A New Integrated Control Algorithm for IEEE 802.11g Standards
Qian Liu, Yongxin Yan
International School Beijing University of Posts and Telecommunications Beijing, China e-mail: ee08b050@bupt.edu.cn, ee08b044@bupt.edu.cn
Abstract—Wi-Fi (Wireless Fidelity) mainly adopts IEEE 802.11g standards. And a binary exponential backoff algorithm is adopted to prevent channel collision, which also introduces reduction of the performance of the network. In this paper, an integrated control mechanism (ICM) is proposed which combines a centrally controlled approach and a distributed access mechanism to control the behavior of the system. Through this mechanism, backoff is eliminated from transmission mode, which can significantly enhance the performance of the system. Through theoretical analysis and simulation, the new mechanism can improve the throughput and reduce delay and packet loss rate. According to the results of simulation, with ICM, the throughput can have an improvement of 83.3% and the packet loss rate remains at 0 under appropriate conditions. Keywords-Wi-Fi, integrated control mechanism, centrally control, distributed access

Jing Tu, Bainan Xia
International School Beijing University of Posts and Telecommunications Beijing, China e-mail: ee08b253@bupt.edu.cn, ee08b217@bupt.edu.cn Many experts have proposed different algorithms and approaches to improve the performance of Wi-Fi. Souvik Sen, Romit Roy Choudhury and Srihari Nelakuditi transformed the contention from time domain to frequency domain using subcarrier and multiple antennas. It migrates from backoff operation to frequency domain so that backoff time is greatly reduced to one OFDM symbol time. [1] Ma Dan and Wen Jun proposed dynamic optimization algorithm based on the conflict weight by replacing the accurate measurement of the number of competition terminals with average collision count and introducing conflict weights into DCF. [2] There are many other proposals, like adaptive backoff and implicit pipelining. However, all these improve performance of Wi-Fi based on a common assumption, the application of DCF. And no matter how many improvements they have made, backoff cannot be eliminated completely. [3] II. BACKGROUND Instead of using distributed coordination function alone, we innovatively combine centrally controlled approach and distributed access mechanism together to regulate the entire system’s behavior, in which case backoff does not exist in competing for the channel. The way we realize that is the flexible use of CTS control frame. The traditional RTS/CTS mechanism is applied in this way: nodes that want to send packets send a RTS to Access Point (AP) firstly and the AP returns a broadcast CTS with the address of the node that is allowed to send packet stored in the Receiver Address (RA) field. So node can determine whether to start to send packet by comparing the RA field with its own address. That is the basic fundamental principle of RTS/CTS mechanism. [4] In our approach, only CTS control frame is needed to announce the node to send packet, which transfers the original two-way handshaking to one-way. The principle is similar. In every AP, there is a list of sequenced nodes it controls. The way nodes are sequenced is based on random number generation to guarantee fairness. When it's time for certain node, let say, N1, to send packet, the AP sends a broadcast CTS with the address of node N1 stored in the RA field. And after N1 receives the CTS, it will know it can use the channel now and then start to send packet. Thus, no backoff is necessary to avoid contention because the channel of certain duration exclusively belongs to certain node to use. Because the order and the utilization

I.

INTRODUCTION

Wi-Fi (Wireless Fidelity) is very popular and useful in WLAN communication since it was put forward. Among the IEEE 802.11 standards, 802.11b is the original standard designed to support Wi-Fi. And 802.11g and 802.11a were proposed afterwards to improve the performance of Wi-Fi with new concepts and techniques. Nowadays, 802.11g has become the most important standard for Wi-Fi, which also has backwards compatibility with 802.11b. Although various techniques are applied in 802.11g, e.g. OFDM and DSSS, to greatly improve the communication speed, one stubborn problem embedded in Wi-Fi has not been resolved yet. In Wi-Fi, to prevent collision caused by concurrent transmission, one mechanism called Distributed Coordination Function (DCF) is adopted. DCF has one important characteristic, that is, Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) protocol. This protocol introduces a binary exponential backoff mechanism which causes the underutilization of media and the waste of resources including time and bandwidth, especially at high transmission rate. Another technique used in Wi-Fi is Request-To-Send/Clear-To-Send (RTS/CTS) mechanism, which involves a two-way handshaking behavior to allow nodes to access the media. This mechanism leads to delay in transmission process. Because of these two mechanisms, performance of Wi-Fi, including throughput and delay, does not meet the expectations of people.

