A survey of MAC based QoS implementations for WiMAX networks

Computer Networks xxx (2009) xxxndashxxx

ARTICLE IN PRESS

Contents lists available at ScienceDirect

Computer Networks

journal homepage wwwelsevier com locatecomnet

A survey of MAC based QoS implementations for WiMAX networks

Y Ahmet Sekercioglu a Milosh Ivanovich a Alper Yegin b

a Department of Electrical and Computer Systems Engineering Monash University Australiab Standards and Industry Initiatives Group Samsung Electronics Korea

a r t i c l e i n f o a b s t r a c t

Article historyReceived 9 September 2008Received in revised form 20 April 2009Accepted 10 May 2009Available online xxxx

Responsible Editor L Jiang Xie

KeywordsWireless networksWiMAXQuality of ServiceQoSMACMedia Access Control

1389-1286$ - see front matter 2009 Elsevier BVdoi101016jcomnet200905001

Corresponding author Tel +61 3 99053503 faE-mail address Asekerciieeeorg (Y A SekercioURL httptitaniactiemonasheduau (Y A Seke

Please cite this article in press as Y A Sekercio(2009) doi101016jcomnet200905001

We present a comprehensive survey of proposed Quality of Service (QoS) mechanisms inthe Media Access Control (MAC) sublayer of WiMAX based wireless networks QoS supportin WiMAX is a fundamental design requirement and is considerably more difficult than inwired networks mainly because of the variable and unpredictable characteristics of wire-less links

We discuss various QoS architectures signaling mechanisms and admission control tech-niques proposed in the WiMAX research literature summarizing the operation of each andproviding comparative evaluations that include advantages and limitations

2009 Elsevier BV All rights reserved

1 Introduction

IEEE 80216 Wireless Metropolitan Area Network AirInterface Standard [1719] provides the details of physicallayer and Media Access Control (MAC) sublayer of an ad-vanced wireless communication system which aims tobuild a cost effective multi-service network with WiMAX(Worldwide Interoperability for Microwave Access) tech-nology The standard published in 2004 [17] describesthe physical and MAC sublayer specifications for fixedwireless access systems supporting multiple services Itconsolidates the IEEE Standards 80216 80216a and80216c The WiMAX Forum describes WiMAX as lsquolsquoa stan-dards-based technology enabling the delivery of last milewireless broadband access as an alternative to cable andDSL (Digital Subscriber Line)rdquo The newer version of thestandard IEEE 80216e-2005 [19] published in 2005 con-tains numerous revisions adds higher layer handovers be-

All rights reserved

x +61 3 99053454glu)rcioglu)

glu et al A survey of MAC

tween base stations as well as support for mobileterminals at vehicular speeds

This standard is the outcome of a convergence of themarket need and current wireless technological achieve-ments and is considered a benchmark solution for wirelessmetropolitan area networks (WMANs) as opposed to Wi-Fiwireless local area networks (WLANs) High data rateslarge area of coverage ease and cost effectiveness ofdeployments makes WiMAX suitable for a number ofapplications This includes connecting multiple Wi-Fi hot-spots backhaul services and high speed mobile datacommunication

When the aim is to provide a multi-service wireless net-work a key challenge is the optimal allocation and utiliza-tion of the available raw data transmission capacity ofshared wireless links among users and services In this sur-vey we use the term lsquobandwidthrsquo to refer to the data trans-mission capacity of the links Bandwidth utilization isconsidered optimum when there is no over- or under-allo-cation of capacity for a particular service type Data trans-mission requirements depend on the type of servicesrequested by a subscriber and suboptimal distribution of

based QoS implementations for WiMAX networks Comput Netw

Table 1Network performance parameters and their characteristics maintained byQoS

Network performanceparameter

Characteristics

Latency Delivery delay of a packet from source todestination

Jitter Variation in latency

Reliability The percentage of traffic that should besuccessfully delivered from source to destinationto maintain the service quality

Data transmissionrate

The amount of data that should be carried fromsource to destination in a given period of time tomaintain the service quality

2 Y A Sekercioglu et al Computer Networks xxx (2009) xxxndashxxx

ARTICLE IN PRESS

available transmission capacity inevitably affects the ser-vice quality Another factor is latency Some services havegreater tolerance for latency (eg FTP or e-mail) whileothers (like VoIP or video-conferencing) have strict delaybounds Taking such QoS requirements into account pack-et flows need to be prioritized via appropriate QoS man-agement and scheduling methods In broad terms thesealways seek to achieve the optimal trade-off between theconflicting goals of maximized user performance and max-imized system utilization

WiMAX has a built-in QoS framework for real-timeapplications as well as data which can take advantage ofits polling architecture and dynamically adaptable modu-lation in the physical layer It should be noted that whilethe standardized WiMAX QoS framework provides the de-tails about the types of service flows that are supported itdoes not explicitly define the actual packet mechanisms forachieving QoS differentiation in the MAC sublayer As inother standards of this kind such mechanisms are leftopen for vendor implementation as long as they conformto the stated WiMAX QoS framework

In Section 2 we provide an overview of this WiMAX-specific QoS framework and separately consider point-to-multipoint and Mesh Network variants Our workexamines QoS implementation in WiMAX by subdividingthe relevant issues into three distinct categories packetscheduling and admission control (Section 31) signallingand internetworking (Section 32) and Mesh Network (Sec-tion 4) In each of these three sections we summarize thestate of the art research activity along with results possi-ble implementation drawbacks and scope for furtherdevelopment In Section 6 an overall analysis is providedalong with our concluding remarks

2 QoS and the MAC sublayer of WiMAX networks

21 Definitions of Quality of Service

There are two broad definitions of Quality of Service(QoS)

User-Centric QoS is lsquolsquothe collective effect of service per-formances which determine the degree of satisfactionof a user of the servicerdquo [22]Network-Centric QoS comprises lsquolsquothe mechanisms thatgive network managers the ability to control the mix ofbandwidth delay variances in delay (jitter) and packetloss in the network in order to deliver a network service(eg voice over IP)rdquo [9]

Our paper is primarily concerned with the second defi-nition of QoS and in the rest of this work the term QoSshall be taken to mean lsquolsquoNetwork-Centric QoSrdquo

Network engineers can use the existing resources effi-ciently by implementing a QoS mechanism Early packetnetworks typically catered for one service type and allpackets were treated equally There was no QoS differenti-ation or guarantee of reliability minimum latency jitter orother performance characteristics (Table 1) for any set ofpackets As a result of such a regime a single bandwidth

Please cite this article in press as Y A Sekercioglu et al A survey of MAC(2009) doi101016jcomnet200905001

intensive application may cause the performance of otherapplications to degrade significantly In a multi-servicenetwork the QoS mechanism has to ensure that it can pro-vide preferential delivery service to packets according totheir performance requirements and QoS priority levelwhile maintaining a high network utilization QoS differen-tiation can be implemented on either a per-application orper-user basis With this in mind QoS mechanisms canbroadly be grouped under the following two categories

Admission Control determines how and when the traf-fic generated by a given application or user can haveaccess to the network resources Typically operates ata session or flow timescale (ie decisions relate toadmission of user sessions or flows)Traffic Control determines how packet marking sched-uling and shaping (flow rate control) is performed forpacket traffic generated by a given application or userTypically operates at packet timescales (ie decisionsrelate to which packet from which flow is to be trans-mitted next)

By implementing a well functioning QoS mechanism (ora combination thereof) network engineers can controlavailable network resources to suit a particular require-ment model and to ensure that critical services are not af-fected by services of lower-priority The end result isimproved user experience and reduced system cost dueto more efficient and targeted use of available resources

22 QoS architecture of WiMAX

In wireless networks including WiMAX QoS supportusually resides in the MAC sublayer because of the needto interact with radio resource management and physicallayer dynamics Fig 1 shows the WiMAX QoS architectureas defined by the standard [19] The base station (BS) hasthe responsibility of managing and maintaining the QoSfor all packet transmissions The BS manages this bydynamically distributing usage time to subscriber stations(SSs) through information embedded in the transmittedframes (Fig 2) The figure only shows the TDD (time-divi-sion duplex) mode of operation in which BS-to-SS broad-casts (downlink subframe) are followed by SS-to-BS(uplink subframe) transmissions It is also possible to useFDD (frequency-division duplex) mode of operation in


Fig 1 Overall structure of the WiMAX QoS architecture

Y A Sekercioglu et al Computer Networks xxx (2009) xxxndashxxx 3

ARTICLE IN PRESS

which downlink and uplink subframes are transmitted inseparate frequency bands but the QoS management prin-ciples remain the same

The downlink subframe contains two fields for manag-ing allocation of the wireless medium DL-MAP (downlinkbandwidth allocation map) to tell the SSs of the timetableand physical layer properties for transmitting subsequentbursts of packets (the latter is referred to as the lsquolsquoDownlinkBurst Profilerdquo in WiMAX literature) and UL-MAP (uplinkbandwidth allocation map) for regulating the uplink trans-mission rights of each SS That is the UL-MAP controls theamount of time each SS is given access to the channel inthe immediately following or the next uplink subframe(s)A parameter called Uplink Allocation Start Time specifies forwhich uplink subframe the UL-MAP contents should be ap-plied for This flexibility allows an SS to have sufficienttime to schedule uplink transmissions and prepare forthe actual physical stream of data to be filled in the as-signed uplink resource

Fig 2 Simplified WiMAX frame structure em


Uplink subframes contain three categories of fields

Initial ranging contention slot (denoted as lsquolsquoinitial rangrdquoin Fig 2) is used by SSs to discover the optimum trans-mission power as well as timing and frequency offsetto communicate with the BS An SS begins the rangingprocess by sending a ranging request MAC sublayer mes-sage using the minimum transmission power If it doesnot receive a response from the BS it resends the mes-sage in the same field of a subsequent UL subframe usinga higher transmission power

Bandwidth requests contention slot is used by SSs fortransmitting bandwidth request (BW-REQ) MAC sublay-er messages

Slots specifically allocated to the individual SSs fortransmitting data

The overall operation of the system can be summarizedas follows

phasizing its QoS management aspects



ARTICLE IN PRESS

The 80216 MAC protocol is connection oriented Signal-ing messages between the BS and an SS need to be ex-changed in order to establish a lsquolsquoservice flowrdquo1 betweenthem Service flows can be requested by the BS (accordingto the standard this is a mandatory capability) or by an SS(an optional capability) Each service flow is characterizedby a range of parameters including three sets of QoS param-eters indicating the required latency jitter and throughputassurances These correspond to the three possible serviceflow states (provisioned admitted and active) and are thuscalled ProvisionedQoSParamSet AdmittedQoSParamSet andActiveQoSParamSet Furthermore each service flow is as-signed a unique 32-bit long SFID (Service Flow Identifier)by the BS

Service flows can be requested with any of these param-eter sets being null The 80216 standard has two differentkinds of call activation processes a flow can be dynami-cally set up through DSA (dynamic service activation)transactions or through a two-phase activation modelsimilar to telephony applications The former DSA-basedapproach is not expected to be available until dynamicQoS is introduced in the Network Release 15 of the IEEE80216e-2005 standard The latter telephony-based ap-proach supports the notion of static (ie nonprovisioned)QoS and is available from the earlier Network Release10 of the standard Given static QoS an arriving serviceflow typically has a non-null ProvisionedQoSParamSet en-ters the provisioned state and is allocated an SFID by theBS without being able to carry data packets until it islsquolsquoactivatedrdquo

Like a telephony call a service flow goes through a tran-sient lsquolsquoadmittedrdquo state and changes to the lsquolsquoactivatedrdquo stateonly after the end-to-end negotiation is completed Foreach of the three states of a service flow different QoSparameter sets can be defined but the set relationship

ActiveQoSParamSet AdmittedQoSParamSet

ProvisionedQoSParamSet

should always hold Provisioned service flows becomeadmitted or activated when their QoS requirements be-come known through the subsequently sent update mes-sages containing non-null AdmittedQoSParamSet andActiveQoSParamSet fields (though it is not clear in the spec-ifications that whether both sets should be non-null) If theQoS requirements of an active flow are included in an ini-tial request such a flow can be provisioned and immedi-ately admittedactivated by the BS For an activatedservice flow the BS allocates a unique 16-bit connectionidentifier (CID) In this way each BS-to-SS connection willhave assigned to it as many CIDs as it has activated serviceflows (ie typically generated by separate active applica-tions on the SS)

