Fast Mechanism for VoIP in IEEE 802.11e wireless LANs Gyung-Ho Hwang and Dong-Ho Cho Division of Electrical Engineering, Department of Electrical Engineering and Computer Science, KAIST, 373-1 Guseong-dong Yuseong-gu Daejeon, Korea E-mail: gabriel@comis.kaist.ac.kr, dhcho@ee.kaist.ac.kr Abstract The IEEE 802.11 wireless LANs provide cheap and high data rate mobile service. The voice service through wireless LANs has became important issue. The new IEEE 802.11e standard provides a framework to support quality of service(qos) according to the traffic types. The enhanced distributed channel access(edca) mechanism, that is contention-based access, has several access parameters which are able to give higher access probability to the high priority service. Since the real time service such as voice of IP(VoIP) has a delay constraint to guarantee a QoS, the fast collision resolution is required. In this paper, we review the IEEE 802.11e standard and propose a new fast retransmission mechanism for s to reduce the delay and improve the channel utilization by eliminating overhead time for backoff procedure. The simulation results show that the proposed scheme reduces the delay and dropping probability of s. I. INTRODUCTION Wireless LAN is very fascinate technology providing mobility and convenience. In the current world market, the IEEE 802.11-compliant wireless LAN equipments are dominant. IEEE 802.11 wireless LANs equipments have been extensively deployed in the recent years in many different environments. Moreover, many hotspots are covered with wireless LANs by mobile service providers. The current 802.11 devices provide the best effort service with the transmission data rate up to 54Mbps at 2.4GHz and 5GHz bands. The main MAC protocol of IEEE 802.11 is CSMA/CA that is a quite old protocol. But this protocol has got an attention since the IEEE 802.11 wireless LAN is deployed[1]. The original 802.11 DCF scheme does not support QoS, that is, it does not have any concept for the different priorities. So, the IEEE 802.11e task group(tg) was organized to make a scheme for providing QoS. The IEEE 802.11e draft provides a framework that can differentiate QoS between traffic categories according to their priority[2]. Since the IEEE 802.11e is organized, there have been many researches on the differentiation of quality of service(qos) in CSMA/CA-based original DCF scheme. In [7], the performance of the EDCA was simulated using three traffic types. But the access parameters were fixed and the effects of the parameters were not presented. Kuo et al. [8] provided an analytical model to study the expected bandwidth for each traffic class in EDCA scheme. Yamada et al. [9] presented a contention window control scheme to suppress the delay fluctuation of the real-time traffic, where the contention window size at retransmission is monotonically decreased in case that a queue has large waiting time. Romdhani et al. [10] proposed adaptive contention widow size control method in EDCA scheme, where the increase and the decrease factors of contention window size varies according to the service classes in case of retransmission. In [12], Hiraguri et al. proposed priority access scheme for s. But this proposed method is actually similar to current 802.11e EDCA scheme. The EDCA method is based on the original DCF scheme and it can differentiate the service classes by using a different access parameters. The support of voice over IP(VoIP) in wireless LANs should be the cutting edge issue because the price of wireless LAN equipments becomes lower and there is no wireless technology to provide such a high data rate like IEEE 802.11a. When VoIP is implemented in wireless LANs, the quality of voice service should be guaranteed. The rest of this paper is organized as follows. First of all, we describe the MAC protocol of the IEEE 802.11e wireless LANs focusing on EDCA scheme. Next, we propose the fast retransmission mechanism for s in case that VoIP and data traffics collide and we evaluate the performance of the proposed scheme compared with original IEEE 802.11e scheme using extensive simulation. Finally, conclusions and further works are presented. II. IEEE 802.11E EDCA SCHEME There are two main schemes for QoS support in the IEEE 802.11e standard. The hybrid coordination function(hcf), which is a coordination function of IEEE 802.11e, has two access schemes: HCF contention-based channel access, that is named as EDCA and HCF-controlled channel access. Generally, real time service such as voice call, is served through controlled or reservation-based channel access schemes. However, we can give a priority to real time traffic even in contention-based channel access scheme with IEEE 802.11e EDCA scheme. Moreover, in ad-hoc mode where there is no centralized control unit, the stations use the EDCA scheme to communicate s. In EDCA scheme, each user has several queues for different prioritized traffics and an access category(ac) is assigned to each traffic.
