A TRAINING BASED TRANSMISSION PERIOD SETTING PROTOCOL

8/13/2019 A TRAINING BASED TRANSMISSION PERIOD SETTING PROTOCOL

1/13

A Training B

Gua

Graduate School

744 Mot

ji

ky

ABSTRACT

We have proposed Intermittentnsmission (IPT forwarding) a

packet relay method for wirel

In IPT forwarding, a sourcepackets to a destination node

time interval (IPT duration) sinterference between relay no

packets simultaneously are

frequency reuse is realized,about an improvement of sys

put. However, the optimumsetting for each node is a diffiwhich was not solved adequate

paper, we propose a new trainiduration setting protocol wh

some training packets to seoptimum IPT durations for eaalso extend the proposed

wireless backhauls applyingantennas. The proposed proto

ated both with computer simulaexperiments. Evaluation resulthe proposed protocol is noeffective but also practical.

KEYWORDSMulti-Hop, Ad-Hoc, Wirele

IPT Forwarding, Training Pack

1 Introduction

The high-speed data transan order of 100Mbps envisi

next generation wireless co

system will restrict the cell rthan 100m (a class of pico- c

sed Transmission Period Setting Prot

gri Jin, Li Gong, Hiroshi Furukawa

of Information Science and Electrical Engineer

Kyushu Universityoka, Nishi-ku, Fukuoka, 819-0395 Japan

[email protected],

[email protected],[email protected]

Periodic Tra- an efficient

ss backhaul.

node sendsith a certain

o that signales that send

reduced and

hich bringsem through-

PT durationcult problemly yet. In this

ng based IPTich employs

arch for thech node. Weprotocol for

directionalcol is evalu-

tion and withs show thatt only very

ss Backhaul,

et, FDA.

ission withned for the

munication

adius to lessell), which

Fig. 1Wireless backhaul

increases the number of ce

cover the service area. Dmany base nodes consid

infrastructure costs, thus c

must be key success factbroadband systems.

In recent years, wirel

systems have drawn great i

of the key technologies tostructure costs for next gen

band systems [1]2. Inhaul, base nodes have the

relay packets by wireless,

them, called core nodes, sways to connect the wirel

with outside backbone

Internet) by cables. Upwar

ginnated from the mobile tcell phone) which are asso

of the base nodes and dioutside network are relaintermediate relay nodes

until they reach the core

ward packets originated fronetwork and directed to a

inal in the wireless backha

the core nodes and rela

84

col

ing,

system.

lls needed to

ployment ofrably raises

ost reduction

r for future

ss backhaul

terest as one

reduce infra- ration broad-

ireless back- capability to

and a few of

erve as gate- ess backhaul

etwork (i.e.

packets ori-

rminals (e.g.ciated to one

ected to theyed by theslave nodes)

odes. Down-

the outsidemobile term-

l are sent by

ed by slave

International Journal on New Computer Architectures and Their Applications (IJNCAA) 1(1): 84-96The Society of Digital Information and Wireless Communications, 2011 (ISSN 2220-9085)


2/13

85

nodes until they reach the final node to

which the mobile terminal is associated

(Fig. 1). By connecting base nodes bywireless links, flexibility of the base

nodes deployment is realized and total

infrastructure costs are reduced due tothe reduction in cable constructions [2].

Wireless backhauls traditionally have

been studied in the context of SpatialTDMA (STDMA) and the Ad Hoc

network. STDMA can achieve collision

free multi hop channel access by a well

designed time slot assignment for each

cell [3]. However, such planning isnot feasible in real systems because ofthe irregular cell forms in real environ-

ments. Additionally, frame synchroniza-tion must be managed carefully in

STDMA, which induces rather difficultoptimization issues [9]. With regard to

the Ad-Hoc network, many studies have

contributed to improve its performance.In [6], Li et al. have indicated that an

application of IEEE802.11 to wireless

multihop network fails to achieveoptimal packet forwarding due to severe

packet loss. In [7], Zhai et al. have

proposed a new packet scheduling algor-ithm called Optimum Packet scheduling

for Each Traffic flow (OPET) which can

achieve an optimum scheduling of

packets by assigning high priority ofchannel access to the current receiver.

However, the overhead imposed by the

complicated hand-shake process decrea-ses frequency reuse efficiency. In [8],

Bansal et al. have indicated that the

throughput of wireless multihop networkdecreases as the hop count increases.

