Efficient MAC Strategies for the IEEE n wireless LANs

Transcription

1 WIRELESS COMMUNICATIONS AND MOBILE COMPUTING Wirel. Commun. Mob. Comput. 2006; 6: Published online 9 January 2006 in Wiley InterScience ( DOI: /wcm.274 Efficient MAC Strategies for the IEEE n wireless LANs Yang Xiao*,y Computer Science Division, The University of Memphis, 373 Dunn Hall, Memphis, TN 38152, U.S.A. Summary Current IEEE wireless local area network (WLAN) standard products can provide up to 54 Mbps raw transmission rate, while non-standard WLAN products with 108 Mbps have already appeared in the market, and the next generation WLAN will provide much higher transmission rates. However, the medium access control (MAC) was designed for lower data rates, such as 1 2 Mbps, and it is not an efficient MAC. Furthermore, a theoretical throughput limit exists due to overhead and limitations of physical implementations, and therefore increasing transmission rate cannot help a lot. Designing efficient MAC strategies becomes critical and important. In this paper, we introduce and propose a series of efficient MAC strategies to overcome the fundamental overhead, and to improve performance. The protocols and mechanisms include Direct Link Protocol, Without Acknowledgement, Without Retransmissions, Block Acknowledgement Protocol, Concatenation, Packing, Multiple Frame Transmission (versions 1 and 2) and Piggyback. The aim of this paper is to introduce and propose these efficient new MACs not only for current IEEE standards (.11a/.11b/.11g), but also for the next generation WLAN with higher speed and higher throughput, especially for IEEE n. Copyright # 2006 John Wiley & Sons, Ltd. KEY WORDS: IEEE ; medium access control (MAC); distributed coordination function (DCF); IEEE n; IEEE e 1. Introduction The IEEE wireless LAN (WLAN) becomes an essential feature of every day s life and is being accepted for many different environments. The IEEE medium access control (MAC) employs mandatory contention-based channel access function called distributed coordination function (DCF), and an optional centrally controlled channel access function called point coordination function (PCF) [1]. The DCF adopts a carrier sense multiple access with collision avoidance (CSMA/CA) with binary exponential backoff. The PCF adopts polling mechanisms and has never been successful in reality. To support the MAC-level quality of service (QoS), the IEEE Working Group is currently working on the standardization of IEEE e [5,11,12], which is in the final stage. The emerging IEEE e standard provides QoS features and multimedia support to the existing /.11b/.11a/.11g WLAN standards, while maintaining full backward compatibility with these standards. The IEEE e MAC employs a channel access function, called hybrid coordination function (HCF), which includes a contention-based channel access and a contention-free centrally controlled channel access mechanism. *Correspondence to: Yang Xiao, Computer Science Division, The University of Memphis, 373 Dunn Hall, Memphis, TN 38152, U.S.A. y yangxiao@ieee.org Copyright # 2006 John Wiley & Sons, Ltd.

2 454 Y. XIAO The IEEE , b and a/g specifications provide up to 2, 11 and 54 Mbps data rates [1 4] respectively. To provide better QoS, especially for multimedia applications, increasing data rates is also highly desirable. Therefore, the IEEE Working Group is pursuing IEEE n, an amendment for higher throughput and higher speed enhancements, aiming for higher throughput instead of higher data rates with PHY and MAC enhancements. The IEEE n standard process has three phases: the phase 1 is the preparation stage from January 2002 to September 2002; the phase 2 is the stage of the IEEE High Throughput Study Group (HTSG) from September 2002 to September 2003; the phase 3 is the stage of IEEE n Task Group from September 2003 to September 2005 (expected). During the phase 1, the preparation stage, IEEE Working Group was seeking higher data rates over 100 Mbps for IEEE a extension [6 10,13 16]. Some higher data rate approaches were proposed such as bit-loading (BL) approach [14], the double clock rate approach [15], the double subcarrier number approach [15], the 4096-QAM- OFDM approach [15] and the OFDM/SDM (multicarrier MIMO) system approach [15] etc. However, we proved that a theoretical throughput limit exists due to overhead of MAC and PHY [8,9]. In other words, the theoretical throughput limit, about 75 Mbps for IEEE a with the payload size of 1500 bytes [9], upper bounds any obtained throughput even when the data rate goes infinitely high. Therefore increasing transmission rate cannot help a lot. Both reducing overhead and pursuing higher data rates are therefore necessary and important [10]. The IEEE HTSG was established in July 2002 emphasizing higher throughput for higher data rates over 100 Mbps wireless LANs. During the phase 2, the stage of the IEEE HTSG, Project Authorization Request (PAR) and Five Criteria for Standards Development had been established. During the phase 3, the IEEE n Task Group undertakes the following steps: (a) identify and define usage models, channel models and related MAC and application assumptions; (b) identify and define evaluation metrics that characterize the important aspects of a particular usage model; (3) call for proposals, combine them, revise them and finally publish a final standard. It is planned to publish the IEEE n standard in October 2005 [17]. So far, most of the contributions in the IEEE n meetings focus on PHY enhancements. The aim of this paper is the same as IEEE n, and we focus on MAC enhancements instead of PHY enhancements. Therefore, this paper serves a good purpose for MAC enhancements, which can also be proposed in future IEEE n task group meetings. In this paper, we give an overhead analysis to identify the fundamental problem of MAC inefficiency, i.e. overhead (MAC/PHY headers and trailers, inter-frame spaces, backoff time and acknowledgements). We introduce and propose following efficient MAC strategies. Direct Link Protocol (DLP): DLP was proposed in IEEE e [5]. Since stations are not allowed to transmit frames directly to other stations that are not the access point (AP) if an AP is present, DLP allows stations to transmit frames directly to another station by setting up such data transfer when an AP is present. Therefore, efficient transmissions can be achieved. Without Acknowledgement: In the DCF, error control adopts positive acknowledgment. In other words, every transmitted frame needs a positive acknowledgment. For some applications such as frames with audio UDP packets as payload, delay is much more important, whereas some frame-loss is tolerable. Therefore, we can adopt without acknowledgement mechanism to become more efficient. Without Retransmissions: For real-time multimedia traffic with sensitive delay requirements, retransmitted frames may be too late to be useful due to the unexpected delay. With this consideration, for some applications such as real-time transmissions, the retry-limit can be set to be zero to become more efficient. In other word, no retransmission happens. Block Acknowledgement Protocol (BAP): BAP was proposed in IEEE e [5]. It is also called Burst Transmission and Acknowledgment (BTA) [10], Contention-Free Burst (CFB) [11], Group Transmission and Acknowledgements (GTA) [12], Delayed Group Acknowledgement (DGA) and etc. The idea of BAP is that instead of acknowledging each frame, a burst of frames is received first, and then the whole burst is acknowledged one time. Concatenation [18]: The idea of a concatenation mechanism is to concatenate multiple frames into a single transmission. Frames can be concatenated if they are available, and have the same source and destination addresses. The length of concatenation should be smaller than a threshold.