Each data frame has a specific priority and the MAC frame header contains its priority. In 802.11e standard, a station can have four ACs. Each frame arriving at the MAC with a priority is mapped into an AC as shown in Table. I where some abbreviations are referred from [5]. The main difference between EDCA and the DCF scheme is that each AC has its own IFS called as arbitrary IFS(AIFS), and minimum and maximum of contention window size. Namely, each AC has AIF S[AC], CWmin[AC] and CWmax[AC]. The AIF S[AC] is as follows. AIF S[AC] =AIF SN[AC] aslott ime + asif ST ime (1) Fig. 1 illustrates the EDCA scheme. The AIF SN[AC], CWmin[AC] and CWmax[AC] are announced by the QoS access point(qap) in the EDCA parameter set information element within beacons and probe response frames. But in adhoc mode where there is no AP, the stations use the default parameters. The default parameters in IEEE 802.11b physical layer are shown in Table II, where the AIFS s difference between AC VO and AC BE is 1 slot length. Therefore, the first slot after AIFS for voice packet, that is represented as AIF S v, is dedicated to voice traffic. DIFS/ AIFS Busy Medium PIFS SIFS AIFS[j] AIFS[i] DIFS Defer Access Slot time Contention Window Slots Fig. 1. EDCA Scheme in IEEE 802.11e Next Frame The smaller AIF S and CW values make the higher access probability. Therefore, the QAP can adapt these parameters according to traffic conditions. The new concept of a transmission opportunity(txop) is defined with a starting time and a maximum duration. When the medium is determined to be available under the EDCA TABLE I USER PRIORITY TO ACCESS CATEGORY MAPPINGS Priority UP 802.1D AC Designation Designation lowest 1 BK AC BK Background 2 - AC BK Background 0 BE AC BE Best Effort 3 EE AC VI Video 4 CL AC VI Video 5 VI AC VI Video 6 VO AC VO Voice highest 7 NC AC VO Voice rules, the station can transmit packets during TXOP. The duration of a TXOP in EDCA is limited by a TXOP limit distributed in beacon frames. Hence, the TXOP value is very important parameter that could determine the radio resource usage. However, we consider no TXOP concept in this paper and the EDCA-TXOP is defined as one packet transmission. III. VOICE OF IP PACKETS We consider the G.711 codec for generation of voice packets which has source bit rate of 64Kbps. The contains vocoder frame and many protocol headers. In case of IPv4, the RTP/UDP/IP header is 320 bits (40bytes) when the header compression is not used. If the header compression is used, the RTP/UDP/IP header is 16bits. We do not consider the header compression and the voice activity detection(vad). Therefore, we consider the voice traffic generating (320 + 1280) bits per 20 msec. The structure of in IEEE 802.11 wireless LAN is shown in Fig.2. We set the delay bound of wireless access for s to 20 msec. This allows sufficient delay margin for the backbone network for an end-to-end delay budget. The QoS of VoIP traffic is defined as the loss rate of voice packets, P loss which is less than 1%. The P loss is defined as following. P loss = P overdelay + P retrylimit (2) where P overdelay is the dropping probability of voice packet in case that elapsed time for wireless access exceeds a predefined delay bound, and P retrylimit is the discarding probability of voice packet due to retry limit. PHY Preamble + Header 20byte 8byte 12 byte Depends on Codec & VAD MAC Header IP UDP RTP Voice Payload Fig. 2. VoIP Packet in wireless LAN MAC FCS IV. FAST RETRANSMISSION MECHANISM FOR VOIP To reduce the delay of s and collision probability, we propose a new fast retransmission scheme. Basically, the access probability of s can be adjusted by assigning AIF S, CW min and CW max. However, there still exist collisions and retransmissions are needed. To satisfy the delay constraints of s, the retransmission should be executed as quickly as possible. Moreover, the packets are discarded in case that the number of retransmission exceeds the pre-defined retry limit. Therefore, the collision probability should be minimized. TABLE II DEFAULT EDCA PARAMETERS IN IEEE 802.11B PHY Parameters AC VO AC VI AC BE AC BK AIFSN 2 2 3 7 CWmin 7 15 31 31 CWmax 15 31 1023 1023
First, let us consider the general situation where the VoIP packets collided with data packets. Normally, the length of is small compared with data packet. Consequently, user to transmit s may detect busy channel after completing its transmission if there is a longer data packet that is transmitted simultaneously. Using this information, the VoIP user can retransmit the packet without performing a backoff procedure. Fig.3 shows the general EDCA mechanism in case that two packets collide. The station A transmits a and station B transmits a longer data packet. If there s a collision between these two packets, the station A and station B should wait for Ack timeout interval. But, in view of station A, the ack timeout timer may expire