You are on page 1of 5

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts

for publication in the IEEE INFOCOM 2007 proceedings.

Software TDMA for VoIP Applications over
IEEE802.11 Wireless LAN
Fanglu Guo Tzi-cker Chiueh
Computer Science Department
Stony Brook University, NY 11794
{fanglu, chiueh}

Abstract—An emerging killer application for enterprise wire- there is more collision and interference, which increase packet
less LANs (WLANs) is voice over IP (VoIP) telephony, which loss rate, packet delay and packet delay jitter. Second, it
promises to greatly improve the reachability and mobility of is impossible to prioritize traffic flows. Consequently, large-
enterprise telephony service at low cost. Most commercial
IEEE802.11 WLAN-based VoIP products cannot support more volume non-real-time data traffic such as FTP may consume a
than ten voice conversations over a single IEEE 802.11b chan- large proportion of the network capacity, leaving time-sensitive
nel, even though its peak transmission rate is more than two voice traffic to suffer the consequences.
orders of magnitude higher than an individual VoIP connection’s The second barrier associated with IEEE802.11 WLAN
bandwidth requirement. There are two main reasons why these is substantial per-packet transmission overhead. Because the
VoIP systems’ effective capacity is lower than expected: stringent
latency requirement and substantial per-WLAN-packet overhead. maximum transmission rate of an IEEE802.11b WLAN link
Time-Division Multiple Access (TDMA) is a well-known tech- is 11 Mbps, in theory an IEEE802.11b link should be able to
nique that provides per-connection QoS guarantee as well as support hundreds of VoIP connections if each of them requires
maximizes the radio channel utilization efficiency. This paper 8Kbps such as G.729. In practice, several benchmarking
presents a software-based TDMA (STDMA) protocol that is tests [3], [4] reported that existing VoWLAN products on
designed to support VoIP applications and successfully works
on commodity IEEE802.11 WLAN interfaces. The resulting the market cannot support more than ten concurrent VoIP
STDMA prototype can support more than 50 two-way G.729 calls with comparable quality to toll calls over a single
voice conversations over a single IEEE 802.11b channel. IEEE802.11b channel. A major cause for this gap between
theory and practice is the considerable overhead associated
I. I NTRODUCTION with a WLAN packet’s transmission. This overhead includes
Voice over Wireless LAN (VoWLAN) is touted as a killer per-packet header bits, link-layer acknowledgment, back-off
application for enterprise WLANs because it significantly delay to avoid contention, retransmission cost due to interfer-
improves the coverage and mobility of enterprise telephony ence and inter-frame spacing for synchronization.
services. In fact, vendors such as Aruba Networks, Symbol, IEEE 802.11e [5] is designed to provide QoS support for
SpectraLink and Cisco have been shipping VoIP phones specif- time-sensitive applications on IEEE802.11 WLAN. It supports
ically designed for IEEE 802.11-based WLANs. However, two new channel access mechanisms: EDCA (Enhanced Dis-
there are still several technical barriers facing the VoWLAN tributed Channel Access) and HCCA (Hybrid coordination
technology that need to be overcome before it can truly take function (HCF) Controlled Channel Access). EDCA improves
off. upon DCF (Distributed Coordination Function) by introducing
The first barrier is lack of Quality of Service (QoS) sup- four queues, each of which corresponds to a different priority
port. ITU-T G.114 [1] recommends the the maximum one- of accessing the shared radio channel. More specifically, each
way voice packet delay be below 150 msec. Because this queue is assigned a different combination of medium access
corresponds to end-to-end path delay, the delay due to wireless parameters, including AIFS (Arbitrary Inter-Frame Space),
LAN channel access must be considerably less than 150 CWmin and CWmax. Although EDCA can effectively priori-
msec. In addition Most voice codec specifications such as tize voice traffic over data traffic, it cannot guarantee the QoS
G.729 [2] require the packet loss ratio in voice connections of individual VoIP connections when they are competing with
to be less than 1 % to avoid audible errors. The original one another for a shared medium. In fact, when the number
IEEE802.11 WLAN standard supports only best-effort service of voice connections contending for access to a radio channel
