A Fuzzy Set Theory Based Method to Discover Transmissions in Wireless Personal Area Networks

Transcription

1 A Fuzzy Set Theory Based Method to Discover Transmissions in Wireless Personal Area Networks Lennart Isaksson 1, Markus Fiedler 1 and Elisabeth Rakus-Andersson Blekinge Institute of Technology School of Engineering Department of { 1 Telecommunication Systems, Mathematics and Science} Campus Gräsvik, SE Karlskrona, Sweden {lennart.isaksson,markus.fiedler,elisabeth.andersson}@bth.se Abstract The complexity within a Wireless Personal Area Network (WPAN) increases. Several technologies have to share the same radio spectrum. In this paper we take a look at the. GHz Industrial Scientific and Medical (ISM)-band. This paper discusses a method of selecting the best wireless channel within Wireless Local Area Network () when several technologies could be used in the same WPAN range of the needed access point. The issue is to keep away from already occupied channels. The method is divided into four steps: the passive probing of the power level detecting IEEE (ZigBee) channels using a new and affordable hardware, the transformation of the probed data to a linguistic level using Fuzzy Set Theory (FST), the classification of the data, and finally the sorting and selection of channels based on whose power levels. parameters and translate those values into an optimal decision. Every network selection decision is always dependent on the user s perception of the subjective application QoS. The way in which an application perceives the capacity of a certain network affects its performance and thus the users perception of the application itself. In order to yield a desired application performance, a certain applicationperceived throughput is required. In case of several available networks, a selection should be made in such a way that the desired performance is met. Traffic Situation Weighted Choice 1. Introduction Measurements Transformation The evolution of wireless networks and device technology regarding WPAN has posed the challenge to handle the trade off between mobility and performance. Users demand high performance quality and at the same time small and resource efficient devices, which have limited resources. The end user needs help to reach the goal of being Always Best Connected (ABC) or to be connected as good as possible. It is therefore essential that a service, chosen by the end user, provides the best-possible Quality of Service (QoS) regardless of the communication used. The specific problem is the dilemma of selecting the optimized choice of a communication technology together with a specific channel one or several channels for the chosen service when needed. Preferably, such a device should execute a seamless switch between two wireless technologies automatically. If no intelligence was implemented and a manual choice happened unsatisfying QoS might be the consequence. Because of this risk, a method is needed that takes care of couples of Preprocessing Fuzzy System Classification Figure 1. Decision model using Fuzzy Set Theory Sutinen et al. [1] describes the subjective application quality which combines the human perception and the objective application quality throughout the whole system. The objective quality consists of four components, Network QoS is one of them. Each network has its own capacity and constrains, which is to be considered when choosing the ABC network. The Network QoS is affected by disturbance