3 EFFICIENT MAC STRATEGIES 455 Packing: Packing is the act of combining multiple service data units (SDUs) from a higher layer into a single MAC protocol data unit (PDU). It is similar to concatenation, but the difference is that SDUs for the packing mechanism form one PDU, whereas SDUs for the concatenation mechanism form separate PDUs. Multiple Frame Transmission: Multiple frames are transmitted in a burst in order to reduce the overhead relatively. It is similar to concatenation and packing too. The difference is that each frame is transmitted separately with a separate SIFS and/ or an acknowledgement. Piggyback [18]: The idea of the piggyback mechanism is that a destination station is allowed to piggyback a data frame to the source station once during the DCF period if the destination station has a frame to send to the source station. Note that not all of the above schemes are perfect: each scheme has advantages and disadvantages. We will further discuss them in later sections. In this paper, all results adopt IEEE a as an example. However, it is easy to be applied to IEEE b/.11g/.11e. Similar conclusions will hold for IEEE b/.11g/.11e too. The rest of the paper is organized as follows. In Section 2, we briefly introduce the IEEE MAC and IEEE e. In Section 3, we study theoretical throughput limit and provide an overhead analysis. We introduce and propose a series of efficient MAC enhancements in Section 4. Performance evaluations are embedded in the schemes. We conclude this paper in Section Introduction of IEEE MAC and IEEE e We introduce the IEEE MAC and the IEEE e MAC in Subsections 2.1 and 2.2 respectively The IEEE MAC The IEEE MAC employs a mandatory DCF and an optional PCF. These functions determine when a station (STA), operating within a basic service set (BSS) or independent BSS (IBSS), is permitted to transmit. There are two types of networks: infrastructure network (BSS) in which an AP is present and ad hoc network (IBSS) in which an AP is not present. In a long run, time is always divided into repetition intervals called superframes. Each superframe starts with a beacon frame, and the remaining time is further divided into a contention-free period (CFP) and a contention period (CP). The DCF works during the CP and the PCF works during the CFP. If the PCF is not active, a superframe will not include the CFP. The DCF defines a basic access mechanism and an optional request-to-send/clear-to-send (RTS/CTS) mechanism. In the DCF, a station with a frame to transmit monitors the channel activities until an idle period equal to a distributed inter-frame space (DIFS) is detected. After sensing an idle DIFS, the station waits for a random backoff interval before transmitting. The backoff time counter is decremented in terms of slot time as long as the channel is sensed idle. The counter is stopped when a transmission is detected on the channel, and reactivated when the channel is sensed idle again for more than a DIFS. In this manner, stations that deferred from channel access because their backoff time was larger than the backoff time of other stations, are given a higher priority when they resume the transmission attempt. The station transmits its frame when the backoff time reaches zero. At each transmission, the backoff time is uniformly chosen in the range [0, CW 1] in terms of timeslots, where CW is the current backoff window size. At the very first transmission attempt, CW equals the initial backoff window size CW min. After each unsuccessful transmission, CW is doubled until a maximum backoff window size value CW max is reached. After the destination station successfully receives the frame, it transmits an acknowledgment frame (ACK) following a short inter-frame space (SIFS) time. If the transmitting station does not receive the ACK within a specified ACK timeout, or it detects the transmission of a different frame on the channel, it reschedules the frame transmission according to the previous backoff rules. The above mechanism is called the basic access mechanism. In such a mechanism, the hidden node problem may happen: transmissions of a station cannot be detected using carrier senses by a second station but interfere with transmissions from the second station to a third station. To reduce the hidden station problem, an optional four-way data transmission mechanism called RTS/CTS is also defined in the DCF. In the RTS/CTS mechanism, before transmitting a data frame, a short RTS frame is transmitted. The RTS frame also follows the backoff rules introduced above. If the RTS frame succeeds, the receiver station responds with a short CTS frame. Then a data frame