model. Almost all existing WLAN deployments operate in the increases, EDCA actually increases the probability of collision
DCF (Distributed Coordination Function) mode, in which each because of its aggressive medium access parameters such as
wireless station accesses the shared radio channel using an smaller CWmin and CWmax. HCCA is an enhanced version of
Ethernet-like medium access mechanism, which is inadequate PCF (Point Coordination Function), and supports a centralized
for real-time voice applications for the following reasons. polling scheme that could schedule network connections ac-
First, it is impossible to have precise control over the exact cording to their bandwidth demand and priority. Unfortunately,
transmission timings of voice frames because of collision most commodity WLAN interface products do not implement
and random back-off. When the injected traffic load is high, HCCA or PCF. In summary, IEEE 802.11e takes the right first

0743-166X/07/$25.00 ©2007 IEEE

But they work in the RF monitor mode.11e standard [5] enhances WLAN’s QoS by supporting Compression is not that useful either because voice data is traffic prioritization.11-based wireless LAN by announcement slot. which notifies all STAs of their channel time slots during layering a software-based Time Division Multiple Access a schedule cycle. Unfortunately software-based token connection are spaced out by 20 to 30 msec. geared towards VoIP applications. 802. and therefore could not be didn’t test EDCA under the scenario in which all 4 stations applied to VoWLAN phones. In ex- announcement frame. Wang et al. a voice period. Rao and Stoica [14] proposed an transmit at the highest priority. which is the default op- erating mode of most IEEE802. [8] also evaluated the HCCA. and a data period. a voice period. to voice traffic when compared with data traffic. (TDMA) protocol on top of DCF. proposes a simple QoS 10 msec and detects inactive nodes through long timeout. They used 4 stations each transmitting at a different NIC without providing concrete design and implementation priority. However. STDMA needs a per-scheduling-cycle announcement schedule announcement frame specifies how the channel time frame to avoid scheduling the radio channel for STAs that are is allocated among WLAN nodes in the current schedule cycle. which features two basic ideas: AP transmitting voice VoWLAN traffic. This scheme can nearly double the capac- proves an IEEE802. This enhancement mechanism called SpectraLink Voice Priority coarse timing granularity control method cannot be reused in (SVP). frames with backoff value of zero and always queuing voice frames in the head of the transmission queue. 1. then multicasted. Each schedule TDMA uses implicit time slotting to achieve the same effect cycle in turn is broken down into three parts: a schedule without incurring a per-STA polling frame overhead. and found that HCCA can the poor fairness and performance problem due to hidden improve channel utilization and provide better QoS support.11-based WLAN QoS. But these techniques cannot be applicable to VoIP guarantee on 802. and a data period. which relies on explicit polling is set to 20 msec in the current prototype and corresponds to to support contention-free access to the shared radio medium. it is not surprising that many researchers try to improve its 1 If a VoIP application’s packetization interval is different from 20 msec. Furthermore. The change. This solution III. and found that EDCA indeed can provide higher details.11 MAC anomalies. [12].11 WLAN channel’s capacity for VoIP ity of VoWLAN systems because it dramatically decreases traffic because it reduces the collision probability to the mini. It is difficult passing incurs too much overhead to be practical. The voice period consists of multiple time slots. STDMA P ROTOCOL D ESIGN is used to improve the AP’s priority of acquiring the WLAN As shown in Figure 1.11 MAC. The compression feature allows a WLAN card to compress The Wireless Rether project [6] tried to provide hard QoS data frames. It also They also found that EDCA may require significant tuning uses TDMA to regulate frame transmission but its design is not to offer better QoS for high priority traffic. Schedule Announcement Frame Contention Slot This project aims to simultaneously solve the QoS and Fig. Its EDCA mechanism can give better QoS already compressed. The bursting feature allows a for individual VoIP connections