On the other hand, we have proposed

Intermittent Periodic Transmission (IPT

forwarding, [9]) as an efficient packetrelay method with which the system

throughput can achieve a constant value.

In IPT forwarding, source node intermit-tently sends packets with a certain time

interval (IPT duration), and each inter-

mediate relay node forwards the relay

packet immediately after the reception ofit. The frequency reuse space attained by

the method is proportional to the given

IPT duration. In [10], a series ofexperiments have been carried out to

confirm the effectiveness of the method

with real testbed. IPT forwarding isfurther enhanced with the combinations

of MIMO transmission [11] and direc-

tional antenna [12].

IPT duration is the most importantparameter for applying IPT forwarding

method. In [13], a collision free IPT

duration setting method was proposed

and evaluated with computersimulations. However, the method is not

feasible since it introduced some newMAC packets, which makes it difficult

to be implemented with general WLAN

modules without any modifications.

Additionally, system throughput is notguaranteed to be maximized by the IPT

durations attained by the method.

In this paper, we propose a new IPTduration setting protocol which employs

training packets to search the optimum

IPT durations for each slave node. Withthese IPT durations, the end to end

throughputs for each slave node are

maximized. A new metric for the train-

ing process is also presented. Then weextend the proposed protocol for wire-

less backhauls using directional antennas

introduced in [12]. The proposed prot-ocol is evaluated with both computer

simulations and experiments on a real

testbeds.The rest of this paper is organized as

follows. Section 2 explains the principle

of IPT forwarding and the collision freeIPT duration setting protocol proposed

in [13]. Section 3 explains the proposed

protocol in detail. In section 4 we extend

the proposed protocol for a wireless



3/13

backhaul system applying

antennas. In section 5, we

proposed protocol with bosimulations and experiment

concludes this paper.

2 Intermittent Periodic TrIn this section, we expla

ciple of IPT forwarding aloconventional packet relay

collision free IPT duration

hod proposed by [13] is also

2.1 Principle of IPT forwarIn order to clearly explain

of IPT forwarding, we ill

packet relay mechanism oftional CSMA/CA based met

forwarding in Fig. 2 andpectively.

In the two figures, 9 nodes

placed and instantaneous p

on the route are shown inwith time. All the packets t

re-formatted in advance to h

time length.In the case of the conventi

/CA based method, the s

sends packets with a randosion period of P_CNV and

mediate relay node forwar

packets from its preceding

random backoff period. InIPT forwarding, the source

mits packets intermittently

transmission period of P_Iintermediate relay node

forwards the received pack

preceding node withoutperiod. No synchronization

for both the conventional

IPT forwarding method.In the conventional met

transmission space, which i

the distance between relay

transmit packets at the same

directional

evaluate the

h computer. Section 6

nsmissionn the prin-

ng with theethod. The

setting met-

introduced.

dinghe principle

strated the

the conven- od and IPT

Fig. 3, res-

are linearly

acket relays

accordancebe sent are

ve the same

onal CSMA

ource node

m transmis- each inter-

ds received

ode with a

the case ofnode trans-

ith a certain

T and eachmmediately

ts from the

ny waitingis required

ethod and

od the co-

defined as

nodes that

ime, is not

Fig. 2 Packet relay mec

in conventional met

Fig. 3 Packet relay mec

in IPT forwarding

Fig. 4 Performance compar

conventional method and IP

fixed. In such situations, paoccurs due to co-channel i

the co-transmission space i

0 2 4

Throughput

Hop Count

Conventional Method

1.

86

anism

od.

anism

.

ison of the

forwarding

cket collisionterference if

shorter than

6 8

IPT Forwarding



4/13

the required frequency reu

shown in Fig. 2. On the ot

the case of IPT forwardinreadily understood that the

sion space could be contr

transmission period of P_given to the source node,

Fig. 3 in which reuse space

be 3.Reduction of the packet c

help to reduce retransmissi

consequently help to improv

performance. If an adequattion is set in the core node, i

to remove interference b

channel relay nodes that s

simultaneously. If the IPTequal to the threshold, t

throughput observed at thenode can be maximized.

Fig. 4 schematically sho

alized throughput versus

feature of the conventionalIPT forwarding for the syste

and 3. In Fig. 4, constant IP

applied for all slave nodesresultant throughputs ar

same [9].