4 456 Y. XIAO and an ACK frame will follow. All four frames (RTS, CTS, data, ACK) are separated by an SIFS time. In other words, the short RTS and CTS frames reserve the channel for that data frame transmission which follows. The PCF is an optional centrally controlled channel access function, which provides contention-free (CF) frame transfer. The PCF is designed for supporting time-bounded services, which can provide limited QoS. It logically sits on top of the DCF, and performs polling, enabling polled stations to transmit without contending for the channel. It has a higher priority than the DCF by adopting a shorter inter-frame space (IFS) called point inter-frame space (PIFS). If the PCF is implemented, CFP under the PCF and CP under the DCF alternate over time. A CFP and a CP forms a superframe. The AP where the point coordinator (PC) is normally located senses the medium idle for a PIFS interval, and then transmits a beacon frame to initiate a CFP (in other words, to initiate a superframe). After a SIFS time, the PC sends a poll frame to a station to ask for transmitting a frame. The poll frame may or may not include data to that station. After receiving the poll frame from the PC, the station with a frame to transmit may choose to transmit a frame after a SIFS time. When the destination station receives the frame, an ACK is returned to the source station after a SIFS time. The PC waits a PIFS interval following the ACK frame before polling another station or terminating the CFP by transmitting a CF-end frame. If the PC receives no response from the polled station for a PIFS interval, the PC can poll next station or terminate the CFP by transmitting a CF-End frame IEEE E IEEE e provides a channel access function, called Hybrid Coordination Function to support applications with QoS requirements. The HCF includes both a contention-based channel access and a centrally controlled channel access. The contention-based channel access of the HCF is also referred to as enhanced distributed coordination function (EDCF). A new concept, transmission opportunity (TXOP), is introduced in IEEE e. A TXOP is a time period when a station has the right to initiate transmissions onto the wireless medium. It is defined by a starting time and a maximum duration. A station cannot transmit a frame that extends beyond a TXOP. If a frame is too large to be transmitted in a TXOP, it should be fragmented into smaller frames. The EDCF works with four access categories (ACs), which are virtual DCFs, where each AC achieves a differentiated channel access. This differentiation is achieved through varying the amount of time a station would sense the channel to be idle and the length of the contention window during a backoff. The EDCF supports eight different priorities, which are further mapped into four ACs. ACs are achieved by differentiating the arbitration inter-frame space (AIFS), the initial window size, and the maximum window size. For the AC iði ¼ 0;...; 3Þ, the initial backoff window size is CW min ½iŠ, the maximum backoff window size is CW max ½iŠ, and the arbitration interframe space is AIFS[i]. For 0 i < j 3, we have CW min ½iŠ CW min ½jŠ, CW max ½iŠ CW max ½jŠ and AIFS[i] AIFS[j], and at least one of above inequalities must be not equal to. In other words, the EDCF employs AIFS[i], CW min ½iŠ and CW max ½iŠ (all for i ¼ 0;...; 3) instead of DIFS, CW min and CW max respectively. If one AC has a smaller AIFS or CW min or CW max, the AC s traffic has a better chance to access the wireless medium earlier. The values of AIFS[i] (i ¼ 0,...,3), CW min [i] (i ¼ 0,...,3) and CW maz [i] (i ¼ 0,...,3) are referred to as the EDCF parameters, which will be announced by the QoS access point (QAP) via periodically transmitted beacon frames. 3. Theoretical Limit and Overhead Analysis The achievable maximum throughput (MT) can be achieved when the system is at the best-case scenario: (1) the channel is an ideal channel without errors; (2) at any transmission cycle, there is one and only one active station which always has a frame to send, and other stations can only accept frames and provide acknowledgments. The throughput upper limit (TUL) [8,9] is defined as the maximum throughput when the raw data rate goes infinitely high. As indicated in Reference [10], overhead is the major fundamental issue for inefficient MAC, and it includes headers (MAC header, FCS and PHY header), inter-frame spaces, backoff time and acknowledgements. In this section, we assume that control rate is the same as data rate. Define overhead as the difference between data rate and throughput, and define the normalized overhead as the overhead divided by data rate. We further assume that all higher data rates are compatible with IEEE a.

5 EFFICIENT MAC STRATEGIES 457 Fig. 1. Maximum throughput (MT) and throughput upper limit (TUL) versus data rate. Fig. 3. Normalized overhead versus data rate (Mbps). Fig. 2. Overhead (Mbps) versus data rate (Mbps). Figure 1 shows that MT and TUL versus data rate with payload size 100 and 1500 bytes. The TUL depends on the payload size. When the payload size is 100 bytes, the TUL is about 5 Mbps, and when the payload size is 1500 bytes, the TUL is about 75 Mbps. We observe that as the data rate increases, the MT increases, but are bounded by the TUL. Figures 2 and 3 show overhead and normalized overhead versus data rate respectively. We observe that both the overhead and the normalized overhead increase as the data rate increases. The normalized overhead almost reaches 100% after 180 Mbps when the payload size is 100 bytes. The normalized overhead almost reaches 70% after 180 Mbps when the payload size is 1500 bytes. Figures 4 and 5 show overhead and normalized overhead versus payload size respectively. We Fig. 4. Overhead (Mbps) versus payload (bytes). Fig. 5. Normalized overhead versus payload (bytes).