regarding packet loss. which is for voice frames. Neither does it solve the problem of substantial per-packet transmission overhead.11-based WLAN NIC to mitigate al.11 WLAN interfaces. the transmission overhead of downstream voice frames. Its scheduling slot size is a leading VoWLAN phone vendor. voice traffic. each of which Unlike PCF and HCCA. Schedule Cycle step toward solving the IEEE802. SpectraLink [9]. . the per-frame backoff overhead. this STDMA approach provides QoS guarantee the capacity of WLAN links. Mea- surements from a fully operational software-based TDMA voice traffic into a smaller number of larger frames that are (STDMA) prototype show that this approach drastically im. packet WLAN card to transmit frames back to back without incurring delay and delay jitter.11-based WLANs through a software-based traffic because typically consecutive frames within a voice token passing scheme. The IEEE to aggregate them to larger frames or to exploit bursting. which is for non-voice data frames. They also to evaluate its effectiveness in supporting voice connections’ mentioned an TDMA MAC project on 802. Atheros chipsets [11] support several new features to improve In addition. nodes and to mitigate other 802. each of which consists of a schedule an infrastructure-mode IEEE802. Time is divided into cycles. R ELATED W ORK a WLAN card to aggregate two frames into one larger frame. Gu and Zhang [7] simulate the EDCA mechanism controlling ACK and backoff through software. This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2007 proceedings. SoftMAC requires each WLAN card to throughput and lower latency to high-priority traffic. [10] proposes to aggregate downstream voice samples should be grouped into 20-msec chunks before being sent out. [13] in their SoftMAC project mentioned mechanism is not at all effective when all traffic sources carry some techniques to override the 802. idle. The mum and because it largely eliminates the back-off overhead. its capacity. this Neufeld et al. But it is ineffective in the face A 0 1 2 N VDown Up0 Up1 C DDown A Voice Period Data Period of a large number of voice connections. The fast frame feature allows II. one for each Given the limited capacity of existing VoWLAN systems. The channel time allocation of the proposed software-based TDMA per-packet transmission overhead problem associated with protocol. for example. In addition to EDCA. the proposed STDMA protocol channel for downstream voice traffic.11 WLAN’s QoS problem by Schedule Cycle (20 msec) (20 msec) supporting traffic prioritization. divides the channel time into schedule cycles. the packetization interval of VoIP applications1 . Garg et overlay MAC layer on 802.

the AP could accurately keep track of each STA’s downstream voice frames in one voice slot. The Data Period voice data frames. whenever it transmits an upstream packet. Ideally. in Figure 1. STDMA first schedules upstream data traffic from STAs. the bandwidth demand comes the downstream data time slot allocated to the AP. it is inefficient. Then the voice period starts. One computer serves as an . active voice connection’s upstream traffic. STAs that did not transmit data in previous schedule contention slot. A more IV. From an STA’s traffic request frame and its physical Consequently. In addition. the AP could determine the amount of controlled. too many STAs may try slots to transmit upstream voice frames. the first frame in the schedule cycle is the scheduling the data period is how to maximize its channel schedule announcement frame. the following downstream data slot in the same data The AP broadcasts a schedule announcement frame at period can be shortened to accommodate any such stretch. P ERFORMANCE E VALUATION efficient design. DDown. the packet loss. stretch the data voice period. to transmit non-voice data frames to STAs. the beginning of each schedule cycle. Unlike voice traffic.e. Again upstream data traffic is scheduled The data period is used for STAs and the AP to transmit first.11e standard. Up1. upstream non-voice data frames in time slots Up0. [16] showed that state-of-the-art silence in some future contention slot. and to use CPU and 256 Mbytes of memory. channel time allocated to this STA and schedules a data slot for it in the following data period. then a contention slot is scheduled for non-active STAs non-voice data frames. piggybacks into the packet a report of its current traffic load. TDMA is much more efficient a voice connection cannot be