2.2 Collision free IPT duraAs discussed earlier, i

achieve optimal performan

node should set an adduration for each slave nod

the optimum IPT duration fo

node depends on many efactors such as channel ch

node placements, antenna di

so on. To make IPT forwarpractical, an automatic I

setting method is require

problem, a collision freebeen proposed to automatica

durations for each node in [1

e space, as

er hand, in

it can beco-transmis-

lled by the

PT that iss shown in

s assume to

llisions will

ns and will

e the system

IPT dura- t is possible

etween co-

end packets

duration ise resultant

destination

s the norm-

hop count

method ands in Fig. 2

duration is

nd thus theall the

ion settingorder to

e the core

quate IPTe. However,

r each slave

vironmentalracteristics,

rections and

ing methodT duration

d. To this

rotocol haslly find IPT

].

In this subsection, we w

duce the collision free prot

indicate its drawbacks.

Fig. 5Collision Free IPT dur

1) Summary of collision fThree new MAC layer p

(Request to Stop Sending

CTP (Clear to Pilling UPCTPACK (CTP ACKn

packet, are defined in [13

shaking algorithm is emplthe IPT duration for each n

As shown in Fig. 5, w

duration setting started the(node 1) continuously send

to the destination node (

certain IPT duration. If atransmission fails in an

node (e.g. node 4 in Fi

interference, the node se

packet to the source node todata packets. The source n

the sending of data packets

after reception of the RTSsends a CTP packet to th

node. The CTP packet is r

same way as that for dat

therefore the destination nthat all the relaying data

cleared out from the syste

of the CTP packet. The desimmediately sends a CTPA

the source node on receptio

packet. After receiving tpacket, the source node

87

ll first intro-

col and then

tion setting.

ree method

ckets, RTSSpacket and

packet andwledgement)

and a hand

oyed to findde.

hen the IPT

source nodedata packets

ode 7) with

data packetintermediate

. 5) due to

ds a RTSS

stop sendingode suspends

immediately

S packet ande destination

elayed in the

packet and

de can knowpackets are

by reception

tination nodeCK packet to

n of the CTP

e CTPACKncreases the



5/13

88

IPT duration by one step and resumes

the sending of data packets. This process

repeats until no data packet forwardingfailure occurs in the relaying route.

2)

Drawbacks of the collision freemethodAlthough the collision free method

can obtain certain IPT durations forwireless backhaul, it has some severe

drawbacks as described below.

1) Since new MAC layer packets areintroduced, it is difficult to be

implemented by general wireless

interface modules.

2)

The packet transmission state isconfirmed by checking the MAC

state of each node. However,existing MAC drivers (e.g. MAD

WiFi Driver) do not provide such

functions.

3) System throughput is not guaranteedto be maximized by applying the IPT

durations attained by the method.

Any modification to existing standards

will cause extra costs. Since one of the

major advantages for wireless backhaulis the ability to reduce costs, a new IPT

duration setting method which is not

only practical but also exploits the

optimum system performance isrequired.

3 Throughput maximization IPT du-

ration setting protocolIn this section, we propose a new

training based IPT duration settingprotocol which maximizes the end to end

throughput for each slave node. The

proposed protocol employs some train-ing packets and performs a series of

training process to search the optimum

IPT duration for each slave node. During

the training process, core node continu-

ously sends a number of training packets

to each slave with an IPT duration which

increases gradually until the end to endthroughput from the core node to the

slave node reaches the maximum value.

Throughout this paper we assume thatthe route of wireless backhaul is already

decided before the IPT duration setting

starts and will not change during theprocess of the protocol.

3.1Variables and parametersWe defined the following variables

and parameters in the new protocol.

1) Training packet2)

Number of training packets:N3) Training time for each node: T

4) Training metric for each node: TM5) IPT duration for each node: D

(micro second)

6) Training Step in the process: (micro second)

In these variables, the training packetis defined as OSI link layers data packet

with the length of 1450 Byte and

identified by sequence number. Theparameters TM and D are initialized

whenever new training begins for a new

slave node. The training metric TM,

which is described latter in detail, is usedas the criterion for the training process.

3.2Details of the protocolAs shown in Fig. 1, wireless backhaul

can be considered as the union of a few

sub systems each of which is consistedof a core node and several slave nodes

belonging to it (i.e. each slave node is

connected to the outside network via the

other slave nodes intermediately andfinally through the core node by wireless

multihop fashion). We call the sub sys-

tems a mesh clusters (Fig. 6) throughoutthis paper and the IPT duration setting is



6/13

performed for each mesh clu

tively in the same way.

