a-mac: adaptive medium access control for next ...cpn.unl.edu/?q=system/files/amac.pdfwireless...

1

A-MAC: Adaptive Medium Access Control forNext Generation Wireless Terminals

Mehmet C. Vuran Ian F. Akyildiz

Broadband & Wireless Networking LaboratorySchool of Electrical & Computer Engineering

Georgia Institute of Technology, Atlanta, GA 30332Tel: (404) 894-5141 Fax: (404) 894-7883

Email:{mcvuran,ian}@ece.gatech.edu

Abstract— Next Generation (NG) wireless networks are en-visioned to provide high bandwidth to mobile users via band-width aggregation over heterogeneous wireless architectures.NG wireless networks, however, impose challenges due to theirarchitectural heterogeneity in terms of different access schemes,resource allocation techniques as well as diverse quality ofservice requirements. These heterogeneities must be captured andhandled dynamically as mobile terminals roam between differentwireless architectures. However, to address these challenges, theexisting proposals require either a significant modification in thenetwork structure and in base stations or a completely newarchitecture, which lead to integration problems in terms ofimplementation costs, scalability and backward compatibility.Thus, the integration of the existing medium access schemes, e.g.,CSMA, TDMA and CDMA, dictates an adaptive and seamlessmedium access control (MAC) layer that can achieve highnetwork utilization and meet diverse QoS requirements.

In this paper, an adaptive medium access control (A-MAC)layer is proposed to address the heterogeneities posed by theNG wireless networks. A-MAC introduces a two-layered MACframework that accomplishes the adaptivity to both architecturalheterogeneities and diverse QoS requirements. A novel virtualcube concept is introduced as a unified metric to model hetero-geneous access schemes and capture their behavior. Based ontheVirtual Cube concept, A-MAC provides architecture-independentdecision and QoS based scheduling algorithms for efficient multi-network access. A-MAC performs seamless medium access tomultiple networks without requiring any additional modific ationsin the existing network structures. It is shown via extensive sim-ulations that A-MAC provides adaptivity to the heterogeneitiesin NG wireless networks and achieves high performance.

Index Terms— Next Generation Wireless Networks, AdaptiveMedium Access Control, Virtual Cube Concept, HeterogeneousQoS Requirements, Heterogeneous Networks.

I. I NTRODUCTION

NEXT Generation (NG) wireless networks are expectedto provide mobile users with a freedom of roaming

between diverse set of wireless architectures as shown in Fig.1. Since an NG wireless terminal will operate in these varioustypes of networks, the Medium Access Control (MAC) layerwill encounter different protocols that are already deployedin the access points (AP) of these networks. The varioustypes of MAC protocols can be classified in terms of their

This work is supported by the National Science Foundation under contractANI-0117840.

Macro−CellMicro−Cell

Pico−Cell

Fig. 1. Next Generation Wireless Networks.

multiple access schemes. Second-generation systems, e.g., IS-54 and GSM, support TDMA while IS-95, IS-136 and PDCsupport CDMA [2]. Wideband CDMA (WCDMA) is alsodesignated as the access technology for cdma2000 [30] andUMTS/IMT-2000 [21]. IEEE 802.11 standard for local areanetworks (LAN) uses CSMA/CA in addition to polling [12].Different combinations of TDMA, CDMA and WCDMA havealso been proposed in the literature [2], [11], [24], [25], [26].NG wireless terminals must provide seamless access to thesenetworks, considering the specialized services of differentnetworks.

In addition to the heterogeneity in the medium accessschemes of the existing wireless architectures, NG wirelessnetworks are also expected to provide the mobile users withdiverse set of services ranging from high rate data traffic totime-constrained real-time multimedia [15]. In [29], fourQoSclasses are defined for UMTS networks, i.e., conversational,streaming, interactive and background. Conversational andstreaming classes are mainly intended to be used to carry real-time traffic flows. Conversational real-time services, suchasvideo telephony and voice over IP, are the most delay sensitiveapplications so the restrictions on transfer delay and jitter arevery strict. Unlike conversational class, streaming classsuchas live broadcasting is a one way transport and has less strictdelay requirements than the conversational class. Interactiveand background classes are mainly used by traditional internetapplications such as web browsing, email, telnet and FTP.

2

Interactive class is characterized by the request responsepattern of the end user. In this class, the data is expected tobedelivered within a certain time. Examples are interactive webbrowsing, telnet and FTP. On the contrary, background classdoes not require the data be transmitted within a certain time.Examples are background delivery of E-mails, short messageservice (SMS), download of databases and reception of mea-surement records. In addition to the UMTS network, voiceover WLAN (VoWLAN) applications are also gaining interestdue to the high popularity of WLANs. However, assuringhigh quality voice communication through WLANs requireQoS-aware networking protocols. Consequently, the scarcityof the wireless resources necessitates efficient utilization ofthe available bandwidth in NG networks [11]. Therefore, toaddress the diverse QoS requirements, NG wireless terminalsmust be able to adapt to the heterogeneous access schemes.

In this paper, we aim to integrate the existing wirelessarchitectures without requiring any modifications in the basestations. Instead, we achieve adaptivity to the architectural het-erogeneity as well as diverse QoS requirements by deployinga new adaptive MAC framework in the NG wireless terminals.The challenges addressed in this work can be summarized asfollows:

• Heterogeneity in Access Schemes:As explained before,NG wireless terminals encounter different access schemeswhile roaming between different wireless networks. For aseamless integration, the mobile terminal must be capableof accessing each network when needed.

• Heterogeneity in Resource Allocation:Each networkstructure performs resource allocation according to vari-ous techniques such as TDMA slots, CDMA codes, andrandom allocation. A metric to compare the amount ofthe resource allocated by different networks is requiredto achieve high network utilization in accessing differentnetworks. However, there exists no such unified metric forcomparison of the allocated resources by different accessschemes.

• Heterogeneity in QoS Requirements:NG wireless termi-nals are envisioned to provide QoS guarantees accordingto the underlying network structures. Thus, the MAClayer must efficiently evaluate the available resources indifferent networks and perform access such a way thatthe QoS requirements of applications are satisfied.

In order to address these heterogeneities posed by the NGwireless networks, we propose a new two-layer AdaptiveMedium Access Control (A-MAC). A-MAC first providesprocedures for detecting and accessing the available networksthat the NG wireless terminal can access, i.e.,access. Then,the available resources in these various types of networksare modeled based on a unified resource model. Each flowthat is sent to the MAC layer is then served through thenetwork that is most suitable for the QoS requirements of theflow, i.e., decision. Moreover, A-MAC provides QoS-basedscheduling for multiple flows assigned to the same network,i.e., scheduling. As a result, the two-layer A-MAC exploitsthe available resources in the NG wireless networks by pro-viding procedures for serving multiple flows through multiple

��

��

��

��

��

��

��

ACCESSSUB−LAYER

MASTERSUB−LAYER

��

��

��

��

��

��

��

��

��

��

��

SCHEDULING

IEEE 802.11 GSM3G

DECISION

NETWORK MODELS��

��

��

��

��

��

��

��

��

��

��

��

INTERFACESADAPTIVE NETWORK

��

��

��

Fig. 2. Main components of A-MAC.

network architectures available to the terminal simultaneously.The main structure and the components of A-MAC is shown

in Fig. 2. The access sub-layeris specialized for access-ing the network, whilemaster sub-layerperforms decisionand scheduling of various application requests for the mostefficient network. A-MAC achieves adaptivity to both theunderlying network structure and the QoS requirements ofdifferent traffic types. We introduce a novelVirtual Cubeconcept which serves as a basis for comparison of differentnetwork structures. Based on the Virtual Cube concept, A-MAC provides architecture-independent decision and QoSbased scheduling algorithms for efficient multi-network ac-cess. Incompatibility among medium access and resourceallocation techniques are melted into a unified medium accesscontrol framework, providing self-contained decision flexibilityas well as capability to access various networks.

The rest of the paper is organized as follows. In Section II,the related work on MAC in heterogeneous wireless networksis presented. We introduce the Virtual Cube concept in SectionIII. Based on Virtual Cube concept, we explain networkmodeling in Section IV. The A-MAC is introduced in SectionV. We then discuss the performance of A-MAC in Section VIand conclude the paper in Section VII.