For SS-initiated communications an SS on behalf of anapplication first requests a connection The CAC (connec-tion admission control) module located in the BS then

1 A service flow signifies a unidirectional flow of packets that is provideda particular QoS or in 80216 terms it lsquolsquois a MAC transport service thatprovides unidirectional transport of uplink packets transmitted by the SS orto downlink packets transmitted by the BSrdquo


checks whether the requested QoS parameters are withinthe limits of available resources and if this is the case theBS then responds with a unique SFID BS-initiated communi-cations work similarly but in addition to the CAC checks theBS after allocating a unique SFID also waits for the responseof the SS indicating whether it can support the requestedcommunication (the standard does not describe the internalcriteria for an SS supporting or rejecting requests)

Various higher layer packet properties (such as IP ad-dresses and protocol ports) are used for assigning the pack-ets generated by the individual applications to specificSFIDs and following activation to corresponding CIDs Theassigned CIDs are then used to classify the resulting MACframes into appropriate SS transmission queues This func-tionality is encapsulated in the Convergence Sublayer (CS)

The scheduler of an SS visits the queues and selectspackets for transmission Selected packets are transmittedto the BS in the allocated time slots as defined in the UL-MAP which is constructed by the BS Uplink Schedulerand broadcast by the BS to the SSs It should be noted thatin the IEEE 80216e-2005 standard the UL bandwidthgrants do not specify the CID That is an SS is delegateddecisions about scheduling multiple service flows belong-ing to it This approach ensures that scheduling is left tothe most appropriate node ndash an SS has queue state infor-mation which is more timely and accurate than the de-layed lsquolsquovirtual queuerdquo estimates available to the BSImportantly the absence of the CID in the UL bandwidthgrants does not diminish the scheduling effectiveness ofthe SS for the following reasons Firstly the traffic priorityQoS parameter specified at service flow creationmodifica-tion governs the scheduling priority among service flowtypes other than UGS and system signalling messages(ie MAC management) Based on this QoS parameterthe SS and BS always have a common understanding ofthe order in which the stationrsquos flows should be scheduledthus making the CID in the UL bandwidth grant unneces-sary Secondly the BS knows when it expects system sig-nalling messages and so at these times will ensure thatenough UL resources are assigned to an SS to cater for boththe UGS and system signalling traffic In terms of the rela-tive priorities between the two the BS and SS have a sim-ilarly common understanding such that system signallingis generally afforded the highest priority Again the addi-tional specification of CID in the UL grants is not neededSimilarly the BS Downlink Scheduler selects the order inwhich it will transmit packets to the SSs and constructs acorresponding DL-MAP as shown in Fig 2

221 QoS parameters scheduling and data delivery servicesAfter the admission of service flows arguably the most

complex aspect of the provision of QoS to individual packetsis performed by the three schedulers (i) the DownlinkScheduler which manages the BS-to-SS flows and (ii) theUplink and (iii) Subscriber Station Schedulers which to-gether manage the SS-to-BS flows The Downlink Sched-ulerrsquos task is relatively simple as compared to the UplinkScheduler since all downlink queues reside in the BS andtheir state is locally accessible to the scheduler On the otherhand as the queues of uplink packet flows are distributedamong the SSs and their states and QoS requirements need



ARTICLE IN PRESS

to be obtained through bandwidth requests the task of theUplink Scheduler is much more complex (for this reasonwe only include the uplink behavior in Table 2) The infor-mation gathered from the remote queues forms the opera-tional basis of the Uplink Scheduler and is depicted aslsquolsquovirtualrdquo queues in Fig 1 None of the actual algorithms forthe three schedulers are defined in the standard and are in-stead left open to proprietary implementations

In order to deal with the complexities associated withQoS provision to various applications in an ecosystem ofdifferent vendorsrsquo scheduler implementations the standarddefines a number of lsquolsquoscheduling servicerdquo and lsquolsquodata deliveryservicerdquo classes One of these classes is typically requestedby an application when its traffic flow goes through thestages outlined in Section 22 For each scheduling serviceclass there is a corresponding data delivery service class(Table 2) The data delivery service classes are defined forand used with both uplink and downlink flows In contrastscheduling service classes are only used for uplink flowsThe version of the standard published in 2004 [17] onlycovered the definitions of the scheduling services for uplinkflows During the discussions of the 80216e [19] standard-ization process the need to also define a scheduling servicecapability for downlink flows was brought up and resultedin the introduction of data delivery services For backwardcompatibility the scheduling services already defined inthe 2004 version of 80216 [17] were also retained Theset of QoS parameters associated with a scheduling serviceandor a data delivery service are almost identical and theonly reason both still remain in the specifications is that ofhistorical standard evolution

Table 2WiMAX scheduling and data delivery service classes their typical usage and BS anthe piggybacked bandwidth (BW) request method in addition to the special purpothe grant management subheader fields in the generic MAC PDUs [19]

Scheduling service Correspondingdata deliveryservice

Typicalapplications

Uplink

BS behaviour

Unsolicited grantservice (UGS)


Voice (VoIP)without silencesuppression

The BS uplink scoffers fixed sizegrants on a real-periodic basis

Extended real-time pollingservice (ertPS)

Extended real-time variable-rateservice (ERT-VR)

VoIP with silencesuppression

The BS uplink scoffers real-timeUL BW requestopportunities (sUGS but ertPS aare dynamic no

Real-time pollingservice (rtPS)

Real-timevariable-rateservice (RT-VR)

Streaming audioor video

The BS uplink scoffers real-timeUL BW requestopportunities

Non-real-timepolling service(nrtPS)

Non-real-timevariable rateservice (NRT-VR)

File transfers The BS uplink scprovides timelyorder of a secondBW request opp

Best-effort service(BE)


Web browsingemail

The BS does notoffer any UL BWopportunity


When a specific scheduling or data delivery service isassociated with a service flow that flow is further associ-ated with a certain pre-defined set of QoS parametersHowever according to the standard this does not includeassignment of specific values to the parameters which ismanaged using dynamic service addition (DSA) and dy-namic service change (DSC) messages

23 Point-to-multipoint vs mesh WiMAX networks

In a mesh WiMAX network a lsquolsquoMesh BSrdquo (mesh basestation) provides the external backhaul link The backhaullinks connect the WiMAX network to other communicationnetworks There can be multiple Mesh BSs in a networkother nodes are known as lsquolsquoMesh SSsrdquo (mesh subscriberstations)

The sectorized antenna used by the BS in a WiMAX cellis capable of splitting its coverage area into separate sub-fields and managing transmissions simultaneously andindependently in each The MAC sublayer uses these an-tenna properties to control data transmission betweenthe BS and SSs to optimize the channel utilization

As discussed earlier in point-to-multipoint mode theSS transmissions are controlled directly by the BS In Meshmode the uplink and downlink is not clearly separated andSSs can communicate with each other without communi-cating with the BS Fig 3 shows the frame structure inMesh WiMAX networks Similar to point-to-multipointWiMAX networks data transfer is connection orientedConnection setup can be achieved using either of the fol-lowing two scheduling schemes

d SS behaviors for uplink The rtPS nrtPS and BE scheduling services can usese BW request PDUs Piggybacked BW requests are signalled by populating

QoS specifications

SS behaviour

hedulerUL BWtime

An SS does not need to sendany explicit UL BW requests

Maximum sustained rateMaximum latency toleranceJitter tolerance

hedulerperiodic

imilar tollocationst fixed)

An SS uses the offeredopportunity to specify thedesired UL BW grant

Maximum sustained rateMinimum reserved rateMaximum latency toleranceJitter toleranceTraffic priority

hedulerperiodic

An SS can use (a) the offeredopportunity to specify thedesired UL BW grant or (b)piggybacked BW requestopportunities (an SS can notuse contention based BWrequests)

Maximum sustained rateMinimum reserved rateMaximum latency toleranceTraffic priority

heduler(in the

or less) ULortunities

An SS can use (a) offereduplink or (b) contention-based or (c) piggybackedBW request opportunities

Maximum sustained rateMinimum reserved rateTraffic priority

specifically An SS can use (a)contention-based or (b)piggybacked BW requestopportunities

Maximum sustained rateTraffic priority


Fig 3 Frame structure of the mesh WiMAX networks


ARTICLE IN PRESS

Centralized Scheduling (Mesh CS) the Mesh-BS hasthe responsibility of granting resources for each linkin response to resource requests Mesh centralizedscheduling messages transmitted in a scheduled controlsubframe (Fig 3) are used for this purposeDistributed Scheduling (Mesh DS) The neighboringMesh SS responds to a request with a correspondinggrant for a link between two Mesh SSs Mesh distrib-uted scheduling messages are exchanged to performthis operation

In contrast with point-to-multipoint WiMAX networksthe standard does not define scheduling services for MeshWiMAX networks

Network control subframes periodically appear and areused for servicing the new nodes which want to gain ac-cess to the network The transmission opportunities in acontrol subframe and the data minislots in a data subframeare separated The nodes compete for the control channelaccess and the contention outcome does not effect the datatransmission

3 Mechanisms for QoS provision in point-to-multipointWiMAX networks

Research studies conducted in point-to-multipoint Wi-MAX networks can be classified into two broad categories

Admission control and packet scheduling researchfocuses on the implementation of the admission controland scheduling services (Uplink and Downlink Schedul-ers at the BS as well as the SS scheduler)Signaling and internetworking research into methodsthrough which signalling can be improved (Section 32)and internetworking between WiMAX and other net-works (eg fiber backhaul and Wi-Fi access)

In the following sections we discuss the research workproposed for each category

31 Admission control and packet scheduling

A considerable number of studies may be found in theresearch literature on algorithms and methods for schedul-ing services and admission control in the context of Wi-MAX point-to-multipoint networks


311 A study of QoS support in 80216 networksIn one of the early studies on QoS support in WiMAX

networks Cicconetti et al focus on the available QoS sup-port mechanisms in the MAC sublayer and evaluate theireffectiveness through simulation [7] They conduct theperformance evaluation based on two common applicationscenarios conceived by the WiMAX Forum [12] residen-tial and small to medium-size enterprises (SME) The testcase uses 7 MHz channel bandwidth with carrier frequencybetween 2 and 11 GHz and operating in FDD mode In thestudy it is also assumed that frame duration is 10 ms allSSs have full duplex capability and channel conditionsare ideal

Since the actual implementations of the SS schedulerand downlink and uplink schedulers of the BS are not in-cluded in the standard the authors needed to chooseappropriate algorithms for them They note that the basicQoS parameter negotiated for a connection within a sched-uling service is the minimum reserved rate and because ofthis they argue that the class of rate-latency schedulingalgorithms [35] are suitable for implementing theschedulers

Within this class the authors have chosen Deficit RoundRobin (DRR) [34] algorithm for implementation of thedownlink scheduler of the BS They justify this selectionto DRRrsquos ability to maintain fair queueing when packet sizeis variable and its ease of implementation But DRR can notbe used for the uplink scheduler since it needs to know thesize of the packet at the head of each queue for its opera-tion The BS through the virtual queues (see Fig 1) canonly estimate the uplink load but not the packet sizeswhich is not sufficient for operation of DRR Because ofthis the authors have selected Weighted Round Robin(WRR) [25] algorithm (which also belongs to the class ofrate-latency scheduling algorithms) Their choice for theimplementation of the SS scheduler remains as DRR be-cause an SS always knows the sizes of the packets waitingat the head of its packet queues

In the Residential Scenario the BS only provides Inter-net connectivity to the SSs and all traffic is of BE classThe results show that as long the network is lightly loadedthe connection queues are almost empty The average de-lay increases sharply as soon as the system starts to getoverloaded When overloaded the average delay of uplinktraffic becomes greater than the downlink traffic

In the SME Scenario the BS caters for various types ofservices like VoIP video or data It assumes that VoIP and