6 458 Y. XIAO observe that both the overhead and the normalized overhead decrease as the payload size increases. The normalized overhead almost reaches 100% when the payload size is very small. In summary, overhead and normalized overhead are extremely large (relatively) either when the data rate is high or when the frame is small. Therefore, new efficient MAC strategies are especially needed. In next section, we will introduce and propose several efficient MAC protocols to overcome overhead. Fig. 6. Directed-link protocol (DLP). 4. Efficient MAC Protocols In this section, we introduce and propose several efficient MAC protocols: direct link protocol, Without Acknowledgement, Without Retransmissions, Block Acknowledgement Protocol, Packing, Concatenation, Multiple Frame Transmission and Piggyback in the following subsections Direct Link Protocol Direct link protocol (DLP) was proposed in IEEE e [5]. The DLP allows a QoS station (QSTA) to transmit frames directly to another QSTA by setting up such data transfer when a quality access point (QAP) is present. The need for this protocol is motivated by the fact that the intended recipient may be in Power Save Mode, in which case it can only be woken up by the QAP. Another purpose of the DLP is to be efficient. The DLP allows the sender and the receiver to exchange rate set and other information. Furthermore, The DLP messages can be used to attach security information elements. This protocol prohibits the stations going into powersave for the active duration of the direct stream. DLP does not apply in an ad hoc network, where frames are always sent directly from one QSTA to another. A direct link can be built by following sequences: QSTA-1 that has data to send invokes the DLP and sends a DLP-request frame to the QAP, shown in Figure 6 (1a). This request contains the rate set, and (extended) capabilities of QSTA-1, as well as the MAC addresses of QSTA-1 and QSTA-2. If QSTA-2 is associated in the BSS, the QAP shall forward the DLP-request to the recipient, QSTA-2, shown in Figure 6 (1b). If QSTA-2 accepts the direct stream, it shall send a DLP-response frame to the QAP, shown in Figure 6 (2a). The QAP shall forward the DLP-response to QSTA-1, shown in Figure 6 (2b), after which the direct link becomes active and frames can be sent from QSTA-1 to QSTA-2 and from QSTA-2 to QSTA-1, shown in Figure 6 (3). When the direct link is active, QSTA-1 may use DLP-probes to gauge the quality of the link between QSTA-1 and QSTA-2. The direct link becomes inactive when no frames have been exchanged as part of the direct link for the duration of adlpidletimeout. After the timeout, frames with destination QSTA-2 shall be sent via the QAP. In the simulations, we evaluate the DLP in the EDCF for voice streams (AC ¼ 3), video streams (AC ¼ 1) and data (AC ¼ 0) in terms of throughput and delay. The simulation program is implemented with Java using discrete event simulation approach. We have the following parameters unless stated otherwise: AIFS[3] ¼ 25 ms; CW min [3] ¼ 8; CW max [3] ¼ 512; AIFS[1] ¼ 25 ms; CW min [1] ¼ 64; CW max [1] ¼ 2048; AIFS[0] ¼ 34 ms; CW min [0] ¼ 256; 256; CW max [0] ¼ 16384; each video stream is 4.86 Mbps, which is generated by a constant inter-arrival time ms with a mean frame size 1464 bytes. Each voice stream is Mbps, which is generated by a constant inter-arrival time 4 ms with a mean frame size 92 bytes. Each data station has traffic of 2 Mbps, which is generated by an exponential interarrival time 6 ms with a mean payload size 1500 bytes. The data rate is 54 Mbps and the control rate is 24 Mbps. We assume that all the stations are within the transmission range. The simulation time is 100 s. Initially, there are one voice stream, one video stream and one data QSTA in the system. At 10 and 20 s, one voice stream one video stream, and one data station are added each time. At 30, 40 and 50 s, only one voice stream is added each time. There are totally six

7 EFFICIENT MAC STRATEGIES 459 improved for all ACs. For voice, video and data, delay per stream increases as the traffic load increases. In summary, the DLP improves IEEE /.11e/.11a MAC efficiency. However, if two QSTAs have only very few frames to transmit/exchange such as one frame only, the DLP is not recommended since the DLP has overhead of four message exchanges when setting up the direct link. Normally, the DLP is recommended if one QSTA will be active for a time period to transmit frames to another QSTA such as FTP, real-time transmission etc. Fig. 7. Throughputs (Mbps) of voice (a), video (b) and data (c) with and without the DLP versus the simulation time. voice streams, three video streams and three data QSTAs at the end of the simulation. Figure 7 shows throughputs (per stream) for voice, video and data with and without the DLP. As illustrated in the figure, with the DLP, throughput has been greatly improved for all ACs. We observe that when the traffic load is small, video throughput (per stream) is pretty well, whereas when the traffic load is large, some fluctuations happen. On the other hand, voice throughput (per stream) is pretty well since voice needs less bandwidth so that the influence by the heavy traffic load is small. Figure 8 shows delays (per stream) for voice, video and data with and without the DLP. As illustrated in the figure, with the DLP, delay has been greatly 4.2. Without Acknowledgement In the DCF error control adopts positive acknowledgment. In other words, every transmitted frame needs a positive acknowledgment. For some applications such as frames with audio UDP packets as payload, delay is much more important, whereas some frame-loss is tolerable. Therefore, we can adopt without acknowledgement mechanism to become more efficient. Define throughput gain (TG) as the difference of the MT without ACK and the MT with ACK, divided by the MT with ACK. Figure 9 shows TG versus data rate when the payload size is 1500 bytes, TG increases as the data rate increases, and reaches 12% at the data rate 216 Mbps. Therefore, TG is high for high data rates. Figure 10 shows TG versus payload. For both data rates (6 and 216 Mbps), TG decreases as the payload size increases. In summary, the without ACK scheme is more effective when the payload size is small or the data rate is high. Fig. 8. Delays (ms) of video, voice and data with and without the DLP. Fig. 9. Throughput gain versus data rate.