time-multiplexed on a single especially for software implementation. upstream data transmission is largely free of collision. contention slot. However. the end of the voice period. constantly polling STAs incurs a great deal design saves power because each voice traffic-carrying STA of scanning overhead without producing any obvious benefits. they are not likely to stretch A. C. The data period is also divided into time slots. the STA until receiving a traffic request frame from the STA several studies [15]. There are four computers connected statistical multiplexing among multiple voice connections. The key challenge in example. For of data traffic fluctuates considerably. After an STA reports to the AP that its channel slot for both upstream and downstream traffic because transmission queue is empty. which is at consecutively at the physical layer. After an STA starts transmitting B. This sparse and bursty. every active STA transmits its up. which has constant to send their traffic request frames (explained later). The AP aggregates and transmits all reports. The AP transmits 802. etc. This frame serves In the initial state. and schedules their packets in a way similar to weighted suppression works well. The Schedule Announcement Frame the contention slot significantly. suppression cannot effectively suppress most of the back. the AP uses the VDown time slot to transmit period into the next schedule cycle and disrupt the QoS downstream voice frames. utilization efficiency without disrupting the timing of voice Each voice traffic-carrying STA uses one of the N voice time frames. as soon as an STA has data to transmit. even though they could cause collision. in principle only needs to wake up during its associated time slots in each schedule cycle. to through a 100Mbps Ethernet switch. STAs transmit guarantee of subsequent voice frames. Without proper control. To avoid this disruption. In the data period. if silence load. because data traffic load tends to be slot immediately after each STA’s upstream time slot. its queue. One way to transport a voice connection’s associated with software-based polling on WLAN is quite downstream traffic is to allocate a separate downstream time high [6]. STAs that are not scheduled a contention slot C and finally downstream data traffic from in the current cycle can send traffic request frames in the the AP. it In each voice period. VoIP traffic is largely full-duplex the AP to constantly poll all STAs to determine their load in practice and therefore downstream and upstream traffic of and transmission schedule. Even if a contention slot is stretched. which the current STDMA prototype chooses. it the double purposes of synchronizing STAs and telling each submits a traffic request frame in the next available contention STA its associated channel slot times for upstream voice and slot. C. Unfortunately. the AP will stop scheduling for voice communication is half-duplex in nature. stream voice traffic in its own slot according to the schedule Based on load information in traffic requests and piggybacked announcement frame.11e’s TXOP mechanism [5] to transmit a batch of frames all downstream voice traffic in a single time slot. All of them are DELL aggregate multiple voice frames into one physical frame to PowerEdge 400SC. As a result. and finally per-connection bandwidth requirement. The traffic request frame includes traffic load information. then After the upstream time slots. We used the testbed shown in Figure 2 to evaluate the is to aggregate and transmit all downstream voice traffic to all performance of the STDMA prototype and compare it with the STAs in one time slot. each of which is used for transmission of non. This design allows an AP to exploit IEEE802. which features a 2. a voice connection only needs one fair queuing [17]. The Voice Period upstream packets.26 GHz Pentium-4 reduce per-packet transmission overhead [10]. At the end of the to transmit frames during a data period.. the amount of time required to empty the data currently in slots. delay and jitter can be strictly transmission speed. Because each STA is guaranteed its own i. which is used in PCF and requires ground noise. Finally the AP uses the downstream data cycles submit a traffic request frame to the AP during the time slot. Compared with polling. This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2007 proceedings. data transmission. Because traffic request frames are small. because the overhead channel time slot.

