ieee 802.11n mac frame aggregation mechanisms for next-generation high-throughput wlans [medium...

IEEE Wireless Communications • February 200840 1536-1284/08/$25.00 © 2008 IEEE

A-MSDU

MPDUframe 1

MPDUsubframe 2

MPDUdelimiter MPDU

MSDUsubframe 1

MPDUdelimiter PadMPDU

MPDUframe 1

MPDUsubframe 2

MSDUsubframe 2

MEDIUM ACCESS CO N T R O L PR O T O C O L S FOR WIRELESS LANS

INTRODUCTIONOver the last decade, the applications of wirelessand cellular devices have expanded rapidly. Themost popular network in the wireless domain isthe IEEE 802.11 wireless local area network(WLAN) [1] mainly because of the advantagesthat its systems possess. Its key characteristics,such as interoperability, mobility, flexibility, andcost-effective deployment, have led to its gainingvast support across enterprises, the public sector,homes, and data service providers. However, thelow efficiency of its medium access control(MAC) and physical (PHY) layer protocolsrestricts its applications to support high data ratemultimedia services. Current WLAN systemsendure difficulties with the increasing expecta-tions of end users and with volatile bandwidthand delay-boundary demands from new higherdata rate services, such as high-definition televi-

sion (HDTV), video teleconferencing, multime-dia streaming, voice over IP (VoIP), file trans-fer, and online gaming.

In July 2002, the IEEE 802.11 standard work-ing group established the High-ThroughputStudy Group (HTSG) with the aim to achievehigher data rate solutions by means of existingPHY and MAC mechanisms [2, 3]. Its first inter-est was to achieve a MAC data throughput over100 Mb/s using the 802.11a standard. However,the objective proved to be infeasible as the esti-mated throughput bounds are well below thetheoretical maximum link rate because of theexisting MAC and PHY overhead [4, 5]. So, inSeptember 2003, the HTSG set off the IEEE802.11n (“n” represents next-generation) resolu-tion to compose a high-throughput (HT) exten-sion of the current WLAN standard that willincrease transmission rate and reduce compulso-ry overhead. The main goal of the IEEE 802.11nTask Group (TGn) is to define an amendmentthat will have maximum data throughput of atleast 100 Mb/s, as measured at the MAC dataservice access point (SAP), and at the sametime, to allow coexistence with legacy devices.Some of their recent propositions [6] for thePHY include multiple input multiple output(MIMO) antennas with orthogonal frequencydivision multiplexing (OFDM) and various chan-nel binding schemes. For the MAC, the maindevelopments are the introduction of frameaggregation at the same time as multiple protec-tion schemes have been designed to allow coex-istence of “n” with legacy devices.

In this article we mainly focus on the MACframe aggregation methods that TGn has pro-posed in the latest 802.11n standard draft, andhow these can improve WLANs by way of higherthroughput and maximized efficiency. Thus, webegin with a brief outline of the current IEEE

DIONYSIOS SKORDOULIS AND QIANG NI, BRUNEL UNIVERSITY

HSIAO-HWA CHEN, NATIONAL SUN YAT-SEN UNIVERSITY

ADRIAN P. STEPHENS, INTEL, UKCHANGWEN LIU, ENTROPIC COMMUNICATIONS

ABBAS JAMALIPOUR, UNIVERSITY OF SYDNEY

ABSTRACTIEEE 802.11n is an ongoing next-generation

wireless LAN standard that supports a very high-speed connection with more than 100 Mb/s datathroughput measured at the medium access con-trol layer. This article investigates the key MACenhancements that help 802.11n achieve highthroughput and high efficiency. A detaileddescription is given for various frame aggrega-tion mechanisms proposed in the latest 802.11ndraft standard. Our simulation results confirmthat A-MSDU, A-MPDU, and a combination ofthese methods improve extensively the channelefficiency and data throughput. We analyze theperformance of each frame aggregation schemein distinct scenarios, and we conclude that over-all, the two-level aggregation is the most effica-cious.

IEEE 802.11N MAC FRAME AGGREGATIONMECHANISMS FOR NEXT-GENERATION

HIGH-THROUGHPUT WLANS

The authors investigate the keyMAC enhancementsthat help 802.11nachieve highthroughput and highefficiency. A detaileddescription is givenfor various frameaggregation mechanisms proposed in the latest 802.11n draft standard.

CHEN LAYOUT 1/31/08 11:54 AM Page 40

IEEE Wireless Communications • February 2008 41

802.11 WLAN standards, followed by a discus-sion of its maximal throughput limitationsbecause of overhead. We then concentrate onthe latest work of 802.11n on both PHY andMAC, but elaborate more on the latter. We alsoevaluate, via simulations, the improved perfor-mance when using frame aggregation and itsefficiency over distinct scenarios. Finally, weconclude by summarizing this article’s findings.