II. RELATED WORK

There exist several studies in the literature to address thein-tegration of existing wireless systems. In [19], a unified frame-work for the channel assignment problem in time, frequency,and code domains is proposed. The unified (T/F/C)DMAalgorithm consists of labeling and coloring phases. Using thegraph theory solutions, channel assignment problems in het-erogeneous network structures have been addressed. Althoughthis work provides the fundamental theoretical results forchannel assignment in (T/F/C)DMA networks, the frameworkis not applicable to the existing network structures where thechannel assignment principles have already been decided. The

3

application of this framework requires major modificationsinthe NG wireless network components.

An Ad-hoc CEllular NETwork (ACENET) architecture for3.5G and 4G mobile systems is proposed in [25], where aheterogeneous MAC protocol to integrate IEEE 802.11, Blue-tooth and HiperLAN/2 with cellular architectures is presented.The coordination between transmissions of different accessprotocols is provided using beacons from the base stations.ACENET consists of a cellular network and ad hoc network.Although it is argued that ACENET improves the throughputperformance over the existing networks, many modificationsin the base stations are required to achieve this.

In [26], a TCDMA protocol is presented for NG Wirelesscellular networks. A time-code structure is proposed withsubintervals allocated for different functions to enable theintegration of cellular networks, ad hoc networks and wirelessLANs. However, the integration requires modifications in theexisting base stations.

A QoS-oriented access control for the 4G mobile mul-timedia CDMA communications is presented in [11]. Theproposed MAC protocol exploits both time-division and code-division multiplexing. A certain QoS level is guaranteed usingfair packet loss sharing (FPLS) scheduling. The proposedMAC protocol is shown to provide QoS guarantees in hybridTD/CDMA systems. However, the proposed protocol neces-sitates a new wireless network infrastructure with new basestations for 4G communications.

In [17], a multiple access protocol is proposed for cellularInternet and satellite-based networks. The proposed schemeaccesses to the network with a probabilitypaccess to decreasecollision probability based on the network load. However,the QoS requirements of the application are not addressed.Although the proposed scheme is designed for both cellularnetworks and the satellite-based networks, it requires modifica-tions in the base station. Hence, the existing cellular networkscannot be used for integration.

As a result, the existing proposals [11], [17], [25], [26]require either a significant modification in the existing in-frastructure and base stations or a completely new architec-ture. Therefore, these approaches lead to integration problemsin terms of implementation costs, scalability and backwardcompatibility. The NG wireless networks are also expected toprovide diverse range of services. Such diversity in the servicesrequires adaptivity in the MAC layer in terms of guaranteeingQoS requirements in wireless environments.

In addition to the proposed MAC protocols, there hasbeen extensive research on the physical layer of NG wirelessterminals. It is envisioned that NG wireless terminals willbeequipped with multiple-mode radio capabilities. Recent devel-opments in radio receiver and transmitter development haveled the way to mobile hand-held devices that are capable offunctioning in multiple access technologies. In [1], dual band(CELL/PCS) triple mode (CDMA/AMPS/TDMA) transmitterhas been developed. In [27], an RF module with dual banddual mode (GSM/W-CDMA) transmitter and receiver has beendeveloped. In our paper, we assume that NG wireless terminalsare capable of receiving signals from multiple network accesspoints and transmitting signals to different access schemes

simultaneously.

III. T HE V IRTUAL CUBE CONCEPT

NG mobile terminals will encounter different accessschemes during accessing different networks within the NGwireless architecture as shown in Fig. 1. Hence, different re-source allocation units such as CSMA random access, TDMAtime slots, CDMA codes, as well as hybrid types will beallocated to the mobile terminals (MTs). Thus, MTs mustbe able to compare these resources in order to provide QoSguarantees by accessing the most efficient network for theapplication. However, it is impractical to perform comparisonbetween the resources available to the MT due to lack of aunified metric for such comparison. For this purpose, we intro-duce a three dimensional space model for modeling networkresources. Based on this three dimensionalresource-space, wepropose a novelVirtual Cube concept in order to evaluatethe performance of each network in the reach of an MT. TheVirtual Cube concept defines a unit structure based on theresource allocation techniques used in the existing networks.In the following two sections, we describe the resource-spacemodel and the properties of the Virtual Cube, respectively.

A. Resource-Space

We model the resource in a three dimensionalresource-spacewith time, rate, and power dimensions as shown in Fig.31. The three dimensions of the resource-space are as follows:

1) Time Dimension: :The time dimensionmodels the timerequired to transfer information.

2) Rate Dimension: :The rate dimensionmodels the datarate of the network. Thus, the capacity of different networkswith the same connection durations but different data ratesarecaptured in the rate dimension. Furthermore, the bandwidthincrease due to the multi-code transmissions or multi-channelcommunication is also captured in this dimension.

3) Power Dimension: :The power dimensionmodels theenergy consumed for transmitting information through the net-work. Note that, the resource in terms of available bandwidthcan be modeled using the time and rate dimensions. However,the cost of accessing different networks vary in terms of thepower consumed by the wireless terminal. Hence, a thirddimension is required. Each network type requires differentpower levels for transmission of the MAC frames becauseof various modulation schemes, error coding and channelcoding techniques. As a result, the resource differences inthese aspects are captured in the power dimension.

The three dimensional resource space described above en-ables modeling the available resource from heterogeneousnetworks as will be explained in Section IV.

B. Virtual Cube Structure

The virtual cube constitutes the unit structure for modelingand comparing different networks for the appropriate accessdecision by the NG wireless terminal. The resource space and

4

Time

Power/Code

M bits/cube

P

T

Rate

Fig. 3. Virtual cube model.

the virtual cube structure is shown in Fig. 3. The virtual cubestructure is defined by three parameters explained as follows:

Cube Capacity (M bits/cube): The number of bits a cubecarries.

Cube Duration (T sec): The time a cube fills in timedimension. As a result, a virtual cube models a rate ofM/Tbits/sec in the rate dimension.

Cube Power (P Watts): The minimum power for a trans-mission of a cube.

The three parameters that define the virtual cubes, i.e.,M ,T , andP are fixed for each cube and provide a granularity ineach dimension. As will be explained in Section IV, the virtualcubes are used to model the available resource a networkprovides. In order to model both the available bandwidth andthe cost of communication in terms of consumes energy, twotypes of virtual cubes are used in the A-MAC as describedbelow:

Virtual Information Cube (VIC): VICs model the infor-mation sent through the network. They virtually containMbits and serve as a basis to determine the capacity of thenetwork.

Virtual Power Cube (VPC): The required power consump-tion for transmission of a bit is different for every network.VICs model the minimum power consumption required totransmit a bit byP/M . In order to model the additionalpower consumption per bit for each network, VPC is used.VPCs model the additional power needed to transmitM bitsof information. As a result, VPCs, which virtually contain nodata, are used to capture the additional power requirementsofa particular network.

IV. N ETWORK MODELING

The Virtual Cube concept introduced in Section III formsa basis for modeling different access schemes. Based on theVirtual Cube concept, underlying access schemes are modeledas a three dimensional structure calledresource bin. Eachdimension of a resource bin defines the number of virtualcubes that can be filled in that dimension as shown in Fig.

1Note that the resulting structure may not necessarily be a cube. However,we refer to the concept as virtual cube for ease of illustration.

4. In Fig. 4, it is shown that VPCs, i.e., gray cubes, are usedto model the additional power requirement of the network,while VICs, white cubes, model the available capacity inthe network. As a result, resource bins capture the capacityof the network access unit as well as timing information,data rate and power requirements. The modeling of TDMA,CDMA, and CSMA based networks as well as multi-ratecommunication into resource bins are illustrated in Fig. 4 andexplained in the following sections.

A. TDMA Modeling

In TDMA systems, the channel resource is partitioned intoframes. Each frame is further divided into time slots, whichareassigned to different users. A TDMA slot is characterized byits slot durationTs and transmission rateR. The slot duration,Ts, specifies the duration of time the MT has the access to theTDMA network within a time frame. The transmission rate,R, specifies the rate at which the MT should send packetsduring the time slot. Due to different modulation schemes,error coding and channel coding techniques, each interfaceis characterized by an average energy per bitEb to be usedduring the transmission.

Using these properties, the TDMA slot can be modeledin the resource space. LetNt, Nr, and Np be the threedimensions of a resource bin representing the number ofvirtual cubes that can be filled in thetime, rate, and powerdimensions, respectively. Then,Nt, Nr, and Np can be ex-pressed by

Nt =Ts

T, Nr =

T R

M, Np =

EbM

P(1)

whereM is the cube capacity,T is the cube duration andPis the cube power as described in Section III. A sample modelof a TDMA network is illustrated in Fig. 4(a). Note that, ifthe number of cubes in the power dimension exceeds one, i.e.,Np > 1, VPCs are filled in the power dimension.

