Effect of Flat Fading in 802.11 MAC for Cross Layer Evaluation Using Channel Emulator Adriano Almeida Goes [email protected]

Omar Carvalho Branquinho [email protected]

Norma Reggiani [email protected]

Pontifícia Universidade Católica de Campinas Centro de Ciências Exatas, Ambientais e de Tecnologias Faculdade de Engenharia Elétrica Rodovia Dom Pedro I km 136 – Campinas – Brazil ABSTRACT WLAN 802.11 operating in 2,4 GHz are being intensely implanted mainly in public environments. In these environments mobility is always a present characteristic. The objective of this work is to analyze the behavior of 802.11 MAC with flat fading. To reach this objective an emulation system was developed to create the flat fading channel phenomenon. With a WLAN and the emulator was possible to show the cross layer effect, evaluating the WLAN performance analyzes in mobility conditions. The experiments evaluated: rate of transmission in each modulation level, data rate in the transport layer, network efficiency, jitter and throughput. The network information was obtained through SNMP. The results show that the 802.11 MAC had problems to support services that need high performance, as VoIP and Streaming. Therefore when the receiver is in movement the degradation does not allow the quality required by these types of services. Keywords  PWLAN, WLAN, Rayleigh, Weibull, Flat fading, wireless, MAC, IEEE802.11. I. INTRODUCTION Each time more 802.11 WLAN (Wireless Local Area Network) [1], are being used in the market in wide scale. The performance depends on the position of the user and its WLAN used in public environment (PWLAN) mobility. can be had accessed by diverse types of equipment, as for example, cellular, palm tops, PDAs, laptops, among others portable and mobile devices. The 802.11 MAC has an anomaly effect that degraded the performance [2][3]. This type of network supports well services that do not need high performance such as, email, HTTP, ftp, among others. However for services like video-on-demand, VoIP, Streaming, among others, that require high availability and functionality, there are serious doubts if the WLAN performance will be good enough. In this work it is studied

the efficiency of the net and the effect of cross layer through the emulation of the flat fading channel. For mobile applications the attention must be in the environment conditions, once there are diverse phenomena that degrade the WLAN efficiency significantly, for example, multi path. To analyze WLAN operating in these conditions it is necessary a tool to reproduce these phenomena and take into account the requirements of this type of application. This article presents results of WLAN performance using a work bench to emulate signal strength variation and evaluate the efficiency with Weibull distribution [4] that takes in consideration some factors normally not considered in other models for indoors propagation. The radio propagation tests to evaluate performance always were a laborious task requiring sophisticated equipment. The currently WLAN possess ways of measurement of the signal intensity, signal-noise ratio, number of packets in each modulation level, among others, through the protocol of management of net SNMP (Simple Network Management Protocol). The objective of this work is to emulate the flat fading phenomenon through a work bench and evaluate the WLAN performance with different Weibull distribution factor, to determine the cross layer effect. The work is organized in the following form: in section II presents the flat fading emulation process in 802.11 WLAN operating in 2,4 GHz [5,6,7]. In section III, the software and the hardware used are described. Section IV presents the results network efficiency, as well as an analysis of the obtained results. Section V presents a preliminary test in VoIP. Finally, section VI presents the conclusion. II. EMULATION OF FLAT FADING To control the flat fading channel emulation was developed a work bench with software (called SCLan5) and a radio frequency hardware to emulate Weibull distribution, to measure the network efficiency, to analyze the network behavior, among others [6].