OVERVIEW OF IEEE 802.11IEEE 802.11 PHY

The IEEE 802.11 PHY layer specification con-centrates mainly on wireless transmission andconcurrently, performs secondary functions, suchas assessing the state of the wireless medium andreporting it back to the MAC sublayer. The orig-inal specification was first approved in 1997 [1]and includes a primitive MAC architecture andthree basic over-the-air communication tech-niques with maximal raw data rates of 1 and 2Mb/s. Because of their fairly low data band-widths, further amendments have been proposedthroughout the years: IEEE 802.11a [7], 802.11b,and 802.11g [8]. Both 802.11a and 802.11b werefinalized in 1999 and support raw data rates upto 11 Mb/s and 54 Mb/s, respectively. In June2003, a third PHY specification (802.11g) wasintroduced, with similar maximum raw data rateas 802.11a but operating in separate frequencybands. For this period, there were many amend-ments and countless research works for improvedPHY specifications that mostly aim to providereliable connections and higher data rates. Thisis mainly because there is a continuous rapidincrease in user demand for faster connections.

In spite of establishing novel techniques thattheoretically can be used for higher data trans-mission rates, the throughput outcomes at theMAC data SAP are surprisingly low and in mostcases, half of what the underlying PHY rates canoffer [4].

IEEE 802.11 MACThe MAC architecture is based on the logicalcoordination functions that determine who andwhen to access the wireless medium at any time.It supports fragmentation and encryption andacts as an interface between the logical link con-

trol (LLC) sublayer and the PHY layer. In thelegacy IEEE 802.11 standard, there are twotypes of access schemes: the mandatory dis-tributed coordination function (DCF), which isbased on the carrier sense multiple access withcollision avoidance (CSMA/CA) mechanism;and the optional point coordination function(PCF), which is based on a poll-and-responsemechanism. Because these MAC schemes areinadequate to resolve differentiation and priori-tization between frames and multimedia applica-tions such as VoIP and audio/video conferencingwith strict performance constraints have becomewidely popular, a new extension was vital. In late2005, IEEE 802.11 Working Group (WG)approved the IEEE 802.11e amendment [9] toprovide an acceptable level of quality of service(QoS) for multimedia applications. The 802.11eproposes the hybrid coordination function(HCF), which uses a contention-based channelaccess method, known as enhanced DCF chan-nel access (EDCA). EDCA has the ability tooperate simultaneously with a polling-basedHCF controlled channel access (HCCA). Inaddition to the differentiation and prioritizationthat IEEE 802.11e offers, the transmissionopportunity (TXOP), an interval of time inwhich multiple data frames can be transferredfrom one STAtion (STA) to another — alsoknown as bursting — was introduced as a way toimprove MAC efficiency. Along with framebursting, another type of acknowledgment(ACK), known as block ACK, was established soreceivers can acknowledge multiple receiveddata frames sufficiently and economically byusing just a single extended ACK frame. Thetraffic flows are characterized through the trafficidentifiers (TID) and the traffic specification(TSPEC) frames, a set of information that speci-fies the corresponding QoS requirements.

THROUGHPUT LIMITATIONSTo understand the inefficiency of IEEE 802.11over higher data rates, we must briefly describethe legacy DCF. This method operates with afirst-in-first-out (FIFO) transmission queue thatis situated for receiving and buffering incomingdata from the higher layers. The basic operationof DCF is illustrated in Fig. 1. After a frame,also known as a MAC service data unit (MSDU),arrives from the LLC at the head of the trans-

n Figure 1. DCF basic operation.

Direct access(NAV + DIFS) Backoff

Channelbusy ACK

Time

SIFSPLCP

preamblePLCP

headerMAC

header FCSFCSMSDU (payload)DIFS

Packet

AcknowledgmentPPDU

Overhead

The MAC architectureis based on the

logical coordinationfunctions that

determine whomand when to access

the wireless mediumat any time.

It supports fragmentation and

encryption and actsas an interface

between the logicallink control sublayerand the PHY layer.


IEEE Wireless Communications • February 200842

mission queue, the DCF operation instructs theMAC to wait for a global defined interframeinterval called DCF interframe space (DIFS)before any other actions can be taken. If thePHY reports back to the MAC that the wirelesschannel is busy, the STA’s MAC halts until themedium becomes free. On the other hand, if themedium remains idle during DIFS deference,the STA enters a back-off procedure where aslot is selected from a random back-off counterwithin a contention window (CW). Next, thecounter starts a decrement process while thechannel remains idle for each slot interval. Whenthe counter reaches zero, the STA obtains anaffirmation to send the information through thewireless link. Now, each STA that receives adata frame utilizes an error checking processesto detect the presence of errors. If no errors arefound, it sends back an ACK frame after a speci-fied short interframe space (SIFS) to verify thatthe information was successfully received. If thesending STA does not receive an ACK afterSIFS, it will assume that the communication wasbroken or interfered, and it will start a new DCFprocess for retransmission. If there is a case ofcollision, then the MAC extends its CW, selectsa new slot, and repeats the previous steps. Final-ly, there is an optional mechanism known asrequest to send/clear to send (RTS/CTS) thatintends to resolve the so-called hidden andexposed node scenarios that usually occur in adhoc networks. With RTS/CTS, after a STA isgranted access to transmit, it first sends an RTSframe and then holds back for the CTS responsefrom the receiver. Obviously, this situation canbe disadvantageous if actual data frame size issmall because the RTS/CTS exchange producesfurther overhead and consequently reduces theeffective throughput.

