Enhancing the DCF mechanism in noisy environment

Transcription

1 Enhancing the DCF mechanism in noisy environment 1 LICP EA 2175 Université de Cergy-Pontoise 3 Av Adolph Chauvin 9532 Cergy-Pontoise France {adlen.ksentini, mohamed.naimi}@dept-info.u-cergy.fr Adlen Ksentini 1, Abdelhak Gueroui 2, Mohamed Naimi 1 2 PRISM CNRS Université de Versailles 45 Av des Etats-Unis 7835 Versailles France mogue@prism.uvsq.fr Abstract-The escalating deployment of wireless networking technology as well as other wireless technologies in the same unlicensed spectrum is rapidly increasing the radio frequency (RF) interference for IEEE products. These interferences involve a high error rate in the channel, leading to degrade considerably the Wireless LAN (WLAN) performances. This paper focuses on the problems associated with the IEEE in presence of high errors in channel, and ways to improve its performance in this situation. The actual MAC layer has no mechanism to differentiate random losses on wireless link from collisions, and therefore treats all losses as collision. Besides demonstrate link errors effect over IEEE performances, our contribution consists of enhancing the RTS/CTS handshake mechanism in order to improve the IEEE behaviors under a noisy channel. We compare the performance of our proposition with the actual MAC protocol and demonstrate considerable throughput gains with our approach. Keywords: IEEE 82.11, RTS/CTS, Contention Window, noise. I. INRODUCTION Wireless communication has become an essential part of modern life, allowing users to maintain network access without being tied to a particular location by a wire. As the bandwidth and throughput of wireless technologies have increased, it has become possible to support true multimedia applications, including, voice, and video traffic. Nevertheless the IEEE 82.11[1] physical modulations operate in the Industrial Scientific and Medical (ISM) frequency band, 2.4 GHz, which interferes with other products using this unlicensed frequency. Here interferences indicate high Bit Error Rate (BER) in the channel, which degrade considerably the wireless link reliability. For instance, simulation results obtained in [2], where the authors show such a situation: a Bluetooth (using FHSS) slave operating close to a Wireless LAN Access Point (AP) (using IEEE 82.11, DSSS), causes a very high frame drop rate, up to 46% and high access delays on the WLAN side. This paper involves the effect of packet errors on IEEE MAC s performances. Actually, the IEEE MAC layer uses the Carrier Sensing Multiple Access/Collision Avoidance (CSMA/CA) protocol to share the wireless medium among the stations. Before sending a packet, a wireless station first senses the medium for the duration of Distributed Inter-Frame Space (DIFS). If the medium is free for the duration, the wireless station starts sending the packet immediately. Otherwise, if the wireless station detects the medium was busy for the duration, the wireless station backs off for a multiple of time slots (SlotTime). The multiple is randomly chosen between [, 2 i CWmin] (i=, 1, 2,, m). Note that, CW min is called minimum Contention Window (CW) size, and it is set to the same value for all wireless stations. Meanwhile, if the wireless station transmitted a packet and received ACK frame correctly, then i is set to. If the wireless station failed to receive ACK frame, i is incremented by 1. Here, i can be up to m, therefore the maximum CW size is 2 m CW min. On the other hand, packet error means that the packet transmission failures between a pair of wireless stations are due to other than collisions. Thus when the receiver station detects an erroneous packet, automatically this packet is rejected. Accordingly, the sender station assumes that packet loss is an effect of collision and takes measures to avoid further collision in the network by delays the retransmission of the packet loss (i.e. increase its Contention Window). This is obviously sub-optimally: Contention Window should not be increased to avoid collisions when loss is due to noise. Therefore, it is necessary for the sender to distinguish between the two types of packets losses. Currently this limitation is similar to the Transport Control Protocol (TCP) problem over wireless networks [3]. It is well proven that TCP s performances degradation are related to the confusion happened at the sender when a packet is lost. In fact, the sender relies this loss to an increased congestion level, and consequently decreases the TCP window. By doing so, the sender decreases the TCP s data rate despite the loss is caused by a network