Before initiating any measurement it is necessary to calibrate the bench work. This process consists basically of sending definitive levels of attenuation to the control block and observing the power of reception. However, this relation is not enough to perform the calibration, because that the attenuation level that must be characterized by the distribution is the effective normalized tension, through the Weibull distribution. The expression below was used to equate these two values of power:

Veffective =

Pi Pm

(1)

Where, Pi is the received power and Pm is the logarithmic average power of all received powers, both the two express in Watts. After concluding the calibration process, the algorithm that emulates the flat fading channel on the transmitted signal must be initiated, following the parameters described in Table I. TABLE I. PARAMETERS OF EXECUTION OF THE FLAT- FADING Parameters Dref VE DP Vrms

α β

Description Referential Distance Speed of Execution Shunting line Standard Tension RMS Path Loss Environment Coefficient

Unit meters m/s meters Volts Dimensionless Dimensionless

The Rayleigh distribution is a particular case of the Weibull distribution, if we consider in the last one α = 2, where sigma is the variance of a gaussian variate [4]. So, the beta parameter of the Weibull distribution is associated with the variance of the variable considered. The Weibull distribution is used to permit a grade of flexibility to emulate de environment. This assumption is reasonable considering the multi path and differences of obstacles in the indoor environment. With the parameter β it is possible to change from a very stable scenario to a severe one. This flexibility is very interesting to evaluate the cross layer effect. Through the described parameters in Table I, the software generates a Weibull distribution and relates the obtained values with the calibration table obtained in the beginning of the process and stored it in a data base that contains the effective tension. Based on this scenario, the software sends the necessary parameters so that the control block attenuates the RF signal following the characteristics of the Weibull distribution, with a definite environment parameter β . The Figure 1 shows a graphic obtained after sending the values of Weibull distribution for different values of β . The results for β =0,2, β =0,5 and β = 1, are showed by the grey, blue and red lines respectively.

The signal that arrives at the receiver is a result of two effects: one is the attenuation suffered by the transmitted signal along its way to the receiver, and another is the fact that the signal that arrives at the receiver is a result of multipath propagation. In order to describe the attenuation of the signal due to the caracteristics of the environment, we used the shadowing model given by [7]: Figure 1. Weibull Distribution generated for software (SCLan5).

 Pr (d )   d   = −10α log  Pr (d 0 )  dB  d0

  + X b 

(2)

where, Pr(d) is the surveyed power, Pr(d0) is the reference power, α is the path loss factor, d is the distance measured, d0 is the distance of reference and Xb is a log-normal distribution. In this work we do not use the shadowing model and concentrate the tests in the signal strength changing with Weibull distribution. This is equivalent a device performing a circle around the access point with fixed distance with velocity that can be changed. In order to describe the effect of summing the various waves that come from the multipath propagation of the transmitted wave, we used the Weibull distribution, given by: α

x −  α −1  β 

f ( x ) = αβ −α ⋅ x e

(3)

In the Figure 2 it can be verified the attenuation of the transmitted wave described by the Equation (2) with different distances an values of alpha (associated with different environments), with the effect of flat fading described by the Weibull distribution with β = 0,5:

Figure 2. Distribution of Shadowing, with values of β (1,6;2;2,5,3 e 4).

Finally, it is possible to relate the effects caused by the Weibull distribution with the efficiency of measured net with protocol SNMP. In order to arrive in network efficiency it was developed an expression that takes into account the frames transmitted considering the rate used in the physical layer by each frame. The result is a percentile evaluation to monitor in real time the performance. The expression of the effective rate is given by: n

Refetiva [%] = ∑ i =1

NFi * Pri NFT

but also to treat the collected information obtained by the SNMP protocol and to execute the algorithms for generation of the considered effect, between them the Weibull distribution. The Figure 5 shows the bench work mounted for the emulation of the flat fading channel in order to study the effect cross layer caused by the physical layer in the superior layers.

(4)

Where, NFi is the numbers of transmitted frames with rate i, NFT is a total number of transmitted frames in a period of time and Pri is the percent of the data rates compared with the maximum rate. III. TEST SYSTEM In the process of emulation of the flat fading channel, the components shown in the Figure 3 had been used. Sniffer (MAC)

Sniffer (TCP)

Shildet Box Attenuator Capture

Coaxial cable

WNIC AP

Hardware Control

PC2

LPT

PC1

Control Terminal

Figure 5. Organized system to emulate the flat fading channel.