ARTICLE IN PRESS

video traffic is classified as rtPS and data as best-effortVoIP is provided with a greater reserved rate than videoThe results show that as the number of active SSs in-creases the downlink delay increases smoothly for all clas-ses of traffic However as the network gets overloaded (iethe number of subscribers is more than 30) there is a sharprise in the delay for BE traffic but delay for VoIP and videois unchanged This happens because of the way in whichcapacity has been provisioned to different connectionsThe scheduling algorithm is configured such that rtPS con-nections have a reserved rate equal to the mean rate ofVoIP and video traffic respectively The guaranteed ratefor BE is negligible compared to rtPS connections Furtherincreases in the load show a rise for delay in video trafficbut not for VoIP due to its greater reserved rate the samebehavior is observed in the uplink

The uplink traffic delay variation is greater than down-link traffic when the system is not in overload (fewer than24 SSs) but lower when the system is overloaded (numberof SSs is between 24 and 36) This happens for the follow-ing reason when the system is not overloaded the BS is-sues an uplink grant as soon as it receives the bandwidthrequest But when the system is overloaded applicationsat the SSs generate the next packet before the uplink grantarrives from the BS for the previous packet Therefore theSSs are able to piggyback the bandwidth request for thenext packet on the current outgoing packet and reducethe delay (and delay variation) The results show thatwhen the number of subscribers exceeds 36 this phenom-enon cannot compensate further and the delay variationcurve begins to increase

312 A scheduling algorithm and admission control methodWongthavarawat and Ganz propose an implementation

of an uplink packet scheduling (UPS) and admission control

Fig 4 QoS architecture proposed by W


framework [38] at the BS and a Traffic Policing module atthe SS (Fig 4) Using simulation methods the authorsshow that their proposal yields an improvement in sys-tem performance over a lsquolsquodefaultrdquo case without thisfunctionality

As mentioned earlier in the standard WiMAX QoSarchitecture (Fig 1) details of both the admission controland uplink scheduling at the BS are undefined with theirimplementation left to vendors The same holds at the SSwhere neither traffic policing module nor its interactionwith the BS admission control are defined The SS sched-uler receives the UL-MAP from the BS after a bandwidth re-quest is made to the BS UPS module however the specificpolicy that will be used in the UPS module is undefined inthe standard WiMAX QoS architecture

Fig 4 shows a sketch of the proposed implementationin the context of the original WiMAX QoS architecturewith key interactions and information flows clearlymarked At the BS an admission control module and de-tailed UPS module are introduced A traffic policing moduleis included at the SS When an application at the SS origi-nates a connection request to the BS it includes the band-width and delay requirement in the request message Theadmission control module accepts or rejects this requestbased on its traffic policy If the request is accepted it noti-fies the BS UPS module and provides appropriate parame-ters After receiving the parameters the SS traffic policingmodule ensures that traffic is classified based on the trafficcontract The information module of the UPS collects thequeue size information from the BW-request messages re-ceived from the previous time frame This is used by theinformation module to update the scheduling databasemodule The UL-MAP is generated by the service assign-ment module after information is received from the sched-uling database module The UL-MAP is broadcast to all SSs

ongthavarawat and Ganz [38]



ARTICLE IN PRESS

and based on this the UPS of an individual SS transmits thepackets

The authors explain in detail the information schedul-ing database and service assignment modules describingthe steps taken for each class of traffic considering variousfactors like queue size arrival time and delay thresholdAppropriate algorithms are also proposed as implementa-tions of each of these three modules

To admit a new connection the admission controlmechanism enforces procedures based on the schedulingclass of traffic which we summarize below

Procedure for UGS On arrival of a new request itchecks for the available bandwidth There is no checkrequired for delay However it checks whether accept-ing this request will cause any delay violation for theexisting rtPS connections If there is no violation theconnection is acceptedProcedure for rtPS First it checks for the availablebandwidth If the bandwidth is available then checksif delay guarantees can be maintained It also checksfor any delay violations for the existing rtPS connec-tions If these conditions are met then the connectionis acceptedProcedure for nrtPS It only checks for the availablebandwidth There is no need to check for the delay vio-lation for existing rtPS (or the lower-priority BE)connectionsProcedure for BE No admission control process isrequired They are always admitted but do not receiveQoS support

The simulation study only assumes that there are twokinds of traffic rtPS and BE Each connection has specificQoS parameters in terms of average bandwidth require-ment which is equal to the token bucket rate and maxi-mum delay requirement The authors present theoutcomes of the study in three graphs the arrival curvewhich depicts the arrival pattern of the input traffic theservice curve which shows the service pattern providedby UPS and the percentage of packets that miss their dead-line The downlink and uplink capacity is set to 5 MBpseach frame size is set to 10 ms For rtPS there are threesessions each with a bandwidth of 3 Mbps

For the first experiment the combined bandwidth forrtPS and BE connection is 5 MBps and the results show thatnone of the packets miss their deadline The second exper-iment shows the arrival and service curves of all three rtPSconnections The graphs show that the service curve adaptsand follows the arrival curve for all three sessions As noneof the packets miss their deadline the delay is alsoguaranteed

Fig 5 The hierarchical packet scheduling model of the uplink in IEEE80216 as proposed in [32] Each traffic class is assigned to three logicalscheduling servers There is a provision for soft-QoS traffic to bescheduled by the best-effort server to obtain additional bandwidth

313 A hard and soft server scheduling mechanismInspired by an earlier study [3] Shang and Cheng pro-

pose a hierarchical packet scheduling model for WiMAXuplink by introducing the lsquolsquosoft-QoSrdquo and lsquolsquohard-QoSrdquo con-cepts [32] rtPS and nrtPS traffic are classified as soft-QoSbecause their bandwidth requirement varies between theminimum and maximum bandwidth available for a con-


nection UGS traffic is classified as hard-QoS since it re-quires the maximum bandwidth available for theconnection By allowing the BE traffic to be scheduled bythe BS the model is able to distribute bandwidth betweenBE and other classes of traffic efficiently and guaranteesfairness among the QoS-supported traffic (UGS rtPS andnrtPS) A delay comparison performance evaluation is pro-vided between the models

The study by Bennett and Zhang [3] proposes the worstcase fair weighted fair queueing (H-WF2Q+) schedulingframework Based on some criteria (the authors do notspecify the criteria) it distributes weighted bandwidth todifferent sets of flows However this model is not suitablefor multimedia traffic as it does not take into considerationits diverse traffic requirements In the proposed modelpacket scheduling takes place in the BS uplink As shownin Fig 5 each traffic class is assigned to three logical sched-uling servers (hard-QoS server soft-QoS server and best-effort server) UGS traffic is routed through the hard-QoSserver rtPS and nrtPS through the soft-QoS server and BEthrough the best-effort server The capacity of each serveris allocated by a pre-defined algorithm There is a provisionfor soft-QoS traffic to be scheduled by best-effort serverThis enables it to obtain additional bandwidth The packetscheduling algorithm comprises of four parts

(1) hard-QoS server scheduling(2) soft-QoS server scheduling(3) best-effort server scheduling(4) co-scheduling among the above three servers

A detailed algorithm for each server is shown alongwith a delay comparison between the initial and thisdeveloped model The difference between the two modelsis the treatment of the soft-QoS traffic This changes thetree-like structure to a two-level hierarchical structureThe results show an improvement in delay and the soft-QoS and BE traffic is able to obtain greater share of band-width by minimizing bandwidth wastage Based on thenetwork dynamics the servers are able to change theirweights for different traffic loads It also proves that the



ARTICLE IN PRESS

hierarchical model can guarantee lower delay and delayjitter for variable bit-rate traffic unlike the model pre-sented in [3]

314 A method for MAC flexibility exploitation formultimedia streaming

The study of Sengupta et al [33] investigates the mech-anisms through which MAC PDUs can be continuouslymodified based on the feedback obtained through thechannel state information The scheme changes the pay-load size by aggregation or fragmentation of the upperlayer PDUs By adopting the dynamic MAC PDU approachthe study shows performance enhancements for streamingvarious types of media

The idea of a dynamic payload size based on near-instantaneous channel state information has been used inother wireless broadband technologies such as 3GPP HighSpeed Downlink Packet Access (HSDPA) [28] albeit at low-er layers of the protocol stack Most typically this ap-proach is adopted at the physical layer an example beingthe Adaptive Modulation and Coding technique employedin HSDPA

The scheme presented in the study works this waywhen an SS requests media content the media server lo-cated in the core network transmits the raw data to theWiMAX gateway The encoder at the BS receives this rawdata and pushes it to the MAC sublayer Depending onthe channel state of the SS the scheduler at the MAC sub-layer manipulates the MAC SDUs to construct the MACPDUs A feedback mechanism placed at a receiverrsquos MACsublayer is the core of this scheme Based on the feedbacksignals generated the transmitting side modifies the MACPDU payload size By changing the MAC PDU size dynami-cally the system attempts to match packet transmissionsto the underlying radio channel conditions This resultsin reduction of the number of dropped or corrupted pack-ets and retransmissions and eventually achieves reduceddelays and increased overall network throughput In theauthorsrsquo scheme ARQ mechanism is used for recoveringthe corrupted transmissions and is an integral part of esti-mating the channel conditions

Fig 6 illustrates how multiple MAC SDUs can be concat-enated to a single MAC PDU or how a single MAC SDU canbe fragmented and distributed over multiple MAC PDUs

(a) Multiple SDUs (Service Data Units) in

(b) Single SDU split in

Fig 6 Packet aggregation and segregation in the adaptive multimedia strea


The connection setup and transmission takes place inthree phases First the SS makes a connection request Thisenables the BS to detect the initial ranging and measure thetimingpower offset This is followed by the service flowparameter request and at this point the variable lengthMAC SDU indicators are turned on Second the BS confirmsthe connection by responding with a response message thathas the initial ranging power adjustment information forthe SS The service flow adjustments are negotiated andthe SS is provided with a CID Finally MAC SDUs obtainedfrom the MAC convergence sublayer are transmittedthrough the MAC PDU payload Depending on the channelrequirements the MAC SDUs can be fragmented or aggre-gated at the start of transmission Feedback is received afterthe first transmission and the next MAC PDU payload sizeis changed accordingly There are six different feedbackpossibilities and Table 3 shows the actions taken by theBS when each type of feedback is received

Simulation based experiments were conducted over achannel model with various bit error probabilities forexperimenting with a range of channel conditions fromlsquolsquogoodrdquo lsquolsquofairrdquo lsquolsquomediumrdquo and lsquolsquobadrdquo (with simulated bit er-ror rates of 0045 006 007 and 0085 respectively) Theexperiments consider mechanisms with or without feed-back and comparative results are presented in the paperThe authors first compare the packet restore probability(PRP) over time for MAC PDUs whose sizes are either keptconstant or adaptively modified as described in the paperAlthough the authors do not provide a quantitative analy-sis of the results the graphs show 70ndash80 improvementfor the adaptive scheme By studying the graph we can ob-serve that for the non-adaptive scheme the PRP reacheszero in a 30 ms time frame 15 times whereas for the adap-tive scheme the worse case scenario occurs just once

The goodput (the ratio of information bits to total bitstransmitted) for the non-adaptive scheme is about 77when the channel error rate is approximately 1 and thisgradually drops to around 63 as the channel error rate in-creases to 20 For the adaptive scheme the goodput is85 and 82 respectively showing an improvement of8ndash20 The most significant improvement is observed withthe MAC PDU drop rate With the non-adaptive scheme asthe channel error rate increases from 1 to 20 the MACPDU drop increases from 15 to 18 However with the

a single PDU (Protocol Data Unit)

to multiple PDUs

ming scheme for WiMAX networks proposed by Sengupta et al [33]


Table 3Feedback types and associated BS responses in the adaptive multimedia streaming scheme for WiMAX networks proposed by Sengupta et al [33]

Feedback type Feedback classification Action taken by base station

1 MAC PDU received correctly (1) Increase MAC PDU payload(2) Decrease CRC for not so important MAC PDU

2 MAC PDU received with errors and uncorrectable (1) Increase CEC for important MAC PDU(2) Keep payload and CRC fixed for not-so-important MAC PDU