2 on network level, e.g. the existence of other transmissions on the desired channels. Therefore, it is vital for providing ABC to get a clear picture on ongoing transmission. The decision of doing a switch between two or several wireless / non-wireless technologies and to maintain the goal of being ABC tends to be complex. The decision depends upon a whole range of steps, see Fig. 1. First, the traffic situation must be measured. Second, those values are grouped and classified using FST. Finally, the transformation of those linguistic levels are later feed into a Multi- Criteria Decision Making (MCDM) model. The remainder of the paper is organized as follows. Section describes related work and research within Wireless Personal Area Network 0.15.x. In Section 3 the measurement probing is described. Section describes elements of Fuzzy Set Theory. Results are then presented in Section 5. Finally in Section conclusions are given.. Related work for WPAN Several standards are available for Institute of Electrical and Electronics Engineers (IEEE) 0.15.x WPAN; (1) IEEE Bluetooth, () IEEE coexistence of and WPAN, (3) IEEE UWB high rate, () IEEE a Wireless USB, and (5) IEEE ZigBee. All these standards have the ISM-band (.00.5 GHz) in common. An additional non standard interfering technique needs to be considered, which is the microwave oven Bluetooth ZigBee 0.11b America 0.11b Europe.0 GHz.0 GHz Channel Figure., Bluetooth and ZigBee channels in the ISM-band 1. Bluetooth uses Frequency Hopping Spread Spectrum (FHSS), which means that several channels are used during a transmission. It uses 79 channels with a bandwidth of 1 MHz without any space between the channels, see Fig.. Medical monitoring is one of important issues that takes Bluetooth to its boundary []. The bit rate of the Bluetooth technique version 1.1 and lower, is at maximum of 73 kbps. For version 1. and.0 the highest bit rate is.1 Mbps.. The coexistence of and WPAN is essential, and a vast amount of research is undrtaken, e.g. [3,,5,]. All technology of today has to work side-by-side, which is not the case in some situations according to Vendictis et al. [7]. The specification mentions 1 channels, of which 13 are inside the ISM-band using the Direct Sequence Spread Spectrum (DSSS) or the FHSS technique. These channels overlap with each other and therefore only few of the channels are usable, see Fig.. The bit rate of the (0.11b) technique is between Mbps and Mbps. 3. Instead of using specific channels, the Ultra Wideband (UWB) uses a very broad area of the spectrum between 3.1. GHz with a bandwidth of 00 Mbps which is outside our interest. According to Hamalainen et al. [] this is an issue to be considered.. Wireless Universal Serial Bus (WUSB) uses DSSS in the same place as Bluetooth s 79 channels together with Code Division Multiplexing Access (CDMA) spreading sequences called Gold codes [9]. 5. ZigBee uses several free bands like the MHz band with one channel (number 0), 915 MHz band with channels (number 1 ), and the. GHz ISM-band with the DSSS technique. In the. GHz area ZigBee uses 1 channels (number 11 ) with non overlapping channels, see Fig.. The bit rates of the ZigBee technique are 0 kbps, 0 kbps and 50 kbps. 3. Probing The measurements are based on an affordable hardware from Freescale TM MC GHz DSSS radio frequency transceiver and Freescale TM MC9S0GT0 microcontroller using IEEE 0.15., ZigBee TM. The probing technique is passive and non-interfering. A dedicated Zig- Bee device is probing all the 1 channels in the. GHz ISM-band very close to the source device. The measurements is executed by sending a command in hex code: 0x a, the response is retrieved and the power levels of each 1 channels are translated into dbm or in milliwatt, P/dBm = log(p/mw). This is executed 0 times per seconds. For each of the 1 channels the highest value is depicted and saved in a log file together with the date and time. All samples are analyzed off-line. This is the initial step of the knowledge-based classification using FST, see Fig Traffic Generator To generate traffic a device from Lucent Technologies 1 was used. During the measurements an arbitrary file was transmitted over the air using File Trans- 1

3 port Protocol (FTP). The nominal output power for the ORINOCO AP-500 is 15 dbm. For ZigBee, a packet generator was constructed to simulate traffic between two devices. The default output power of 7. dbm was used. Finally, the Bluetooth device uses the power class which is.5 mw (0 dbm) with the built in antenna. 3.. Channel characteristics Before we go any further a description of each technology is needed. Each technology has its own characteristics. uses a bandwidth of MHz with a channel spacing of 5 MHz (peak-to-peak) and covers ZigBee channels as well as Bluetooth channels. To test the behavior of a channel, in this case channel, an access point was set up. The minimum and maximum power level in dbm is presented on the right side of the figure, see Fig. 3. The traffic is scattered all over the ZigBee channel 5,, 7 and during samples 1 to 7. This behavior is natural because the access point broadcasts information periodically to other devices during this time. From samples to 1 a distinct pattern is shown, the channel character is like a lobe, higher in the middle and lower at both sides. This is the transmission behavior of 37 dbm with a relatively high transmission power. The power level are between 37 dbm and 95 dbm. Bluetooth has only a bandwidth of 1 MHz with a channel spacing of 1 MHz (peak-to-peak) and is uniquely using FHSS, in this case no space between the channels. This characteristic is reflecting the behavior of using both Time Division Multiplexing (TDM) and Frequency Division Multiplexing (FDM), see Fig.. The figure shows initially no traffic in any channel from samples 1 to. From samples 7 to 1 the figure shows a random transmission behavior at 57 dbm. The power level are between dbm and 9 dbm. ZigBee itself has a small bandwidth of only 3 MHz which covers 3 Bluetooth channels with a channel spacing of 5 MHz (peak-to-peak) i.e., number 7, see Fig. 5. The figure shows initially no traffic in any channel from samples 1 to. From samples 9 to 1 the ZigBee channel shows a transmission behavior at dbm with a relatively moderate transmission power. The power level are between dbm and 9 dbm. A standard microwave oven with 900 W and with the probe in front of the oven shows a similar behavior as in the Bluetooth case. The deviations between the microwave oven and Bluetooth are as follows: the power is more concentrated at the beginning of the. GHz ISM-band, and the power level of the microwave oven is much higher, see Fig.. Between samples 1 and 7 the power is insignificant. From sample 9 the power is a lot higher as compared to ORINOCO PCE-H-FC with a maximal output power of 15 dbm. other technologies. The power level are between 1 dbm and 9 dbm Figure 3. Power level scenario with contour lines in dbm showing a channel transmission Bluetooth Figure. Power level scenario with contour lines in dbm showing a Bluetooth transmission. Fuzzy set theory First we give a description of Fuzzy Set Theory, which is a part of the decision model, is given. The fuzzy inference method consist of mainly three parts, see Fig. 7, Zimmermann []. First, we fuzzify the model by assigning to values of real input sets by assigning the membership degrees