In the Figure 5 is possible to identify in the right the shielded box where the AP is conditioned. Besides it there are the circuit of control and the changeable attenuator for tension. The laptop works as a sniffer capturing packets TCP/UDP directly in the RF signal. To facilitate the visualization, the Figure 6 shows the shielded box open with the AP connected to a pig tail and the N connector with the coaxial cable that establishes connection with the mechanism of attenuation and control of the signal.

Figure 3. Components used in the emulation of flat fading.

The AP controlled by the PC1 (microcomputer) and confined in an armored box transmits a signal through a coaxial cable. This signal is received by attenuator (AT) that controls its power. The Div block is a RF splitter to permit the sniffer to work directly in the RF signal. To manage this attenuator, a control circuit (control block) was set up, that consists of a digital-to-analogical converter and a circuit that supplies an adjustment of the gain and Offset of the involved signal as showed in the Figure 4.

Figure 4. Control block.

This circuit is controlled by software located in the “Terminal of Control” that supplies a tension to the attenuator. Finally, the PC 2, through a plate WNIC (Wireless Network Interface Card) PCMCIA receives this controlled signal for the attenuator. The software SCLan5 used to the control and management of the system was developed not only to control the tension,

Figure 6. Visualization of the emulation system.

In the Figure 6 it can be verified that the sniffer of the MAC is directly connected with the box of control, that allows to capture the packets of the transport layer (TCP/UDP) and to analyze with software as Ethereal, Fluke (software for VoIP), among others.

IV. RESULTS After the emulation of the Weibull distribution for some values of β as it was shown in Figure 1, graphical of jitter, throughput, SNR, rate of transmission UDP and efficiency of net had been captured. Before, during and after the execution of the algorithm we observed the power of the signal received in Netstumbler software, presented in the Figure 7. To facilitate the visualization of the presented phenomena, each resultant graph will be divided in phases and numbered. 1

2

3

At the same time, through protocol SNMP, the SCLan5 software measures the SNR, signal and noise power, to relate them later to the effect of cross layer. These three graphs are shown in Figure 8. In Figure 8, the power of the noise remains unchanged during all the process, but the power of the signal and consequently the SNR had been degraded significantly during the emulation of the flat fading channel, what influences directly the QoS of the application (VoIP or Streaming, for example). In Figure 9, it is observed that at moments 1 and 3 while the Weibull distribution is not on, the net efficiency is 100%. But when the signal is submitted to the flat fading channel, a high degradation in the net efficiency is observed (equation 3), presented at moment 2. Moreover, it is noticed that there were attempts carried through by the MAC protocol in searching greater transmission rates, however without keeping stability between the possible levels of transmission.

Figure 7. Visualization of power of the signal received for the Netstumbler.

At moment 1, the key of the controlling block is opened without any effect of attenuation on the signal, characterizing a communication without movement. At moment 2, the execution of the Weibull distribution is started. The variation of the signal is clearly observed in this moment. Finally, at moment 3, the similar configuration of the moment 1 is reestablished.

Figure 8. Graphs of the physical layer: Powers of signal, noise and SNR.

Figure 9. Graph of the efficiency for a net 802.11b.

During the execution presented in Figure 9, it was possible to identify a gradual reduction in the net efficiency. This happens because the Access Point looks for lower rates in order to stabilize the transmission. This induces the transmission to become unstable and influences the superior layers degrading and many times directly disabling services as VoIP and Streaming. Another important result obtained by the process is net throughput. The Figure 10 shows a great variation of throughput when the channel is flat fading. In this figure, β was equal to 1,5.

In the Figure 12 the rate variation in a station 802.11b without movement is presented, that is, with a channel that is not flat fading. In this situation, the logarithmic average between rates is 786,04 KB/s and the standard deviation is 78,45 KB/s. However, when the channel is flat fading there is a great variation in the rate, as shown in the Figure 13. Reception rate in flat fading channel 1200 1100 1000

991 909

Rate (KB/s)

900

944

918

910

862

800

794

769

914

910

888

729

700

621

600 518

500

487

400

395

300

285 238

200

Figure 10. Variation of throughput for the flat fading channel.

173

148

118

100

118

115

101

142 89

150 107 51

71

55

0 0,0

Already when the channel is not flat fading the throughput practically remains unchanged, as shown in Figure 11.