Fig. 6Mesh cluste

Now let us consider a

with a core node C and anodes {S1, S2, , Sn}. For

node S {S1, S2, , Sn}, tprocess is executed .

Step1: The core node C in

training metric TMas -1.0

Das D0for the slave nodeD0is a small non-negative

Step2: The core node C send

packets which have th

number of 1, 2, ,Nto thScontinuously with the IP

Step3: Whenever the slareceives a training pack

destined to it, S records t

number and the packet rece

Step4:If the reception of traidestined to itself is finishe

node Ssends a report packe

node C which contains tnumber and reception tim

of the first training packetand the sequence number atime (Seq2, T2) of the l

packet it received. The

training packets receiv

duplication, Num, is alsothe report packet.

ster respect-

.

esh cluster

set of slaveeach slave

e following

itializes the

nd initialize

S, in whichalue.

s N training

sequence

slave nodedurationD.

e node St which is

e sequence

tion time.ing packets

d, the slave

t to the core

e sequence(Seq1, T1)

it receivedd reception

ast training

number of

d without

included in

Step5: When the core nod

report packet from the sla

estimates actual training tSas below.

Fig. 7Core node behavior du

/ 2

According to the estimtime T, a new training me

ated as below.

After the computation of

metric, the training processtwo cases based on the v

TM.

a) If , thfinishes the training

the IPT duration ofand move to the trai

slave node.b) If , th

increases the IPT durreplace the training m

and repeat

Step2Step5.

89

e C receives

ve node S, it

me spent for

ing training.

1 1

2 (1)

ated trainingric is calcul-

new training

branches intolue of New_

core node Cor S and set

as ning of next

core node C

tion D by ,tric TMwith

s the above



7/13

Step6:Thecore node Crepea

Step2Step5 until the traithe slave nodes is finished.

In Fig.7, we showed the be

core node during training pr3.3Features of the proposeWe make some explanat

new proposed protocolsubsection.

1) Throughout the trainingassume that the syste

provide packet relay seronly training related pac

the system.2) During the training proc

packet losses occur, the

packet received by the s

could be different with t

by the core node C. Forthe actual training time

estimated as in Step5. In

is the average traitransmission time and

start time and end time

mented by , which ccomplements T.

3) The protocol repeatstraining process by grad

sing the IPT duration fo

node until its training mthe maximum value. Si

slave node the training fi

moment when TMbegin

we set the IPT durationof one preceding step as

of Step5.4) It can also be easily r

training metric TM is

nected to the end to en

from the core node Cnode S. Since the proto

the IPT duration for eac

which maximizes its traithe end to end through

slave node is also maxi

ts the above

ning for all

avior of the

cess.d protocolons on the

in this

process, we

does not

vice so thatets exist in

ess, if someirst and last

lave node S

e ones sent

this reason,must be re-

formula (1),

ing packetthe training

are comple-

onsequently

the sameally increa-

each slave

tric reachesce for each

ishes at the

to decrease,

as the valueshown in a)

ealized thatlosely con-

throughput

o the slaveol searches

slave node

ning metric,ut of each

ized by the

calculated IPT durations

4. Extension of the proposIn this section we extend

protocol to a wireless bac

which employs directionalspecified as FDA system [1

Fig. 8 AnF-3node with 3 i

4.1 Introduction of FDA sFDA system was propos

the application of directionwireless backhaul. In the

first briefly introduce the

FDA system.InF-nFDA system (Fixe

Antennas), each node conta

interfaces and each interfac

with one directional antdirectional antennas are s

spaced and each antenna i

one of the n uniform direcover the whole space toge

assumed that the beamw

directional antenna is equthan 2/n. For illustration,

an F-3 node which contai

interfaces and 3 directionalIn FDA, packet relays

between the two interfaces

and receiver nodes whicantennas to each other, thefor switching antennas dur

During packet relay servic

receives packets from oneforwards the received pac

one of the n interfaces.

