

terms of throughput. For instance, in the case of 1200 Kbps

load, the ACS obtained the throughput of 9.0 106 bytes, theDCF obtained the throughput of 2.81 106 bytes, the KMHobtained the throughput of 3.53 106 bytes, and the multi-

frequency forwarding obtained 0 bytes. The multi-frequencyforwarding mechanism failed to transmit inter-subnet traffic

since PC1 and PC2 had no overlapped coverage area. In DCF,

the throughput did not increase with the offered load since the

packets belonging to the same flow frequently contended with

each other for channel access, reducing end-to-end throughput.

B. The transmission range of each PC is overlapped

This topology was almost the same as the topology shown

in the upper portion of Fig. 5, but the transmission range

of each PC was overlapped (the distance between PC1 andPC2 was 60m). WT4 and WT5 in the upper portion of Fig.

5 were replaced by an MF-WT in the overlapped area of

300 400 500 600 700 800 900 1000 1100 1200 1300 1400 15002

3

4

5

6

7

8x 10

6

Offered load (Kbps)

Totalthroughput(bytes)

ACS

DCF

MultiF

KMH

Fig. 7. The cumulative throughput of the inter-subnet traffic vs. offered loadin the second scenario.

PC1 and PC2. the ACS mechanism was compared with othermultihop mechanisms by varying the load of each dataflow

from 300Kbps to 1500Kbps. Figure 7 illustrates the variation

of the cumulative throughput on Dataflow 1. The ACS still

achieved excellent performance while the other mechanisms

had similar performance in terms of the throughput of inter-subnet traffic. For instance, in the case of 1500 Kbps load,

the ACS obtained the throughput of 7 106 bytes, the DCFobtained the throughput of3.0106 bytes, the multi-frequencyforwarding obtained the throughput of3.36106 bytes, and theKMH obtained the throughput of3.53106 bytes. A minimum

gain of 3.64 106 bytes was achieved in this case.1) The WTs and PCs are distributed in a line: The lower

portion of Fig. 5 shows the network topology in this scenario.

The topology comprised three PCs and one dataflow going

from WT1 to WT4. The ACS mechanism was compared withother multihop mechanisms in terms of throughput by varying

the load of Dataflow 1 from 300Kbps to 1500Kbps. Figure 8

illustrates the variation of the cumulative throughput on the

dataflow. The overall throughput was higher than that in the

previous two topologies since there was no competitor intra-

subnet traffic. The ACS again outperformed other mechanisms

while the DCF had the lowest throughput. For instance, in the

case of 1500 Kbps load, the ACS obtained the throughput of11.5106 bytes, the DCF obtained the throughput of4.5106

bytes, the multi-frequency forwarding obtained the throughput

of 6.46 106 bytes, and the KMH obtained the throughput of8.44106 bytes. The KMH achieved the same performance asthe ACS did when the offered load was less than 1100 Kbps,but it failed to catch up to the ACS after 1100 Kbps.

V. CONCLUSION

This work proposed the ACS mechanism for centralized

MAC protocols to eliminate the restriction on single-hop

transmission. The ACS mechanism that allowed the central-ized MAC protocol to transmit data flows across subnets

and alleviated the hidden terminal problem can be adapted



7/7


300 400 500 600 700 800 900 1000 1100 1200 1300 1400 15002

3

4

5

6

7

8

9

10

11

12x 10

6

Offered load (Kbps)

Totalthro

ughput(bytes)

ACS

DCF

MultiF

KMH

Fig. 8. The cumulative throughput of the inter-subnet traffic vs. offered loadin the third scenario.

to various network topologies. The ACS mechanism alsoeliminated the need for both complex initialization procedures

and synchronization between subnets. Experimental results

indicated the optimal configuration on the ACS mechanism,

and the comparison between the ACS mechanism and other

multihop forwarding mechanisms. The comparison results

showed that our mechanism outperformed other multihop

mechanisms in terms of adaptability, efficiency, and bandwidthutilization.

REFERENCES

[1] Institute of Electrical and Electronics Engineers, Wireless LAN mediumaccess control (MAC) and physical (PHY) layer specifications, IEEE

802.11 Standard, Nov. 1997.[2] A. Muir and J. J. Garcia-Luna-Aceves, A channel access protocol

for multihop wireless networks with multiple channels, in Proc. IEEEICC98, Atlanta, GA, USA, June 1998, pp. 1617-1621.

[3] J. So and N. H. Vaidya, Multi-channel MAC for ad hoc networks:handling multi-channel hidden terminals using a single transceiver, inProc. ACM MobiHoc, Tokyo, Japan, May 2004, pp. 222-233.