4 1 1 ZigBee 5 rule. The output is later matched toward a technology i.e. x {, microwave oven, Bluetooth, ZigBee} with a classification degree. To obtain the aggregated output we apply a formula α r f r (x 1,x ) u Sugeno r =. (1) α r r 75 0 Defuzzyfication Data Fuzzification Inference Matching Classes Figure 5. Power level scenario with contour lines in dbm showing a ZigBee channel transmission Linguistic Approximation Figure 7. Knowledge-based classification 1 1 Microwave oven 0.1. Parameter characteristics Figure. Power level scenario with contour lines in dbm showing a Microwave oven behavior Two types of parameters are used: x 1 = standard deviation, see Fig., and x = power level, see Fig. 9, in dbm. First the power level are measured for each technology which are as follow: ZigBee = 79 dbm, Bluetooth = 71 dbm, = 5 dbm and microwave oven = 3 dbm. Second, the standard deviation value is depicted to match each technology the behavior is analyzed by using standard deviation window of five samples delta value and a moving window of one sample i. N is calculated by the size of the vector minus delta and plus one according to: Size(vector) delta +1. The equation N i=1 Std(P (i, i + delta 1)) is used to calculate the standard deviations which are as follow: = dbm, ZigBee = 0.50 dbm, microwave oven = 1.0 dbm and Bluetooth = 1.70 dbm... Cut points Second, we create fuzzy decision rules, called inference rules. These, via appropriate operations, evaluate a common impact of input values on a single decision. Finally, we collect single decisions generated by inference rules to make a final output decision. The output could also be translated into a linguistic level. The output method could be chosen between the Mamdani method [11] or the Sugeno method [, 1]. In this paper the zero-order Sugeno model is depicted, constant output together with weighted average (1). The output level i = f(x 1,x ), i 0,, of each rule is weighted by the strength α r = min(µ(x 1 ),µ(x )) of the Each technology has its unique characters of standard deviation and power level. The arithmetic mean µ = 1 n n i=1 x i was used to calculate the cut point between the two adjacent membership functions which occur at cross point level of µ(x) =0.5. The method used is called the condition width which is used to find our cut points for each membership function. With this method the total value of the membership function is always µ(x) =1. Another advantage is that all changes in the fuzzy control is handled smoothly.

5 Degree of Membership Low Middle High Standard deviation Figure. Membership functions of three fuzzy sets characterizing the intensity of standard deviation Degree of Membership Low Middle High Power level Figure 9. Membership functions of three fuzzy sets characterizing the intensity of power level.3. Rules Each rule is connected to a group of membership functions (mf) [] α r = min j=1,...,n {µ ji i (xinput i )}. In this paper three levels are selected as follows: low, middle and high. Together they are formed by the AND operator IF(Std. supp (µ mf (x 1 ))) AND (P owerlevel supp (µ mf (x ))) THEN Output = f(µ mf (x 1 ),µ mf (x ))... Defuzzyfication As mentioned before, the Sugeno model is depicted using singleton sets with the linguistic output levels of low, middle and high. Thelow value is set to 0, middle is set to and high is set to. If the value is lower than the classification of the technology is low. If the value is equal to the classification of the technology is middle. Ifthevalue is higher than the classification of the technology is high. The output scale, which uses the singleton sets, is simpler to handle compared to other methods like the min-max or the center of gravity method suggested by Mamdani in the case of transforming the linguistic output to a fixed x-point scale. That s why the minimum boundary is set to represent the value of zero and the maximum boundary is set to represent the value of eight which transform us to the a 9-point scale..5. An example The crisp control action, suggested by Sugeno, is based on the weighted average (1) which is called the singleton set. In our example the low boundary is placed at position zero, the middle singleton at position four and finally the high boundary at position eight, which gives us a 9-point scale. Now, each membership degree must be extracted on the basis of inference rules for the standard deviation and power level. If the standard deviation value is 0. dbm and the power level is 7 dbm the extracted membership degree for rule number one: IF std. supp (µ middle (x 1 )) AND power supp (µ middle (x )) THEN output = µ high (f( )) is α 1 = min(0.730, ) = 0.730, and for rule number nine: IF std. supp (µ high (x 1 )) AND power supp (µ middle (x )) THEN output = µ low (f( )) is α 9 = min(0.370, ) = The weighted average, according to equation (1), is later used to calculate the output level as follow: (0.730 ) ( ) = This means that we have a technology present, which must be considered with a degree of 5.3 between 0 and on a 9-point scale... Matching Another example of matching and classification is illustrated in Fig.. From samples 0 to 17 no indication of any technology is detected. After sample 17 the technology is over level and no indication detected as one possible technology used at the moment. Still nothing else is detected. Now, the matching and classification is established. If is the target and if the output membership value is higher than, the next step is to sort all channels beginning