90

.

ed protocolthe proposed

haul system

antennas and].

nterfaces.

stemd to simplify

l antennas inollowing, we

principles of

d Directional

ns nwireless

e is equipped

nna. The nymmetrically

s oriented to

ctions whichher. It is also

dth of each

al to or lessFig. 8 shows

s 3 wireless

antennas.only occur

f transmitter

direct theireby no need

ing run-time.

e, each node

interface andkets through

This feature



8/13

significantly simplifies the a

directional antennas in wirel

The authors have proprouting protocol for FDA s

specifies not only the uplin

link interfaces used in eacalso the corresponding inter

destination nodes. The autho

extended traditional IPT forhod for FDA system as belo

(a)Core node applies IPT feach of its wireless interf

(b)Core node waits for a(referred to as inter-pat

tion in the paper) when

wireless interface during

It was shown that by (a) ancollisions in FDA system ca

However, the optimization

IPT duration and inter-path

in (a) and (b) remained to be

4.2 IPT and inter-path I

setting for FDA systemWe extend the training ba

for FDA system. Throughou

we assume that system routdecided before the training p

Since in FDA system ant

ing is not required during r

IPT duration setting methodsection 3 can be directly

decide the optimum IPT dur

for each slave node (e.g. Sonly needs to execute the pr

the interface corresponding

node S.Now we consider the in

duration setting problem in (

the inter-path IPT durationdecided for each pair of sla

which core node uses differe

to relay packets. We consid

nodes S1and S2to which co

plication of

ss backhaul.

sed a newstem which

and down-

node, butfaces in the

rs have also

arding met- .

rwarding in

ace.certain time

IPT dura-

it switches

packet relay.

(b), packetbe avoided.

roblems for

PT duration

solved.

T duration

sed protocol

this section,

e is alreadyocess.

nna switch-

n time, the

proposed inapplied. To

tions in (a),

, core nodeotocol using

to the slave

er-path IPT

b), in which

should beve nodes to

nt interfaces

r two slave

e node uses

interface M1 and M2 to

respectively. The inter-path

between S1and S2is denowe assume that core n

interfaces from M1 to M2

Fig. 9. We also assume thaations for S1 and S2 are alr

and denoted asD1andD2.

Fig. 9Inter path IPT durati

The inter-path IPT dura

be large enough, so that tra

node to slave node S2does

with the traffic from core nshould also be small enou

the optimum system perfor

on this consideration, we ptraining based inter-path

setting method.

We defined the followiand parameters for the met

1) Training packet2) Number of training pa3) Training time for each4) Training metric for ea

TM1, TM2.

5) Training Step in the prAll of these variables h

meanings as defined in sethe pair of slave nodes S1

node repeats the following

Step1: Core node initializes

metric TM1 and TM2initializeDiasDi0.

91

elay packets

IPT duration

ed asDiandde switches

as shown in

the IPT dur-eady decided

n setting.

ion Di must

fic from core

not interfere

ode to S1.Digh to ensure

ance. Based

opose a newPT duration

ng variablesod:

ckets:Nnode: T

ch node: TM,

ocess:

ve the same

tion 3.1. Forand S2, core

rocess.

the training

as -1.0 and



9/13

Step2:Core node sendsNtrai

continuously with the sequ

of 1, 2, ,Nto slave nodeIPT duration of D1. Then

waits forDitime. After tha

sends N training packetswith the sequence number

to slave node S2with the

ofD2.

Step3: Whenever slave nod

receive a training packet,

the sequence number and

reception time.Step4: When the reception

packet is finished, slave no

send report packets to cor

the sequence number antime (Seq1, T1) of the f

packet and the sequencereception time (Seq2, T2)

training packet they rec

number of training pack

without duplication, Num1are also included in the rep

Step5: When core node rec

packets from S1 and S2,TM1 and TM2 in the sa

shown in Step5 of section 3

calculatesNew_TM = TM1that the training process b

two cases based onNew_ T

a) If , thefinishes the training and

path IPT duration as

b) Else If increases the inter-path

Di by , replace the tra

TM with

and

above Step2Step5.

It can be easily underst

above method is an exten

training based protocol dsection 3, which searches t

inter-path IPT duration for t

ing packets

nce number

S1with theCore node

t Core node

ontinuouslyf 1, 2, ,N

PT duration

s S1 or S2

they record

the packet

of training

e S1and S2

node with

d receptionrst training

umber andof the last

eived. The

ts received

and Num2,rt packets.

ived report

t calculatesme way as

.2, and then

TM2. Afteranches into

.

core nodeet the inter-

.