during the data packet transmission and it can detect busy channel. If the ack timeout timer expires before completing data packet transmission, the station A starts backoff procedure as soon as the channel becomes idle. The station B performs a normal backoff procedure in this case. For other listening stations, there are a few situations. The station C listens a data packet s preamble but the frame check sequence is failed due to collision. So, the station C waits for EIFS time. On the contrary, the station D only detects the collision as busy signal. Therefore, the station D just waits for AIFS time before starting backoff procedure when it has pending voice or data packets. If it has pending s, the station D will wait for AIF S v time before decreasing backoff counter. Considering above all situations, the new fast retransmission is presented in Fig.4. The main motivation of this scheme is that the collision probability between VoIP and data packets is higher than collision probability between s because generally the concurrent number of VoIP sessions is small in one cell. The station A detecting the busy signal after transmitting its own, retransmits the after AIF S v time without backoff procedure. If either VoIP or data packets are received correctly, the station A can detect the Ack frame and the retransmission does not occur. Since voice traffic has delay bound, we should restrict the number of retransmission. We set the retry limit to three in this paper. As shown in Fig. 4, the first retransmission is done after channel becomes idle. The first slot after AIF S v is only dedicated to voice packets. Therefore, the collision with data packets does not appear. In case that the first retransmission fails, the second retransmission performs the original backoff procedure. Failing of the first retransmission indicates that there is a collision between s. When the station A detects no busy channel and ack timeout timer expires, original backoff procedure follows because it means that there may be other VoIP transmission of the same length. We apply this new fast retransmission scheme to only VoIP packets because the data users with small packet size may have high access probability and it can affect the delay of s against providing QoS. V. EVALUATION AND DISCUSSION For the performance evaluation, we use the OPNET tool. The simulation parameters are shown in the Table III. We consider only the ad-hoc mode where there is no AP. We use two access categories for performance evaluations. One is packet voice traffic that is delay sensitive service and the other is data traffic that is best effort service. Namely, both AC VO and AC BE in Table I are considered. The VoIP traffic has constant inter-arrival time of 20msec. The best effort traffic has exponential inter-arrival time with 12msec on the average that makes 1Mbps data load. We used the IEEE 802.11b physical specification which is widely deployed in wireless LAN equipments which supports up to 11Mbps data rate. The control packet such as ACK frame is also transmitted using the same data rate as data packet. s are able to be retransmitted up to 3 times and the retry limit for data packets is set to 7. The delay bound is 20 msec. When two or more packets collide, other receiving stations detect the transmissions as busy signal or frame check sequence(fcs) error according to receiving conditions. If the receiver detects busy signal, it starts the backoff procedure after detecting idle channel during AIF S time. But in case of FCS error, the receiver should wait for EIFS time. In this simulation, we give each case a half probability, that is, a half of listening stations experience FCS error due to the collision, which could be seen in the station C shown in Fig. 3 and a half of listening stations recognize the collision as just busy signal which could be seen in the station D shown in Fig. 3. The performances of the proposed scheme are shown in Fig.5 to Fig.8. We evaluate the performance with fixed VoIP users as increasing the number of data users. Fig.5 shows the average delay of s. The average delay means the elapsed time from arrival at MAC queue of the higher layer packet to correctly received time, that it, the sum of queueing delay and MAC transmission delay. As the number of data users increases, the delay also increases because the probability of collision with data packets becomes high. But the proposed fast retransmission scheme has shorter delay since the first retransmission is fast and avoids collision with data packets. In fact, the data user is saturated when the number of stations is over 5 which means that the data users always have a packet to transmit. The case of 10 voice users TABLE III PARAMETERS USED IN SIMULATIONS Parameters Symbol Value PHY(Data rate) 11Mbps Slot time T slot 20µsec SIFS time T sifs 10µsec AIFS for VoIP T aifs,v 50µsec AIFS for Data T aifs,d 70µsec Propagation delay T prodely 1µsec Delay Bound T delaybound 20µsec size L voip 200 byte Data packet size L data 1500 byte