We can comprehend clearly the consequencesof that hefty overhead on the system throughputfrom Fig. 1 (RTS/CTS has been omitted for sim-plicity). The figure illustrates the required proce-dure that each single packet traverses from thetime it arrives at the MAC until it is successfullyreceived by the receiver, with different headersand tails added on the original payload over dif-ferent sublayers. Note that [1] states that allphysical layer convergence protocol (PLCP)preambles and PLCP headers shall be transmit-ted using the basic (most of the time, the mini-mum) link rate, which is much less than the rateused for data transmission. A complete transmis-sion cycle of a simple DCF consists of DIFSdeferral, back off, data transmission, SIFS defer-ral, ACK transmission, and propagation delay.As shown in Fig. 1, to transmit a data packet, alarge overhead is associated. Be aware that theoverhead shown does not correspond to realtime lengths, as payload varies, but it shows theadditional time that is required to have a suc-cessful transmission. So, the higher the packetrate — meaning the number of packets that areinjected to the MAC per second — the higherthe relative overhead the system introduces. Atthis point the analysis in [5] has shown that themaximum ideal throughput is bounded at 50Mb/s when the frame size is 1 KB (or 1024 B),and the link rate increases to infinite. In general,when a legacy device uses a 54 Mb/s link, it can

provide only a maximum relative MAC through-put that is less than 50 percent of the peak PHYor around 25 Mb/s. Consequently, it was con-firmed that to reach the TGn target of an achiev-able MAC layer throughput higher than 100Mb/s, it is essential to adopt innovative MACtechniques. Because most of these types of MACand PHY overheads are essential, we must focuson how we can reduce the frequency of themduring the transmission of multiple packets. Thiscan be achieved by concatenating or packingthem together and adding the overheads over agroup of packets rather separate ones, and thisis known as aggregation. Various methods ofaggregation (e.g., [5, 6, 10–13]) have been pro-posed, but all of them follow a similar logic. Inthe next section, we will describe the TGn pro-posal for frame aggregation.

IEEE 802.11N ARCHITECTURE

IMPROVEMENTS ON PHY LAYER

The key underlying model defined in the specifi-cation for the PHY layer operates multiple anten-nas for both transmitter and receiver. The use ofMIMO provides many benefits, such as antennadiversity and spatial multiplexing. In general,MIMO technology increases the spectral efficien-cy of a wireless communication system. Tradi-tional single-input single-output (SISO) systemstend to immobilize certain propagation phenom-ena, such as multipath propagation. Multipath istypically perceived as interference degrading theability of a receiver to recover intelligent infor-mation. However, by using multiple antennas, anexploitation of the multipath phenomena canturn advantageous as the data throughput andrange increases, and the bit error rate decreases.MIMO technology has the ability to simultane-ously resolve information from multiple signalpaths using spatially separated receive antennas.

Another important proposition increases thebandwidth of the current channel from 20 MHzto 40 MHz. Using a wider channel bandwidth,we improve the theoretical capacity limits; thiscan easily be seen from Shannon’s capacity equa-tion C = B log2(1+SNR). Thus, if properlyimplemented, the 40 MHz channels can becomemore desirable than two times the usable chan-nel bandwidth of two 802.11 legacy channels.Diversity in a multi-antenna system is achievableeither through space-time coding or using thechannel state information at the transmitterintelligently. Before transmission, a sessionexchange of sounding PHY protocol data units(PPDUs) is required to calibrate the radio chan-nel. Based on the information elements collectedfrom sounding and calibration sequences, beamforming can be used to boost signal quality byselecting the proper power and coding schemefor the spatial streams. Higher signal qualitymeans that a given data rate can be available atlonger range. For a given signal-to-noise ratio(SNR), a beam-formed transmission can carry ahigher data rate. Finally, transmission uses dif-ferent coding schemes as they perform analo-gously, and thus MIMO has two coding schemes,space time block coding (STBC) and low densityparity check coding (LDPC).

The key underlyingmodel defined in thespecification for thePHY layer operatesmultiple antennas forboth transmitter andreceiver. The use ofMIMO providesmany benefits, suchas antenna diversityand spatial multiplexing.



Some of the improvements have been classi-fied as optional because they cause technicalhitches when non-802.11n devices are present inthe same WLAN. Nonetheless, this matter doesnot demote their importance; on the contrary ifthese mechanisms are employed over a networkthan includes only HT nodes, the PHY rate canincrease as much as 600 Mb/s [6].

MAC ENHANCEMENTSIn this subsection, we briefly mention the mostimportant MAC enhancements that are pro-posed in [6] with a more detailed explanation onframe aggregation, which maximizes throughputand efficiency. There are two methods availableto perform frame aggregation: aggregate MACprotocol service unit (A-MSDU) and aggregateMAC protocol data unit (A-MPDU). The maindistinction between an MSDU and an MPDU isthat the former corresponds to the informationthat is imported to or exported from the upperpart of the MAC sublayer from or to the higherlayers, respectively, whereas the later relates tothe information that is exchanged from or to thePHY by the lower part of the MAC. Aggregateexchange sequences are made possible with aprotocol that acknowledges multiple MPDUswith a single block ACK in response to a blockacknowledgment request (BAR). Another keyenhancement that the 802.11n specifies is thebidirectional data transfer method over a singleTXOP, known as reverse direction. This featurepermits the transportation of data frames, evenaggregates, in both directions in one TXOP.Until now, when the sender STA is allocatedwith a TXOP, it informs surrounding STAsabout how long the wireless medium will beengaged. However, this approximation of chan-nel use is not always accurate, and often thetransmission ends sooner. As a result, contendedSTAs assume that the channel is still occupiedwhen this is not the case. With reverse direction,the initial receiver STA is allowed to send anypackets available that are addressed to thesender for the remaining TXOP time. This fitsespecially well with TCP because it allows a TCPlink to piggyback TCP ACK collection onto TCP