B. CDMA Modeling

In the direct sequence-CDMA (DS-CDMA), each bit ofduration Tb is coded into a pseudo-noise code ofN chipsof duration Tc = Tb/N . The pseudo-noise codes used inCDMA are characterized by the processing gain or so-calledthe spreading gainGp which is expressed asGp = W/R [23],whereW is the bandwidth of the CDMA system andR is thedata rate. Since each user interferes with other users at thebasestation in the reverse link, the base station performs powercontrol and assigns a transmission signal power for each MTto utilize the network. The transmitted energy per bit,Eb, canbe expressed as [23],Eb = Ps/R, wherePs is the transmittedsignal power andR is the data rate.

Although the physical channel is identified by a specific rateof chips, the actual bit rate of the channel varies accordingto the spreading factor used in coding. The MT is interestedin the actual data transmission rateR rather than the channeltransmission rate. The data rate of the connection is calculatedaccording to the code assigned to the MT. Moreover, the MT isinformed of the power level associated with the transmission of

5

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

Virtual Information Cube

Virtual Power Cube

pr

tN

N N

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

(a)

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

Nr

Nt

Np

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

(b)

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

rN (code2)��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

tN

pNrN (code1)

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

(c)

t

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

N (ch 1)p

N (ch 2)p

N (ch 2)r

N (ch 1)r

N

(d)

Fig. 4. (a) TDMA and CDMA modeling, (b) CSMA Modeling, (c) multi-code CDMA modeling, and (d) multi-channel modeling.

the signal by the base station. In TDD CDMA systems suchas UMTS, the time is divided into radio frames, slots andsub-frames. Using the length of the allocated slot,Ts can bedetermined. In FDD systems,Ts is determined by the durationof the connection. Using this information,Nt, Nr andNp aredetermined by (1).

In CDMA networks, two nodes may lead to intracell in-terference when nonorthogonal codes are assigned. However,it is clear that this interference cannot be deterministicallycalculated before APs assign the codes to nodes and nodes starttransmitting information. As a result, in A-MAC, the CDMAnetwork is modeled based on the assignment informationprovided by the BSs to the adaptive network interfaces (ANIs)as discussed in Section V-C. However, as will be explainedin Section V-B, the wireless channel aware scheduling blockadjusts the rate of each flow based on the wireless channelstate. As a result, the effects of intracell interference isaddressed by the A-MAC protocol.

C. CSMA Modeling

CSMA based systems such as IEEE 802.11 are character-ized by the randomness in the access to the network. The mo-bile terminals contend for the channel using CSMA/CA. TheMT that captures the medium transmits information. Hence, itis not possible to deterministically calculate the transmissiontime of a specific MT in CSMA based systems. In a recent

work, it has been shown that in wireless networks, the pastthroughput value has a strong influence on the future through-put value [4]. Moreover, it has been found that irrespectiveofthe velocity of wireless node, throughput prediction basedonthe past values is feasible. In this respect, various predictionmethods can be used for predicting the connection opportu-nities in CSMA based network. Consequently, for CSMA-based networks, we use the last transmission information tomodel the resource bin as illustrated in the Fig. 4(b). Basedon the previous transmission information, e.g., data rate,con-nection time, consumed power, the dimensions of the modelare determined using (1). Moreover, the generated model isdynamically modified to account for wireless medium errorsand actual changes in the connection information as explainedin Section V-B. As a result, the CSMA-based networks aremodeled based on the previous connection information withadaptive updates.

D. Multi-Rate Networks

In addition to the modeling schemes for each type of net-work presented above, a specific network may provide multiplerates to a wireless terminal through various techniques such asadditional codes or multi-channel communication. Using thevirtual cube concept, multi-rate networks can also be modeled.

In CDMA-based networks, in addition to fixed code as-signment, the MT can be assigned multiple codes throughoutthe communication. Two techniques used in CDMA-basedsystems for multi-code transmission and modeling of thesecases are discussed next.

In IS-95-B and cdma2000, an MT is provided the supple-mental code channels (SCC) in addition to a fundamental codechannel (FCC) [14], [30]. In IS-95-B, the SCCs have the samespreading gain as the FCC. As a result, multi-code channelsprovide an aggregate bandwidth for an MT using the samevariable spreading gain. In modeling these systems, the ratedimension is extended to include the additional bandwidthprovision of multi-code channels as shown in Fig. 4(c).

In addition to code aggregation, variable spreading gain isalso used in the cdma2000 system [14], [30]. By reducingthe spreading gain, the transmission gain is increased tosupport high data rate applications. In the cdma2000 system,in addition to an FCC, one SCC is assigned to an MT perdata service. Multiple SCCs are used to support multiple datastreams with different QoS requirements [14]. The increaseinthe data rate due to variable spreading gain is also capturedin the rate dimension of the resource bin for SCCs as shownin Fig. 4(c).

The number of virtual cubes in the rate dimension,Nr,of the CDMA frame model increases with the data rate. Onthe other hand, for a specific bit-error rate (BER), a decreasein the spreading gain increases the average power level [24].The power levelPv for a CDMA signal with a spreading gainGv can be expressed asPv = Pf (Gf/Gv), wherePf is thepower level of the fundamental spreading gain andGf is thefundamental spreading gain. Since the rate dimension capturesthe effect of multi-code transmission in CDMA networks, thenumber of VPCs in the power dimension,Np, remains thesame.

6

Recently, multi-channel communication has gained interestdue to the recent advances in transceiver technology. Espe-cially, in cellular networks and IEEE 802.11 standard, multi-channel communication is possible. Note that the number ofvirtual cubes in the rate dimension,Nr, depends on manychannel and transceiver dependent parameters such as thechannel bandwidth, modulation and channel coding schemes.Hence, the resource in each channel needs to be calculatedseparately as shown in Fig. 4 (d). Over a fixed interval, theavailable aggregate resource through multi-channels is thenconsidered for decision as explained in Section V.

V. A-MAC: A DAPTIVE MEDIUM ACCESSCONTROL

A-MAC accomplishes the adaptivity to both architecturalheterogeneities and diverse QoS requirements using a two-layer structure as shown in Fig. 2. Theaccess sub-layerisspecialized for accessing the network, whilemaster sub-layerperforms coordination of the application requests. The accesssub-layer consists of ANIs that are responsible for the adaptiv-ity of the MT to the underlying heterogeneous architectures.The MT communicates with different networks through ANIs.In addition to accessing functionalities, the ANIs providethe environment-awareness to the master sub-layer. Usingthe information gathered from the underlying network fromaccess points (APs), an ANI models the access scheme andinforms the master sub-layer. The network modeling, which isexplained in Section IV, is performed according to the VirtualCube concept, which is introduced in Section III.

The master sub-layer consists of scheduler and decisionblocks. In NG wireless networks, multiple flows with variousquality of service (QoS) requirements can be forwarded tothe MAC layer simultaneously. The master sub-layer aims toforward these flows to appropriate interfaces to effectivelyutilize the wireless medium and guarantee the QoS require-ments of each flow. In addition, as the MT traverses throughdifferent networks, the most efficient access point for specificQoS requirements is chosen. When a flow is forwarded to theMAC layer by upper layers, the QoS-based decision blockfirst performs decision according to the information about theunderlying networks gathered from the ANIs. On top of eachinterface, a QoS-based scheduler handles fair share of thebandwidth if multiple flows are serviced through the sameinterface. The functions of decision and scheduler blocks,and the adaptive network interfaces (ANIs) are explained inSections V-A, V-B, and V-C, respectively.

A. Decision

NG wireless networks are envisioned to provide diverseset of services which require the MAC layer to efficientlyaddress the QoS requirements of these services. In order toguarantee QoS requirements of each traffic forwarded to theMAC layer, A-MAC performs QoS-based decision for theappropriate interface to forward a specific traffic. This decisionis performed by thedecision block. The ANIs associated witheach network models the available resources based on theprinciples introduced in Section IV. The available connectionsare then informed by the associated ANIs to the decision

block as shown in Fig. 2. This information consists of thestart time of the network slot and the three dimensions of theresource bin,Nt, Nr andNp, which are expressed by (1). Thedecision algorithm performs decision in eachdecision intervalTd where, A-MAC decides which traffic flow will be servedon which interface according to the resource bins provided bythe ANIs.