error. Our main concern in this paper is to enhance the IEEE in noisy channel. To this end, we provide the wireless station means to differentiate between collision and noise loss. This is achieved by adapting the RTS/CTS (Request and Clear to Send) handshake mechanism to noisy channel environments. The main idea is, when a packet following an RTS/CTS exchange is not acknowledged, the sender station goes into retransmission routine of the loss packet, however, instead of increasing the CW, it maintains the current CW value and proceeds as normal. This paper is organized as follows: section II shows the impact of packet errors on IEEE performances. In section III, we give details of our proposition. We present performance evaluation in section IV, and conclude the paper in tion V. II. THE IMPACT OF ERRORS ON IEEE NETWORKS PERFORMANCES /5/$2. (c) 25 IEEE

2 A. Legacy IEEE MAC Protocols: The IEEE MAC defines two transmission modes for data packets: the Distributed Coordination Function (DCF) and the Point Coordination Function (PCF). The DCF provides contention-based access to the medium. The PCF is a polling scheme, allowing an access point to control all transmissions in order to provide contention-free access defines a repetition interval called a superframe, which is subdivided into contention-free (PCF) and contention (DCF) periods, as well as a beacon or management frame. Although the PCF was designed to support real time traffic, it has several drawbacks. It requires considerable control overhead, and has difficulty scaling to support large numbers of nodes. Additionally, due to its reliance on a central node for coordination, it does not adapt well to ad hoc configuration scenarios. As a result, the PCF is considered optional, and is rarely implemented in based devices. Immediate access when Medium is free for >= DIFS DIFS DIFS PIFS SIFS Busy Medium Defer Access Contention Window random from to CWmin Backoff-Window Slot time Figure 1. DCF access mechanism Next Frame As stated, the DCF utilizes a basic CSMA/CA approach to control medium access (Figure 1). Each station senses the medium prior to initiating a packet transmission. If the channel is sensed as idle, for a time interval greater than a Distributed Inter-Frame Space (DIFS), then the station can begin to transmit. Otherwise, the station is forced to defer, as a transmission would cause a collision with the transmission already using the channel. This deferral is controlled by the Backoff process. Each station maintains its contention window, a range of values from to the CW size. A Backoff interval is randomly selected on this interval, and the Backoff timer is initialized with the value. After each DIFS, the timer is decremented for each time slot that the medium remains idle, until the timer expires. B. Problem Analysis Thereby, the CSMA/CA algorithm increases the contention window each time a frame is lost, whether it results from noise or collision. However this procedure is not efficient when the frame losses are due to noise only. In this context it is important to study the CSMA/CA behavior in presence of noisy channel and compare these results to a perfect channel. Previous works have improved the well known Bianchi Markov model [4], aiming at analytically analysis the IEEE s behavior under noisy channel. In [5], the authors draw the saturation throughput in respect to the Packet Error Rate (PER) values. From the results obtained, it is obvious that performances in noisy channel are very poor, particularly when the PER is too high. Now in order to complete these analytic results, we conduct simple simulation through ns2 (Network Simulator)[7]. These simulations focus on the performance of single sender when the channel is both clear and noisy. Here by using one sender we are able to eliminate the effect of collision, and show only the effect of network errors. Further, the simulation consists of two Wireless Terminal WT1 and WT2. WT1 sends its packets to WT2 using constant bit rate (CBR) traffics over User Datagram Protocol (UDP). At first, we use a clear channel (i.e. PER =%). After that, we introduce noise to the channel by increasing the PER from % to 1% then to 3%. Note that channel throughput is 2 Mbps. TABLE.1 WT1 s CW distribution PER = % PER = 1 % PER = 3% CW = CW = CW = CW = CW = Table 1 shows the WT1 s CW distribution before and after increasing the PER by 1% and 3%. It is clearly seen that the CW s value does not exceed 31 when the channel is clear, meaning that all packets are sent correctly. This is expected since the network is both empty (P c = ), and operates in perfect conditions (P e = ). On the other hand, when we introduce noise in the channel, we see that the CW s value is highly increased, reaching 132 times and 5299 times the CW max (123), when the PER is equal to 1% and 3%, respectively. Altough no collisions can occur in the channel. We argue this by the wrong interpretation made by WT1 when no ACK is received. Indeed, WT1 interprets the packet loss as a sign of collision, and automatically increases the CW, despite the packet loss is due to errors. 