data transmission. The long network allocationvector (long-NAV) is another enhancement thatimproves scheduling, given that a station thatholds a TXOP may set a longer NAV valueintended to protect multiple PPDUs. Anothermandatory feature is phased coexistence opera-tion (PCO) which protects stations using either20 MHz or 40 MHz channel spectrum at thesame time. Finally, the reduced IFS (RIFS) isproposed to allow a time interval of 2 µs betweenmultiple PPDUs, which is much shorter thanSIFS as defined in the legacy standards.

A-MSDU — The principle of the A-MSDU (orMSDU aggregation) is to allow multiple MSDUsto be sent to the same receiver concatenated in asingle MPDU. This definitely improves the effi-ciency of the MAC layer, specifically when thereare many small MSDUs, such as TCP acknowl-edgments. This supporting function for A-MSDUwithin the 802.11n is mandatory at the receiver.For an A-MSDU to be formed, a layer at the topof the MAC receives and buffers multiple pack-ets (MSDUs). The A-MSDU is completed eitherwhen the size of the waiting packets reaches themaximal A-MSDU threshold or the maximaldelay of the oldest packet reaches a pre-assignedvalue. Its maximum length can be either 3839 or7935 bytes; this is 256 bytes shorter than the max-imum PHY PSDU length (4095 or 8191 bytes,respectively), as predicted space is allocated forfuture status or control information. The size canbe found in the HT capabilities element that isadvertised from an HT STA in order to declareits HT status. The maximal delay can be set to anindependent value for every AC but is usually setto 1 µs for all ACs. There are also certain con-straints when constructing an A-MSDU:• All MSDUs must have the same TID value• Lifetime of the A-MSDU should correspond

to the maximum lifetime of its constituent ele-ments

• The destination address (DA) and senderaddress (SA) parameter values in the sub-frame header must match to the same receiveraddress (RA) and transmitter address (TA) inthe MAC header.

n Figure 2. One-level frame aggregation: a) A-MSDU; b) A-MPDU.

PHY HDR A-MPDU

Subframe 1 Subframe NSubframe 2

Reserved MPDUCRC

PHY HDR MAC HDR FCSA-MSDU

Subframe 1 Subframe 2 Subframe N

DA SA Length PaddingMSDUDAReserved Delimiter

signature MPDUMPDUlength PaddingCRC

SA Length

Subframe HDR

PaddingMSDU

Subframe 1 Subframe 1 Subframe NSubframe 2

PHYHDR MACHDR FCS PHYHDRA-MSDU A-MPDU

PSDU

(a)

Subframe 2 Subframe N

6Bytes: Bytes:

Bits: B0 B3 B4 B15 B16 B23 B24 B31MPDUdelimiter Variable 0-3

6 2 0-30-2304

PSDU

(b)

CHEN LAYOUT 1/31/08 11:54 AM 3


Thus, broadcasting or multicasting is notallowed.

Figure 2a describes a simple structure of acarrier MPDU that contains an A-MSDU. Eachsubframe consists of a subframe header followedby the packet that arrived from the LLC and 0 –3 bytes of padding. The padding size depends onthe rule that each subframe, except for the lastone, should be a multiple of four bytes, so theend receiver can approximate the beginning ofthe next subframe. A major drawback of using A-MSDU is under error-prone channels. By com-pressing all MSDUs into a single MPDU with asingle sequence number, for any subframes thatare corrupted, the entire A-MSDU must beretransmitted. This situation has been addressedin [5, 13], where additional frame structures oroptimum frame sizes have been proposed toimprove performance under noisy channels.

A-MPDU — The concept of A-MPDU aggregationis to join multiple MPDU subframes with a sin-gle leading PHY header. A key difference fromA-MSDU aggregation is that A-MPDU func-tions after the MAC header encapsulation pro-cess. Consequently, the A-MSDU restriction ofaggregating frames with matching TIDs is not afactor with A-MPDUs. However, all the MPDUswithin an A-MPDU must be addressed to thesame receiver address. Also, there is no wait-ing/holding time to form an A-MPDU so thenumber of MPDUs to be aggregated totallydepends on the number of packets already in thetransmission queue. The maximum length thatan A-MPDU can obtain — in other words themaximum length of the PSDU that may bereceived — is 65,535 bytes, but it can be furtherconstrained according to the capabilities of theSTA found in the HT capabilities element. Theutmost number of subframes that it can hold is64 because a block ACK bitmap field is 128bytes in length, where each frame is mappedusing two bytes. Note that these two bytes arerequired to acknowledge up to 16 fragments butbecause A-MPDU does not allow fragmentation,

these extra bits are excessive. As a result, a newvariant has been implemented, known as com-pressed block ACK with a bitmap field of eightbytes long. Finally, the size of each subframe islimited to 4095 bytes as the length of a PPDUcannot exceed the 5.46-ms time limit; this can bederived from the maximum length divided by thelowest PHY rate, which is 6 Mb/s and is thehighest duration of an MPDU in 802.11a.