[4] W. C. Hung, K. L. E. Law, and A. L. Garcia, A dynamic multi-channel MAC for ad-hoc LAN, in Proc. 21st. Biennial Symp. Commun.,Kingston, Apr. 2002, pp. 31-35.

[5] A. Nasipuri and S. R. Das, A multichannel CSMA MAC protocol formobile multihop networks, in Proc. IEEE Wireless Commun. Netw. Conf.(WCNC), New Orleans, LA, USA, Sep. 1999.

[6] R. Carces and J. J. Garcia-Luna-Aceves, Collision avoidance and resolu-tion multiple access for multichannel wireless networks, in Proc. IEEE

Infocom 2000, Tel Aviv, Israel, Mar. 2000, pp. 595-602.

[7] J. So and N. H. Vaidya, A multi-channel MAC protocol for ad hocwireless networks, Technical report, Dept. Comp. Science, Univ. ofIllinois at Urbana-Champaign, 2003.

[8] A. Raniwala, K. Gopalan, and T. C. Chiueh, Centralized channel assign-ment and routing algorithms for multi-channel wireless mesh networks,

Mobile Computing Commun. Review, vol. 8, no. 2, pp. 50-65, Apr. 2004.

[9] Y. D. Lin and Y. C. Hsu, Multihop cellular: a new architecture forwireless communications, in Proc. IEEE INFOCOM, Tel Aviv, Israel,Mar. 2000, pp. 1273-1282.

[10] H. Li, D. Yu, and Y. Gao, Spatial synchronous TDMA in multihopradio network, in Proc. IEEE 59th Veh. Technol. Conf., Milan, Italy,May 2004, pp. 1334-1338.

[11] B. Walke, et al., Relay-based deployment concepts for wireless andmobile broadband radio, IEEE Commun. Mag., vol. 42, no. 9, pp. 80-89, Sep. 2004.

[12] K. Mizuno, M. Katayama, and H. Suda, A new QoS-guaranteed

multichannel MAC protocol for multihop wireless networks, in Proc. IEEE 55th Veh. Technol. Conf., Birmingham, Alabama, May 2002, pp.967-971.

[13] J. Peetz, HiperLAN/2 multihop ad hoc communication by multiple-frequency forwarding, in Proc. Veh. Technol. Conf., Rhodes, Greece,May 2001, pp. 2118-2122.

[14] M. Johnson, HiperLAN/2 - the broadband radio transmission tech-nology operating in the 5GHz frequency band. [Online]. Available:http://paginas.fe.up.pt/ hmiranda/cm/whitepaper.pdf. [Accessed Sep. 15,2009]]

[15] G. Bianchi, Performance analysis of the IEEE 802.11 distributedcoordination function, IEEE J. Sel. Areas Commun., vol. 18, no. 3, pp.535-547, Mar. 2000.

[16] D. B. Johnson, D. A. Maltz, and J. Broch, DSR: the dynamic sourcerouting protocol for multi-hop wireless ad hoc networks, Ad Hoc

Networking, Chap. 5. Addison-Wesley, 2001, pp. 139-172.

Sheng-Tzong Cheng received the B.S. (1985) andM.S. (1987) in Electrical Engineering from theNational Taiwan University, Taipei, Taiwan. He re-ceived the M.S. (1993) and Ph.D. (1995) in Com-puter Science from the University of Maryland,College Park, Md. He is currently a Professor inthe Department of Computer Science and Informa-tion Engineering, National Cheng Kung University,Tainan, Taiwan. His research interests are in de-sign and performance analysis of mobile computing,wireless communications, multimedia, and real-time

systems.

Ming-Hung Tao received the B.S. (2001) and Ph.D.(2007) in Computer Science and Information Engi-neering from the National Cheng-Kung University,Tainan, Taiwan. He is currently an Engineer inthe SoC Technology Center, Industrial TechnologyResearch Institute, Hsinchu, Taiwan. His researchinterests are in wireless broadband network, wirelessresource management.

Chun-Yen Wang received the B.S. (2001) in Math-ematics, and Ph.D. (2006) in Computer Science andInformation Engineering from the National Cheng-Kung University, Tainan, Taiwan. He is currently anEngineer in the Information and CommunicationsResearch Lab, Industrial Technology Research In-stitute, Hsinchu, Taiwan. His research interests arein wireless broadband network, quantum computing.

Documents

Adaptive Channel Switching for Centralized MAC Protocols in Multi-hop Wireless Networks