are determined by access point, this is a centrally controlled approach. As for new node to access the media, it has to advertise itself to the access point to let AP add it to the list. And this is realized by the use of RTS control frame, the only process RTS is used with the possibility contention and backoff may happen. This is a distributed access mechanism. Since our approach is a combination of centrally controlled mechanism and distributed access mechanism, it is an integrated control mechanism (ICM). Next, we will describe our approach in details. III. NEW ALGORITHM Since access point is the core of centrally controlled approach, we focus on the behavior of AP to explain this innovative mechanism. We ideally divide the behavior of AP into two modes, access mode and transmission mode. The former is for the newly-coming nodes to add to the list of AP. The latter is for the nodes existing in the list of AP to send packet. These two modes happen alternately in the behavior of AP. And one access mode and one transmission mode constitute a period of AP’s behavior. Next, they are introduced separately. A. Access Mode At the initial stage, no nodes exist in the list of AP. So AP has to special broadcast CTS with specific address stored in the RA field to announce all the nodes that they can advertise themselves to AP now. When nodes receive the CTS, they will return with a RTS to tell the AP that they want to access the media. When the AP receives the RTS, it will put the node sending the RTS to the first place of the list and enter into the transmission mode. If there are two or more nodes needing to access the media concurrently, the RTS they send to AP will get corrupted. At this time, AP will ignore the corrupted RTS and enter into the transmission mode directly without adding any node to the list. If the node does not receive a normal

broadcast CTS for it immediately (explained later), the node knows it does not obtain the access to the media and it will generate a random number, v, in a range, let say, [1, M]. The random number indicates the next time the node tries to access the media. When the node receives v special broadcast CTSs from AP, it tries to access the media again. B. Transmission Mode AP selects the first node, let say, N0, from the node list and sends normal broadcast CTS with address of N0 stored in the RA field. After comparing the address in RA with its own address, N0 knows it’s time for it to send packet and other nodes just ignore the CTS. If N0 has packet(s) to send, it will send a packet of data to AP along with a randomly generated number in a range. If N0 still has another packet to send, the range of the random number is [0, 2n-1]. If this is the last packet, the range is [2n, 2n+m-1]. If AP successfully receives the data packet, it will send an ACK back to N0. And the number received with the data replaces the original value of N0. If N0 is a newly-added node from last access mode, the number is the value of N0. According to the amount of value of each node in the list, AP reorders the nodes. And then a new normal broadcast CTS with the address of the first node in the new list stored in the RA field is sent out for another packet transmission. If N0 has no packet to send, it will send an empty data packet along with a randomly generated number in a range, [2n, 2n+m-1]. Similarly, AP sends back an ACK if the empty data packet is received successfully. And the number with the packet replaces the original value of N0. A new ordering of nodes happens and a new normal broadcast CTS with the address of the first node in the new list stored in the RA field is sent out for another packet transmission. Sending normal CTS, waiting SIFS, receiving data packet and sending ACK constitute a cycle of transmission (Shown in Fig1). And one transmission mode contains several such cycles. When the transmission mode ends, AP enters into a new access mode to accept new nodes.

Figure 1. Access procedure in the new mechanism

The different settings of the ranges of randomly generated number with data packet and empty data packet

enforce the node with no data to send to be ordered to the end of the list in AP. And when a node sends a certain

number of empty data packets, it is removed from the list. To enter the list of AP again, it needs to wait for the access mode and retry. IV. EVALUATION According to the description of our integrated control mechanism, it can be easily seen that our approach eliminates the possibility of backoff in transmission mode and thus saves channel resources and enhances the overall throughput. Although contention may happen in access mode, it is still a great improvement compared with the original contention all the time. The reduction from twoway handshaking to one-way handshaking simplifies the transmission process. All these will result in an enhancement of performance of Wi-Fi. Besides, we can have a high control over the entire network by setting several key values, like M, n, m mentioned above. Hence, the system based on ICM is a highly controllable system with high performance. Next, simulation about this approach and a comparison between it and the original mechanism are introduced. V. SIMULATION AND COMPARISION To validate our new mechanism, we use MATLAB to simulate the performance, including throughput, system delay, and packet loss rate, of the network under our mechanism with different parameter settings. Besides, we compare it with the performance of the network under the original old mechanism to examine the improvement. A. Standard Parameter Value Variables
Scale of users Average data length(DATA) A Packet Max waiting time(LIMIT) Sending data request period(NEXT) Waiting period of access retry Random range for data packet Random range for blank packet

In Fig. 2, the overall throughputs of the network under new mechanism and old mechanism with different intervals of sending packet are presented. When the number of nodes is small, throughput is likely to increase as the number of nodes increases. But the speed of increasing under new mechanism is larger than that under old mechanism. When the number of nodes increases to some extent, the throughput will remain stable, in which case the throughput of the new mechanism have an improvement of 83.3% compared with throughput of the old mechanism. When we change the interval of sending a packet, the same conclusion can be obtained from the comparison between new mechanism and old mechanism. Besides, the smaller the interval of sending a packet, the faster the throughput reaches its peak.

Figure 3. Throughputs of new VS old by changing packet size

Range
0-50 19500/29500/9500 byte 50/100/400 ms 0-2 seconds 1-4T(default T=100ms) 0-1024 1025-2048

Constants
Data rate Control bit rate Length of ACK message Length of SIFs message in time Physical header length in time Mac header length Max packet length (Ethernet)

Value
54Mbps[5] 11Mbps 112 bit 10 us[6] 192 us[7] 224 bit 1500 byte[8]

In Fig. 3, we still simulate the trend of throughputs of new mechanism and old mechanism but with different sizes of packets. No matter how large the packet is, the throughput of the new mechanism is always larger than those of the old mechanism with the same size of packet. And the improvement is nearly 90% when the packet is 9500-byte long. Under the same mechanism, the throughput has slight difference with different sizes of packets. When the packet is large, the throughput is a bit larger than small packet.