The basic structure is shown in Fig. 2b. A setof fields, known as delimiters are inserted beforeeach MPDU and padding bits varied from 0 – 3bytes are added at the tail. The basic operationof the delimiter header is to define the MPDUposition and length inside the aggregated frame.It is noted that the cyclic redundancy check(CRC) field in the delimiter verifies the authen-ticity of the 16 preceding bits. The padding bytesare added such that each MPDU is a multiple offour bytes in length, which can assist subframedelineation at the receiver side. In other words,the MPDU delimiters and PAD bytes determinethe structure of the A-MPDU. After the A-MPDU is received, a de-aggregation process ini-tiates. First it checks the MPDU delimiter forany errors based on the CRC value. If it is cor-rect, the MPDU is extracted, and it continueswith the next subframe till it reaches the end ofthe PSDU. Otherwise, it checks every four bytesuntil it locates a valid delimiter or the end of thePSDU. The delimiter signature has a uniquepattern to assist the de-aggregation processwhile scanning for delimiters.

Two-Level Aggregation — A two-level frame aggre-gation comprises a blend of A-MSDU and A-MPDU over two stages. In Fig. 3 we illustratehow this new scheme can be achieved.

The basic operation is explained as follows:In the first stage, if any MSDUs that arebuffered in the A-MSDU provisional storagearea justify the A-MSDU constraints explainedin the previous related subsection, these dataunits can be compacted into a single A-MSDU.If the TIDs are different, all these aberrantframes can move to the second stage where theywill be packed together with any A-MSDUsderived from the first stage or other singleMSDUs by using A-MPDU aggregation. Howev-er, it must be mentioned that given that themaximum MPDU length for an A-MPDU dataframe is limited to 4095 bytes, then A-MSDUsor MSDUs with lengths larger than this thresh-old can not be transmitted. Conjointly, any frag-ments from an A-MSDU or MSDUs also cannotbe included in an A-MPDU. In the followingsection, we evaluate how this synthesis is moreefficient in most of the cases than A-MPDU andA-MSDU aggregation operating alone.

PERFORMANCE EVALUATION

In this section, we compare the performance ofthe latest draft of the IEEE 802.11n A-MSDUand A-MPDU aggregation schemes along withthe two-level aggregation. For the simulations,we used a simulation model implemented byIntel, based on the OPNET Modeler [14] withthe latest 802.11n PHY and MAC enhance-ments.

n Figure 3. Two-level frame aggregation.

A-MSDU

PSDU (A-MPDU)

PHY HDR MPDUsubframe 1

MPDUsubframe 2

MPDUsubframe N

MPDUdelimiter MPDU

MAC HDR FCSMSDUsubframe 1

MSDUsubframe 1

MSDUsubframe N

MSDUSubframeHDR

MPDUdelimiter PaddingMPDU

MACHDR

PHYHDR MPDUsubframe 1

MPDUsubframe 2

MPDUsubframe N

FCSMSDUsubframe 1

MSDUsubframe 2

MSDUsubframe N

PaddingMSDU


SIMULATION SCENARIOS

The IEEE 802.11n Task Group has predefinedspecific usage models [15] based on various mar-ket-based use cases. The usage models intend tosupport the definitions of network simulationsthat will allow them to evaluate performance ofvarious proposals in terms of, for example, net-work throughput, delay, packet loss, and othermetrics. In this subsection, we test the efficiencyof aggregation over two high throughput sta-tions. This scenario also is known by the TGn asScenario 17 (point-to-point goodput test). Here,we study the maximum throughput for each pro-posed aggregating scheme while increasing theoffered load or when the load remains constantbut the packet size is increasing. From these sce-narios, we also can observe the degrading chan-nel efficiency when aggregation is disabled, andthe system uses in full its latest PHY techniques(e.g., MIMO).

The scenario is situated in an infrastructureservice area that operates under the EDCAmode. It includes a fixed HT AP and a fixed HTSTA, both operating in a 20-MHz channel andusing the 64-QAM 3/4 modulation codingscheme (MCS) as their operational data ratewith two antennas. The devices are placed over adistance of ten meters with line of sight (LOS).The first station has a synthetic data source thatprovides varying offered loads (in Mb/s) of UserData Protocol (UDP) traffic. These UDPsources have no timeout values specified, andthey all have the same TIDs. To understand infull the remarkable improvement of frame aggre-gation, we assume that transmission proceedswith no interference or channel fading and thatall frames are received successfully, and noretransmission is required. The simulation timeis ten seconds, which is an adequate period formultiple packets to be transmitted over the twoSTAs, taking into consideration that packetarrival at the MAC layer sets off immediately.