3 MAC PDU received with errors but correctable (1) Decrease payload for MAC PDU(2) Increase CRC of MAC PDU

4 MAC PDU dropped timeout in receiver MAC occurred Same as feedback type 3 but the incrementdecrement is morepronounced

5 Receiver MAC buffer full last stored frame is important Stall transmission until further request received

6 Receiver MAC buffer full last stored frame is not soimportant

(1) Skip transmission of next few not so important frames(2) Important frame(s) isare transmitted


ARTICLE IN PRESS

adaptive scheme the MAC PDU drop rate never increasesbeyond 15

315 A two-tier scheduling algorithmA hierarchical QoS architecture is proposed in Chan

et alrsquos study [4] that implements a two-tier scheduling algo-rithm (2TSA) at the BS The first tier is based on the connec-tion category and the second tier is weight-based The studyconsiders TDD operating mode of WiMAX physical layer andassumes the uplink subframe occupies half a frame time AsUGS connection is automatically allocated per frame 2TSAdoes the scheduling for rtPS nrtPS and BE traffic

2TSA implements a simple service category for eachconnection that is based on the allocated bandwidth Thecategories are

Unsatisfied A connection receives less bandwidth thanthe minimum requirement or reserved rateSatisfied A connection receives bandwidth that is morethan the minimum requirement but less than the max-imum bandwidth sustained rateOver-Satisfied A connection receives more bandwidththan the specified maximum requirement

Based on the service category each connection is givena weight between 0 and 1 For example if the allocatedbandwidth of a connection is less than its minimum de-mand its weight indicates the shortage compared to thisdemand Similarly weights of the other two categoriesindicate the corresponding satisfaction degree Fig 7shows the flowchart of the proposed 2TSA The functional-ity of each tier can be summarized as below

First-Tier Allocation The BS classifies all connectionsinto three categories based on the collected bandwidthrequest and updated weights 2TSA then allocates thebandwidth first to the lsquolsquounsatisfiedrdquo followed by thelsquolsquosatisfiedrdquo and finally to the lsquolsquoover-satisfiedrdquo categoriesSecond-Tier Allocation For each specific category thereceived bandwidth is further distributed to the connec-tions based on the value of the weight parameter Con-nections with smaller weights are given higher priority

After completing the two-tier bandwidth allocation theBS generates the corresponding UL-MAP and broadcasts toall SSs


The authors investigated the performance of thescheme through the simulation of a WiMAX networkwhich has 5 UGS and 7 rtPS nrtPS and BE connectionsserved by a BS The simulation has two scenarios

In the first scenario the total available uplink bandwidthis 8 Mbps and the sum of all connectionsrsquo maximumsustained rate exceeds 8 Mbps The results show thatno matter how much traffic a connection generates2TSA can guarantee each connection its minimum band-width demand and fairly distribute the residual band-width to all connections (01 Mbps to each) This is incontrast with strict-priority scheduling (SPS) proposedin [37] where nrtPS and BE connections begin to starveunder same conditions This is because SPS always allo-cates rtPS connections first

In the second scenario the total available bandwidth is12 Mbps (which is greater than the total maximum sus-tained rate of the connections) This experiment wereconducted to evaluate how fairly the residual band-width is allocated in 2TSA compared to SPS The resultspresented demonstrate that the residual bandwidth isdistributed to all connections after maximum sustainedrates are allocated In contrast nrtPS and BE connectionsget starved when SPS algorithm is used

316 A scheduling architecture for improving delay andthroughput

In the study [31] the authors propose a schedulingarchitecture in order to improve the delay and throughputfor rtPS connections which is an extension of an earlierresearch work [6] The previous work implemented atwo-layer scheduling structure for bandwidth allocationto support all types of service flows Direct Fair PriorityQueue (DFPQ) was used in the first layer to distribute totalbandwidth among flow services in different queues (6 intotal depending on service class and direction) as shownin Fig 8 In the second layer of [6] various scheduling algo-rithms are used for each class of traffic For rtPS connec-tions the packet with the earliest deadline is scheduledfirst [14] Weight based scheduling algorithm [8] is usedfor nrtPS connections and round robin scheduling algo-rithm [15] for BE traffic The paper then proposes thenew scheduling technique and presents three different sce-narios for its implementation


Fig 7 Operational flowchart of the 2TSA scheduling algorithm [4]

Fig 8 Deficit Fair Priority Queue (DFPQ) bandwidth allocation methodproposed by Chen et al [6] For supporting all types of service flows ahierarchical scheduling structure of the bandwidth allocation is proposedfor TDD mode (RR round robin EDF earliest deadline first WFQweighted fair queue)


ARTICLE IN PRESS

The proposed architecture is designed to provide rtPSservice flow packets more chance to meet their deadlineand decrease the delay Apart from checking if the avail-able bandwidth is enough for granting a request the sys-tem has to monitor nominal polling interval nominalpolling jitter and reference time (the time used as a refer-ence to calculate both the generation time and the dead-line of the rtPS data grants) related to the rtPS serviceflows that are admitted The information gathered fromthis monitoring is used to approximate the expected delayof each rtPS connection and the proposed scheduling algo-rithm similar to [16] is used to calculate the deadline Thisdeadline is used by the scheduler to determine if an rtPSpacket is critical or not Preemptive Direct Fair PriorityQueue (PDFPQ) is proposed for the first layer schedulingand total bandwidth distribution The structure is almostidentical to the one shown in Fig 8 the only differencebeing the DFPQ in the first layer is replaced by PDFPQmaintaining four lists


Active List contains non-empty queues whose deficitcounter values are greater than zeroBlocked List contains non-empty queues whose deficitcounter values are either zero or negativeWaiting List contains queues that are empty and theirdeficit counter values are greater than zeroNon-active List contains queues that are empty andtheir deficit counter values are either zero or negative

The rtPS queues both uplink and downlink are non-preemptive queues Other queues can be preempted undercertain conditions If an rtPS packet has a deadline to meetbut will probably fail then that packet is considered criti-cal The PDFPQ defines a lsquolsquoquantum criticalrdquo value for eachnon-preemptive queue Queues are allowed to use this va-lue to serve critical packets only This gives a queue an-other chance to service critical packets There are threescenarios that are not handled in the original DFPQmethod

Scenario 1 A critical packet arrives to the waiting list ofthe non-preemptive queue while the scheduler is servingpackets from one of the preemptive queues Under this sit-uation the DFPQ algorithm will most likely service thecritical packet with the next frame This will cause thepacket to fail meeting its deadlineScenario 2 The deficit counter becomes less than or equalto zero while the scheduler is processing the packets of anon-preemptive queue If a critical packet is waiting tobe serviced at the head of the queue the DFPQ algo-rithm will not service the packet in the current roundScenario 3 A critical packet arrives to the inactive list ofthe non-preemptive queue while the scheduler is servingpackets from one of the preemptive queues The packetwill be served by the DFPQ algorithm However PDFPQwill not serve packets in the inactive list

These scenarios are addressed in the Preemptive DFPQalgorithm proposed by the authors

The simulation compares the improvement in delay andthroughput when using PDFPQ over DFPQ Some assump-tions are made such as total bandwidth is 10 Mbps andeach frame duration is 1 ms The authors simulated thebehavior for four frames each divided into several rtPSand BE packets DFPQ and PDFPQ were applied to all theabove mentioned scenarios and the minimum maximumand average delay were measured and reported for 4 ms(four frames) There is no change in the maximum delayfor both the algorithms Minimum delay improves by800 ls in frame number two and four when PDFPQ is usedThis 800 ls is a significant amount considering the maxi-mum delay recorded is 3600 ls For the first and thirdframes the minimum delay improves by 200 ls Thischange in minimum delay naturally affects the average de-lay accordingly Consequently the results show thatPDFPQ algorithm reduces the delay of critical packets thatcould not have possibly been serviced using the DFPQalgorithm

Throughput of rtPS and BE service flows were also com-pared for both DFPQ and PDFPQ algorithms The results



ARTICLE IN PRESS

show that for DFPQ the throughput for rtPS and BE serviceflows are almost at a constant level (negligible change) forthe simulation duration However when PDFPQ algorithmis implemented throughput for rtPS in the first and thirdframes increases This increase is directly proportional tothe decrease in throughput of BE service flow for therespective frames The authors claim that this decrease inBE service flow is insignificantly small and it will neverexperience starvation

The simulation results are convincing at face value butthe simulation is run only for four frames To observe theimprovement in average delay simulations should be con-ducted over a large number of frames Experiments con-ducted over a longer period will also demonstrate if BEservice flow actually survive starvation when PDFPQ isimplemented

32 Signaling and internetworking

In this section we discuss the research efforts focusingon the QoS signaling mechanism in the MAC sublayer andinternetworking issues with other networks (such as opti-cal and Wi-Fi) The studies covered here propose variousways to improve QoS signaling and create hybrid architec-tures for improving inter-connectivity with existingnetworks

321 An integrated signaling mechanismA fast signaling mechanism proposed by Chen et al [5]

modifies the default signaling mechanism of WiMAX to en-able the system to reduce the initial connection setup timeThe WiMAX standard specifies that service flows can bedynamically added changed or deleted (DSA DSC andDSD messages) and these actions can take a number ofhandshakes between an SS and the BS In contrast to thedefault architecture in the authorsrsquo proposed system theSS sends the DSA message embedded with the BW requestmessages This is illustrated in Fig 9 where in the contextof IntServ architecture [2] the sender initially transmits aPATH message that includes the traffic specification(TSpec) information consisting of bandwidth jitter anddelay requirements This information then can be embed-ded in the subsequent DSA request message Similarlythe DSA response message can contain additional informa-tion such as allocated bandwidth When a new service flowarrives the admission control mechanism accepts it if therequested bandwidth is less than the available bandwidth(the difference between the total capacity and the sum ofall current connections) Under the default architecture

Fig 9 Traffic specification (TSpec) information is embedded in PATH and


the negotiation of QoS parameters between the BS andan SS takes place twice ndash a situation which is avoided inthe authorsrsquo enhanced signalling proposal

The authors developed a simulation platform for evalu-ating their proposal The simulated network used for eval-uation consists of one BS and three SSs The totalbandwidth is 10 Mbps and frame duration is 10 ms whichis divided into 256 minislots For management basic pri-mary and secondary connections 1 Mbps of bandwidth isreserved DSA DSC and DSD message transmission delaysare set to 10 ms (even though admission control and reser-vation related processing time can vary due to perfor-mance of the BSs and SSs for the purpose of thissimulation work it is fixed at 10 ms)

The graphs presented in the study illustrate that thesetup time for the proposed signalling implementation isinsensitive to offered load it remains unchanged atapproximately 75 ms as the rate of frame arrival increasesfrom one to three per time unit Conversely with unmod-ified (traditional) WiMAX signalling the minimum setuptime starts at just over 100 ms and increases to around200 ms and 700 ms for frame arrival rates of two and threeper time unit respectively This shows a significantimprovement in the setup time

322 WiMAX and optical network integrationThe study presented in [27] proposes a bandwidth allo-

cation scheme for Video-on-Demand (VoD) services overan experimental integrated optical and WiMAX networkThe end-to-end connection between the VoD client andserver is distributed over Synchronous Optical Network(SONET) and WiMAX links The SONET ring is the backboneused for connecting the WiMAX BSs and VoD clients

As WiMAX BSs can cater for up to 75 Mbps data rate(shared among all users) if only one STS-1 link is providedto each BS node congestion will be experienced whenevertotal user demand per BS exceeds the STS-1 data rate of5184 Mbps If two links are provided that will make thesystem less efficient and not cost effective This researchproposes a solution that overcomes these obstacles touse one STS-1 link per BS and shift system operation be-tween an Erlang-C and an Erlang-B queueing modeldepending on the network load The three possible scenar-ios are

(1) Average offered load is less than the link capacity(single STS-1 circuit) All requests are queued andserved accordingly The behavior of the BS subnetis characterized by the Erlang-C delay model

DSAreq messages in the proposed mechanism by Chen et al [5]


Fig 10 The architecture proposed by Gakhar et al [13] The radiogateway module functions as an SS for the IEEE 80216 network and anaccess point for the IEEE 80211e wireless LAN