but the upstream voice traffic experiences program to generate traffic at a specific packet rate (packet much fewer losses. per-packet payload size.11 Wifi driver madwifi-ng-r1457 [18] and the High Res standard can actually support up to 18 simultaneous voice POSIX timer i386-hrt-2.11e 23 Down 0. There are totally 36 WLAN cards associated TABLE I with one AP. To emulate 802.11b TXOP limit to transmit multiple frames once it acquires the WLAN.11.11a/b/g mini-PCI card uses the AR5004X chipset. In this experiment. 2.e. This through three 4-port mini-PCI to PCI adaptors.11e. This is why it is the STAs rather than the AP that computers transmits to the AP packets of the total size of 98 become the bottleneck when a WLAN channel is fully loaded. which includes the Mad. unlike the difference between maximum and minimum packet delay IEEE802. Packet delay jitter is nections. This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2007 proceedings.23 23 30 802.97 26 54 12 WLAN cards 12 WLAN cards 12 WLAN cards 802.01 3 30 STA Computer1 STA Computer2 STA Computer3 STDMA 50 Up 1. rate is 11 Mbps. we use the Atheros chipset’s voice queue and use mini-PCI cards. each WLAN card on the STA channel.6. The testbed used to evaluate the performance of the STDMA prototype and the IEEE802. To emulate IEEE802. sequence number.11.11e 23 Up 3. IEEE802.10. The 36 Comparison among IEEE802.11. each of which logically corresponds to a distinct STA. and use NTP to syn.11e 22 Up 0. Each STA computer hosts 12 Wistron NeWeb IEEE802. AP and the other three serve as STAs. bytes and at a rate of 50 packets per second (pps).1 30 121 802. When the input traffic load of a WLAN channel is UDP receiver program measures the throughput and packet close to its capacity. we set the UDP payload priority than STAs as far as accessing the shared channel is size to 32 bytes (12 bytes RTP + 20 bytes G.11b mode with short preamble support enabled. IEEE802. the AP uses the same AIFS per second). and and CWmin as the STAs. at the socket send/receive interface to reflect the end-to-end For IEEE802.11e and STDMA can encoding rate of 11Mbps.11 18 Up 0. and therefore is not given a higher time stamp. which is smaller than DIFS. IEEE802.4 20 6 STDMA 50 Down 0. When the 23rd voice connection is added.. transmit/receive every 10 msec rather IEEE802. which are connected to the computer the STA computers with the same packet size and rate. we use the Atheros chipset’s IEEE802.11b WLAN assuming each call is a two-way 8Kbps voice connection. At the STDMA can support 50 voice calls successfully. CM9).11e standard.11. Because .11e and STDMA in terms of the WLAN cards transmit voice frames to the AP and the QoS of each voice number of voice calls that they can support over a single channel of an connection in terms of packet delay and delay jitter is measured and reported. In addition. In IEEE802. Our measurement shows that the IEEE802. To emulate more than 36 STAs working as an RF monitor to verify that the STDMA prototype using 36 WLAN cards.11 18 Down 0.patch [19]. the AP’s voice queue uses a large they can support over a single channel of an IEEE802. To emulate VoIP traffic.11e.11.11a/b/g mini-PCI cards same time. we use the Atheros chipset’s best- which consists of an AR5213 MAC controller chip supporting effort queue. All WLAN cards operate in the voice connections.11 19 Down 6 40 36 802. and an AR5112 dual-band radio.26 18 44 802. the AIFS of the AP’s voice throughputs of STDMA.6. one of them working as an AP while the other a 200-µsec voice slot size. the AIFS of STAs’ terms of the number of two-way 8Kbps voice connections voice queue. IEEE802.10. For AP computer hosts two Wistron NeWeb IEEE802. the AP transmits packets to each WLAN card on (Model No. it can support 22 concurrent voice con- delay perceived by the VoIP application. connections. The NeWeb set-up emulates a constant-rate two-way voice communication. some of them need to support two indeed works as expected. When the 19th voice connection is added.11a/b/g STDMA.11e 22 Down 0. We use a UDP sender goes up drastically.4 20 6 Fig. These computers are located Table I summarizes the number of concurrent 8Kbps voice within a 5-meter range and thus always transmit frames at the connections that IEEE802.8 14 28 AP RF monitor 802. ACK than every 20 msec. which is the same as in the baseline test.11 19 Up 0.729 voice). In IEEE802.11e and IEEE802. An concerned. the packet loss ratio of upstream traffic from measurements. IEEE802. frames are encoded in 11Mbps. AP Computer MAC Cards Direction Loss Ratio Latency Jitter (%) (msec) (msec) 802. The voice queue. support on a single IEEE802. the chronize their clocks over the wired Ethernet link so that we packet loss ratio of downstream voice traffic from the AP can measure the one-way packet latency.11e’s EDCA. STAs goes up while the downstream traffic experiences much The goal of this experiment is to compare the effective fewer losses. i. the AP is given less channel time share loss ratio based on the sequence number.11e.11 in queue is PIFS.01 1 6 802. The packet delay than it needs to transmit downstream packets and thus becomes and delay jitter are computed based on the time stamps taken the bottleneck.11b channel whose transmission All computers run Linux 2.