SIMULATION RESULTS AND DISCUSSIONFigure 4 illustrates the throughput results (inMb/s) obtained from the MAC SAP while theoffered load (OL) from the associated HT STAaccumulates gradually. For the frame aggrega-tion, we define four different settings:• Both A-MPDU and A-MSDU are enabled• Only the A-MPDU algorithm is used• Only the A-MSDU algorithm is used• No aggregation at all

In the first simulation, the OL is incrementedby just varying the packet size while keeping theconstant packet generation interval, also knownas the constant packet rate (CPR), which is 40µs. The initial OL is 25 Mb/s, and it increases upto 300 Mb/s with the increase of packet size (i.e.,125/250/500/750/1000/1500 bytes). For each case,the traffic generation rate is configured highenough to saturate the air link rate that corre-sponds to the “PHY peak” (144 Mb/s) on eachfigure, and the maximal A-MSDU length is 4KB. As shown in Fig. 4, we observe that allthroughputs increase according to the load. Ingeneral, as the packet size increases, the A-MSDU stays below 75 Mb/s, while A-MPDUand the two-level aggregation achieve maximum

throughputs of 136 and 134 Mb/s, respectively.When the packet size maintains values of 125 B(see OL = 25 Mb/s) and 250 B (see OL = 50Mb/s), the corresponding throughputs for anytype of aggregation are similar. This is becauseA-MSDU can aggregate several small packetswithin a single MPDU, even if the length is lim-ited to 4 KB; the same way that A-MPDU canplace multiple MPDUs in a single PSDU. Thus,for small packet sizes, we can choose any type ofaggregation. On the other hand, when the pack-et size is larger than 250 KB, the throughput ofA-MSDU distinguishes significantly from A-MPDU and the two-layer aggregation, withmuch lower values because the number ofMSDUs that fit into a single A-MSDU is becom-ing less than the other cases. We can monitorthis behavior even more closely when packet sizeincreases from 1000 B (see OL = 200 Mb/s) to1500 B (see OL = 300 Mb/s). The A-MSDUthroughput drops slightly for the reason that wehad four packets of 1000 B, fitting in one A-MSDU, where in the case of 1500 B packet size,only two of them can occupy the same space.The same behavior occurs with the two-levelaggregation but only because of the A-MSDUstage. This is also why the A-MPDU throughputincreases further when two-level aggregationremains at the same levels. The throughput forthe no-aggregation case always increases withthe increase of the OL by varying the packet sizeeven after the channel is saturated. However, byonly increasing the packet size up to the maxi-mum Ethernet transmission unit (1500 bytes),without aggregation, will achieve throughputabout three times lower than that of the A-MPDU and the two-level aggregation. Thisclearly demonstrates that small packet size is thekey factor that lowers the throughput efficiency.

In the second simulation, we increase the OLby altering the packet interarrival rate, that is,variable packet rate (VPR), instead of increasingthe packet sizes. So the packet size remains con-stant at 1,000 B during the simulation test. From


n Figure 4. Throughput vs. increased offered load by varying the packet size.

Offered load (Mb/s)

250

20

MA

C t

hrou

ghpu

t (M

b/s)

40

60

80

100

120

140

50 100 150 200 250 300

Two-levelA-MPDUA-MSDUNonePeak PHY



Fig. 5, we can observe that the MAC throughputperformance of all the schemes first increaseswith an increase of the OL. However, after chan-nel is saturated, the throughput of all theschemes remains constant even when the OLkeeps increasing, which is different from the firstsimulation. This behavior is characterized as nor-mal because of channel saturation, meaning theresources are limited to the impending demand.Furthermore, the throughput achieved by the A-MPDU and the two-level aggregation after satu-ration is approximately 4.5 times higher than theno aggregation scheme.

The last simulation represents a scenario fora fixed OL of 100 Mb/s with variant packet sizesand appropriate interval times. In this simula-tion, there is no channel saturation. Along withthe throughput values for each type of aggrega-

tion, we investigate the degree of aggregationthat the two-level aggregation performs, by com-paring the number of indicated MSDUs and thereceived MPDUs. In Fig. 6, the A-MPDU andthe two-level aggregation achieve the 100 Mb/sgoal that TGn has set, whereas A-MSDU fallsbelow that threshold at around 75 Mb/s. In con-clusion, although the A-MSDU mechanism canachieve higher throughput than the legacy 802.11standards, it does not utilize the channel as fullyas A-MPDU and the two-level aggregation do.However, there is an exception for the A-MPDUfunction when packet size is 125 B and packetrate is 10 µs each. This shows that this type ofaggregation cannot handle the bulk traffic effec-tively, and thus, a lower throughput than thetwo-level aggregation because the A-MSDUalgorithm facilitates to overcome that issue atthe first stage of aggregation. It is very importantto understand how this blend of A-MSDU andA-MPDU, in most cases, is capable of improvingthe effectiveness of the MAC layer, specificallywhen there are many small MSDUs, such asTCP acknowledgments. For example, when thepacket size is set for 125 B, there are approxi-mately 999,800 MSDUs generated during thesimulation period. In spite of this, the number ofMPDUs received at the receiver is 34,500MPDUs, and that suggests that a huge MACand PHY overhead was avoided. For the largestpacket size of 1500 B, there are approximate83,300 indicated MSDUs and about 41,650received MPDUs, a 50 percent improvement inefficiency.