, core nodePT duration

ning metric

repeats the

od that the

sion of the

iscussed ine optimum

e pair of S1

and S2. With the attained i

duration, the sum of the

throughput for S1and S2is

5 Evaluation

In this section, we evalposed protocol with both c

ulations and experiments o

bed under indoor environm

Fig. 10 Simulation site 1, stri

system.

Fig. 11 Simulation site 2, tree to

Table 1. Simulation ParMAC Model IEEE802.11a,

Retry Count =PHY Model Packet recepti

SINR level bethan 10dB.

Propagation Model 2 Ray Groun

Model.No Fading Ef12dB Attenua

Routing Method Minimum Pat

Data Packet Length 1500 Byte.

5.1Evaluation by SimulatWe assume IEEE802.11

less interface of each nodetwo simulation scenarios

topology and tree topologas shown in Fig.10 (Sce

Fig.11 (Scenario 2). The si

are models of West Buil

Campus, Kyushu-Universit

92

ter-path IPT

end to end

maximized.

ate the pro- mputer sim-

n a real test-

nt.

ng topology

ology system.

ameters.Basic Mode.

3.on fails whencomes lower

Reflection

ect.tion by a Wall.

Loss Routing.

ons

as the wire-

and deployedwith string

respectivelyario 1) and

ulation sites

ding of ITO

, Japan.



10/13

93

Each system in the Scenarios is

consisted of only one mesh cluster. The

simulation parameters are shown inTable 1 and IPT forwarding is applied.

5.1.1

Simulation scenariosIn the first, we measured the end to

end throughput from core node to each

slave node with different IPT durationsin the two simulation scenarios using the

following formula (2).

Fig. 12 IPT durations and end to end throughput

for node 4, 5, 6 in simulation scenario 1.


for node 7, 8, 9, 10 in simulation scenario 1.

Table 2 Optimum IPT durations in simulation

scenario 1 ().

Node 4 Node 5 Node 6 Node 7

1000 1300 1600 1900

Node 8 Node 9 Node 10

1900 1900 1900

2

In the above formula This throughput,

Nr is the number of received packetswithout duplication, PL is packet length

and Tmis transmission time.

In this first evaluation, IPT durations

are set manually for the purpose of

searching an optimum IPT duration foreach slave node. Manually taken optim-

um IPT durations are compared with the

ones to be found by the proposed prot-ocol, afterward.

We assume that no extra traffic occurs

during the measurement and the numberof transmitted packets in the above

formula is 2000.

Fig. 14Automatically calculated IPT durations

in simulation scenario 1.


for node 4, 5, 9, 10 in simulation scenario 2.

Table 3 Optimum IPT durations in simulation

scenario 2 ().

Node 4 Node 5 Node 9 Node 10

1000 1300 1300 1300

After that, we performed the proposed

protocol to calculate the IPT durations

for each slave node with the initial valueD0to be 0, to be 100.

5.1.2 Simulation resultsThe throughput vs. IPT duration for

each slave node is shown in Fig. 12, 13for scenario 1 and in Fig. 15 for scenario

2. The optimum IPT durations with

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200EndtoE

ndThroughput(Kbps)

IPT Durations (micro sec)

Node 4

Node 5

Node 6

0

1000

2000

3000

4000

5000

6000

0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200 2400

Endto

EndThroughput(Kbps)


Node 7Node 8

Node 9 Node 10

0

200

400

600

800

1000

1200

1400

1600

1800

2000

0 1 2 3 4 5 6 7 8 9 10 11

IPTDuration(microsec)

Node

N = 300, 500, 1000

N = 100 N = 50

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

0 100200300 400500600 70080090010001100120013001400150016001700

Throughput(kbps)


Node 4

Node 5

Node 9

Node 10



11/13

which the end to end th

maximized are shown in

scenario 1 and in Table 3 foThe IPT durations obtai

protocol for each slave nod

in Fig. 14 for scenario 1 anscenario 2.

As shown in Fig. 14 and.

duration calculated by thezero for node 1, 2, 3 in sce

for node 1, 2, 3, 6, 7, 8 in sce

Fig. 16Automatically calculated

in simulation scenario

According to the feature of

ding, it can be easily undersIPT durations for the slave

are located within the CSthe core node could be

because for these nodes noinal problem occurs and thu

purposely adjust the packet

time in the core node. Forwe deleted the throughput

results of these nodes in Fig

15.In Fig. 14 and 16, the a

calculated IPT durations

optimum ones in Table 2 an300, 500, 1000. However, wi

small values of N(50, 100 ithe calculated IPT durati

match to the optimum obecause with such small N,

received training packets nu

varies intensively each tiestimation of training time i

enough.