no. T.11 networks. Greene. downstream voice traffic can effectively leverage the TXOP out much modification.11 or IEEE802. (WWC).spectralink. “Evaluation of EDCF Mechanism for QoS in IEEE link-layer ACK is disabled.4% loss ratio under Network World. this paper proposes a software-based TDMA for VoIP System. we ran the same test with [1] V. “Service disciplines for guaranteed performance service in efficiency problems of commodity IEEE802. based medium access control and ACK elimination.729: Coding of speech at 8 kbit/s using conjugate- with link-layer ACK disabled experiences a packet loss ratio structure algebraic-code-excited linear prediction (CS-ACELP).. “Using IEEE 802. C. Grunwald. For IEEE802. Gopalan. the contention slot in each schedule cycle. Zhang.. that 802. ac- pang/course/voip 2005 fall/presentation/B2.ntu. C. S.. offering comparable packet delay and smaller delay jitter. M. IEEE802.” http://www. channel. 2005. an offered load of 1 Kpps from 4 WLAN cards. Fifield. “Soft- voice traffic. The result for IEEE802. [Online].html.csie. [4] showed that state-of-the.” http://www. vol. Real world experiments [3]. R.” http://madwifi.” http://www. Union. Doshi.11b link. “Aruba conquers challenge of Wi-Fi scalabil- In summary. hardware poses substantial challenges. the same 4 WLAN cards can [3] D.” around 10 concurrent voice calls over a single IEEE802. C ONCLUSION [13] C. packet-switching networks. The prototype could be easily ported to other VoWLAN appliances with- .html [4] ——. while [5] IEEE. timer.3 msec. N. cker Chiueh. Available: http://www. specifically. Sheth. 2005. V. “MultiMAC . 2005. January 1995. Fifield. 2004. and T.7 msec for the schedule announcement frame and to 50. De. [8] is somewhat surpris.pdf.11 mainly comes from the fact per G: Maximizing Wireless Performance. MAC . calls on a vanilla IEEE802.” http://standards.voip-info. thus significantly reduces the packet loss ratio of downstream [12] M. International Performance Computing and Communications Conference. can support on an IEEE802. K. The performance difference between STDMA and [10] W.” achieve around 3 Kpps throughput with 0. art IEEE802.” in Proceedings of ACM/SPIE Multimedia Computing and Net- working (MMCN 2002). how to [19] M. 802. The throughput improvement of STDMA over IEEE802. O. 01/10/05. LAN.729 Codec (8 Kbps).” http:www. “IEEE Standard 802.atheros. D.networkworld. 1374–1396. C.11 WLAN channel is disappoint. Available: (22 voice calls).html voice capacity by a factor of To isolate CNS-0410694 and CCF-0342556.11e [2] I. “VOIP Qos Requirements.11e mainly arises from two optimizations: white paper.Per Call Bandwidth Consumption. [Online].” Proceedings of the IEEE. re.11 and IEEE802. 