EDCA mechanism Station A Detect BUSY signal Station B Data packet AIFS_d Listening Station C FCS error Wait EIFS time Listening Station D BUSY signal AIFS Data/VoIP Packet Fig. 3. s s collision with Data packets in EDCA scheme Fast Mechanism Station A Detect BUSY signal No First No ACK & Detect BUSY signal Normal Second Collided with other VoIP packet No BUSY signal Station B Data packet Fig. 4. Fast Mechanism for s has longer delay than that of 5 voice users, which is analogical. We classify the dropping rate into two types. One is due to retry limit and the other is due to delay bound. The dropping rate of s due to retry limit is shown in Fig. 6. The fast retransmission scheme has lower dropping rate since the first retransmission makes collision probability low and reduces the number of retransmissions. Fig. 7 shows the dropping rate of s due to delay bound. In simulation, the s are discarded in cases that either waiting time in queue or delay of received VoIP packet exceeds the pre-defined delay bound. Like Fig.6 the proposed scheme shows better performance. We consider that the sum of two dropping rates determines the QoS of VoIP service and use 1% packet loss as QoS constraint. Satisfying the constraint, the original scheme supports about 15 data users but fast retransmission scheme is able to support 40 data users in case that 5 voice users exist. Finally, we evaluate the channel utilization of data packets, which means the used time for successfully transmitted data packets over total simulation time, which is shown in Fig. 8. The channel utilization decreases as the number of data users increases because the collision probability becomes high which wastes the channel resource. The data utilization with 5 voice users is much higher than that with 10 voice users. One interesting point is that the voice users affect the data utilization severely though the voice traffic itself is not high load. The proposed scheme shows more data utilization than the original scheme because of low collision probability. VI. CONCLUSIONS AND FURTHER WORKS In this paper, we reviewed the IEEE 802.11e standard and proposed a new fast retransmission mechanism for VoIP packets to reduce the delay and dropping rate of s. The proposed scheme eliminates overhead time for backoff procedure in case of collision between VoIP and data packets. The simulation results showed that the proposed scheme has better performance compared with original 802.11e EDCA scheme in view of delay and dropping rate. Since there is
Fig. 5. Average delay of s Fig. 7. Dropping rate of s due to Delay Bound Fig. 6. Dropping rate of s due to Retry Limit Fig. 8. Channel Utilization of Data Packets not a centralized control unit performing AP s function in adhoc mode, it is difficult to guarantee the QoS of VoIP service. We are going to research on providing guaranteed service on s in ad-hoc mode. REFERENCES [1] Part11:Wireless Medium Access Control(MAC) and Physical Layer(PHY) specifications, IEEE 802.11 Std, 1999. [2] Part11:Wireless Medium Access Control(MAC) and Physical Layer(PHY) specifications; Medium Access Control(MAC) Enhancement for Quality of Service(QoS), IEEE 802.11e/D8.0, Feb. 2004. [3] Part11:Wireless Medium Access Control(MAC) and Physical Layer(PHY) specifications; High-speed Physical Layer in the 5Ghz Band, IEEE 802.11a Std, Sep. 1999. [4] Part11:Wireless Medium Access Control(MAC) and Physical Layer(PHY) specifications; High-speed Physical Layer Extension in the 2.4Ghz Band, IEEE 802.11b Std, Sep. 1999. [5] Part3: Media Access Control (MAC) bridges, ANSI/IEEE 802.1D Std, 1998. [6] Yang Xiao, -based Priority Schemes for IEEE 802.11, Proc. IEEE ICC 2003, pp. 1568-1572, 2003. [7] Sunghyun Choi, Javier del Prado, Sai Shankar N and Stefan Mangold, IEEE 802.11e Contention-Based Channel Access(EDCF) Performance Evaluation, Proc. IEEE ICC 2003, pp. 1151-1156, 2003. [8] Yu-Lian Kuo, Chi-Hung Lu, Eric Hsiao-Kuang Wu, Gen-Huey Chen and Yi-Hsien Tseng, Performance Analysis of the Enhanced Distributed Coordination Function in the IEEE 802.11e, Proc. IEEE VTC fall 2003, 2003. [9] Hiroyuki Yamada, Hiroyuki Morikawa and Tomonori Aoyama, Decentralized Control Mechanism Suppressing Delay Fluctuation in Wireless LANs, Proc. IEEE VTC fall 2003, 2003. [10] Lamia Romdhani, Qiang Ni and Thierry Turletti, Adaptive EDCF: Enhanced Service Differentiation for IEEE 802.11 Wireless Ad-Hoc Networks, Proc. IEEE WCNC 2003, pp. 1373-1378, 2003. [11] Stefan Mangold, Sunghyun Choi, Guido R. Hiertz, Ole Klein and Bernhard Walke, Analysis of IEEE 802.11e for QoS Support in Wireless LANs, IEEE Wireless Communications, pp. 40-50, Dec. 2003. [12] Takefumi Hiraguri, Takeo Ichikawa, Masataka Iizuka and Masahiro Morikura, Novel multiple access protocol for Voice over IP in Wireless LAN, Proc. IEEE ISCC 02, 2002.