8 460 Y. XIAO Fig. 10. Throughput gain versus payload size. Fig. 11. Delay gain (DG) and frame dropping loss (FDL) Without Retransmissions In the DCF error control adopts positive acknowledgment and retransmission to improve transmission reliability in the wireless medium. If an acknowledgement frame for a transmitted frame has not been received for a timeout period, the transmitted frame is assumed to be corrupted and the frame will be retransmitted for many times until a positive acknowledgement is received or the number of retransmissions reaches a limit. In the later case, the frame is dropped. Therefore, the MAC is a very robust protocol for the best-effort service in the wireless medium. For realtime multimedia traffic with sensitive delay requirements, retransmitted frames may be too late to be useful due to the unexpected delay. With this consideration, for some applications such as real-time transmissions, the retry-limit can be set to be zero to become more efficient. In other word, no retransmission happens. Define delay gain (DG) as the difference of the delay with the retry limit as 7 and the delay without retransmissions, i.e. the retry limit as 0, divided by the delay with the retry limit as 7. Define frame dropping loss (FDL) as the difference of the frame dropping ratio with the retry limit as 0 and the frame dropping ratio with the retry limit as 7, divided by the frame dropping ratio with the retry limit as 0. Figure 11 shows the DG and the FDL over the number of active stations. As illustrated in the figure, as the number of active stations increases, the DG increases and the FDL decreases. We observe that the DG can be as large as % when the number of active stations is 45; and the FDL can be one the range from 99 to 100%. In other words, the without retransmission scheme can significantly improve delay with some degree of degraded frame dropping ratio Block Acknowledgement Protocol Block acknowledgement protocol (BAP) was proposed in IEEE e [5]. It is also called burst transmission and acknowledgment (BTA) [10], contention-free burst (CFB) [11], group transmission and acknowledgements (GTA) [12], delayed group acknowledgement (DGA) and etc. The idea of BAP is that instead of acknowledging each frame, a burst of frames is received first, and then the whole burst is acknowledged at one time. A MAC service data unit (MSDU) is the information that is delivered as a unit between MAC service access points (SAP s). A MAC protocol data unit (MPDU) is the unit of data exchanged between two peer MAC entities using the services of the physical layer (PHY). We use MPDU s and frames interchangeably in this section. Since wireless medium (WM) is error-prone, transmitted frames can easily be corrupted, even without collisions. In the IEEE MAC protocol, each frame is acknowledged. This approach is very natural and robust, but it introduces quite an amount of overhead. In order to reduce the acknowledgement overhead, BAP is currently being discussed in the IEEE e task group [5]. The BAP mechanism allows a burst of frames to be transmitted before any acknowledgement. After sending a burst of frames, the sender sends a burst acknowledgement request (BlockAckReq) frame,

9 EFFICIENT MAC STRATEGIES 461 Table I. Comparison. Data rate (Mbps) Payload size: 1000 bytes Basic access BAP BAP model MT (Mbps) improvement (%) MT (Mbps) MT increase % % % % TUL % and the receiver must respond by sending the burst acknowledgement (BlockAck) frame, in which the correctly received frames information is included. All the frames, including BlockAckReq frame and BlockAck frame, are separated by an SIFS period. The sender should first win a transmission opportunity (TXOP) using a channel access mechanism before starting a burst. A TXOP is a time period when a station has the right to initiate transmissions onto the WM. A TXOP is defined by a starting time and a maximum duration. The burst length is limited, and the amount of state that must be kept by the receiver of the receiving frames is bounded. The sequence control field is 16 bits in length and consists of two subfields, the fragment number (4 bits) and the sequence number. Therefore, an MSDU could be decomposed into 2 4 ¼ 16 fragment frames maximum if the size of the MSDU is larger than the fragmentation threshold. The Ack bitmap field is 32 octets in length, where each bit can acknowledge one fragment frame (MPDU). Some of the MSDU s in the burst may be fragmented, and it is hard to know a priori if fragmentation will ever be used or not. Therefore, in order to accommodate fragmentations, 16 bits in the Ack Bitmap field are needed to acknowledge an MSDU, where each bit acknowledges one potential fragment frame (MPDU) of the MSDU. In other words, to acknowledge one MSDU, two octets in the Ack Bitmap field are needed. Therefore, the Bitmap in the Ack Bitmap field can acknowledge up to 32/2 ¼ 16 MSDU s. The Sequence Control field contains the fragment number and sequence number corresponding to bit 0 of the Ack Bitmap. The sequence control field defines an MPDU sequence number equal to (sequence number 16) þ fragment number. Bit position n, if set to a 1, acknowledges MPDU with MPDU sequence number equal to (sequence control þ n). If the BlockAck indicates that an MPDU was not received correctly, the sender shall retry that MPDU subject to MPDU s appropriate retry limit. Retransmitted burst data MPDU s shall preserve their original relative order. The receiver shall maintain a burst acknowledgement record consisting of a transmitter address and a 32-octet bitmap of received MPDU sequence numbers. These hold the acknowledgement state of the burst data received from that sender. Table I summarizes and compares the maximum throughputs when the payload size is 1000 bytes. It also shows the performance improvement after using the BAP mechanism. The maximum throughput increase is calculated by (MT BAP -MT)/MT in percentage. The TUL-increase is calculated in a similar way. As illustrated in the table, after using the BAP mechanism, MT increases 61.54, 97.6, %, for % for data rates 54, 108, 162 and 216 Mbps respectively. The TUL increases % after using the BAP mechanism. One observation is that such increase (in percentage) increases as the data rate increases after using the BAP mechanism. In other words, the BAP mechanism particularly works better for higher data rates Packing, Concatenation and Multiple Frame Transmission In this section, we introduce and propose three similar efficient MAC strategies: Packing, Concatenation and Multiple Frame Transmission, shown in Figure 12. The ideas of these three approaches are similar, i.e. to put multiple frames into a single (or approximately single) transmission. These mechanisms have many benefits. First of all, since transmitting longer frames may have a better throughput than transmitting shorter frames, adopting these mechanisms, the system can achieve the throughput of transmitting longer frames. The second and the most important benefit is that these mechanisms can reduce overhead. Without these mechanisms, each frame transmission needs a separate set of overhead (headers, inter-frame spaces, backoff time and/or acknowledgements). With these mechanisms, instead of several sets of overhead for different frames, only one set of overhead will be used. Finally, these mechanisms can reduce the average delay. Without these mechanisms, the second or a later frame is transmitted in a much later time. With these mechanisms, it will be transmitted almost at the same time or at an earlier time.