2 TID value 1 and 2 are assigned to access category queue 0 0 and 3 toqueue 1 4 and 5 to queue 2 and 6 and 7 are assigned to queue 3 [20 Table20i]


ARTICLE IN PRESS

(2) Average offered load is greater than the capacity of asingle STS-1 circuit Queueing of the infinite numberof requests makes the system unstable Hence extrapackets need to be dropped and the behavior of theBS subnet is characterized by the Erlang-B delaymodel

(3) Average offered load is greater than the capacity of asingle STS-1 circuit but less than two STS-1 circuitsIt is reasonable now to queue all unsatisfiedrequests The behavior of the BS subnet then followsthe Erlang-C model

The experiment looks into two possible heuristicsolutions

Maximum Utilization The algorithm picks the BS inthe non-increasing order of utility and sequentiallyallocate sufficient bandwidth to itMaximum Efficiency The algorithm picks the BS withthe maximum cost effectiveness first Cost effectivenessis defined by the larger ratio of the utility over thebandwidth needed between the two types of band-width allocation

In the simulation 1000 nodes are used and the arrivalrequests follow the MMm models and each VoD requestconsumes 1 Mbps The results are presented in a graph(aggregate utility value against the VoD server capacity)The results show that when the capacity is small simplegreedy approach does not work well but the algorithm pro-posed optimizes the utility function and performs betterAlthough the authors did not provide a quantitative analy-sis our study of the graph shows an improvement by 25when the capacity is small The results also show that Max-imum Efficiency heuristic is not sensitive to capacity vari-ations and outperforms other greedy algorithms

323 WiMAX and Wi-Fi integration via mapped QoS classesThe study of Gakhar et al [13] proposes an architecture

to achieve differentiated QoS for end-to-end services in ahybrid WiMAX and Wi-Fi (80211e) network It maps QoSrequirements of an application that originates from a Wi-Fi network to a WiMAX network and assures transfer ofdata with appropriate QoS

80211 abg offers best-effort service only In contrastthe 80211e [18] was designed to ensure QoS differentia-tion among packet flows generated by applications Itintroduces the Hybrid Coordination Function (HCF) whichenhances the DCF and PCF access schemes of 80211 HCFmultiplexes between two channel access methods for shar-ing the medium Enhanced Distributed Channel Access(EDCA) which is a decentralized algorithm and a central-ized algorithm called HCF Controlled Access (HCCA) fortightly controlled frame transmissions Varying degreesof QoS at the MAC sublayer of 80211e can be providedby either of these mechanisms [24]

Prioritized QoS through service differentiation withEDCA Frames are segregated into classes and framesbelonging to the same class receive best-effort-within-class service while different classes receive different


grades of service in aggregate Absolute guarantees ofQoS parameters like delay or loss are not providedThus this is also called lsquolsquobetter than best-effortrdquo serviceand is suitable for elastic trafficParameterized QoS through per-flow time reserva-tion with HCCA MAC-level flows are defined and eachflow is guaranteed a certain fraction of time duringwhich a node (AP or Wi-Fi client) can transmit Thestandard also includes means for admission controland reservation signaling at MAC level between a Wi-Fi client and AP This mechanism provides tightly con-trolled QoS

In 80211e classification of traffic is achieved throughthe introduction of access categories for EDCA and trafficstreams for HCCA HCF defines four access category queuesand eight traffic stream queues at MAC sublayer When aframe arrives at MAC sublayer it is tagged with a trafficpriority identifier (TID) according to its QoS requirementsA frame with TID value between 0 and 7 is assigned to oneof the four2 access category queues Similarly a frame with aTID value of 8 to 15 is assigned to one of the eight trafficstream queues [26]

The authors in their study create a mapping mecha-nism between the traffic parameters of the IEEE 80216and IEEE 80211e networks Their approach is depicted inFig 10 The Radio Gateway simultaneously performs thefunctions of an IEEE 80216 SS and an IEEE 80211e accesspoint and the Mapping Module is responsible for choosingthe most suitable class for traffic flowing between the twosystems For the QoS mapping the authors propose twoapproaches The first one called lsquolsquoprioritized mappingrdquo issimilar to the Differentiated Services architecture [1] Inthis kind of mapping application flows coming from anIEEE 80211e network are mapped to a corresponding traf-fic class in an IEEE 80216 network and vice versa In thesecond kind of mapping called per-flow lsquolsquoparameterizedmappingrdquo which resembles the Integrated Services archi-tecture [2] optionalmandatory traffic parameter require-ments for a traffic stream are used to find the mostsuitable traffic class (C1 to C4 as shown in Table 4)


Table 4Parameterized mapping function performed by the Mapping Module traffic classes and their typical usage

Traffic class Typical usage IEEE 80211e IEEE 80216 Remarks

C1 Constant bit rate (CBR)with real-time traffic

Peak data rate Maximum sustained traffic rate Applications like real-time audiovideo Thedesirable characteristics for this class are verylimited packet losses minimum latency delaysand very little jitter

Delay bound Maximum latencyData rate + delay bound Tolerated jitter

C2 Variable bit rate (VBR)with real-time traffic

Maximum data rate Minimum reserved traffic rate Examples of traffic for this class include videoon demand (streaming) and variable ratevoice-over-IP Packet loss minimum latencydelay and jitter limits apply to such trafficwithin more relaxed bounds as compared toClass C1

Peak data rate Maximum sustained traffic rateDelay bound Maximum latencyBurst size Maximum traffic burst

C3 VBR with precious data Minimum data rate Minimum reserved traffic rate Can be used for traffic types like large data filetransfersPeak data rate Maximum sustained traffic rate

User priority Traffic priorityBurst size Maximum traffic burst

C4 Unspecified type Peak data rate Maximum sustained traffic rate Caters for best-effort type traffic such as Webaccess email communication etcUser priority Traffic priority

Fig 11 User equipment protocol stack proposed in [30]


ARTICLE IN PRESS

The authors do not elaborate the circumstances underwhich each mapping model should be used pointing outthat handling of these mappings is implementation depen-dent It may be assumed that the type of traffic expected tobe carried in such a hybrid network (eg predominantlyelastic or a mix with a sizeable component of delay-sensi-tive traffic) would be the determining factor for the choiceof mapping model

In the paper the authors also discuss the furtherimprovements required for their approach Unfortunatelythere are no experimental results provided in the studyto quantify the performance of the architecture proposed

324 A QoS integration model for WLANs and WiMAX usingmedia independent handover

Focusing on a heterogeneous network consisting of IEEE80211e WLAN and IEEE 80216d WiMAX nodes Roy et al[30] propose a mechanism that supports Always Best-Con-nected (ABC) QoS integration In this ABC QoS integrationmethod a client can seamlessly switch between WLANand WiMAX networks and vice-versa without compromis-ing QoS during the handover The work also incorporatesthe IEEE 80221 draft standard where a lsquolsquoL25rdquo layer is de-fined to execute the media independent handover (MIH)that occurs between multiple access networks The stan-dard also specifies MIH to attain service continuity withguaranteed QoS during handover The paper proposes amechanism where a drop in measured user QoS parame-ters in one network will trigger a MIH to switch to theother network The architecture places a Generic VirtualLink Layer (GVLL) to reside over the MAC sublayer TheGVLL triggers the MIH based on various user QoS parame-ters such as throughput packet loss and delay The user isalways connected to the network with the best QoSsupport

Fig 11 illustrates the proposed architecture the userequipment is equipped with multiple interfaces to supportboth access networks Layers including Mobile IP andabove do not have the knowledge that there are multipleMAC sublayers with their corresponding MAC addressesWhen a higher layer packet arrives destined for the MAC


sublayer the GVLL sends the packet to the appropriateMAC sublayer depending on the best available QoS Packetloss and delay are the chosen QoS parameters to determinethe superior network The GVLL has three primaryfunctions

(1) It is the virtual MAC sublayer interface to the higherlayers

(2) Collects information from multiple MAC sublayersand triggers the MIH if the handover condition isachieved

(3) Receives higher layer packets and forwards them toany particular MAC to which it is attached at thatmoment

According to the proposed architecture the GVLL trig-gers MIH under two circumstances

Whenever a new connection (data or voice) has beenadmitted When a new connection admission isrequested the GVLL simultaneously sends requests toboth access networks If the response is from only onenetwork then the user decides if the connection shouldbe established If the response is from both networksthen the GVLL decides between the available interfacesdepending on the best QoS support If the QoS supportfrom both networks is the same then the signal to noiseratio (SNR) is determined to select the interface As thestudy does not implement SNR in the simulation undersuch circumstances it defaults to WLAN network



ARTICLE IN PRESS

Whenever the QoS guarantee falls beyond thethreshold QoS parameters are continuously monitoredand if they fall below the threshold a request to otheravailable networks is sent out the one with the bestQoS support is selected and MIH is triggered to initiatethe handover

HCF MAC functionality is used to guarantee QoS support inthe WLAN domain and TDMA based MAC has been used inthe WiMAX domain Admission control policy accepts acall if there are sufficient resources available to supportthe mean data rate of the call which is assumed to be ofVBR type

The simulation scenario in the study consists of twoWLAN access points inside a single WiMAX coverage areawith 18 user stations Each network is connected to a back-bone individually Link capacities for each WLAN and Wi-MAX network are 12 Mbps and 24 Mbps respectively

In the first simulation scenario handovers betweenWLAN and WiMAX are not supported Users can only at-tach to their respective networks and WLAN users canroam between the two WLAN access points but a WLANcall can not be handed over to the WiMAX BS and vice-ver-sa 9 users are directly connected to the WiMAX networkand the remaining 9 are WLAN users This is the singleinterface scenario where GVLL is not implemented

In the second simulation scenario the GVLL scheme isadopted to support interoperability between the two tech-nologies During the call initiation the GVLL sends requestto both a WLAN access point and WiMAX BS and choosesthe best among them according to the responses it getsHandovers between WLAN and WiMAX are also supported

The results compare the performance of the twoscenarios and show noticeable improvement on GVLLimplementation when the network is heavily loaded animprovement of roughly 18 As the number of calls admit-ted is more in case of GVLL the system throughput reflectslikewise The results further show that as the network getssaturated with the implementation of GVLL the averagedelay improves by approximately 10

4 Mechanisms for QoS provision in WiMAX based meshnetworks

In a mesh WiMAX network a lsquolsquomesh base stationrdquo(Mesh BS) undertakes the role of a BS and provides theconnection to other communication networks There canbe multiple mesh BSs in a network and other nodes areknown as mesh subscriber stations (Mesh SSs) In contrastwith point-to-multipoint WiMAX networks the standarddoes not define scheduling services for mesh WiMAXnetworks In point-to-multipoint mode the SSs are underthe direct control of the BS In Mesh mode the uplinkand downlink is not clearly separated and SSs can commu-nicate with each other without communicating with theBS The transmission opportunities in the control subframeand the data minislots in the data subframe are separatedThe nodes compete for the control channel accessand the contention outcome does not effect the datatransmission


The QoS provision in mesh WiMAX networks is morechallenging and very few researchers have thus far focusedtheir efforts on this area In the following sections we pres-ent a couple of representative studies which propose waysto improve QoS signaling mechanisms and create hybridarchitectures for improving inter-connectivity with exist-ing networks

41 Routing and admission control for mesh WiMAXnetworks

In [36] Tsai and Wang propose a routing method usingShortestndashWidest Efficient Bandwidth (SWEB) as a metricfor distributed coordinated WiMAX mesh mode alongwith a token bucket based admission control (TAC) algo-rithm The study uses the token bucket mechanism as itworks well for smoothing the burstiness of packet flowsand helps in estimating the required bandwidth

The SWEB metric considers three parameters

Packet Error Rate can be retrieved by exchanging theMSH-DSCH messages Each MSH-DSCH message isassociated with a unique sequence number there anylost or damaged messages can be detectedLink Capacity can be determined by the burst profileindicated in the MSH-NCFG messageHop Count is included also in the MSH-NCFG messagesfrom a station to the BS

Based on these parameters SWEB is retrieved and thepath with the largest SWEB is chosen