CONCLUSION

This article investigates several frame aggrega-tion schemes proposed in the latest IEEE802.11n draft, namely, A-MSDU, A-MPDU, andthe combined two-level aggregation. Our simula-tion results demonstrated that concatenationmechanisms performed over different sublayerscan actually increase the channel efficiency andreduce the overhead of the future 802.11 MAC.

All types of aggregation schemes are highlyrecommended as they resolve the fundamentalproblem of inefficiency due to the PHY/MACoverhead. However, the IEEE 802.11n draft onlyidentifies the basic concepts and data framestructures. In a flawless environment, it candeliver attractive results but in terms of its func-tionality in a real working environment, thereare still many issues that require further investi-gation. For example, the processing timerequired to compute these aggregates canincrease the overall delays. Actually, as the effi-ciency of aggregation increases, its operationbecomes more complex (e.g., two-level aggrega-tion). Another unresolved issue is how large aconcatenation threshold the devices should set.Ideally, the maximum value is preferable but in anoisy environment, short frame lengths are pre-ferred because of potential retransmissions. TheA-MPDU concatenation scheme operates onlyover the packets that are already buffered in thetransmission queue, and thus, if the CPR datarate is low, then efficiency also will be small.There are many ongoing studies on alternativequeuing mechanisms different from the standard

n Figure 5. Throughput vs. increased offered load by varying the packet arrivalinterval (packet size = 1 kbyte).

Offered load (Mb/s)

250

20

MA

C t

hrou

ghpu

t (M

b/s)

40

60

80

100

120

140

50 100 150 200 250 300 320


n Figure 6. Throughput vs. increased MSDU size.

MSDU size (bytes)

1250

20

MA

C t

hrou

ghpu

t (M

b/s)

40

60

80

100

120

140

250 500 750 1000 1250 1500




FIFO. A combination of frame aggregation andan enhanced queuing algorithm could increasechannel efficiency further.

Future work includes examining multidestina-tion aggregation, frame aggregation performingover error-prone channels, and the co-existenceof HT stations with the legacy ones.

ACKNOWLEDGMENTDionysios Skordoulis and Qiang Ni are gratefulfor the support of the U.K. EPSRC (Engineer-ing and Physical Sciences Research Council)Doctoral Training Account. Hsiao-Hwa Chengratefully acknowledges the research grants(NSC96-2221-E-110-035 and NSC96-2221-E-110-050) from the National Science Council of Tai-wan that partially supported the work presentedin this article. We also would like to thankanonymous reviewers for their invaluable com-ments.

REFERENCES[1] IEEE Std. 802.11 WG, “Part 11: Wireless LAN Medium

Access Control (MAC) and Physical Layer (PHY) Specifi-cations,” Aug. 1999 (reaffirmed June 2003).

[2] V. Jones, R. DeVegt, and T. Jerry, “Interest for HigherData Rates (HDR) Extension to 802.11a,” IEEE 802.11nworking doc. 802.11-02-081r0, Jan. 2002.

[3] J. Rosdahl, “Draft Project Authorization Request (PAR)for High Throughput Study Group,” IEEE 802.11nworking doc. 802.11-02/798r2, Mar. 2003.

[4] Y. Xiao and J. Rosdahl, “Throughput and Delay Limitsof IEEE 802.11,” IEEE Commun. Letters, vol. 6, no. 8,Aug. 2002, pp. 355–57.

[5] Q. Ni et al., “AFR partial MAC proposal for IEEE802.11n,” IEEE 802.11n working doc. 802.11-04-0950-00-000n, Aug. 2004.

[6] IEEE P802.11n, Draft 2.0, “Part 11: Wireless LAN Medi-um Access Control (MAC) and Physical Layer (PHY)Specifications: Enhancements for Higher Throughput,”Feb. 2007.

[7] IEEE Std. 802.11a, “Part 11: Wireless LAN MediumAccess Control (MAC) and Physical Layer (PHY) Specifi-cations: High-Speed PHY in the 5 GHz Band,” 1999(reaffirmed June 2003).

[8] IEEE Std. 802.11g WG, “Part 11: Wireless LAN MediumAccess Control (MAC) and Physical Layer (PHY) Specifi-cations: Further High Date Rate Extension in the 2.4GHz Band,” June 2003.

[9] IEEE Std. 802.11e WG, “Part 11: Wireless LAN MediumAccess Control (MAC) and Physical Layer (PHY) Amend-ment 8: Medium Access Control (MAC) Quality of Ser-vice Enhancements,” Nov. 2005.

[10] Y. Kim et al., “Throughput Enhancement of IEEE802.11 WLAN via Frame Aggregation,” IEEE VTC, vol. 4,Sept. 2004, pp. 3030–34.

[11] Y. Xiao, “Packing Mechanisms for the IEEE 802.11nWireless LANs,” IEEE GLOBECOM ’04, pp. 3275–79.

[12] Y. Chen, S. Emeott, and R. Choudhury, “PerformanceAnalysis of Data Aggregation Techniques for WirelessLAN,” Motorola Labs, 2005.

[13] Y. Lin and V. Wong, “Frame Aggregation and OptimalFrame Size Adaptation for IEEE 802.11n WLANs,” Proc.IEEE GLOBECOM, San Francisco, CA, Nov. 2006, pp. 1–6.