0

200

400

600

800

1000

1200

1400

0 1 2 3 4 5 6

IPTDuration(microsec)

Node

N = 300, 500, 1000

N = 100

oughput is

able 2 for

scenario 2.ed by the

are shown

Fig. 16 for

16, the IPT

protocol isnario 1 and

nario 2.

IPT durations

2.

IPT forwar-

ood that theodes which

A range ofset to zero

idden term- no need to

ransmission

this reason,easurement

. 12, 13 and

tomatically

atch to the

3 withN=th relatively

n this case),ns do not

es. This isthe ratio of

mber and N

e and thenot precise

However, with the inc

(larger than 300 in this cas

iations are suppressed andthe calculated IPT duration

the optimum values for eac

The simulation results sadequate parameter setting

sed protocol can find the

durations for each slave nothe end to end throughput is

Fig. 17 Picomesh Lu

Table 4 Specificatio

CPU AMD

Memory D

Backhall Wireless IF IEEE8

Access Wireless IF IEEE8

OS Linu

Fig. 18 Experimental

Fig. 19Throughput measuring

5.2Evaluation by ExperiIn order to further confi

mance, we implement the

tocol into a testbed and evformance under real indoor

7 8 9 10 11

N = 50

94

ement of N

e), these var-

consequentlyconverge to

slave node.

ow that withs, the propo-

ptimum IPT

e with whichmaximized.

chBox

of LB.Geode LX800R 512MB02.11b/g/a 2

02.11b/g/a 1

x kernel 2.6

site.

system.

entsm its perfor-

roposed pro-

luate its per- environment.



12/13

95

5.2.1 Testbed and Throughput Meas-uring Tool

The testbed is called PicoMesh

LunchBox (LB, Fig. 17). LB is the first

product of MIMO MESH Project,which the authors are working on [14].

LB is equipped with three IEEE802.11

modules, two of which are used forrelaying packets between base nodes and

the other one is used for mobile terminal

access.

Each module of LB is assigned withdifferent spectrum so that the inter-

ference between these modules could be

avoided. The hardware specification of

LB is shown in Table 4.In this experiment, we use IPerf to

measure the throughputs [15]. IPerf isfree software which can measure the end

to end throughput in various networks

with a pair of server and client. Add-

itionally, we adopt UDP mode of its twooperational modes (TCP mode and UDP

mode) and measure the throughput from

client to server.

5.2.2 Experimental ScenarioWe deployed a wireless backhaul

system with one core node and six slave

nodes in West Building of ITO Campus,

Kyushu-University, Japan (Fig. 18).

At first, we measured the end to endthroughput of each slave node with

different IPT durations by IPerf. Speci-

fically, we set up the IPerf client in a PCwhich is connected to the core node and

the IPerf server in a PC which is

connected to the slave node (Fig. 19).During the measurement, traffic flows

from IPerf client to sever and each

measuring continues 30 seconds. Wealso assume that no extra traffic occurs

during the measurement. In this first

experiment, IPT durations are set man-

ually for the purpose of searching an

optimum IPT duration for each slave

node. The optimum IPT durations are

compared with the ones to be found bythe proposed protocol, afterward.

In the next, we performed the prop-

osed protocol to calculate the IPT dura-tions for each slave node with the

training packet numberNto be 1000 and

the initial value D0 to be 1000, to be

100.


for Node 3, 4, 5, 6 in experiment.

Table 5 IPT durations searched by the

protocol () and its run time.Node 1 Node 2 Node 3 Node 4

0 0 1000 1300

Node 5 Node 6 Run Time

1300 1300 11 (sec)

5.2.3 Result of the ExperimentsThe throughput vs. IPT duration for

each slave node is shown in Fig. 20 and

the IPT durations calculated by the prop-

osed protocol and the protocols run timeare shown in Table 5.

The calculated IPT durations of node

1 and 2 are zero in Table 5, which meansthat the two nodes are located within the

CSMA range of the core node and thus

we deleted the corresponding through-puts of the two nodes in Fig. 20.