18 x 15 WT1 Data rate Clear Channel Channel with PER=1% Channel with PER=3% 1- Figure 2. Contention Window transition Actually Figure 2 represents the CSMA/CA s Contention window transition, where a transition is triggered with a P f probability. It is worth mentioning that P f depends on the P e (Probability of error) and P c (Probability of collision) (see 1). Bits/sec P f = P + P P P (1) e c e f Figure 3. WT1 s Data rate

3 The direct consequence of this behavior is seen in Figure 3, which represents the WT1 s data rate. We observe large throughput difference between the clear channel and noisy channel. Here, the gain is constant and represents 1.2 Mbps and 1.8 Mbps, when the PER is equal to 1% and 3%, respectively. This is principally caused by both, the high packet loss rate resulting from the PER, and the fact that WT1 increases frequently the CW s value. Indeed by increasing the CW, WT1 waits additional time before retransmitting the erroneous packet, although the network is free. Accordingly, WT1 s data rate is highly decreased. Based on this simple simulation we illustrate the poor behaviors of the CSMA/CA algorithm in presence of transmission errors as well as the packet errors results over the global performances. Thus it is important to provide the sender stations ways to differentiate between network losses and collisions. III. ENHANCED RTS/CTS MECHANISM To improve the CSMA/CA performances in noisy environment, we propose to enhance the RTS/CTS mechanism. Usually, the RTS/CTS mechanism is used to solve the hidden terminal problem. When this mechanism is applied, the contention winner does not transmit the data immediately, instead, it sends an RTS frame to which the receiver answers with a CTS frame. This guarantees that all terminals in the range of either the sender or the receiver know that a packet will be transmitted. In this case, terminals remain silent during the entire transmission by initiating the NAV variable with the duration of the ongoing transmission, so only the sender is allowed to transmit frames. Consequently it is obvious that the RTS/CTS packets are the only packets which collide in the networks, meanwhile the data packets are apart of collision loss. ( ) L PER = 1 1 BER (2) where L is the packet length in bits On the other hand, if we consider Figure 4 that represents the PER of both RTS/CTS and Data packets in respect to equation (2), we notice that the probability that either RTS or CTS packets are erroneous is very negligible by report to Data packets, whatever the BER s value. This means that Data packets are highly affected by network error in contrast of RTS/CTS packets. Therefore, we can affirm that in case of noisy channel, the RTS/CTS packets are lost due to collision with others RTS/CTS packets while DATA packets are dropped due to errors introduced by the channel. PER(%) RTS (16 bits) CTS (112 bits) Data Packet (8 bits) Data Packet (16 bits) BER(1e x) Figure 4. PER versus BER and Packet length Based on these two important observations we propose to improve the RTS/CTS mechanism in order to differentiate between network loss and collision. Thus after RTS/CTS exchange, if no ACK is received for a data packet, indicating that packet loss is caused by network error (first observation). The sender station reacts by invoking the retransmission routine. However, instead of increasing the current CW value, we propose that sender station uses the same CW s value. Furthermore, if no CTS packet is received after sending RTS packet, the sender calls the RTS retransmission routine and increases the CW s value. In this situation, the RTS loss is due to collision (second observation). Therefore, the CW s value is increased only when collision occurs. Thus by avoiding the confusion between network increased load and bad wireless channel, the adaptation of RTS/CTS mechanism limits the reason of increasing CW s value only to collisions. In other word maintain P f always equal to P c. IV. PERFORMANCE EVALUATION In order to evaluate the advantage of the proposed scheme, we have constructed a simulation of the enhanced RTS/CTS protocol