5,0

10,0

15,0

20,0

25,0

Time (min)

30,0

Figure 13. UDP transmission rate for 802.11b net with flat fading channel.

With the emulation of the flat fading channel a great variability in the received rate can be verified. With this a station 802.11 that it is in movement, can influence all the others stations in the net because of its constant instability. In this case, the logarithmic average of the received rate decreased to 468,62 KB/s and the standard deviation increased to 359,52KB/s, showing a great degradation in the service. Another important information is the jitter, that determines the viability of transmissions UDP with a minimum of QoS. In other words, it is the essential information to determine the delay held in a VoiP connection in a flat fading channel. With the same captured packets by the sniffer the graph of Figure 14 was generated, which shows the behavior of the jitter in a flat fading channel. Flat Fading channel - Jitter

Figure 11. Variation of trhoughtput for the channel without any movement.

650 600 550 500 450

Jitter (ms)

With the same parameters of emulation UDP traffic was created and these packets had been captured by the sniffer using the Ethereal software. Through the analysis of these packets, it was possible to determine the variation in the UDP transmission rate for the normal channel and flat fading. The Figure 12 shows the variation of the UDP transmission rate captured in the channel without mobility.

400 350 300 250 200 150 100 50

Reception rate in channel without flat fading

0 0

Rate (KB/s)

1000

800

600

956 847852 755

755768774 725 713

892 841

10000

15000

20000

25000

30000

35000

723736

703

Figure. 14. Jitter for net 802.11b with flat fading channel.

947 883874

782

797 750 693

864 816 715

762 723 721

819 773

622

400

200

0 0,0

5000

Packages sequences

1200

5,0

10,0

15,0

20,0

25,0

30,0

Time (min)

Figure 12. UDP transmission rate for 802.11b net with normal channel.

35,0

The logarithmic average of jitter joined in this situation was 49,11ms and its shunting line standard was 151ms. With the jitter extracted from station 802.11b during the emulation of flat fading channel, it can be verified that there is a more variability and delay between the packets in this case than the jitter extracted from a station without movement. It is shown in the Figure 15.

independently of the power of received signal. In the Figure 17 it is possible to analyze the quality of the call presenting the coefficient MOS LQ.

Channel without flat fading - Jitter 65 60 55 50

Jitter (ms)

45 40 35 30 25 20 15 10 5 0 500

5500

10500

15500

20500

25500

30500

Packages sequences

Figure 15. Jitter for net 802.11b with channel not flat fading.

As shown in Figure 15, in a station without movement, the jitter, besides being small, didn’t have a variability so accented as the station submitted to the flat fading channel. In this situation, the jitter was stable with a shunting line standard of 4ms and a logarithmic average of 0,39ms.

Figure. 17. Quality of the call graph MOS LQ obtain with software Fluke before and during the emulation with a β equal 1,5.

An acceptable value for this coefficient MOS LQ is between 3,5 and 4. With the emulation of flat fading channel, the quality of call is bad because MOS LQ reaches values below 3,5. This shows that a linking in these conditions would not be possible. Finally, the Figure 18 shows the factor R. This factor defines the viability of to have or no a VoIP call.

V. RESULTS ON VOIP PLATFORM

As an extension of this work, services of VoIP had been tested on the basis of a coder of available voice in the market and some coefficients of environment. When submitted to the flat fading channel, the results about quality of the call, quality of net, jitter, loss of packets and discarded packets, could have been compared with the variation of the relation signal-noise and the variability of the system when it is in movement. The Figure 16 shows three graphs for the drawn up QoS of VoIP with two different moments: before and during the emulation of the flat fading channel. These tests had been executed under a platform 802.11b and a environment coefficient equal 1,5.

Figure 18. Efficiency of net with controlled attenuation.

This graph was generated from Fluke software and shows that in the conditions presented and mainly with a value of β equal 1.5, the degradation of the quality in the call is enough to interrupt it and to make implantable the conversation. Beyond the “cuts in the voice” was identified a great delay and deformity in the transmitted sound. VI. CONCLUSION

Figure 16. Jitter graphs, discarded packets and lost packets obtained from software Fluke before and during the emulation of flat fading channel.