10 462 Y. XIAO the packing mechanism form one PDU, whereas SDUs for the concatenation mechanism form separate PDUs. In other words, packing is to combine all concatenated frames into a real big frame with one header instead of many concatenated frames. As shown in Figure 12(b), the result of packing is a real frame with MAC header called packing header, and multiple payloads: each payload has a length field in front. The packing mechanism is more efficient than the concatenation mechanism, but sacrifices complexity and processing time of combining and decomposing frames. Fig. 12. Diagrams for Concatenation, Packing and Multiple Frame Transmission (CH, concatenation header; PH. packing header; L1 Ln. length of the following payload). Note that in Figure 12, physical overhead is not shown, and diagrams are not drawn with scales. Furthermore, only basic access mode is showed. The diagrams for the optional RTS/CTS mode are omitted Concatenation Frames can be concatenated if they are available, and have the same source and destination addresses. Multiple frames are concatenated into a super frame, shown in Figure 12(a). A super frame includes a concatenation header (CH) frame and concatenated frames. In the CH, the frame control type field indicates that it is a concatenation super frame, and the payload includes the count of concatenated frames (2 bytes), and a total length field (2 bytes). When the destination station receives the super frame, it decomposes the super frame into normal frames, and acknowledges the last concatenated frame only. The destination station can easily identify boundaries of concatenated frames using preambles and FCS Packing Frames can be packed if they are available, and have the same source and destination addresses. Packing, shown in Figure 12(b), is the act of combining multiple service data units (SDUs) from a higher layer into a single MAC protocol data unit. It is similar to concatenation, but the difference is that SDUs for Multiple frame transmission The idea of Multiple Frame Transmission is similar to packing and concatenation. There are two kinds of multiple fame transmission. We called them multiple frame transmission version 1 (MFT-1) (Figure 12(c)) and multiple frame transmission version 2 (MFT-2) (Figure 12(d)). In MFT-1, frames are separated by SIFS. In MFT-2, each framed are acknowledged separately. MFT-1 is similar to BAP. Frames can be conducted with MFT-1, if they are available, and have the same source and destination addresses. However, for MFT-2, such requirements are not necessary. For Packing, Concatenation and MFT-1, the acknowledgement scheme needs to changed, whereas for MFT- 2, the acknowledgement scheme remains the same Thresholds One question is how big a super/real/virtual frame should be. One possible solution is that the number of concatenated/packed/multiple frames should not be larger than a threshold (such as 2, 3 or 4), and the total length should be smaller than another threshold, which is smaller than or equal to the fragmentation threshold. The threshold is called the concatenation threshold, the packing threshold or the multiple frame transmission threshold. For example the threshold can be 500 bytes. The purpose of these mechanisms is not to build a huge frame, but a reasonable size of frame since huge frames may cause a bad effect on fairness, and when collided, a longer frame is lost. Note that for IEEE802.11a, the length field of PLCP frame can indicate less value than 4096 bytes, and the maximum size of MAC frame is generally 2346 bytes [3]. The threshold should be much smaller than 2346 bytes. The threshold will be further discussed in simulation results.

