21st Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communiations 1 RAC 2 E: Novel Rendezvous Protool for Asynhronous Cognitive Radios in Cooperative Environments Valentina Pavlovska, Daniel Denkovski, Vladimir Atanasovski and Liljana Gavrilovska Faulty of Eletrial Engineering and Information Tehnologies Skopje {valenpav; danield; vladimir; liljana}@feit.ukim.edu.mk Abstrat This paper introdues a novel rendezvous protool (RAC 2 E) for dynami ontrol hannel provisioning among ognitive radios in a deentralized and ooperative ad-ho environments. Unlike previous work in the field, RAC 2 E does not require synhronization in the radio environment. Its main harateristis are lightweight design and salability. The paper elaborates RAC 2 E operation, gives a numerial analysis of its performane behavior under different irumstanes and demonstrates its funtionalities on a testbed platform realized with four Universal Software Radio Peripheral 2 (USRP2) nodes. Index Terms RAC 2 E, Rendezvous Protool, Cognitive Radio, Cooperation, Control Channel, USRP2. I. INTRODUCTION ognitive Radio (CR) networks omprise wireless nodes Cable to learn, adapt and optimize their transmission parameters to urrent radio environmental onditions. One of their key attributes is the ability to sense the spetrum and dynamially use spetrum opportunities ensuring high spetrum effiieny. CR nodes are able to san a ertain portion of the wireless spetrum, identify temporary unused spetrum holes and aommodate their transmission to the urrently available frequeny band [1]. However, when several CR nodes opportunistially try to aess spetrum, it beomes diffiult to manage and oordinate their behavior in a manner that guarantees ollision free and optimal opportunisti spetrum sharing. The foal point of the dynami spetrum sharing among several CR nodes is the problem of exhanging information for ontrol hannel establishment. The ontrol hannel is used to exhange relevant ontext data (e.g. spetrum holes seen by a ertain CR et.) and is of paramount importane for the proper operation of CR networks. Stemming from their nature, CR environments yield dynami ontrol hannel provisioning (as no frequeny an be statially assigned for this purpose) [2-6]. Additionally, the problem beomes more omplex if a entral CR entity ontrolling the entirely available spetrum is absent. Therefore, appropriate rendezvous protool that will failitate dynami ontrol hannel establishment among CRs is needed. In an opportunisti and deentralized dynami spetrum sharing environment, individual CR nodes may sense the loal spetrum and hoose their operating frequenies and bandwidth in ollaboration with other partiipating nodes. The ooperation failitates the spetrum aess and spetrum sharing information and deisions among CR nodes [7-12] leading to improved spetrum management in ognitive environments. However, most of the work done in this area assumes an already established ontrol hannel or perfet synhronization among the CR nodes ontending for spetrum opportunities. This paper introdues a novel Rendezvous protool for Asynhronous Cognitive radios in Cooperative Environments (RAC 2 E). Its main advantage is the ability to establish ontrol hannel among asynhronous CR nodes, thus not restriting the CR nodes behavior to a ertain predefined synhronous pattern of operation. The protool itself is lightweight (i.e. is easily appliable to real CR nodes) and salable showing performane improvements as the number of CR nodes involved inreases. The paper is organized as follows. Setion II reflets on the related work in the field of rendezvous protools. Setion III introdues and elaborates RAC 2 E, whereas setion IV provides performane evaluations of RAC 2 E through a numerial and testbed analysis. Finally, setion V onludes the paper. II. RELATED WORK This setion briefly desribes some rendezvous approahes that address the problem of dynami ontrol hannel provision in a fully distributed network with CR devies. The rendezvous algorithm proposed in [2] operates through a ombination of reeiver pilot tones, a tone sanning protool and transmitter/reeiver handshaking proess. Based on the information for urrent frequeny oupany, all of the potential reeivers and transmitters are broadasting pilot tones at unoupied frequenies. The transmitter attempts to find the target reeivers to form a network and exhange data, but first it should determine at whih pilot tone frequenies the reeivers are loated. Therefore, it searhes the target potential 978-1-4244-8015-9/10/$26.00 2010 IEEE 1846