A specific network interfacek with k ∈ C, whereC isthe set of active connections, is modeled by the followingparameters:

Nkt - Number of information cubes in the time dimension

of the network model of interfacek during decisioninterval Td.Nk

r - Number of information cubes in the rate dimensionof the network model of interfacek during decisioninterval Td.Nk

p - Number of power cubes in the power/code dimen-sion of the network model of interfacek during decisioninterval Td.

For a specific traffic flowf, the decision block chooses theinterface such that the utility functionUf given by

Uf =rkfNk

r Nkt

Nkp

(2)

is maximized. Note that, the utility function,Uf , aims tofind the interface with maximum throughput capability forthe minimum transmission power. The utility function (2) ismaximized subject to the constraints

∑

k∈Ca

Nkp ≤ Nmax

p , (3)

rkfNk

r ≥ Nr,f , (4)

whererkf is the bandwidth share of the trafficf in interfacek, if

multiple traffic types are interleaved into one interface,Nmaxp

is the maximum power dissipation allowed for the decisioninterval T, andNr,f is the minimum required bandwidth interms of VICs for the traffic typef in order to guarantee itsQoS requirements.Ca in (4) is the set of active interfaces thatare chosen by the decision block. The constraints in (3) and(4) are chosen such that the chosen interface will not exceedthe maximum allowed power consumption of the MT and theminimum required bandwidth is met.

Depending on the number of traffic flows to be servicedand the number of network interfaces available, A-MAC canbe in one of four states. The actions taken in each state issummarized in the following sections.

1) Single Type Traffic - Single Interface (STSI):In this case,A-MAC simply forwards the incoming packets from the upperlayers to the active interface as long as (3) and (4) are satisfied.

2) Single Type Traffic - Multiple Interfaces (STMI):Thedecision layer performs the decision algorithm for each of theinterface. If the (3) and (4) are met by one or more interfaces,the one that maximizes (2) is chosen. The incoming packetsare then fragmented into MAC frames of the specific interfaceand sent to the ANI. The specified ANI performs access to thenetwork and transmits the frames.

7

If the QoS requirements of the traffic type cannot be fulfilledusing one interface, the decision block attempts to use multipleinterfaces for the flow. The interfaces are sorted in a decreasingorder of their utility functions. The decision block tries tofulfill the constraints (3) and (4) using multiple interfaces. Inthis case, the constraint (4) becomes

∑

k∈Cfa

rkf ∗ Nk

r ≥ Nr,f , (5)

where Cfa is the set of active interfaces chosen for traffic

type f . If the QoS requirements are guaranteed by multipleinterfaces, the decision block forwards incoming packets in areverse round-robin fashion to the appropriate network. Ac-cording to the bandwidth share of the traffic in each interface,a weight,wk

f is associated with the interfacek such that

wkf =

rkf∑

k∈Cfa

rkf

. (6)

The packets are then fragmented into MAC frames accord-ing to the properties of the network interface and sent throughANIs.

3) Multiple Type Traffic - Single Interface (MTSI):In thiscase, A-MAC first checks if the decision constraints hold foreach traffic type, assuming they were sent alone through theinterface. The traffic types that cannot be satisfied by thecurrent interface are rejected and the upper layers are notified.A-MAC then performs scheduling among the eligible traffictypes in order to serve multiple services through the singleinterface.

According to the QoS requirements of the traffic types, eacheligible traffic typef is assigned a bandwidth sharerk

f such

that∑

f rkf <

Nkr ∗M

T. The decision block then checks if each

flow satisfies (3) and (4). If the requirements are guaranteedfor each of the traffic type, the scheduler is informed of thebandwidth shares and MAC frame size of the interfaces. Thepackets from each flows are then fragmented into MAC framesand sent to the scheduler associated with the ANI responsiblefor the transmission.

If the requirements cannot be fulfilled, the traffic type withthe lowest priority is rejected and the bandwidth sharesrk

f

are updated. The algorithm performs decision algorithm untilit finds a set of traffic types that can be serviced throughthe interface. Since traffic types which can be served throughthe interface alone are taken into account, the decision blockeventually finds a set of traffic types.

4) Multiple Type Traffic - Multiple Interfaces (MTMI):In this case, A-MAC performs decision according to thepriorities of traffic types. The traffic typef with the highestpriority is forwarded to appropriate interface or interfacesaccording to the principles described in Section V-A.2. Whenthe connectionsCa

f for the traffic type f is decided, thecapacities of each interfacek ∈ Ca

f are updated such thatthe bandwidth occupied by the traffic typef in each interfacek is deducted. The same procedure is repeated for other traffictypes in decreasing priority among the remaining capacity.This process is performed until either all traffic types are

served or all interfaces are occupied. Remaining lower prioritytraffic types are rejected.

After the decision process, the bandwidth shares of eachtraffic type in each interface is provided to the appropriateschedulers; and accordingly the scheduling is performed inthe ANIs where multiple flows are directed.

One exception in the decision algorithm is the case whereWLAN interfaces are evaluated. As explained in section V-C,CSMA based systems are modeled reactively, i.e., based onthe last transmission information, since the access conditionscannot be predicted in advance. However, it might be the casethat this previous transmission information may not be presentin the node. This occurs when a node enters the coverage areaof a WLAN. Since generally, WLAN provides much betterperformance than the other network structures in terms of datarate, the decision layer gives precedence to the WLAN overother network structures. When a WLAN interface is informedto the master sub-layer, high priority traffic is forwarded to theWLAN. The decision block then checks if the requirementsof the traffic type are guaranteed in each decision interval.Ifthe requirements are not guaranteed, the decision algorithm isperformed according to the models generated by the ANI asexplained in the previous sections.

B. Scheduling

In NG wireless terminals, multiple flows with various QoSrequirements can be forwarded to the MAC layer at the sametime. Due to these various QoS requirements, these flows canbe served using different types of network architectures thatare most suitable for each flow. During the decision process,the decision block may end up selecting a single networkstructure for multiple services. Since these services cannot beinterleaved into a single MAC frame, a QoS-based scheduler isused for each ANI in order to guarantee the QoS requirementsof each flow.

Scheduling in wireless networks is a highly explored areabecause of its importance in the design of base stations [8],[10], [16]. The recent work aims to perform base stationcentered scheduling to various access requests from multipleMTs. In NG wireless networks, a scheduler is also necessary atthe wireless terminal responsible for transmission of multipleflows interleaved into a single interface. However, the require-ments of the scheduler differs from the conventional usage ofschedulers in base stations. In designing a scheduler for NGwireless terminals, the following requirements are considered:

• QoS Guarantee:The QoS requirements of each flowshould be guaranteed throughout the connection.

• Channel Dependent Scheduling:Since wireless channelconditions change throughout the connection, schedulershould be able to adapt to these changes in order toprovide fairness to each traffic flow.

• Dynamic Behavior:Since the number of traffic flowsassigned to a network AP may vary during the courseof scheduling, the scheduler has to be easily configurableto adapt to these dynamic changes.

• Implementation Complexity:Since the number of simul-taneous flows to be served in one connection is limited,

8

a scheduler with low implementation complexity is suf-ficient in NG wireless terminals.

In order to address the requirements of the scheduler inNG wireless terminals, we propose a scheduling algorithmbased on the ideas presented in [5]. We extend the Bin SortFair Queuing (BSFQ) scheduler in order to accommodate theunique characteristics of the wireless medium. The schedulingalgorithm is presented as follows.

In our scheduling algorithm, the output buffer is organizedinto N scheduling bins. Each scheduling bin is implicitlylabeled with a virtual time interval of length∆. The schedulingbins are ordered according to their virtual time intervals andserved in that order. Only the items from the current schedulingbin is transmitted. A virtual system clockτ(t), which isequal to the starting point of the virtual time interval of thecurrent scheduling bin at timet, is maintained in the scheduler.Therefore, the virtual time clock is a step function andτ(t) isincremented by∆ whenever all the packets in the schedulingbin are transmitted.

As explained in Section V-A, for each interfacek, eachtraffic flow f is assigned a guaranteed raterf

2 such that∑f rf < Nr M

T, whereNr is the number of VICs in the rate

dimension of the resource bin of the interface expressed by (1)in Section IV,M is the cube capacity andT is cube durationas presented in Section III.

It is important to note that, the output link rate may changeduring the connection time due to wireless channel conditions.This leads to unfairness for the flows which experience channelerrors. Thus, our scheduling algorithm adaptively varies theoverall output rate according to the available bandwidth. Thereexist many work on opportunistic wireless link scheduling[3], [6], [9], [20]. The main principle of the wireless linkawareness in A-MAC is based on the credit method introducedin [3]. Each flow raterf is updated if the interface experiencesunfairness. The scheduler keeps track of the bandwidth degra-dation of each flow according to wireless channel errors. Thisdegradation is added to the bandwidth share of the flow. Inorder to prevent fluctuations in the bandwidth share of flows,the scheduler updates each bandwidth share in each decisioninterval.