TAC has two essential components

Bandwidth Estimation It is estimated using the tokenbucket based admission control and it is dependent ontoken rate and bucket size associated with a given con-nection and frame lengthAlgorithm Determination The estimated bandwidth isused to determine the admission control algorithm Toprevent starvation of lower-priority traffic minimumusage of timeslots by each connection is defined Thealgorithm is determined through the followingprocedure

(1) When a new bandwidth request occurs the sourcenode computes its available bandwidth as the total

empty slot number

(2) The station that handles the request checks ifrequested bandwidth is less than available band-width If yes it goes to next step otherwise goes toStep 4

(3) By comparing the current and minimum usage ofother traffic classes the station determines if theflow should be downgraded

(4) If the current usage exceeds the minimum usage ofthe traffic class the station rejects the flow Or elseit goes to next step

(5) The station checks the timeslots used by down-graded flows in the order of BE VBR or CBR Therequest is rejected if there are no such timeslots



ARTICLE IN PRESS

Else it sets these timeslots empty which means topreempt these timeslots It then grants the timeslotsand updates the value of available bandwidth

The study reports the results of simulation based exper-iments conducted on a 16 node topology with varioustypes of traffic (BE VBR and CBR) In terms of the physicaland data link layer parameters QPSK modulation is as-sumed the simulation area is 16 km2 the radio range ra-dius is set to 15 km while frame length is chosen to be8 ms The data rate used for CBR traffic is 64 kbps with960 bit packet size and a packet interval of 15 ms VBR traf-fic data rate is 400 kbps with a mean packet size of 16000bits and a packet interval of 40 ms Finally the offered BEdata rate is 1000 kbps with a packet size 8000 bits andpacket interval of 8 ms The results based on SWEB arecompared against previous studies that used ExpectedTransmission Count (ETX) and Shortest Path First algo-rithm The graphs show that the throughput for ETX ishighest because it selects a route with the lowest packeterror rate However as ETX does not take the hop countinto account it causes higher packet delays SWEB hadthe best performance with respect to jitter but from ourobservation the improvement is significant only after thenumber of flows is more than 20

Under normal circumstances best-effort traffic experi-ences preemption from higher priority traffic classes Butwhen TAC is used the best effort flows gain the advantageof having the guaranteed minimum throughput Howeverthe observations on the simulation results presented as agraph in the study reveal that to prevent the starvationof best-effort flows variable bit rate traffic throughput issacrificed The results further show that when TAC is notused 12 of variable bit rate packets exceed the delayrequirements when the number of flows is 25 This is re-duced to 7 when TAC is used

42 A QoS differentiation scheme for mesh WiMAX networks

Zhang et al propose a scheme to achieve QoS differen-tiation in the WiMAX mesh mode [40] In their work theauthors introduce the distributed scheduling conceptand also develop a new formula for its theoretical evalua-tion in random topologies

In distributed scheduling a node can transmit in anyslot during the eligibility interval and has to contend withother nodes This contention is irrespective of the servicetype and its priority To overcome this drawback the studyproposes a scheme to prioritize traffic and enable the QoSdifferentiation by varying the eligibility intervals for differ-ent traffic classes

The formula for evaluating the scheme is derived fortwo different topologies co-located scenario (all nodesare one-hop neighbors of each other) and general topology(multihop neighborhood) The numerical results show theeffectiveness of achieving differentiated QoS in both ofthese topologies with all nodes equally partitioned intothree priority classes (1 2 and 3) the proposed scheme isable to ensure that class 1 has the shortest and 3 the lon-gest delay


5 QoS Issues in evolutions of the WiMAX standard

Several important evolutions of the WiMAX standardare currently in progress and for each of these we nowexamine the issues relating to QoS support

51 WiMAXrsquos road to 4G

With the goal of improving performance of the currentrelease of Mobile WiMAX [19] two separate evolution ef-forts have been under way since the beginning of 2007

511 WiMAX Forum Release 15The WiMAX Forum with its Release 15 evolution pro-

ject is aiming for a short time horizon (targeting systemsdeployed in 200910 timeframe) by trying to minimize thechanges to the current IEEE Network Release 10 specifica-tion which supports the IEEE 80216e-2005 standard [39]

With regards to QoS support Release 10 only offers ba-sic functionality in the form of static (ie pre-provisioned)QoS and an optional rather than mandatory radio resourcemanager Static QoS implies that the SS may not modifyparameters of the service flows already provisioned bythe system nor create any service flows dynamically Thisissue is addressed in Release 15 through the incorporationof dynamic QoS functionality whereby an SS may dynam-ically set up a flow through DSA transactions as discussedin Section 22 [10]

Another QoS enhancement in Network Release 15 is thepolicy and charging (PCC) functionality planned to be fullycompatible with the 3GPP Release 7 specification [29] Pol-icies are rules which are triggered by certain types of trafficor user behavior in the network Combining such rules withthe ability to dynamically assign QoS to user flows PCC be-comes a powerful enabler of differentiated QoS featuressuch as (i) QoS based on accumulated usage and (ii) QoSbased on aggregate network load In (i) a particular useror application could be dynamically assigned an inferiorQoS class (eg lower-priority or a small traffic-shaped datarate) after reaching a volume threshold in bytes Similar dy-namic de-prioritization of a targeted user or applicationcould be undertaken in (ii) with the policy trigger in thiscase being aggregate network load (eg protecting higherpriority users when network utilization exceeds 80)

The final enhancement in Network Release 15 whichmay be considered a direct enabler in the provision of dif-ferentiated QoS is the inclusion of telephony VoIP On theair interface this is supported in the 80216REV2 revisionof the standard by a VoIP specific optimization called lsquolsquoper-sistent schedulingrdquo [11] More broadly speaking all of thevarious fixes and minor amendments necessary to supportthe Network Release 15 specification are incorporated inthe 80216REV2 revision of the mobile WiMAX standardwhich lsquolsquocombines the IEEE 80216-2004 base standard plusIEEE 80216efg amendments and related corrigendardquo[10] Compared to use of the default MAC protocol theauthors of [11] report an increase in WiMAX VoIP usercapacity of approximately 16 under this MAC sublayerpersistent allocation modification due to the significantreduction of signalling message overheads



ARTICLE IN PRESS

In summary while Network Release 10 was primarilyaimed at carriage of BE data traffic or static QoS-enabledflows the Network Release 15 and associated80216REV2 air interface enhancements described abovedirectly enable the dynamic provision of application- anduser-based QoS differentiation while maintaining efficientnetwork utilization

512 IEEE 80216 Working Group 80216mThe IEEE 80216 Working Group with its IEEE 80216m

project and proposed future standard has its sights set ona more ambitious longer-term (201112) goal to funda-mentally enhance the performance of mobile WiMAX sothat it meets the requirements of the ITUrsquos international4G standard known as IMT-Advanced (the successor ofthe IMT-2000 3G standard) [23] As such most of the focusin IEEE 80216m is on deriving raw physical layer perfor-mance improvements [10] which would only indirectlyimpact QoS by improving the performance of all QoS clas-ses Such performance improvements would be achievedusing techniques such as increased spectral efficiencythrough more advanced and higher-order Multiple InputMultiple Output (MIMO) antenna systems lower framingoverheads at the physical and data link layers and widerband carriers (eg 20 MHz)

From the set of IEEE 80216m enhancements which willdeliver better performance for all users we single out twoin particular [1039] because they may be viewed as lsquolsquodi-rect enablersrdquo for improved QoS differentiation in WiMAXThe first of these is lower latency which will be achieved inIEEE 80216m by a leaner faster MAC and signalling frame-work The expectation is that this will lead to less complexand lower-latency QoS signalling and hence a better abilityto provide differentiated QoS The second of these enablersis the planned provision of seamless low-latency hand-overs between WiMAX and other radio access technologiessuch as Wi-Fi allowing true lsquolsquomultimedia session continu-ityrdquo ndash an important aspect of providing ent-to-end QoS formultimedia services

52 WiMAX Multihop Relays IEEE 80216j

The goal of the emerging IEEE 80216j WiMAX MultihopRelay (MR) standard [21] is to increase radio coverage userthroughput and capacity of traditional 80216e-2005 Wi-MAX networks The proposed 80216j standard aims toachieve this goal by specifying PHY and MAC sublayerenhancements for licensed bands of spectrum that enablethe operation of relay stations (RS) Note that the SS spec-ifications are not changed

The two main difficulties which are found to be com-mon to each of the impacted QoS features of the proposed80216j standard (discussed below) are (i) the increasedcomplexity of the whole-of-path signalling as opposed toa single message exchange in single-hop networks and(ii) the increased latency associated with relaying informa-tion (both signalling and user data) across multiple hops

A challenge for designers of future MR WiMAX net-works will be to compute near-optimal tradeoffs betweenthe increased complexity and latency associated with


relaying information across multiple hops and the benefitsafforded by increased radio coverage without the cost ofrolling out full base stations

What follows is a summary of the impacts of multihoprelaying on the key WiMAX QoS features as described inthe latest baseline document of the proposed IEEE80216j standard

521 Impact on scheduling servicesUnlike in single-hop networks where bandwidth is

granted by a BS directly to its attached SSs in an MR sys-tem this allocation is cascaded down in hierarchical fash-ion In the case of the UGS scheduling service this meansthat to meet a UGS service flowrsquos need the Multihop Relaybase station (MR-BS) and RSs along the path have to grantfixed size bandwidth to their subordinate nodes on a real-time periodic basis

Similarly unlike single-hop networks where a BS di-rectly polls its attached SSs an MR system requires thatthe polling must be cascaded down in hierarchical fashionIn the case of the rtPS and ertPS scheduling services thismeans that in order to meet an rtPSertPS service flowrsquosneed the MR-BS and RSs along the path must poll theirsubordinate nodes on a real-time periodic basis

522 Impact on bandwidth allocation and requestmechanisms

Another distinguishing feature of MR WiMAX networksis that an RS may combine (i) bandwidth requests arrivingfrom its subordinate neighbor RSs during a given period oftime and (ii) bandwidth needs of packets in its local queueinto one lsquolsquoaggregatedrdquo bandwidth request header per QoSclass In order to minimize the additional delays intro-duced by this relay-based procedure the RS is allowed totransmit a bandwidth request header shortly after it re-ceives a bandwidth request header from one of its down-stream stations instead of waiting for the actual packetsto arrive The timing is chosen to yield an uplink allocationat the RS which immediately follows the arrival of the re-layed packets from the downstream station

523 Impact on dynamic QoS proceduresIn an MR WiMAX network with distributed scheduling

a BS cannot immediately admit a service flow and send aDSA-RSP message to the requesting SS as in the case of tra-ditional single-hop WiMAX networks Instead the proce-dure becomes considerably more complex due to theneed for the BS to discover if all of the RSs in the path tothe SS have sufficient resources to support the dynamicallyrequested QoS The discovery procedure begins with the BSsending a DSA-REQ message to its subordinate RS This RSthen sends its own DSA-REQ message to its subordinatedneighboring RSs with this hierarchical cascade continuingdown until the access RS is reached

6 Analysis and concluding remarks

The studies discussed in this paper examine various as-pects of QoS architecture and QoS differentiation for two



ARTICLE IN PRESS

key types of WiMAX networks point-to-multipoint andmesh The paper by Cicconetti et al [7] provides an imple-mentation of a QoS mechanism with basic traffic manage-ment Significant improvement with regards to trafficmanagement and admission control is proposed by Wongt-havarawat and Ganz [38] with a focus on uplink packetscheduling and traffic policing at the SS Although the sim-ulation results only take rtPS and BE traffic into consider-ation the research provides adequate information fordealing with other classes of traffic However one aspectof the admission control implementation in [38] whichhas room for improvement is maintaining fairness be-tween all classes of traffic The current implementationfails to prevent instances where one service class can dom-inate the entire link bandwidth

A successful implementation of a WiMAX-customizedWFQ2+ algorithm is reported by Shang and Cheng [32]Their approach of implementing lsquolsquohardrdquo or lsquolsquosoftrdquo QoS canbe integrated with [38] for further optimization There isample scope for further research into an optimal schedul-ing algorithm from the many available candidates