2 reeivers by sweeping the spetrum of interest. When a target reeiver radio is deteted, the transmitter starts with the handshaking proedure for exhanging onnetion speifi parameters and thus the rendezvous proess is finished. Referene [2] also onsiders the problem of optimal sanning rule. It proposes three sanning rules and analyzes their performanes in terms of shortest searh time for target reeiver. The rules have minor differenes in the average sanning time that inrease linearly with the number of reeivers. Therefore, the algorithm is not able to ope with dense networks environment. Another approah for dynami ontrol hannel seletion is proposed in [3]. The proess begins by dividing the spetrum into regularly spaed frequeny bins. The node that aims to establish onnetion senses the spetrum in eah of the bins. The bins with RF energy below some threshold are assumed as unoupied and they are flagged as potential hannels for establishing ommuniation where the transmitter emits attention signal with a very low power. The reeiving node ollets all of the attention signal ourrenes it finds within its sanning range. Then, it hooses whih frequeny to use in response based on its knowledge about the spetrum oupany and the reeived signal strength. Evaluation of the protool performanes in terms of probability of false positives and negatives alarms are given in [4]. The onnetion time is also alulated and there is a linear dependene with the number of network nodes thus the salability issue again is not well omprised. A frequeny hopping based method for ontrol hannel establishment in ognitive radio networks is presented in [5]. The frequeny hopping sequene is reated based on the urrent sensing results from eah node in order to adapt to the primary users ativities. The protool is based on the probability to ahieve rendezvous on a hannel with low primary user ativities for both nodes, simultaneously. When the rendezvous is ahieved, the nodes exhange paket for synhronization whih ontains information for reating the pseudo random hopping sequene. Referene [6] presents a rendezvous algorithm for entralized ognitive networks. This approah is suitable for ognitive networks that operate under a seondary base station ontrol. The main ontribution is the possibility to establish the ontrol hannel between a base station and a CR node, even if diret spetrum band between them is not available (usage of a relay CR node). However this approah is not apable of operation in a fully distributed network senario. Most of the related work in the area of rendezvous protools shows salability problems or is not targeted at deentralized or ooperative environments. Following setion introdues RAC 2 E, a novel rendezvous protool speifially tailored to address the dynami ontrol hannel provisioning among CR nodes operating in an asynhronous and ooperative manner. III. A NOVEL RENDEZVOUS PROTOCOL This setion gives details on the RAC2E protool. It differs from previous approahes in terms of allowing omplete asynhronous behavior of the CRs. RAC2E allows dynami ontrol hannel seletion and negotiation of a mutually suitable data hannel for ommuniation among the CRs. The first part of this setion gives more insight into the senario where RAC 2 E is supposed to operate, while the seond part provides extensive details on the protool itself. A. Senario setup The RAC 2 E envisioned senario omprises a seondary network of CSMA/CA based CR nodes with spetrum sensing apabilities that operates in the 2.4 GHz ISM band and oexists with a primary IEEE 802.11 based network. All CR nodes are able to reate a loal spetrum map (a top-down power ranking of the available hannels). If two CR nodes want to establish diret link ommuniation by opportunistially using the temporary unused hannels from the primary network terminals, they should exhange their spetrum maps in order to selet the mutually best hannel (i.e. the hannel having the lowest level of interferene for both nodes). Afterwards, the initiator node should request onnetion and the destination node should onfirm the requested onnetion allowing the data hannel to be established. All these ontrol messages should be exhanged through on demand dynamially established ontrol hannel. RAC 2 E fouses on this aspet providing a lightweight and salable solution for dynami ontrol hannel provisioning among the CR nodes. The ontrol hannel is mutual for all simultaneous ative seondary nodes and serves for ooperative spetrum sensing info exhanges as well as for ommuniation parameters negotiation. When a new seondary node beomes ative it first searhes for ontrol hannel where it an obtain the spetrum maps from other nodes and find some node for ommuniation. B. Protool desription The proposed ooperative rendezvous protool (RAC 2 E), has