11 EFFICIENT MAC STRATEGIES 463 When the RTS/CTS mode is used, the RTS threshold (defined as dot11rtsthreshold [1]) is turned on, i.e. if the total length of a frame is larger the RTS threshold, the RTS/CTS mode is automatically used. We could use the total length of super frame as the size of a frame to use the RTS/CTS mode Processing time One tradeoff of the concatenation and packing mechanisms is that the processing delay will increase a little. A good processing time is an idle time such as backoff time. Some frames could be concatenated/ packed beforehand, but at most times, frames must be concatenated/packed at real-time when frames are in the queue waiting for transmission. The super frameformat for the concatenation mechanism is designed in such a way that processing time of combining and decomposing frames is insignificant, whereas the packing mechanism is more efficient than the concatenation mechanism, but sacrifices complexity and processing time of combining and decomposing frames. MFT-1 and MFT-2 do not need additional processing time Implementation issue and other issues The concatenation/packing mechanism can be implemented in both stations and access points in which a queue is implemented so that when higher layer data arrives and cannot be sent immediately, the waiting frames are put into the queue. Such a scenario happens since the frame in the head of the queue often experiences backoff, collisions and waiting for an ACK. For all four mechanisms, examples for available frames can be formed from short UDP packets, realtime voice/audio packets, TCP-ACK packets and so on. The above-mentioned packets are all small in size, and therefore, good throughputs are expected by using these mechanisms. One small frame can also be concatenated/packed/transmitted with a relatively large frame, e.g. a TCP ACK packet follows a relatively large frame. In case that there are no available frames in the queue (in other words, concatenation/ packing rarely occurs), normal DCF procedures will be used by default and the system will have the same performance as before. These mechanisms are not a reversed mechanism of fragmentation. In fact, the proposed mechanisms require that the total length of the super/real/virtual frames is less than the fragmentation threshold. Therefore, there will be no concatenated/packed/ multiple frame that was originally generated by a previous fragmentation mechanism. On the other hand, the super/real/virtual frame will not be fragmented since the total length is less than the fragmentation threshold. Since these mechanisms do not change any system behavior such as the backoff behavior except that frames become larger, it is easy to see that these mechanisms should have a better performance untuitively Simulation results with single transmission station In this subsection, we study MT for the proposed schemes. In the simulation, the data rate is 54 Mbps. The number of packed/concatenated/multiple frames is 2. Figure 13 shows MT for different mechanisms. All mechanisms have better MTs than that of the original MAC. It also shows that packing is the most efficient mechanism, and concatenation is better than MFT-1, which is better than MFT Simulation results with multiple transmission stations In this subsection, we study the throughput with multiple transmission stations for the proposed schemes. In the simulation, the data rate is 54 Mbps and the payload size is 100 bytes. The number of packed/concatenated/multiple frames is 3. In the simulations, each station always has frame ready to send. Fig. 13. MT for different mechanisms (Payload ¼ 1000 bytes.).

12 464 Y. XIAO Fig. 14. Normalized throughput versus number of stations. Figure 14 shows normalized throughput versus number of stations, where the normalized throughput is defined as the portion of time used to transmit payload when stations have frames ready to transmit in any time. We observe that the packing mechanism has the highest throughput. Concatenation is a little better than MFT-1, which is better than MFT-2. All mechanisms are better than the original MAC in terms of throughput. In summary, the decreasing order in terms of efficiency is packing, concatenation, MFT-1 and MFT-2. However, it is also a decreasing order in terms of complexity Piggyback The idea of the piggyback mechanism (PM) is that a receiver station is allowed to piggyback a data frame to the sender station once if the receiver station has a frame to send to the sender. Such a mechanism is beneficial for cases such as TCP ACK packets. TCP over IEEE DCF is considered as an example. After a TCP receiver entity successfully received a TCP packet from a TCP sender entity, it needs to send a TCP ACK back to the TCP sender entity. However, the backward TCP ACK also needs to compete with the channel access. If the access delay of the TCP ACK is very large, the TCP sender entity will misjudge that the original TCP packet did not go through to the receiver, and will retransmit the original TCP packet again. With the proposed PM, the above backward TCP ACK will be piggybacked to the sender immediately after receiving the TCP packet. The overall performance will be improved greatly. For the original MAC, if the destination has a frame to send to the source after receiving a frame, it needs Fig. 15. MT over payload. to compete the channel again by at least a DIFS time and a random backoff. On the other hand, in the PM, the destination can piggyback a frame to the source with the ACK information included. The source then sends an ACK to acknowledge the piggybacked frame after a SIFS time. The benefit of the proposed piggyback mechanism is that the piggybacked frame does not need to wait DIFS and complete the channel again. Piggyback will reduce overhead, and improve performance. If there is not a frame available to piggyback, similar to the concatenation mechanism, normal DCF operations will be performed by default, and the system will have the same performance as before. In the simulations, the data rate is 54 Mbps. Figure 15 shows MT for the PM and the original MAC. The PM is better than the original MAC in terms of MT. Figure 16 shows normalized throughput versus number of stations. In the simulations, each station Fig. 16. Normalized throughput versus number of stations.