using ns-2 [8]. We compare Enhanced RTS/CTS with the classical RTS/CTS mechanism used by the CSMA/CA algorithm. Both mechanisms were implemented atop NS2 implementation of Orinoco 82.11b card [9] that uses more realistic physical channel. In fact through this implementation we are able to verify the proposed scheme s ability to adapt to the location s dependant characteristics as well as to the wireless channel fluctuation. On the other hand, to avoid the 82.11b anomalies [1], involved by the physical data rate adaptation, we propose rather to fix this data rate to 11Mbps. A. Simulation Model For the simulation, we have created a network based on an Access Point (AP) and different Wireless Terminals (WT i ). The physical channel is based on the Lucent Orinoco 82.11b propagation model (see table 2). Here we focus particularly on the Two Ray Ground. Actually this propagation model allows communication in open space with 16 meters of range. Thus higher is the distance from the AP, higher is the BER value. TABLE 2. Orinoco s propagation model Parameters Open Open + Movement Closed Propagation Model TwoRayGround Ricean Shadowing Range Further, each WT generates CBR traffic that represents 7.27% of the network load (Packet size = 2 Byte, Interval =.2 sec). B. Simulation Results Through the two figures (5, 6) we give the results after evaluating the global performance of the proposed scheme. In fact, Figure 5 represents the overall throughput by report to the network load when using our scheme and the classical RTS/CTS mechanisms. The PER is fixed at 3% throughout the simulations. We notice that the enhanced RTS/CTS outperforms the RTS/CTS mechanism at both network load. The goodput gain of the enhanced RTS/CTS reaches 32%,

4 when the load is about 13 % (18 WT). This roughly represents 4 kbps of realized throughput gain. This large throughput gain is principally due to the differentiation made between network losses and collisions when using Enhanced RTS/CTS. In parallel, in classical RTS/CTS, high error rate means an increased CW s value, which forces the wireless station to delay packet s retransmission involving a large number of idle slots in the channel that decreases naturally the channel utilization Enhanced Throughput PER=3% Figure 7 represents WT1 s data rate; clearly when WT1 is placed between 13 and 16 meters (PER = 2%), the enhanced RTS/CTS shows high performance than RTS/CTS. In this situation, the throughput gain represents roughly 1 kbps. This gain results from the differentiation made by WT1, between collisions and network errors. Moreover, when WT1 s distance from the AP is smaller than 12 meters, the two schemes achieve roughly the same performances Enhanced Load=22% Stations 6 Figure 5. Overall throughput versus network load Throughput load=1% Enhanced Meters Figure 7. WT1's data rate according to the distance from the AP Load=22% 1.4 Enhanced PER (%) Figure 6. Overall throughput versus PER In the other side Figure 6 shows the overall throughput of each mechanism in presence of different PER. Thus when the PER is high, both mechanism achieved throughput are mainly the same, the discrepancy is clearly apparent when the PER is lower than 3%. The weak gain observed when the PER is too high is mainly due to the packets dropped at the MAC layer when the maximum retry limit is reached. Here we must keep in mind that uses a retry counter in order to count the failing transmission s number for each packet at the MAC layer. As the PER is high, the packet retransmission number is rapidly reaching the maximum retry, conducting to drop the failing packets. Moreover through this figure it is clearly seen that our mechanism outperforms the classical RTS/CTS throughout the error rate values, which means that our proposal can adapt to the dynamic fluctuation of the wireless channel (dynamic change of the PER value). This last feature is also confirmed through Figures 7, 8, 9 and 11. In order to evaluate the performance of the two schemes with more realistic scenarios, we provide in Figure 8, 9, 1, 11, 12 and 13, simulation results using a dynamic channel characteristics. In these simulations we change the WT1 emplacement aiming at increase the BER at this station; meanwhile the others WTs operate with perfect channel (PER= %). Here the traffic load represents 22% of the network capacity Meters Figure 8. WT1's packet delay according to the distance from the AP In Figure 8 we show the WT1 s packet delays according to the distance from the AP. As for the throughput, the improvement is high when the WT1 is far from the AP (high PER). The Enhanced RTS/CTS greatly