Each packet forwarded to the scheduler is fragmented intoMAC frames according to the MAC frame structure of thenetwork. Thejth frame of flow f, denoted bypj

f , is assignedwith the virtual time stampvts(pj

f ) such that

vts(pjf ) = max(τ(A(pj

f )), vts(pj−1

f )) +lfrf

, j ≥ 0 ,

whereτ(A(pjf )) is the system virtual time at the arrival time

of frame pjf , and lf is the length of the framepj

f which isspecified by the resource bin, i.e.,lf = MNtNr. Arrivingframes are then stored in their corresponding scheduling binsin the FIFO order. The indexijf of the scheduling bin to storeframepj

f is equal to

ijf = ⌊vts(pj

f ) − τ(A(pjf ))

∆⌋, j > 0 .

2Since each scheduler is associated with one ANI, we drop the superscriptk introduced in Section V-A in our analysis.

If ijf = 0, thenpjf is stored in the current scheduling bin.

If ijf < N , it is stored in theijf -th scheduling bin followingthe current scheduling bin. Ifijf ≥ N, the frame is discarded.As a result, packets from different flows are scheduled in aframe-by-frame basis.

The original BSFQ algorithm provides fairness guaranteebased on the values of maximum packet lengthlmax

f andguaranteed data raterf (Please refer to Theorem 2 in [5]). Notethat, in A-MAC, the BSFQ algorithm is enhanced to integratethe procedures for determination the packet size and data ratewith our virtual cube concept. Other than this modification,the functionality of the BSFQ algorithm is preserved. Sincethefairness guarantee is a function of thelmax

f andrf , the fairnessguarantee provided by [5] is still valid in our scheduler.

C. Adaptive Network Interfaces

When a flow is forwarded to the A-MAC protocol, thedecision block performs QoS-based decision to select theappropriate interface for the data transfer as explained inSection V-A. In the case of multiple flows assigned to a singleinterface, the QoS-based scheduler explained in Section V-Bin each interface orders the packets accordingly. Finally,theMAC frames are forwarded to the Adaptive Network Interfaces(ANIs) in the access sub-layer as shown in Fig. 2. Accessingthe networks is accomplished by the ANIs. With the aid ofthe underlying physical capabilities of the MT, each interfaceis capable of performing the following functions:

1) Network Structure Awareness:ANIs provide the MTwith environment-awareness by gathering information aboutthe underlying network structures. When the MT enters thecoverage area of an access point (AP) of a network, theassociated ANI gathers network information from the beaconsperiodically sent from the AP. The environment awarenessis, hence, achieved in different wireless systems, i.e., GSM,UMTS, cdma2000, and WLAN, as follows:

GSM: The base stations use Broadcast Control Channel(BCCH) to periodically broadcast information about the prop-erties of a cell for MTs. BCCH is mapped to a specific radiofrequency for each cell and it does not hop between radiofrequencies. MTs access the network using this channel bydecoding the base station identification code and frequency[29], [18]. UMTS: The Broadcast Channel (BCH) is usedto broadcast system- and cell-specific information. The BCHis always transmitted over the entire cell and has a singletransport format [28].cdma2000:The paging channel is usedto send common channel information that is used for accessingthe network [30].WLAN: The MT is assumed to listen to thechannel in passive mode for a beacon from a WLAN AP. Ifsuch a beacon is received, the master sub-layer is informed.When a beacon is received from one of the networks by thespecified ANI, the master sub-layer is informed about theexistence of an access network.

2) Network Modeling: Once the existence of a networkhas been informed to the master sub-layer, ANIs model thenetwork MAC structure as explained in Section IV and informthe master sub-layer as shown in Fig. 2. When a mobileterminal enters the coverage area of an AP, the network

9

TABLE I

TRAFFIC MODEL PARAMETERS

Parameter ValueAvg. Talk Duration 180 sAvg. Principal Talkspurt Duration 1.0 s

Voice Avg. Principal Gap Duration 1.35 sTraffic Avg. Minispurt Duration 0.275 s

Avg. Minigap Duration 0.05 sPacket Length 260 bits

CBR Avg. Talk Duration 360 sVideo Bit Rate 64 kbpsTraffic Packet Length 1280 bits

Avg. Talk Duration 180 sVBR Avg. State Holding Duration 160 msVideo Minimum Bit Rate 120 KbpsTraffic Maximum Bit Rate 420 Kbps

Avg. Bit Rate 239 KbpsPacket Length 840 bitsAvg. Packet Generation Rate 20 Msgs/s

Best-Effort Avg. Idle Time 2 s(ABR) Avg. Message Length 30 kBTraffic Avg. Frame Length 150 bits

specific information, such as slot information, data rates andmultiplexing information is captured by the ANI. The ANIthen, models the access structure according to a minimumresource allocation that is achievable from the network. Asthe MT is connected to the network, the BS may allocatemore bandwidth to the MT according to the QoS requirementsof the traffic. In this case, the associated ANI updates thecorresponding resource bin according to the allocated resource.Furthermore, as the modeled properties, i.e., data rate, spread-ing gain or frame dimensions vary during the connection, theANI updates the resource bin and informs the master sub-layer.

3) Access and Communication:An ANI associated witha network structure performs access to the network if itis selected for a transmission by the master sub-layer. Thecommunication with the AP is performed according to thenetwork specific procedures. Each network has specific pro-cedures for access phase and resource allocation. The packetstructures, message formats and signaling procedures alreadyexist for each network. Due to implementation and scalabilityconcerns, it is not feasible to propose modifications to theexisting MAC protocols that are deployed in the access pointsof these networks. Since there exists no unified standard foreach network, we assume that the ANIs are embedded withthe functionalities of the MAC protocols already deployed foreach network required for access and communication with theAPs. Once the underlying network is chosen to access, thecorresponding ANI performs the required access and resourceallocation procedures for accessing that specific network andtransmission.

VI. PERFORMANCEEVALUATION

In this section, we investigate the performance of A-MACfor different scenarios in CDMA, TDMA and CSMA basednetworks. In these scenarios, the throughput performance ofthe proposed framework is analyzed. We investigate the per-formance of an adaptive NG terminal equipped with A-MAC,

TABLE II

NETWORK MODEL PARAMETERS

TDMA SystemBS radius 25mFrame Length 4.6 msSlot Length 577 µsSlot rate 115.2 kbpsOperating Frequency 900 Hz

Voice CBR ABRMax. Slots 8 8 8Max. Slots for a MT 1 2 1Timeout 10ms 50ms 1 s

CDMA SystemBS radius 40mOperating Frequency 1.8GHz

Voice CBR VBR ABRBER 10

−310

−410

−610

−9

Max. Codes 16 8 4 4Max. Codes for a MT 1 1 2 1Timeout 20 ms 50 ms 40 ms 1 s

WLAN SystemAP radius 15mTransmission rate 1MbpsOperating Frequency 2.4GHz

in the presence of heterogeneous network architecture. In thesimulations, we simulate background traffic using designatednodes for each traffic model and network type. For the A-MAC protocol, decision interval,Td, is chosen as1s. Thetraffic models used in the simulations, the network structuresand the results of the performance evaluations are explained inSection VI-A, Section VI-B, and Section VI-C, respectively.

A. Traffic Models

Four types of traffic models are used in the simulations.We describe each traffic model in the following sections. Thevalues of the traffic model parameters are shown in Table I.Moreover, in Table I, the requirements for each traffic type ineach network in terms of BER and timeout are shown.

1) Voice Traffic:Voice traffic is modeled based on a three-step Markov model [2]. The voice traffic is assumed tohave main talkspurts and gaps denoted asprincipal talkspurtsand principal gaps, respectively. The principal talkspurt alsoconsists ofminispurts and minigaps. The duration of eachspurt and gap is exponentially distributed and statistically inde-pendent of each other. The talk duration is also exponentiallydistributed. The average duration of each distribution is shownin Table I. The stations generate IP packets with length of 260bits in every 20 seconds achieving a constant rate of 13kbpsduring the talkspurts.

2) CBR Video Traffic:CBR video traffic is modeled witha constant rate of packets generated with talk duration expo-nentially distributed.