13 EFFICIENT MAC STRATEGIES 465 always has frame ready to send. We observe that the PM has better throughput than the original MAC. 5. Conclusions and Future Work In this paper, we have identified that overhead is the fundamental problem of inefficiency. Increasing transmission rate alone cannot help a lot. The overhead is very large either when the data rate is high or when the frame size is small. Therefore, new efficient MAC strategies are especially needed. We have introduced and proposed several MAC enhancements for higher throughput: Direct Link Protocol, Without Acknowledgement, Without Retransmission, Block Acknowledgement Protocol, Concatenation, Packing, Multiple Frame Transmission Version 1 and Version 2 and Piggyback. Our simulation results show that all these MAC enhancements can improve throughput greatly. The without ACK scheme is more effective when the payload size is small or the data rate is high. The without retransmission scheme can significantly improve delay with some degree of degraded frame dropping ratio. The ideas of packing, concatenation, MFT-1 and MFT-2 are similar. The above order is a decreasing order in terms of efficiency, but a decreasing order in terms of complexity. In this paper, we only consider an ideal channel in our performance evaluations. Performance evaluations under a noisy channel are our current and future work, and we have already reported some preliminary results in this respect in the recent IEEE n meeting [19,20]. Acknowledgement The author thanks Mr. Haizhon Li for providing two simulation figures (Fig. 7 and Fig. 8) of the DLP. References 1. IEEE WG, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specification, Standard, IEEE, August IEEE b WG, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specification: High- Speed Physical Layer Extension in the 2.4 GHz Band, IEEE, September IEEE a WG, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specification: High- Speed Physical Layer in the 5 GHz Band, IEEE, September IEEE g-2003, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications Amendment 4: Further Higher-Speed Physical Layer Extension in the 2.4 GHz Band. 5. IEEE e/D3.3.2, Draft Supplement to Part 11: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications: Medium Access Control (MAC) Enhancements for Quality of Service (QoS), November Jones VK, DeVegt R, Terry J. Interest for HDR extension to a, IEEE Interim Meeting, doc. IEEE r0, January Tan TK. Call for Interests on a Higher Rates Extension. IEEE Interim Meeting, January Xiao Y, Rosdahl J. Throughput Limit for IEEE IEEE 802 Interim Meeting, May 13 17, 2002, doc.: IEEE / 291r0. 9. Xiao Y, Rosdahl J. Throughput and Delay Limits of IEEE IEEE Communications Letters 2002; 6(8): Xiao Y, Rosdahl J. Performance Analysis and Enhancement for the Current and Future IEEE MAC Protocols. ACM SIGMOBILE Mobile Computing and Communications Review (MC2R), special issue on Wireless Home Networks 2003; 7(2): Choi S, Prado J, Shankar S, Mangold S. IEEE e Contention-Based Channel Access (EDCF) Performance Evaluation. In Proceedings IEEE ICC 03, Anchorage, Alaska, USA, May Xiao Y. IEEE e: A QoS Provisioning at the MAC layer. IEEE Wireless Communications 2004; 11(3): Xiao Y, Rosdahl J. Throughput Analysis for IEEE a Higher Data Rates. IEEE r0, March Tzannes M, Cooklev T, Lee D. Extended Data Rate a. IEEE r0, March Hori S, Inoue Y, Sakata T, Morikura M. System capacity and cell radius comparison with several high data rate WLANs. IEEE r1, March Coffey S. Suggested criteria for high throughput extensions to IEEE systems. IEEE r0, March Shoemake MB. Task Group N Schedule. IEEE / 488r Xiao Y. Concatenation and Piggyback Mechanisms for the IEEE MAC. Proceedings of IEEE WCNC Ni Q, Li T-J, Turletti T, Xiao Y. AFR Partial MAC Proposal for IEEE n. IEEE n Working Group Document: IEEE n, September Ni Q, Li T-J, Turletti T, Xiao Y. AFR Partial MAC Proposal for IEEE n: Presentation. IEEE n Working Group Document: IEEE n, September Author s Biography Yang Xiao received his Ph.D. in Computer Science and Engineering from Wright State University, Dayton, OH, U.S.A. He had been a software engineer a senior software engineer, and a technical lead working in the computer industry for 5 years in early 1990s. He worked at Micro Linear-Salt Lake City Design Center as an MAC

14 466 Y. XIAO architect involving the IEEE (Wireless LAN) standard enhancement work before he joined The University of Memphis as an assistant professor of computer science. Dr. Xiao is a voting member of the IEEE Working Group, a senior member of IEEE, and a member of ACM. He is an associate editor of EURASIP Journal on Wireless Communications and Networking, and currently serves on the editorial boards of (Wiley) Journal of Wireless Communications and Mobile Computing, International Journal of Wireless and Mobile Computing and International Journal of Signal Processing. He serves a lead guest editor for (Wiley) Journal of Wireless Communications and Mobile Computing, special issue on Mobility, Paging and Quality of Service Management for Future Wireless Networks in , a lead guest editor for International Journal of Wireless and Mobile Computing, special issue on Medium Access Control for WLANs, WPANs, Ad Hoc Networks and Sensor Networks in , and an associate guest editor for International Journal of High Performance Computing and Networking, special issue on Parallel and Distributed Computing, Applications and Technologies in He serves as a symposium co-chair for Symposium on Data Base Management in Wireless Network Environments in IEEE VTC 2003 Fall. He serves as a TPC member for many conferences such as ICC, GLOBECOM, ICDCS, WCNC ICCCN, PIMRC, WMASH etc. Dr. Xiao s current research interests include wireless local area networks, wireless personal area networks and mobile cellular networks. He has published many papers in major journals and refereed conference proceedings related to these research areas, such as IEEE Transactions on Mobile Computing, IEEE Transactions on Wireless Communications, IEEE Transactions on Parallel and Distributed Systems, IEEE Transactions on Vehicular Technology, IEEE Communications Letters, IEEE Communications Magazine, IEEE Wireless Communications, ACM/Kluwer MONET etc.