[14] OPNET Technologies, Inc., OPNET Modeler: Accelerat-ing Network R&D; http://www.opnet.com/products/modeler/home-2.html (accessed Feb. 20, 2007).

[15] A. P. Stephens et al., “IEEE P802.11 Wireless LANs:Usage Models,” IEEE 802.11n working doc. 802.11-03/802r23, May 2004.

BIOGRAPHIESDIONYSIOS SKORDOULIS ([email protected])received his B.Sc. in computer science with distributed sys-tems from City University, London, in 2005 and his M.Sc.in wireless communication systems from Brunel University,London, in 2006. Currently, he is working toward his Ph.D.,supervised by Dr. Qiang Ni at Brunel University. His mainresearch interests are QoS and performance optimizationfor current and next-generation WLANs. He has received

the IEEE Communications prize for best project and theRohde & Schwarz award for best overall PG student.

QIANG NI [M’04] ([email protected]) received his Ph.D. in1999 from Huazhong University of Science and Technology(HUST), China, and subsequently spent two years as a post-doctoral fellow at the Wireless Communication ResearchLaboratory, HUST. He is currently a lecturer in the Schoolof Engineering & Design, Brunel University, London. Priorto that he was a senior researcher at Hamilton Institute,National University of Ireland Maynooth. His research inter-ests are wireless networking and mobile communications.He has published over 40 refereed papers in his researchfields. He worked with INRIA France as a researcher forthree years (2001–2004). He has been active as an IEEE802.11 Wireless Standard Working Group voting membersince 2002.

HSIAO-HWA CHEN [SM’00] ([email protected]) is currentlya full professor and was the founding director of the Insti-tute of Communications Engineering at National Sun Yat-Sen University, Taiwan. He has authored or co-authoredover 200 technical papers in major international journalsand conferences, five books and several book chapters inthe areas of communications, including the following: NextGeneration Wireless Systems and Networks and The NextGeneration CDMA Technologies (Wiley, 2005 and 2007).He has been an active volunteer in various IEEE technicalactivities for over 20 years. Currently, he is serving as chairof the IEEE Communications Society Radio CommunicationsCommittee. He served or is serving as symposium chair/co-chair of many major IEEE conferences, including VTC, ICC,GLOBECOM, and WCNC. He has served or is serving asassociate editor and/or guest editor of numerous technicaljournals in communications. He is serving as chief editor(Asia and Pacific) for Wiley’s Wireless Communications andMobile Computing Journal and International Journal ofCommunication Systems. He is editor-in-chief of Wiley’sSecurity and Communication Networks (http://www.inter-science.wiley.com/journal/security).

ADRIAN P. STEPHENS [M] ([email protected])received his B.A. and Ph.D. from Cambridge University. He isa principal engineer in Intel’s Wireless Standards and Tech-nology group (part of the Mobile Wireless group). His focusis on developing next-generation IEEE 802.11 WLAN stan-dards. He is Technical Editor of IEEE 802.11n and coordinat-ed (internally and externally) Intel’s MAC proposal for IEEE802.11n. He has over 20 years of product development andresearch experience working for the industry and govern-ment. He is a member of the IEE and is a chartered engineer.

CHANGWEN LIU ([email protected]) received his B.S. inmathematics from Hunan Normal University, his M.S. inapplied mathematics from HUST, and his Ph.D. in comput-er science from the University of Illinois at Urbana Cham-paign in 1995. He is now a staff MAC engineer withEntropic Communications Inc., San Diego, California. Hisresearch interests include network protocol design, simula-tion, and analysis for wireless and wired data communica-tions. From 2002 to 2007 he was a senior research scientistin Intel’s Communication Technology Lab.

ABBAS JAMALIPOUR [S’86, M’91, SM’00, F’07] ([email protected]) holds a Ph.D. from Nagoya University, Japan. Heis the author of the first book on wireless IP and two otherbooks. He has co-authored five books and over 180 journaland conference papers, and holds two patents, all in thefield of wireless networks. He is an IEEE Distinguished Lec-turer and a Fellow of Engineers Australia. He was chair ofthe Satellite and Space Communications Technical Commit-tee (2004–2006) and is currently vice chair of the Commu-nications Switching and Routing Technical Committee andchair of the Asia-Pacific Board, Chapters CoordinatingCommittee. He is a technical editor of IEEE Communica-tions Magazine, Wiley’s International Journal of Communi-cation Systems, and several other journals. He is a votingmember of the IEEE GITC and was a vice chair of IEEEWCNC ’03–’06, chair of IEEE GLOBECOM ’05 (WirelessCommunications), and a symposium co-chair of IEEE ICC’05–’08 and IEEE GLOBECOM ’06–’07, among many otherconferences. He is the recipient of several internationalawards, most recently the Best Tutorial Paper Award andDistinguished Contribution to Satellite CommunicationsAward, both from the IEEE Communications Society in2006.

There are manyongoing studies onalternative queuing

mechanisms differentfrom the standard

FIFO. A combinationof frame aggregation

and an enhancedqueuing algorithm

could increase channel efficiency

further.


ieee 802.11n mac frame aggregation mechanisms for next-generation high-throughput wlans [medium...

Documents