3) VBR Video Traffic:VBR Video traffic models the video-phone and videoconference transmissions with a multi-statemodel [2]. The multi-state model generates continuous packetsfor a certain duration in each state. The bit rate in each stateis determined independently using a truncated exponentialdistribution. Holding duration of each state is exponentially

10

distributed and statistically independent of each other. Theduration of the traffic is also determined according to anexponential distribution.

4) Best-Effort (ABR) Traffic: ABR traffic models thenonreal-time best-effort data traffic. During the ABR connec-tion, multiple packets are sent. The length of each packet isgeometrically distributed and independent of each other. Thetotal length of the packets in a connection and the idle timebetween consecutive connections are exponentially distributed.

B. Network Models

Three different network structures are simulated throughoutthe performance evaluation. We describe these network struc-tures in the following sections.

1) TDMA System:A TDMA based system is simulatedwith time partitioned into time frames. Each time frame isfurther partitioned into time slots. The time frame length is4.6s with each frame containing 8 time slots. The time slot sizeis 577µs with a capacity of 156 bits. We simulate a TDMAbased system based on GPRS, such that multiple slots areassigned to the MT during one time frame to provide QoS ofdifferent traffic types. Each MT is assigned multiple slots ineach frame according to its traffic requirements. The systemparameters used in the simulations are shown in Table II. TheMaximum Time Slotsfield denotes the maximum number ofslots that can be allocated to a specific traffic type by the BS,whereMaximum Time Slots for a MTdenotes the maximumtime slots that can be allocated to a single MT in a frame.

2) CDMA System:An MC-CDMA system is simulatedwith time partitioned into uplink and downlink frames. Eachuplink and downlink frames is 10ms. A frame is furtherpartitioned into 15 slots. Each slot contains 2560 chips. TheCDMA network has a transmit rate of 3.84 MChips/s. Basedon the BER requirements and target SINR of each traffic type,multiple codes are assigned to an MT. The network providesthe MT up to 960kbps bandwidth. The base station is assumedto perform power control in each connection and the CDMAsystem is assumed to perform successful handoff as the mobileterminal roams through the network. The system parametersused in the simulations are shown in Table II.

3) WLAN System:The WLAN system is simulated basedon the IEEE 802.11 MAC layer [12]. The MT communicateswith the access point of the WLAN network throughout thecommunication. The transmission rate in the WLAN systemis assumed to be 1Mbps.

C. Simulation Results

The A-MAC protocol is simulated using the traffic modelsand network systems presented in Section VI-A and SectionVI-B, respectively. In a200mx200m grid, nodes are placedwith uniform distribution. Each node is assumed to be ableto connect to only one network. In addition, a node equippedwith A-MAC is simulated in the grid. We refer to this nodeas theAdaptive Nodethroughout our discussions. We vary thenode number in each network and the percentage of trafficdistribution to analyze the performance of A-MAC in theheterogeneous network structure. Each simulation lasts230s

and the results are average of5 trials for each5 randomtopologies.

Note that, although a single A-MAC node is investigatedthroughout the simulations, since A-MAC provides seamlessaccess and communication to each of the networks, eachconnection in the background traffic can also be consideredto be generated by another A-MAC node. A-MAC aims toprovide a seamless MAC protocol for accessing to hetero-geneous networks without requiring any modifications to theexisting medium access schemes. As a result, the resourceassignment for each connection attempt is handled by thespecific principles of the network. Hence, for a single A-MACnode, the effects of other A-MAC nodes are not distinguishablefrom any node inside a network.

As explained in Section I, in NG networks, adaptive nodesencounter heterogeneity in both network structure and traffictypes. In order to investigate the effects of these hetero-geneities on the performance of the A-MAC and evaluatethe adaptivity of the protocol, we performed two sets ofsimulations as explained in the following sections.

1) Fixed Topology: In this set of experiments, we areinterested in the adaptivity of the A-MAC protocol to theheterogeneity in the traffic load in different networks. Hence,we simulate a fixed topology where all the nodes, including theadaptive node, are stationary. The adaptive node is assumedtobe in the coverage area of three of the network structures, i.e.,TDMA, CDMA and WLAN access points3, throughout thesimulation. Each node in the network creates a specific typeof traffic and tries to send it to its designated access point.Thetraffic type distribution (Voice, ABR, CBR, VBR) is chosen as(65%, 15%, 10%, 10%). The number of nodes in each networkis defined by their percentage of the total number of nodesin the simulations. The distribution of the number of nodesin each network is defined as triple (nT , nC , nW ), wherenT ,nC , andnW denote the percentage of nodes in TDMA, CDMAand WLAN networks, respectively. The adaptive node createsfour types of the traffic according to the models describedin Section VI-A and tries to send it to one of the networks.We performed various simulations with varying load on thenetworks. The average throughput performance of A-MAC isshown in each case.

In Figs. 5(a) and 5(b), the average throughput of the A-MAC for voice traffic is shown with two node distributions,i.e., (nT , nC , nW ) = (10%, 10%, 80%) and (50%, 30%,20%). The horizontal axis represents the total number of nodescreating traffic throughout the simulations in the coverageareaof the three network access points. As shown in Fig. 5(a),voice traffic is served through the TDMA network when thetraffic load is low in the TDMA network. This decision is dueto the capability of TDMA structure to support voice traffic.However, when the load on the TDMA network is increased,as shown in Fig. 5(b), A-MAC uses multiple networks in orderto provide the required throughput of the voice traffic. Whenthe number of nodes inside the TDMA network exceeds10,(i.e. 21 ∗ 50%), 30% of the voice traffic is served through the

3We use the termaccess pointfor both the base stations and the accesspoints throughout the section

11

5 9 13 17 21 25 29 33 37 41 45 490

1000

2000

3000

4000

5000

6000

7000

8000

Number of Nodes

Avg

. Thr

ough

put (

bits

/s)

TDMA NetworkWLAN NetworkCDMA Network

5 9 13 17 21 25 29 33 37 41 45 490

1000

2000

3000

4000

5000

6000

7000

8000

Number of Nodes

Avg

. Thr

ough

put (

bits

/s)


(a) (b)

5 9 13 17 21 25 29 33 37 41 45 490

1

2

3

4

5

6

7

8x 10

4

Number of Nodes

Avg

. Thr

ough

put (

bits

/s)


5 9 13 17 21 25 29 33 37 41 45 490

1

2

3

4

5

6

7

8x 10

4


Number of Nodes

Avg

. Thr

ough

put (

bits

/s)

(c) (d)

Fig. 5. Average throughput of voice traffic with (nT , nC , nW ) = (a)(10%, 10%, 80%), (b)(50%, 30%, 20%), and CBR traffic with (nT , nC , nW ) =(c)(10%, 10%, 80%), (d)(50%, 30%, 20%).

WLAN access point. As the load is increased, TDMA networkcannot provide the required capacity and A-MAC switches toWLAN network. Fig. 5(b) shows that as the traffic load inWLAN network is increased, CDMA network is chosen inorder to achieve the throughput guarantee of the voice traffic.As shown in Fig. 5(b), as long as there is available capacity inthe reach of the adaptive node, A-MAC exploits this capacityto serve the required traffic.

In Figs. 5(c), and 5(d), the average throughput of CBRtraffic is shown. In both figures, the required throughput ismet for increasing traffic load. In Fig. 5(c), A-MAC accessesWLAN network to serve the CBR traffic when the traffic loadis low. When the WLAN traffic load is increased, the CBRtraffic requirements cannot be guaranteed, and the CDMAnetwork is chosen. The same behavior is also observed forthe VBR traffic, which is not shown here. A-MAC exploitsboth networks in order to provide the throughput required bythe CBR and VBR traffic, i.e., 64kbps and 239kbps. In Figs.6(a) and 6(b), the throughput of the best effort traffic for bothof the load distributions is shown. Since the best effort traffichas no bandwidth requirements, it is served as the resourcesin a network is available. As shown in Fig. 6(b), A-MACmainly serves best-effort traffic through the TDMA network.However, as the traffic load is increased, CDMA and WLAN

networks are used as they have available bandwidth. Thedecision is made mainly based on the decision requirements ofthe mobile station instead of traffic requirements as describedin Section V-A. Note that, achieved throughput is actuallylower than the average transmission rate of these traffic types.This is due to the channel errors and access collisions, whichalso increase with increasing traffic load. However, A-MACexploits the available bandwidth in the remaining networkstoguarantee the requirements of the traffic types.