As we can see from Fig. 20 and Table

5, the calculated IPT durations match to

the optimum ones measured by IPerfwith which the end to end throughputs

reach the maximum values. With 6 slavenodes and 1000 training packets, the

0

1

2

3

4

5

6

7

8

0 200 400 600 800 1000 1 200 1400 1600 1800 2 000 2 200

Throughput(kbps)

IPT Duration (micro sec)

Node 3

Node 4 Node 5 Node 6



13/13

96

protocol spent 11 seconds to finish,

which makes it practical enough in real

applications.

5 Conclusion

In this paper we proposed a new IPTduration setting protocol which can

calculate the optimum IPT duration for

each slave node automatically. The pro-posed protocol is evaluated both with

computer simulations and experiments

on a real testbed in an indoor enviro-

nment.Evaluation results show that with the

calculated IPT durations the end to end

throughput of each slave node is max-

imized. Since the protocol does notdemand modification of the existing sta-

ndards, it could be easily implementedwith general WLAN modules.

Additionally, we left the evaluation of

inter-path IPT duration setting method as

future work.

6 ACKNOWLEDGMENT

This work was supported by MIMO-MESH PROJECT under Grant of Know-

ledge Cluster Initiative implemented by

Ministry of Education, Culture, Sports,Science and Technology, Japan.

7 REFERENCES

[1] G. Narlikar, G. Wilfong, L. Zhang,Designing Multi-hop Wireless BackhaulNetworks with Delay Guarantees, in procINFOCOM 2006. 25th IEEE InternationalConference on Computer Communications,pp.1-12, 2006.

[2] R. Pabst et al, Relay-Based DeploymentConcepts for Wireless and Mobile Broad-

band Radio, IEEE Communication Maga-zine, pp.80-89, September 2004.

[3] R. Nelson and L. Kleinrock, Spatial TD-MA: A Collision Free Multihop ChannelAccess Protocol, IEEE Trans. Comm., Vol.33, No.9, pp. 934-944,September1985.

[4] J. Gronkvist, J. Nilsson and D. Yuan,Throughput of Optimal Spatial ReuseTDMA for Wireless Ad-Hoc Networks,Proc. VTC04 Spring, 11F-3, May 2004.

[5] H. Li, D. Yu and Y. Gao, Spatial Synch-ronous TDMA in Multihop Radio Network,Proc. VTC04 Spring, 8F-1, May 2004.

[6] J. Li, C. Blake, D. S. J. De Couto, H. I. Leeand R. Morris, Capacity of Ad Hoc Wire-less Network, in Proc. ACM MobiCom2001, July 2001.

[7] H. Zhai, J. Wang, Y. Fang, and D. Wu, "ADual-channel MAC Protocol for Mobile AdHoc Networks, " in Proc. IEEE Workshopon Wireless Ad Hoc and Sensor Networks,in conjunction with IEEE GlobeCom, pp.27-32, 2004.

[8] S.Bansal, R.Shorey, A.Misra, Energy effi-ciency and throughput for TCP traffic inmulti-hop wireless networks, Proc INFO-COM2002, 23-27, pp. 210-219, 2002.

[9] H. Furukawa, Hop Count IndependentThroughput Realization by A New WirelessMultihop Relay, Proc. VTC04 fall, pp.-2999-3003, September 2004.

[10]Y. HIGA, H. Furukawa, ExperimentalEvaluation of Wireless Multihop NetworksAssociated with Intermittent PeriodicTransmit, IEICE Trans. Comm., vol. E90-B, no.11, November 2007.

[11]E. M. Mohamed, D. Kinoshita, K. Mitsu-naga, Y. HIGA and H. Furukawa, "AnEfficient Wireless Backhaul Utilizing MI-MO Transmission and IPT Forwarding",International Journal of Computer Net-works, IJCN, vol. 2, No. 1, pp.34-46,March 2010.

[12] Guangri Jin, Ximing Huang, HiroshiFurukawa, Performance Enhancement ofIPT forwarding for Wireless Backhaul by

Fixed Directional Antennas, Proc. IEEEAPWCS 2011, Aug. 2011, Singapore.

[13]Y. HIGA and H. Furukawa, Time IntervalAdjustment Protocol for the New Wirelessmultihop Relay with Intermittent PeriodicTransmit. IEICE 2004, B-5-180, Sept-ember 2004.

[14]http://mimo-mesh.com/en/.[15]http://iperf.sourceforge.net/.


Documents

A TRAINING BASED TRANSMISSION PERIOD SETTING PROTOCOL