2005.An Adaptive MAC Framework for Dynamic The number of VoIP calls that existing VoWLAN products Radio Networking. We have successfully implemented the first known STDMA prototype on Linux and commodity WLAN NIC without modifying the WLAN device driver. Malek. M. Although STDMA is conceptually of concurrent VoIP calls on an IEEE802. R EFERENCES based medium access Li. 2006. G. [14] A.html. Wiki.networkworld. “Voice Over IP .org/getieee802/802. S. This research is supported by NSF Awards CNS-0435373. Q. P. assuming each VoIP call uses G. Under 2005. Baker.11e ity.11b WLAN channel leaving 0. “SpectraLink Voice Priority.11 and IEEE802. Available: IEEE802. Sicker.. To overcome this VoIP [16] Yiying Zhou and others. “MadWifi . T. A. 83.11e over IEEE802. “A Multiplex-Multicast these commercial systems is mainly attributed to careful Scheme that Improves System Capacity of Voice-over-IP on Wireless LAN by 100%. and D. IEEE802.11e MAC for QoS over Wireless.” in Proceedings of ing because previous tests [3]. and D.” in Proceedings of the Ninth IEEE Symposium on reduction in per-packet transmission overhead and downstream Computers and Communications (ISCC).11b http://www. mainly because of the mysterious collision behavior when [7] D. [9] SpectraLink Inc. [4] [15] Cisco Systems Inc. Neufeld.11 WLAN in. Stoica. 2003. 2002.11e.729-199603-I/en of more than 25%! In contrast. The performance improvement [11] Atheros Communications Inc. the STDMA prototype can transmit downstream STDMA prototype show that STDMA can increase the number packets associated with 50 voice calls within 9. Liew. Inc. and X.11e without disabling link-layer ACK. Doerr. showed that they can support up to ten concurrent VoIP general/bwidth consume. “Performance Analysis capacity problem. spectively). “G.” in ingly small because of lack of QoS support and inefficient Proceedings of ACM Mobisys Conference 2005. Rao and I.pdf. 2005. “An Overlay MAC layer for 802.11. approach to simultaneously solve the QoS and transmission [17] H. 2005. voice traffic batching at the AP. [18] packet loss ratio of the upstream traffic under STDMA is “Implementation Experiences of Bandwidth Guarantee on a Wireless slightly higher than those in IEEE802. out the throughput improvement contribution due to TDMA.Flexible Wireless Research Platform. compared with IEEE802. The [6] S. terface hardware.8 and 2.11e-2005.11-based VoWLAN products can only support 2003. pp. WLAN packet transmission.” in Proceedings of Fourth Workshop on Hot Topics in Networks (HotNets) ’05. Peng. Wang. [Online].11e. X.11e but disabled link-layer ACK this time.” in First IEEE Symposium on New Frontiers in Dynamic Spectrum Networks (DySPAN) ’05. Gu and J.3. 2005. “Su- of Performance measurements on the mechanism.Multiband Atheros Driver for Wireless Fidelity implementing it on commodity operating system and WLAN (WiFi). STDMA (50 voice calls) can improve the http://www.” 1999.” synchronize WLAN nodes and how to support fine-resolution http://sourceforge. Neufeld. 2006. superg whitepaper. project. Doerr. 10. http://www. Sharma. 2006.11e gives the AP a higher priority over the STAs and 2005.ppt. M. whose length ACKNOWLEDGMENT is 20 msec. respectively. Garg. Cheng. 2005. This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2007 proceedings. Zhu.” in Proceedings of World Wireless Congress they also show decent capacity (18 and 22 voice calls. Pang.11 Wireless Networks. R. Grunwald.” Network World. “High Resolution POSIX timers. “Review: Voice over Wireless LAN. 11/06/ and V.