Through these graphs, it is possible to notice the degradation in the parameters when the channel is flat fading, what will influence the QoS of VoIP. Depending on the environment this it can make the service unfeasible

In this work we presented the results of performance evaluation of 802.11 WLAN through a work bench developed specially to test the behavior of the MAC in different environments to analyze the cross layer effect. To realize this test it was developed an emulator flat fading. The developed system is all automatic allowing any type of simulation and test with flat fading, emulating the movement of users of a WLAN, predominant characteristic found in PWLAN. Through the results it was verified that there is a reaction of 802.11 MAC to the variation of the signal and does not present stability when submitted the wireless device to move in high speed, that is, when a mobile receiver is in movement the network efficiency presented was inadequate for applications with high rates and real time requirements.

Another interesting point is that the network efficiency is better when the receiver is stopped than when it is in movement, even if the received power in the stopped receiver was smaller than the average power received in movement. A type of timeout was identified on the stabilization process of the signal in the sub layer MAC when it is submitted to the high variance, or either, high mobility. During the execution of the tests it was identified a low effectiveness in working with WLAN in environments with other systems and with mobility. This work will have its evolution in the direction to evaluate the parameters necessary for identification of the points of deficiency in many services with high performance and to consider solutions to WLANs that work with this type of service. It will also be used to assist implantation of PWLAN standard 802.11 in public environments with high mobility. This paper is not important because the results found in 802.11 WLAN, although this work shows a important schema to build a hardware for emulate flat fading effects in channel in wireless networks. VII. ACKNOWLEDGEMENT The authors are thankful the support of laboratory WCN (Wireless Competence Network) of the INTEL in the Institute of Computation of UNICAMP. REFERENCES [1] IEEE Std 802.11 Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications. ANSI/IEEE Std 802.11, Information technology , 1999 Edition. [2] Heusse, Martin e Rousseau, Franck; Berger-Sabbatel, Gilles; Duda, Andrzej. Performance Anomaly of 802.11b, IEEE INFOCOM 2003. [3] O. C. Branquinho, N. Reggiani, C. E. Correa, and D. M. Ferreira, “Mitigating 802.11 MAC Anomaly Using SNR To Control Backoff Contention Window”, in Proc. of ICWMC2006, Bucharest, Romania, 29-31, Jul., 2006. p. 55 [4] LAW, Averill M. Kelton, W. David. Simulation Modeling & Analysis. 2nd edition, McGraw-Hill, 1001 [5] Crow, Brian e Widjaja, Indra; Prescott, Sakai. IEEE802.11 “Wireless Local Area Networks”, IEEE Communications Magazine, September 1997. [6] Atheros Communications, Inc “Methodology for Testing Wireless LAN Performance”, 2003. [7] T.K. Sarkar et al, “”A survey of Various Propagation Models for Mobile Communications””, IEEE Antennas and Propagation Magazine, vol.45, n. 3 pag. 51 (2003)YACOUB, Michel D. Foundations of Mobile Radio Engineering. CRC Press. 1993. [8] IEEE Std 802.11b-1999. Supplement to ANSI/IEEE Std 802.11, 1999 Edition [9] Sklar, Rayleigh Fading Channels in Mobile Digital Communication System. Part I: Characterization, IEEE Communications Magazine, pag. 90, July (1997). [10] J. Padhye, V. Firoiu, D. Towsley, J. Kurose: Modeling TCP Throughput: A Simple Model and its Empirical

Validation. Proceedings of SIGCOMM’98, Vancouver, Canada, 1996. [11] Intersil Corporation: Wireless LAN Integrated Medium Access Controller with Baseband Processor. Data Sheet of the ISL3873, File Number 8015.2, USA, 2001. [12] Y. C. Tay, K. C. Chua: A Capacity Analysis for the IEEE 802.11 MAC Protocol. Wireless Networks, 7, 2001.