Using the fragmentation and aggregation capabilities ofMAC SDUs in multiple PDUs Sengupta et al [33] provide avery good solution for maintaining differentiated QoS forstreaming media Their approach of rearranging MAC SDUsbefore transmission along with a feedback mechanismprovided significant improvement in performance

The two-tier scheduling algorithm (2TSA) proposed byChan et al [4] improves network performance significantlycompared to earlier approaches which used strict-priorityscheduling (such as [37]) In 2TSA the first-tier allocationalgorithm is category based and the second-tier allocationis weight based When compared with the [37] algorithmthe simulation results show that 2TSA can guarantee con-nectionsrsquo bandwidth demands avoid starvation of lower-priority service class and achieve a better degree of fair-ness Other QoS metrics such as delay or delay jitter are leftfor future research

The Preemptive Direct Fair Priority Queue (PDFPQ)scheduling method implemented by Safa et al [31] im-proves minimum and average delay for rtPS traffic as com-pared to a previous proposal [6] that used the non-Preemptive version (Direct Fair Priority Queue DFPQ)However one significant drawback is the drop in through-put of BE traffic Although BE traffic does not face starva-tion PDFPQ will cause a slower BE traffic response thanin DPFQ The study therefore leaves room for future workon methods which simultaneously seek to minimize thethroughput degradation of BE traffic while still improvingdelay for rtPS traffic

Chen et al [5] presented a technique embedding DSADSC and DSD messages inside the BW-Request messagewhich showed a significant improvement in connectionsetup time However their approach can compromise otherpotential capabilities of the network If a network is to pro-vide multiple services like VoIP video and data it is impor-tant for admission control to know the service request fromeach SS before it receives the BW-Request To provide mul-tiple services the admission control needs to consider fair-ness for all classes of traffic Therefore if a SS has multipleservice requests it should be able to partially accept some


of the requests (to maintain fairness) As BW-Request mes-sages only deal with aggregates it will not be possible forthe WiMAX admission control to partially accept some ofthe requests This would prevent the system from providingdifferentiated admission control running contrary to ourstated goal of QoS differentiation in a multi-service wirelessnetwork Conversely in a network that caters for only oneclass of traffic such an embedded signaling approachwould work without any problems

A successful internetworking solution between SONETand WiMAX is provided by Lin et al [27] They overcomethe problem of bandwidth over- or under-utilization (dueto mismatch of a WiMAX BS and an STS-1 backhaul link)by implementing a heuristic approach The heuristic isbased on maximizing utilization and efficiency dependenton the measured network saturation level

A WiMAX and QoS-enabled Wi-Fi (IEEE80211e) inter-networking solution is illustrated in a paper by Gakharet al [13] The QoS management facility provided by80211e is successfully exploited by implementing a Map-ping Module Although the paper does not provide anysimulation results to verify the possible outcome in theorythe solution sounds plausible The implementation is notavailable for other popular Wi-Fi variants which do notsupport QoS at the MAC sublayer (ie 80211abg)

QoS integration model for WLAN and WiMAX of Royet al [30] is another promising WiMAX internetworkingstudy that provides scope for further developments Thestudy shows how Generic Virtual Link Layer (GVLL) canbe used for interoperability between multiple standardsFactors such as high speed mobility and coverage presentopportunities for future work The study can also be ex-panded by including other wireless networks such asHSDPA EDGE and EV-DO

Zhang et al [40] make a significant contribution withtheir QoS Differentiation Scheme for WiMAX mesh modeThe probabilistic methodology evaluating the schedulingperformance in a general topology is a novel idea Thenumerical results illustrate performance improvements inboth the collocated and general topologies

In closing in this survey paper we illustrated the gen-eral framework as well as many specific approaches forimplementing QoS differentiation in the MAC sublayer ofa WiMAX network A brief explanation of the WiMAXMAC architecture was given before a number of researchstudies were explored Each of these studies was placedinto one of three categories The lsquolsquoPacket scheduling andadmission controlrdquo category looked into the way QoSimplementation improves user service quality and networkefficiency The lsquolsquoSignaling and integrationrdquo category focusedon how WiMAX networks can be deployed alongside othernetworks to meet various requirements The third categorylsquolsquoQoS in WiMAX mesh networksrdquo focused on research intothe distributed methods of signalling and scheduling re-quired to achieve QoS differentiation in the mesh variantof WiMAX networks We also examined the issues associ-ated with provision of differentiated QoS services in futureevolution of the WiMAX standard Finally we comparedand contrasted the various studies analyzing the potentialand limitations of each including options for future work inthis important area of networking research



ARTICLE IN PRESS

Acknowledgement

Many thanks to Hyoung-Kyu Lim and Jungshin Park ofSamsung for their valuable comments on improving thecontents of the paper

References

[1] S Blake D Black M Carlson E Davies Z Wang W Weiss RFC 2475an architecture for differentiated services 1998 URL referencelthttpwwwietforgrfcrfc2475txtgt

[2] R Braden D Clark S Shenker Integrated services in the internetarchitecture an overview 1994 URL reference lthttpwwwietforgrfcrfc1633txtgt

[3] JCR Bennett H Zhang Hierarchical packet fair queueingalgorithms IEEEACM Transactions on Networking 5 (5) (1997)675ndash689

[4] L Chan H Chao Z Chou Two-tier scheduling algorithm for uplinktransmissions in IEEE 80216 broadband wireless access systems inProceedings of the International Conference on WirelessCommunications Networking and Mobile Computing (WiCOMrsquo06)September 2006 pp 1ndash4

[5] J Chen W Jiao Q Guo An integrated QoS control architecture forIEEE 80216 broadband wireless access systems in Proceedings ofthe IEEE Global Telecommunications Conference (GLOBECOMrsquo05) StLouis USA IEEE Communications Society November 2005

[6] J Chen W Jiao H Wang A service flow management strategy forIEEE 80216 broadband wireless access systems in TDD mode inProceedings of the 2005 IEEE International Conference onCommunications (ICCrsquo05) Seoul Korea IEEE CommunicationsSociety May 2005 pp 3422ndash3426

[7] C Cicconetti L Lenzini E Mingozzi C Eklund Quality of servicesupport in IEEE 80216 networks IEEE Network 20 (2006) 50ndash55

[8] A Demers S Keshav S Shenker Analysis and simulation of a fairqueueing algorithm in Proceedings of the CommunicationsArchitectures and Protocols Symposium ACM September 1989 pp1ndash12

[9] H Dewing S Potter Implementing QoS solutions in enterprisenetworks February 2002 URL reference lthttpwwwtmcnetcomit02020202inimhtmgt

[10] K Etemad Overview of WiMAX technology and evolution IEEECommunications Magazine 46 (10) (2008) 31ndash36

[11] M Fong R Novak S McBeath R Srinivasan Improved VoIP capacityin mobile WiMAX systems using persistent resource allocation IEEECommunications Magazine 46 (10) (2008) 50ndash56

[12] WiMAX Forum Business case models for fixed broadband wirelessaccess based on WiMAX technology and the 80216 standardOctober 2004 URL reference lthttpwwwwimaxforumorgtechnologydownloadsWiMAX-The_Business_Case-Rev3pdfgt

[13] K Gakhar A Gravey A Leroy IROISE a new QoS architecture forIEEE 80216 and IEEE 80211e interworking in Proceedings of theSecond International Conference on Broadband Networks(Broadnetsrsquo05) Boston USA October 2005 pp 607ndash612

[14] L Georgiadis R Guerin A Parekh Optimal multiplexing on a singlelink delay and buffer requirements IEEE Transactions onInformation Theory 43 (5) (1997) 1518ndash1535

[15] EL Hahne RG Gallager Round Robin scheduling for fair flowcontrol in data communication networks in Proceedings of the IEEEInternational Conference on Communications (ICCrsquo86) TorontoCanada IEEE Communications Society March 1986 pp 103ndash107

[16] M Hawa DW Petr Quality of service scheduling in cable andbroadband wireless access systems in Proceedings of the 10th IEEEInternational Workshop on Quality of Service IEEE May 2002 pp247ndash255

[17] IEEE IEEE standard for local and metropolitan area networks Part16 Air interface for fixed broadband wireless access systems 2004URL reference lthttpstandardsieeeorggetieee802download80216-2004pdfgt

[18] IEEE Wireless LAN medium access control (MAC) and physical layer(PHY) specifications Part 11 Amendment 7 medium access control(MAC) quality of service (QoS) enhancements 2004

[19] IEEE IEEE standard for local and metropolitan area networks Part16 Air interface for fixed and mobile broadband wireless accesssystems (amendment and corrigendum to IEEE Std 80216-2004)2005 URL reference lthttpstandardsieeeorggetieee802download80216e-2005pdfgt


[20] IEEE IEEE standard information technology ndash telecommunicationand information exchange between systems ndash local andmetropolitan area networks ndash specific requirements ndash Part 11Wireless LAN medium access control (MAC) and physical layer(PHY) specifications ndash amendment 8 medium access control(MAC) quality of service enhancements 2005 URL referencelthttpstandardsieeeorggetieee802download80211e-2005pdfgt

[21] IEEE Baseline document for draft standard for local andmetropolitan area networks Part 16 Air interface for fixed andmobile broadband wireless access systems (Multihop Relayspecification) 2007 URL reference lthttpwwwieee802org16relaydocs80216j-06_026r4zipgt

[22] European Telecommunications Standards Institute General aspectsof quality of service and network performance in digital networksincluding ISDN Technical report ETR 003 ed1 ETSI 1990

[23] ITU ITU-R recommendation M1645 framework and overallobjectives of the future development of IMT-2000 and systemsbeyond IMT-2000 2003 URL reference lthttpwwwituintrecR-REC-M1645egt

[24] A Kumar D Manjunath J Kuri Wireless Networking MorganKaufmann (2008)

[25] M Katevenis S Sidiropoulos C Courcoubetis Weighted round-Robin cell multiplexing in a general-purpose ATM switch chip IEEEJournal on Selected Areas in Communications 9 (8) (1991) 1265ndash1279

[26] H Labiod H Afifi C De Santis Wi-Fi Bluetooth Zigbee and WiMAXSpringer 2007

[27] P Lin C Qiao T Wang J Hu Optimal utility-based bandwidthallocation over integrated optical and WiMAX networks inProceedings of the Optical Fiber Communication Conferenceand the 2006 National Fiber Optic Engineers Conference March2006

[28] Third Generation Partnership Project 3GPP TS 25308 high speeddownlink packet access (HSDPA) overall description stage 2 URLreference lthttpwww3gpporgftpspecshtml-info25308htmgt

[29] Third Generation Partnership Project Technical specification groupservices and system aspects policy and charging controlarchitecture (release 7) 3GPP TS 23203 V750 (2007-12) URLreference lthttpwww3gpporgFTPSpecshtml-info23203htmgt

[30] RJ Roy V Vaidehi S Srikanth Always best-connected QoSintegration model for the WLAN WiMAX heterogeneous networkin Proceedings of the First International Conference on Industrialand Information Systems August 2006 pp 361ndash366

[31] H Safa H Artail M Karam R Soudah S Khayat New schedulingarchitecture for IEEE 80216 wireless metropolitan area networkin Proceedings of the IEEEACS International Conference onComputer Systems and Applications (AICCSArsquo07) May 2007 pp203ndash210

[32] Y Shang S Cheng An enhanced packet scheduling algorithm for QoSsupport in IEEE 80216 wireless network in Third InternationalConference on Networking and Mobile Computing (ICCNMCrsquo05)Zhangjiajie China August 2005 pp 652ndash661

[33] S Sengupta M Chatterjee S Ganguly R Izmailov Exploiting MACflexibility in WiMAX for media streaming in Proceedings of theSixth IEEE International Symposium World of Wireless Mobile andMultimedia Networks (WoWMoM 2005) Taormina Italy IEEEComputer Society June 2005 pp 338ndash343

[34] M Shreedhar G Varghese Efficient fair queuing using deficit roundRobin IEEE Transactions on Networking 4 (3) (1996) 375ndash685