2) Dynamic Topology:In order to investigate the adaptivityof the A-MAC protocol to the heterogeneity in the networkenvironment, we performed a second set of experiments witha dynamic topology. The network topology used in the simula-tions is shown in Fig. 7. The various points in Fig. 7 representthe TDMA, CDMA and WLAN access points with the circlesrepresenting their coverage areas. The topology is designedsuch that the left of the grid is covered by a TDMA networkand the right part is in the coverage area of a CDMA network.WLAN access points are spread over the topology includingthe overlapping area of TDMA and CDMA networks. Thistype of topology can simply be thought of as the hot spotsin a metropolitan network. In our simulations, we used100mobile nodes and a simple mobility pattern for the adaptivenode in order to observe the response of the A-MAC to

12

5 9 13 17 21 25 29 33 37 41 45 490

500

1000

1500

2000

2500

3000

3500

4000

4500

5000

Number of Nodes

Avg

. Thr

ough

put (

bits

/s)

TDMA NetworkCDMA NetworkWLAN Network

(a)

5 9 13 17 21 25 29 33 37 41 45 490

500

1000

1500

2000

2500

3000

3500

4000

4500

5000

Number of Nodes

Avg

. Thr

ough

put (

bits

/s)


(b)

Fig. 6. Average throughput of best-effort traffic with (nT , nC , nW ) =(a)(10%, 10%, 80%), (b)(50%, 30%, 20%).

0 20 40 60 80 100 120 140 160 180 2000

20

40

60

80

100

120

140

160

180

200

x (m)

y (m

)

Fig. 7. Sample topology used in the mobility simulations.•: TDMA BS, �:CDMA BS, �: WLAN AP.

varying network structures as shown in Fig. 7. The adaptivenode roams through TDMA to the CDMA network, passingthrough the coverage area of WLAN APs. The simulation lasts230s. The instantaneous throughput, calculated by averagingthe throughput for intervals of5s, is shown for each of thetraffic types in Fig. 8.

The adaptive terminal serves the voice traffic throughout

the simulation, through TDMA, WLAN and CDMA networksas shown in Fig. 8 (a). As the adaptive node enters theoverlapping area of the networks, the voice traffic is handedover to the WLAN network, since the adaptive node leavesthe coverage area of the TDMA network. In addition, part ofthe traffic is served through the CDMA network in order toachieve the traffic requirements. However, since the TDMAnetwork cannot allocate enough bandwidth to CBR and VBRtraffic due to high load in the network, the CBR and VBRtraffic is served only through WLAN APs while the adaptivenode is in the coverage area of the TDMA network as shownin Figs. 8 (b), and 8(c). As the adaptive mobile node entersthe coverage area of the CDMA network, the CBR and VBRtraffics are handed over to the CDMA network. In Fig. 8(c), the average throughput of the best-effort traffic is shown.A-MAC serves the best effort traffic through three types ofnetworks throughout the simulations. However, note that thetime the adaptive node resides in the WLAN network is alsoevident from the significant decrease in the best effort trafficthroughput. As shown in Fig. 8 (b) and 8(c), CBR and VBRtraffic are only served through the WLAN APs while theadaptive node is in the coverage area of the TDMA network.Hence, in order to serve these traffic types and meet thedecision requirements at the same time, A-MAC suppresseslowest priority best-effort traffic. The decrease in throughputin Fig. 8 (c) is due to the priority mechanism of A-MAC. Fig. 8(b) and Fig. 8(c) show that A-MAC provides smooth transitionbetween three network structures as the adaptive node roamsthrough them. The voice traffic requirements is met throughoutthe simulation, where CBR and VBR traffic is served throughthe WLAN APs and the CDMA network in the second partof the network, where required bandwidth is provided. Asshown in Fig. 8, A-MAC exploits the available bandwidth inits vicinity by performing necessary switches from networks tonetworks and aggregating available bandwidth via its multipleaccess capabilities.

In addition to adaptivity to heterogeneities of the nextgeneration wireless networks, A-MAC also provides a memoryefficient access capability to the next generation wirelessterminals. It is evident from the discussion in Section I thatnext generation wireless terminals will be equipped withphysical capabilities to access different types of multipleaccess technologies [1], [27]. Although the physical layerforheterogeneous wireless access is being developed, efficientnetwork protocols are still required for next generation wire-less terminals. The main concern in developing protocols forthe NG wireless terminals is the memory constraints posedby the mobile devices due to limited memory and processingcapabilities. Hence, the research is focused on developingsoftware tools for implementation of protocols with minimummemory footprint [22], in addition to consistent, platformindependent software interface layer for mobile applicationprocessors, i.e., managed runtime environments (MRTEs) [7],[13]. It is clear that implementing separate protocol stacks foraccessing each type of network is contradictory to the memoryefficiency of the MAC layer. Such an approach would requireexcessive memory requirements, independent access protocolsand inefficient resource management since these protocols

13

0 50 100 150 2000

1000

2000

3000

4000

5000

6000

7000

8000

Time(s)

Avg

. Thr

ough

put (

bits

/s)


0 50 100 150 2000

0.5

1

1.5

2

2.5

3x 10

5

Time (s)

Avg

. Thr

ough

put (

bits

/s)


(a) (b)

0 50 100 150 2000

0.5

1

1.5

2

2.5

3x 10

5

Time (s)

Avg

. Thr

ough

put (

bits

/s)


0 50 100 150 2000

500

1000

1500

2000

2500

3000

3500

4000

4500

5000

Time (s)

Avg

. Thr

ough

put (

bits

/s)


(c) (d)

Fig. 8. Throughput of (a) voice, (b) CBR, (c) VBR, and (d) ABR traffic in dynamic topology.

are not developed to coexist in the same device. However,it is also clear that limited access functionalities such as,synchronization procedures, packet structures, are required foraccess to each network scheme. As a result, A-MAC avoidswhole protocol stacks to be implemented to the NG wire-less terminals by implementing the required functionalities inthe ANIs and performs adaptive access decision by simpledecision and scheduling algorithms which provide seamlessmedium access to the upper network layers.

VII. C ONCLUSION

In this paper, we proposed an adaptive medium accesscontrol (A-MAC) for NG wireless terminals. To the best of ourknowledge, this is the first effort on designing a MAC protocolthat achieves adaptation to multiple network structures withQoS aware procedures without requiring any modificationsto the existing network structures. A-MAC provides seamlessaccess to multiple networks using a two-layered architectureto achieve adaptation to heterogeneous access networks andQoS requirements from traffic types. The access sub-layerconsists of ANIs, which are specialized for network accessstructures for accessing and communication. The master sub-layer coordinates incoming traffic packets and provides QoSbased service using the available access networks in the reach

of the MT.

We introduced a novel virtual cube model as a unit structureto compare different access schemes. According to the virtualcube model, access networks are modeled into three dimen-sional resource bins by the ANIs. Using these resource bins,the master sub-layer performs QoS-based decision in orderto choose the best access network for a specific traffic type.The schedulers associated with each ANI are invoked in caseof multiple traffic flows assigned to a single interface. Thescheduler performs QoS-based scheduling achieving fairnessand delay guarantees on traffic flows according to their pri-orities. The simulation results confirm that A-MAC providesQoS requirements of different traffic types by adapting tothe heterogeneity in the network structure and the availableresources in each networks in its reach.

Note that in our framework, we do not consider the cost ofswitching between network technologies. However, this factcan easily be incorporated into our decision framework. Morespecifically, if the cost of switching from a specific networkishigh, the decision interval can be increased such that frequenthandovers between networks is prevented. Another approachcould be to incorporate a cost function into the network mod-eling framework provided in Section VI-B. More specifically,a cost function can be incorporated into the network modeling

14

framework as a fourth dimension in the virtual cube conceptsuch that decision is performed accordingly. These possiblesolutions definitely introduce tradeoff between performanceand cost and this tradeoff is a further research topic in ourapproach.

REFERENCES

[1] S. Aggarwal, et al., “A Highly Integrated Dual-band Triple-mode Trans-mit IC for CDMA2000 Applications,” inProc. Bipolar/BiCMOS Circuitsand Technology Meeting, 2002, pp. 57 -60, 2002.

[2] I. F. Akyildiz, D. A. Levine, I. Joe, “A slotted CDMA protocol with BERscheduling for wireless multimedia Networks,”IEEE/ACM Transactionson Networking, vol. 7, no. 2, pp. 146-158, April 1999.

[3] P. Bhagwat, P. Bhattacharya, A. Krishna, S. K. Tripathi,“Enhancingthroughput over wireless LANs using channel state dependent packetscheduling,” inProc. IEEE INFOCOM ’96, vol. 3, pp. 1133-1140, March1996.