B. Results of simulation

Figure 4. Average Delay new VS old by changing intervals

Figure 2. Throughputs of new VS old by changing intervals

Fig 4 plots the trend of the change of system delay as the number of nodes increases under different mechanisms and different intervals of sending packet. When the interval of sending packet is large, say 100ms or 50 ms, the system delay of the new mechanism is always smaller than that of the old mechanism. Although system delay tends to increase with increasing nodes, the slope of the new mechanism is flatter than the slope of the old mechanism,

which proves a better performance. When the interval of sending packet is small, say 10 ms, the system delay of the new mechanism is larger than that of the old mechanism in the case of a few nodes. However, when the nodes of the network increase, the system delay of the new mechanism becomes smaller than that of the old mechanism because of its relative flat slope.

whether a packet is lost. From Fig. 6, we can find out that when the timing value is large, say 400 ms or 100 ms, the packet loss rate remains at 0.00%. But when the timing value is reduced to 50 ms, some packets become lost when the number of nodes exceeds 50 and the trend of the increase of packet loss rate is rather fast. Referring to the old mechanism, the packet loss rate starts to increase from 0 when the number of nodes only reaches 18 and suffers a slow increasing afterwards. Therefore, our new mechanism performs better than the old mechanism in packet loss rate when the timing value determining whether a packet is lost is large. VI. CONCLUSION The binary exponential backoff algorithm used by DCF causes underutilization of channel and waste of resources in spite of the advantages it brings. Our integrated control mechanism flexibly applies CTS control frame and innovatively let access point determine the order of nodes to transmit packets, which eliminates backoff in transmission mode. Although contention and backoff may exist in the access mode, it does not influence the whole performance too much. And the results of simulation prove the superiority of the new mechanism over the old mechanism. The overall throughput of the network under new mechanism has an improvement of 83.3% compared with the old one. And the packet nearly suffers a loss rate of 0 under appropriate conditions. Although some drawbacks exist in our new mechanism which we will solve in future work, the ICM provides a great improvement to the performance of the network compared with the old mechanism. REFERENCES
[1] Sakurai,T. and Vu,H.L,“MAC Access Delay of IEEE 802.11 DCF” Wireless Communications, IEEE Transactions on, 2007, 6(5), p1702-1710 Wu Ruixiao and Liu Qian “A Novel Centrally Controlled Channel Access MAC Protocol for 802.11 Wireless Networks”. 2010 walking paper MIAO Yi，FENG Hao-ran. “Energy-efficient reduced collision MAC for wireless sensor networks“.Computer Engineering and Applications，2010，46（35）：97-100. Feng Haoran and Zhao Qianchuang, “Modeling a typical enhanced 802. 11 DCF-based protocol DIDD in nonsaturated conditions“,Tsinghua Univ(Sci&Tech), 2009, Vol 49,No.1 pp157150 Listen (on the frequency domain) before you talk Souvik Sen, Romit Roy Choudhury, Srihari Nelakuditi Research and optimize in IEEE802.11 MAC protocol. Ma Dan, Wen Jun WU Ya-jun, HU Ai-qun, SONG Yu-bo,” Design and Implementation of WLAN Protocol Analysis System” Computer Engineering, 2008.11, Vol 34, No.22, pp 140-142 H. S. Chhaya and S. Gupta, “Performance modeling of asynchronous data transfer methods of IEEE 802.11 MAC protocol,” Proc. IEEE Symp. Wireless Networks,, IEEE Press, Dec. 2007, pp. 217-234, doi:10.1109/SCIS.2007.357670.

Figure 5. Packet loss rate of new VS old by changing intervals

To examine the influence of the new mechanism to packet loss rate, we plot the relationship between the packet loss rate and the number of nodes under different mechanisms and different intervals of sending packet in Fig. 5. With the same interval of sending packets, the packet loss rate of the new mechanism is obviously smaller than that of the old mechanism. Especially when the number of nodes gets large, the packet loss rate of the old mechanism becomes intolerable while the packet loss rate of the new mechanism keeps small (no more than 0.10%). When the interval of sending packet is large, say 100 ms or 50 ms, the packet loss rate of the new mechanism stays at 0.00%, which indicates a big improvement compared with the old mechanism. When the interval of sending packet becomes small, say 10 ms, the packet loss rate of the new mechanism becomes larger than 0.00% when the nodes increase, but no more than 0.10%. From this result, we can conclude that our new mechanism provides a great improvement to packet loss rate.

[2]

[3]

[4]

[5] [6] [7]

[8] Figure 6. Packet loss rate of new VS old by changing timing values

Further, we examine the performance of packet loss rate under different timing value which determines