[35] D Stiliadis A Varma Latency-rate servers a general model foranalysis of traffic scheduling algorithms IEEEACM Transactions onNetworking 6 (5) (1998) 611ndash624

[36] TC Tsai CY Wang Routing and admission control in IEEE 80216distributed mesh networks in IFIP International Conference onWireless and Optical Communications Networks (WOCNrsquo07)Singapore 2007 pp 1ndash5

[37] K Wongthavarawat A Ganz IEEE 80216 based last mile broadbandwireless military networks with quality of service support inProceedings of the IEEE Military Communications Conference vol 2October 2003 pp 779ndash784

[38] K Wongthavarawat A Ganz Packet scheduling for QoS support inIEEE 80216 broadband wireless access systems InternationalJournal of Communication Systems 16 (1) (2003) 81ndash96

[39] F Wang A Ghosh C Sankaran P Fleming F Hsieh S Benes MobileWiMAX systems performance and evolution IEEE CommunicationsMagazine 46 (10) (2008) 41ndash47

[40] Y Zhang J Zheng W Li A simple and effective QoS differentiationscheme in IEEE 80216 WiMAX mesh networking in Proceedings of



ARTICLE IN PRESS

the IEEE Wireless Communications amp Networking Conference(WCNCrsquo07) Hong Kong China March 2007

Ahmet Sekercioglu is a researcher at theCentre for Telecommunications and Informa-tion Engineering (CTIE) and a Senior Lecturerat the Department of Electrical and ComputerSystems Engineering of Monash UniversityHe was the leader of the Applications Programof Australian Telecommunications CRC untilthe end of the centrersquos research activities(December 2007) He has completed his PhDdegree at Swinburne University of Technol-ogy and BSc MSc (all in Electrical andElectronics Engineering) degrees at Middle

East Technical University He has lectured at Swinburne University ofTechnology for 8 years and has had numerous positions as a researchengineer in private industry

His more recent work focuses on distributed algorithms for self-organi-

zation in wireless networks He is also interested in application of intel-ligent control techniques for multi-service networks as complexdistributed systems

Milosh Ivanovich fills the role of SeniorEmerging Technology Specialist within theChief Technology Office of Telstra and is anHonorary Research Fellow at Melbourne andMonash Universities in Australia A SeniorMember of IEEE Miloshrsquos interests lie inqueuing theory teletraffic modeling perfor-mance analysis of wireless networks and thestudy and enhancement of TCPIP in hybridfixedwireless environments He obtained aBE (1st class Hons) in Electrical and Com-puter Systems Engineering (1995) a Master of

Computing (1996) and a PhD in Information Technology (1998) all atMonash University Australia He is an author of two edited book chaptersa patent and over 40 international journal and conference publications


Alper Yegin is an architect at the Standardsand Industry Initiatives Group of SamsungElectronics He currently chairs IETF PANAWorking Group and Security Team of WiMAXForum Network Working Group In the pasthe served as members of IETF WirelessDirectorate and IPv6 Forum Technical Direc-torate He has received his MSc in ComputerScience degree at University of IllinoisUrbana-Champaign and BSc in ComputerEngineering at Bogazici University His recentwork focuses on IP-based end-to-end 4G

architectures especially in the network security and mobility manage-

ment areas


Table 1Network performance parameters and their characteristics maintained byQoS

Network performanceparameter

Characteristics

Latency Delivery delay of a packet from source todestination

Jitter Variation in latency

Reliability The percentage of traffic that should besuccessfully delivered from source to destinationto maintain the service quality

Data transmissionrate

The amount of data that should be carried fromsource to destination in a given period of time tomaintain the service quality


ARTICLE IN PRESS

available transmission capacity inevitably affects the ser-vice quality Another factor is latency Some services havegreater tolerance for latency (eg FTP or e-mail) whileothers (like VoIP or video-conferencing) have strict delaybounds Taking such QoS requirements into account pack-et flows need to be prioritized via appropriate QoS man-agement and scheduling methods In broad terms thesealways seek to achieve the optimal trade-off between theconflicting goals of maximized user performance and max-imized system utilization

WiMAX has a built-in QoS framework for real-timeapplications as well as data which can take advantage ofits polling architecture and dynamically adaptable modu-lation in the physical layer It should be noted that whilethe standardized WiMAX QoS framework provides the de-tails about the types of service flows that are supported itdoes not explicitly define the actual packet mechanisms forachieving QoS differentiation in the MAC sublayer As inother standards of this kind such mechanisms are leftopen for vendor implementation as long as they conformto the stated WiMAX QoS framework

In Section 2 we provide an overview of this WiMAX-specific QoS framework and separately consider point-to-multipoint and Mesh Network variants Our workexamines QoS implementation in WiMAX by subdividingthe relevant issues into three distinct categories packetscheduling and admission control (Section 31) signallingand internetworking (Section 32) and Mesh Network (Sec-tion 4) In each of these three sections we summarize thestate of the art research activity along with results possi-ble implementation drawbacks and scope for furtherdevelopment In Section 6 an overall analysis is providedalong with our concluding remarks

2 QoS and the MAC sublayer of WiMAX networks

21 Definitions of Quality of Service

There are two broad definitions of Quality of Service(QoS)

User-Centric QoS is lsquolsquothe collective effect of service per-formances which determine the degree of satisfactionof a user of the servicerdquo [22]Network-Centric QoS comprises lsquolsquothe mechanisms thatgive network managers the ability to control the mix ofbandwidth delay variances in delay (jitter) and packetloss in the network in order to deliver a network service(eg voice over IP)rdquo [9]

Our paper is primarily concerned with the second defi-nition of QoS and in the rest of this work the term QoSshall be taken to mean lsquolsquoNetwork-Centric QoSrdquo

Network engineers can use the existing resources effi-ciently by implementing a QoS mechanism Early packetnetworks typically catered for one service type and allpackets were treated equally There was no QoS differenti-ation or guarantee of reliability minimum latency jitter orother performance characteristics (Table 1) for any set ofpackets As a result of such a regime a single bandwidth


intensive application may cause the performance of otherapplications to degrade significantly In a multi-servicenetwork the QoS mechanism has to ensure that it can pro-vide preferential delivery service to packets according totheir performance requirements and QoS priority levelwhile maintaining a high network utilization QoS differen-tiation can be implemented on either a per-application orper-user basis With this in mind QoS mechanisms canbroadly be grouped under the following two categories

Admission Control determines how and when the traf-fic generated by a given application or user can haveaccess to the network resources Typically operates ata session or flow timescale (ie decisions relate toadmission of user sessions or flows)Traffic Control determines how packet marking sched-uling and shaping (flow rate control) is performed forpacket traffic generated by a given application or userTypically operates at packet timescales (ie decisionsrelate to which packet from which flow is to be trans-mitted next)

By implementing a well functioning QoS mechanism (ora combination thereof) network engineers can controlavailable network resources to suit a particular require-ment model and to ensure that critical services are not af-fected by services of lower-priority The end result isimproved user experience and reduced system cost dueto more efficient and targeted use of available resources

22 QoS architecture of WiMAX

In wireless networks including WiMAX QoS supportusually resides in the MAC sublayer because of the needto interact with radio resource management and physicallayer dynamics Fig 1 shows the WiMAX QoS architectureas defined by the standard [19] The base station (BS) hasthe responsibility of managing and maintaining the QoSfor all packet transmissions The BS manages this bydynamically distributing usage time to subscriber stations(SSs) through information embedded in the transmittedframes (Fig 2) The figure only shows the TDD (time-divi-sion duplex) mode of operation in which BS-to-SS broad-casts (downlink subframe) are followed by SS-to-BS(uplink subframe) transmissions It is also possible to useFDD (frequency-division duplex) mode of operation in




ARTICLE IN PRESS













ARTICLE IN PRESS

















ARTICLE IN PRESS





Typicalapplications

Uplink

BS behaviour


















Web browsingemail









QoS specifications

SS behaviour

hedulerUL BWtime



hedulerperiodic




hedulerperiodic



heduler(in the









ARTICLE IN PRESS



















ARTICLE IN PRESS













ARTICLE IN PRESS

















ARTICLE IN PRESS















to multiple PDUs













ARTICLE IN PRESS



















ARTICLE IN PRESS











ARTICLE IN PRESS





















ARTICLE IN PRESS




























ARTICLE IN PRESS















ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas




ARTICLE IN PRESS













ARTICLE IN PRESS

















ARTICLE IN PRESS





Typicalapplications

Uplink

BS behaviour


















Web browsingemail









QoS specifications

SS behaviour

hedulerUL BWtime



hedulerperiodic




hedulerperiodic



heduler(in the









ARTICLE IN PRESS



















ARTICLE IN PRESS













ARTICLE IN PRESS

















ARTICLE IN PRESS















to multiple PDUs













ARTICLE IN PRESS



















ARTICLE IN PRESS











ARTICLE IN PRESS





















ARTICLE IN PRESS




























ARTICLE IN PRESS















ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas



ARTICLE IN PRESS

















ARTICLE IN PRESS





Typicalapplications

Uplink

BS behaviour


















Web browsingemail









QoS specifications

SS behaviour

hedulerUL BWtime



hedulerperiodic




hedulerperiodic



heduler(in the









ARTICLE IN PRESS



















ARTICLE IN PRESS













ARTICLE IN PRESS

















ARTICLE IN PRESS















to multiple PDUs













ARTICLE IN PRESS



















ARTICLE IN PRESS











ARTICLE IN PRESS





















ARTICLE IN PRESS




























ARTICLE IN PRESS















ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas



ARTICLE IN PRESS





Typicalapplications

Uplink

BS behaviour


















Web browsingemail









QoS specifications

SS behaviour

hedulerUL BWtime



hedulerperiodic




hedulerperiodic



heduler(in the









ARTICLE IN PRESS



















ARTICLE IN PRESS













ARTICLE IN PRESS

















ARTICLE IN PRESS















to multiple PDUs













ARTICLE IN PRESS



















ARTICLE IN PRESS











ARTICLE IN PRESS





















ARTICLE IN PRESS




























ARTICLE IN PRESS















ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas




ARTICLE IN PRESS



















ARTICLE IN PRESS













ARTICLE IN PRESS

















ARTICLE IN PRESS















to multiple PDUs













ARTICLE IN PRESS



















ARTICLE IN PRESS











ARTICLE IN PRESS





















ARTICLE IN PRESS




























ARTICLE IN PRESS















ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas



ARTICLE IN PRESS













ARTICLE IN PRESS

















ARTICLE IN PRESS















to multiple PDUs













ARTICLE IN PRESS



















ARTICLE IN PRESS











ARTICLE IN PRESS





















ARTICLE IN PRESS




























ARTICLE IN PRESS















ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas



ARTICLE IN PRESS

















ARTICLE IN PRESS















to multiple PDUs













ARTICLE IN PRESS



















ARTICLE IN PRESS











ARTICLE IN PRESS





















ARTICLE IN PRESS




























ARTICLE IN PRESS















ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas



ARTICLE IN PRESS















to multiple PDUs













ARTICLE IN PRESS



















ARTICLE IN PRESS











ARTICLE IN PRESS





















ARTICLE IN PRESS




























ARTICLE IN PRESS















ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas












ARTICLE IN PRESS



















ARTICLE IN PRESS











ARTICLE IN PRESS





















ARTICLE IN PRESS




























ARTICLE IN PRESS















ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas





ARTICLE IN PRESS











ARTICLE IN PRESS





















ARTICLE IN PRESS




























ARTICLE IN PRESS















ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas



ARTICLE IN PRESS





















ARTICLE IN PRESS




























ARTICLE IN PRESS















ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas





ARTICLE IN PRESS




























ARTICLE IN PRESS















ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas















ARTICLE IN PRESS















ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas



ARTICLE IN PRESS



















empty slot number







ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas



ARTICLE IN PRESS




















ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas



ARTICLE IN PRESS























ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas



ARTICLE IN PRESS
















ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas



ARTICLE IN PRESS

Acknowledgement


References












































ARTICLE IN PRESS











ment areas



ARTICLE IN PRESS











ment areas


Date post:	14-Nov-2023
Category:	Documents
Upload:	independent
View:	0 times
Download:	0 times

A survey of MAC based QoS implementations for WiMAX networks

Documents