[4] L. Cheng, I. Marsic, “Modeling and prediction of sessionthroughputof constant bit rate streams in wireless data networks,” inProc. IEEEWCNC 2003, vol. 3, pp. 1733-1741, March 2003.

[5] S. Y. Cheung, C. S. Pencea, “BSFQ: Bin Sort Fair Queueing,” in Proc.IEEE INFOCOM 2002, vol. 3, pp. 1640 -1649, 2002.

[6] R. Cusani, F. D. Priscoli, G. Ferrari, M. Torregiani, “A novel MAC andscheduling strategy to guarantee QoS for-the new-generation WIND-FLEX wireless LAN,” IEEE Wireless Communications Magazine, vol.9, no. 3, pp. 46-56, June 2002.

[7] P. Drews, D. Sommer, R. Chandler, T. Smith, “Managed runtime envi-ronments for next-generation mobile devices,”Intel Technology Journal,vol. 7, no. 1, pp. 77 -82, Feb. 2003.

[8] A. El Gamal, C. Nair, B. Parabhakar, E. Uysal-Biyiklioglu, S. Zahedi,“Energy-efficient scheduling of packet transmissions overwireless net-works,” in Proc. IEEE INFOCOM 2002,vol. 3, pp. 1773 -1782, 2002.

[9] G. Holland, N. Vaidya, and P. Bahl, “A rate-adaptive MAC protocolfor multi-hop wireless networks,” inProc. ACM MOBICOM ’01, pp.236-251, Aug. 2001.

[10] V. Huang, W. Zhuang, “Optimal resource management in packet-switching TDD CDMA systems,”IEEE Personal Communications,vol.7, no. 6, pp. 26 -31, Dec. 2000.

[11] V. Huang, W. Zhuang, “QoS-Oriented access control for 4G multimediaCDMA communications,”IEEE Communications Magazine, vol. 20, no.3, pp. 118-125, March 2002.

[12] IEEE 802.11, “Wireless LAN medium access control (MAC)andphysical layer (phy) specifications,” 1999.

[13] Intel, “The Intel(R) XScale(TM) microarchitecture Technical Summary,”http://www.intel.com/design/intelxscale/

[14] D. N. Knisely, S. Kumar, S. Laha, S. Nanda, “Evolution ofwirelessdata services: IS-95 to cdma2000,”IEEE Communications Magazine,vol. 36, no. 10, pp. 140 -149, Oct. 1998.

[15] J. W. Lee, R. R. Mazumdar, N. B. Shroff, “Downlink power allocationfor multi-class CDMA wireless networks,” inProc. IEEE INFOCOM2002, vol. 3, pp. 1480 -1489, 2002.

[16] S. Lu, V. Bharghavan, R. Srikant, “Fair scheduling in wireless packetnetworks,” IEEE/ACM Transactions on Networking, vol. 7, no. 4, pp.473-489, Aug. 1999.

[17] Z. Naor, H. Levy, “A centralized dynamic access probability protocolfor next generation wireless networks,” inProc. IEEE INFOCOM 2001,vol. 2, pp. 767 -775, 2001.

[18] T. T. Nielsen, J. Wigard, “Performance enhancements ina frequencyhopping GSM network,” Boston: Kluwer Academic Publishers,2000.

[19] S. Ramanathan, “A unified framework and algorithm for (T/F/C)DMAchannel assignment in wireless networks,” inProc. IEEE INFOCOM’97, vol. 2, pp. 900-907, April 1997.

[20] B. Sadeghi, V. Kanodia, A. Sabharwal, and E. Knightly, “Opportunisticmedia access for multirate ad hoc networks,” inProc. ACM MOBICOM’02, Atlanta, GA, Sept. 2002.

[21] O. Sallent, J. P. Romero, R. Agusti, F. Casadevall, “Provisioningmultimedia wireless networks for better QoS: RRM strategies for 3GW-CDMA,” IEEE Communications Magazine, vol. 41, no. 2, pp. 100-106, Feb 2003.

[22] Sun Microsystems, “The Connected Limited Device Configuration(CLDC) HotSpot Implementation Virtual Machine (JAVA 2 Plat-form, Micro Edition),” Sun Microsystems White Paper[Online].http://java.sun.com/products/cldc/

[23] A. J. Viterbi, “Cdma: Principles of Spread Spectrum Communication,”Prentice Hall PTR; 1st edition, June 1995.

[24] X. Wang, “An FDD wideband CDMA MAC protocol for wirelessmultimedia networks,” inProc. IEEE INFOCOM 2003, vol. 1, pp. 734- 744, March 2003.

[25] C.-H. Yeh, “ACENET: architectures and protocols for high throughput,low power, and QoS provisioning in next-generation mobile communi-cations, ” inProc. IEEE PIMRC 2002, vol. 2, pp. 779 -786, 2002.

[26] C.-H. Yeh, “A TCDMA protocol for next-generation wireless cellularnetworks with bursty traffic and diverse QoS requirements,”in Proc.IEEE PIMRC 2002, vol. 5, pp. 2142 -2147, 2002.

[27] X. Zhu, et al., “The RF module design for W-CDMA/GSM dualband and dual mode handset,”IEEE MTT-S International MicrowaveSymposium Digest, 2001, vol. 3, pp. 2215 -2218, 2001.

[28] 3GPP TS 25.211: “Universal Mobile TelecommunicationsSystem(UMTS); physical channels and mapping of transport channels ontophysical channels (FDD),” version 5.2.0, release 5, Sept. 2002.

[29] 3GPP TS 05.02: “Digital cellular telecommunications system (Phase2+); Multiplexing and multiple access on the radio path,” version 8.8.0,Jan. 2001.

[30] 3GPP2 C.S0003-0-2: “Medium Access Control (MAC) standard forcdma2000 spread spectrum systems,” version 1.0, release 0,addendum2, April 2001.

Mehmet C. Vuran (M’98) received his B.Sc. degreein electrical and electronics engineering from BilkentUniversity, Ankara, Turkey, in 2002. He received hisM.S. degree in electrical and computer engineeringfrom School of Electrical and Computer Engineer-ing, Georgia Institute of Technology, Atlanta, GA,in 2004. He is currently a Research Assistant in theBroadband and Wireless Networking Laboratory andpursuing his Ph.D. degree at the School of Electri-cal and Computer Engineering, Georgia Institute ofTechnology, Atlanta, GA under the guidance of Prof.

Ian F. Akyildiz. His current research interests include adaptive and cross-layer communication protocols for heterogeneous wirelessarchitectures, nextgeneration wireless networks, and wireless sensor networks.

Ian F. Akyildiz (M’86-SM’89-F’96) received theB.S., M.S., and Ph.D. degrees in Computer Engi-neering from the University of Erlangen-Nuernberg,Germany, in 1978, 1981 and 1984, respectively.

Currently, he is the Ken Byers DistinguishedChair Professor with the School of Electrical andComputer Engineering, Georgia Institute of Tech-nology, Atlanta, and Director of Broadband andWireless Networking Laboratory. He is an Editor-in-Chief of the Computer Networks (Elsevier)and ofthe AdHoc Network Journal (Elsevier). His current

research interests are in sensor networks, IPN Internet, wireless networks, andsatellite networks.

He received the “Don Federico Santa Maria Medal” for his services tothe Universidad of Federico Santa Maria, in 1986. From 1989 to 1998,he served as a National Lecturer for ACM and received the ACM Out-standing Distinguished Lecturer Award in 1994. He receivedthe 1997 IEEELeonard G. Abraham Prize Award (IEEE Communications Society) for hispaper entitled “Multimedia Group Synchronization Protocols for IntegratedServices Architectures” published in the IEEE Journal of Selected Areasin Communications (JSAC) in January 1996. He received the 2002 IEEEHarry M. Goode Memorial Award (IEEE Computer Society) with the citation“for significant and pioneering contributions to advanced architectures andprotocols for wireless and satellite networking”. He received the 2003 IEEEBest Tutorial Award (IEEE Communication Society) for his paper entitled ”ASurvey on Sensor Networks,” published in IEEE Communications Magazine,in August 2002. He also received the 2003 ACM Sigmobile OutstandingContribution Award with the citation “for pioneering contributions in the areaof mobility and resource management for wireless communication networks”.He has been a Fellow of the Association for Computing Machinery (ACM)since 1996.

a-mac: adaptive medium access control for next ...cpn.unl.edu/?q=system/files/amac.pdfwireless...

Documents