6 Level Microwave oven Bluetooth ZigBee Figure. WPAN matching and classification with the lowest power level first. Choose the first channel in the sorted list that is recommended by the specification and we have finally a channel which is most likely to be free without any disturbance. 5. Results In this paper we showed that it is very cheap and simple to measure a wireless network traffic close in range nowadays. In many cases the different types could also be classified. We also showed that the Sugeno model is highly usable to transform different metric values to a distinct linguistic level. As described in the introduction this type of transformation is important when a decision is to be made regarding choosing the Always Best Connected network.. Conclusions We have discussed an approach to classify specified technologies within a Wireless Personal Area Network. The classification is derived by using Fuzzy Set Theory. Two types of parameters are used: standard deviation of the power level and power level itself in dbm. A new and inexpensive hardware is used to retrieve the power level in range. The approach of using Fuzzy Set Theory to translate metric measurements into different linguistic level is promising. This classification of linguistic level could later be fed into a more complex decision making model. References [1] T. Sutinen and T. Ojala. Case Study in Assessing Subjective QoS of a Mobile Multimedia Web Service in a Real Multi-access Network. In Thirteenth International Workshop on Quality of Service, Passau, Germany, 005. [] M.J. Morón, E. Casilari, R. Luque, and J.A. Gázquez. A Wireless Monitoring System for Pulse-oximetry Sensors. In Proceedings of the Advanced Industrial Conference on Wireless Technologies, pages , 1-17 Aug., 005. [3] Khaira and S. Manpreet. Bluetooth can coexist with In Electronic Engineering Times, pages 90 9, 001. [] A. Contiand, D. Dardariand, G. Pasolini, and O. Andrisano. Bluetooth and IEEE 0.11b coexistence: analytical performance evaluation in fading channels. In IEEE Journal on Selected Areas in Communications, volume 1(), pages 59 9, 003. [5] N. Golmie, N. Chevrollier, and O. Rebala. The evolution of wireless LANs and PANs - Bluetooth and coexistence: challenges and solutions. In IEEE Personal Communications, volume (), pages 9, 003. [] J. Lansford, A. Stephens, and R. Nevo. Wi-Fi (0.11b) and Bluetooth: enabling coexistence. In IEEE Network, volume 15(5), pages 0 7, 001. [7] A.D. Vendictis, F. Vacirca, and A. Baiocchi. Experimental Analysis of TCP and UDP Traffic Performance over Infra-structured 0.11b s. COST 79, Technical Document 79 TD(0)033, 11 th Management Committee Meeting, Ghent, Belgium, 00. [] M. Hamalainen, J. Saloranta, T. Patana, J.-P. Makela, and I. Oppermann. Ultra Wideband Signal Impact on IEEE 0.11b and Bluetooth Performances. In IEEE Proceedings on Personal, Indoor and Mobile Radio Communications, volume, pages 93 97, 003. [9] R. Gold. Optimal Binary Sequence for Spread Spectrum Multiplexing. In Proceedings of IEEE Transaction on Information Theory, volume 13(), pages 19 1, Oct [] H.-J. Zimmermann. Fuzzy Set Theory and Its Applications, fourth edition. Kluwer Academic Publishers, 001. ISBN [11] E.H. Mamdani. Twenty years of fuzzy control: experiences gained and lessons learnt. In Second IEEE International Conference on Fuzzy Systems, volume 1, pages 339 3, March-1 April [1] M. Sugeno. Industrial applications of fuzzy control. In Elsevier Science Pub. Co., 195.