reduces delay to minimal levels, indicating that the packets are transmitted almost immediately. Actually, under Enhanced RTS/CTS, WT1 increases the CW only when collision occurs. Otherwise it transmits the packet without increasing the CW. In this last case, the Backoff timer is initialized with low values, leading to minimize the waiting time at the MAC level. In contrast, under classical RTS/CTS, WT1 increases the CW whatever the cause of loss. Hence, the CW reaches high value as illustrated in table1, which leads to delay the retransmission of the erroneous packet. Note that when WT1 s distance from the AP is larger than 16 meters, the delays are.1 sec and 1.32 sec in Enhanced RTS/CTS and classical RTS/CTS, respectively. Now we reorganize the network in order to increase the network load (to 54%), and show the impact of the two mechanisms over both WT1 placed at 16 meters from the access point and WT2 operating in perfect channel (BER=%). Besides illustrating the effectiveness of both mechanisms when high collisions are mixed with network errors, this reorganisation allows us to study the fairness of the Enhanced RTS/CTS, by showing that although WT1 increases

5 its data rate, the others stations operating in perfect channel do not suffer sec. The main reason of this, is actually the fact that within Enhanced RTS/CTS, WT1 differentiates between collision and network load. Moreover, given that WT1 is placed far from the AP, the majority of the packets losses are caused by the high BER. Thus under CSMA/CA the CW reaches high value leading to increase the queuing delay at the MAC layer. In contrast, in Enhanced RTS/CTS mechanism the CW is mainly low, which maintains the delays with minimum values x 1 3 Enhanced 2 1 Enhanced Figure 9. WT'1 data rate (Distance = 16 meters) Figure 9 illustrates WT1 s data rate achieved, when using the two schemes. The Enhanced RTS/CTS maintains a consistent delivery bit rate throughout the simulation time, despite the high BER perceived by WT1. When using Enhanced RTS/CTS, WT1 s mean data rate is 65 kbps. In contrast, with classical RTS/CTS, WT1 s mean data rate is 6kbps Enhanced Figure 1. WT2's data rate Through Figure 1, we plot the data rate obtained by a station near to the AP (WT2). Clearly, WT2 s data rate is the same in both Enhanced RTS/CTS and classical RTS/CTS. Although the Enhanced RTS/CTS improves the performance of station with poor channel conditions, it also maintains fair performances for the others stations that operate with good channel conditions Enhanced Figure 11. WT1's packet delays (Distance = 16 meters) Figure 11 gives the delays achieved by WT1. As expected, Enhanced RTS/CTS maintains lower delays (.12 sec) compared to RTS/CTS mechanism. Here the delays oscillate around Figure 12. WT2's packet delays At this point, the only weakness of the Enhanced RTS/CTS by report to the classical RTS/CTS mechanism is seen in Figure 12, which represents the WT2 s packet delays. In fact, the RTS/CTS mechanism outperforms the proposed scheme, however, the improvement is very negligible since we observe a mean gain of.1 sec. IV. CONCLUSION In this paper, we presented an enhanced RTS/CTS mechanism that improves the performances of the DCF mechanism under noisy channel. This mechanism provides means to the sender stations, in order to differentiate between collisions and network losses. Thus, the enhanced RTS/CTS reduces the number of events triggering the CW increase, which therefore reduces the average CW size, leading to better performances without increasing the collision rates. REFERENCES [1] IEEE, Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, IEEE Standard 82.11, [2] M. Gerla and P. Johansson, Bluetooth: technology, Applications, performance, in ACM Mobicom tutorial, July 21. [3] Y. Tian et al. TCP in Wireless Environments: Problems and solutions, IEEE Communication Magazine, Vol 43, No 3, March 25. [4] G.Bianchi Performance analysis of the IEEE distributed coordination function, IEEE Journal on selected Area In communications, Vol. 18. No. 3, March 2. [5] P. Chatzimisios et al. Performance Analysis of IEEE DCF in Presence of Transmission Errors, in proc. of IEEE ICC 4, IEEE International Communication Conference, Paris, France, Mai 4. [6] IEEE 82.11b-1999 Supplement to , Wireless LAN MAC and PHY specifications: Higher speed Physical Layer (PHY) extension in the 2.4 GHz band. [7] Network Simulator 2, [8] Network Simulator 2 with realistic channel operations, [9] Orinoco 11b client Pc specifications, [1] M. Heusse et al. Performance anomaly of 82.11b, in proc. of IEEE Infocom 3. San Francisco. USA, 3.