two phases, i.e. initialization and exhanging ontrol information phases. The former phase is used to selet the ontrol hannel if suh pre-exists in the seondary network from prior CR ommuniation. In this ase, the initialization phase requires that the node listens long enough on eah hannel frequeny in order to detet any ontrol message. However, if a ontrol hannel is not deteted, then the node selets the best hannel from its spetrum map and delares it to be the ontrol hannel. The seond phase is used for ooperative exhange of ontrol information (sensing reports exhange) on the seleted ontrol hannel. Fig. 1 depits the operational phases of the protool. It should be stressed that RAC 2 E operates upon an appliation request, i.e. when two CR nodes in the seondary network want to establish a data session between. The seond phase of the protool, after detetion of the ontrol hannel, operates in a time division mode, i.e. the node spends T s seonds for spetrum sensing and reating a spetrum map and T seonds for sending and listening on the ontrol hannel. The total period of (T s + T ) seonds is ontinuously repeated until the ommuniation link on an appropriate data hannel between the two CR nodes that want to ommuniate is established. The duration of the spetrum sensing intervals T s should be fixed to a long enough value to 1847
3 get aurate sensing information for all hannels. The time duration of the ontrol hannel attendane T is fixed for eah node and is randomly hosen from an interval [T 1, T 2 ], where the values for T 1 and T 2 an be fine tuned depending on the number of nodes. The value of T remains onstant for eah node until data ommuniation is established (Fig. 2) and is hanged for eah subsequent ontrol hannel establishment proedure. node, it heks its own spetrum map and the destination node s spetrum map and selets mutually aeptable hannel. Then, it sends a onnetion request message to the other node with all proposed onnetion parameters and waits for onnetion reply message. When affirmative onnetion reply is reeived, the pair of nodes is swithed to a seleted data hannel and starts ommuniation. In ase of a negative reply, the node that initiated the ommuniation ontinues to sense the spetrum, searhes for spetrum maps and sends new onnetion requests. Fig. 2. RAC 2 E time operation Fig. 1. RAC 2 E operational phases The total asynhronies among users, stemming from the randomly hosen T periods, provides lower or higher overlapping of the different nodes ontrol hannel periods, depending on T and T s values. It is obvious that the ends of the T periods will be overlapped most frequently (Fig. 2). Consequently, this imposes sending the ontrol hannel messages at the start and at the end of the ontrol hannel period in order to maximize the probability of paket reeption from other nodes. In the period between the two ontrol messages in one T interval, the nodes swith to a listening mode in order to detet other nodes potential ontrol messages. If the nodes were to operate in a synhronous way, then the ollisions would our more frequently sine all nodes send messages at the two ends of the ontrol hannel duration. This would lead to degraded protool performane; therefore the randomization in RAC2E is introdued in order to ensure asynhrony among the nodes. If an already established ontrol hannel is being interrupted (e.g. primary user appearane), the ontrol hannel band must be hanged. Sine every node has other nodes spetrum maps, all nodes know whih hannel is mutually the best suitable to be the new ontrol hannel. This will result in faster ontrol hannel reestablishment. RAC 2 E envisions several ontrol messages suh as sensing report message, onnetion request message and onnetion reply message. These messages are sent twie from eah node in a T period. The sensing report message arries sensing results obtained during the T s period and is the most frequently sent ooperative ontrol message. Upon reeption of suh message, a node stores it loally. Therefore, eah node an keep information of other ative nodes in the network and their spetrum maps. Moreover, before initiation of any ommuniation, a node must first detet the sensing report message. When a node wants to ommuniate with another When an already established onnetion is finished, the nodes return to the previous mode when T s seonds sense the spetrum and reate spetrum maps and T seonds send their sensing maps and listen for other nodes sensing maps. If, for some reason, the ontrol hannel is swithed to other band during the data ommuniation, then the nodes start to searh the ontrol hannel as in the initialization phase, but now first looking into the mutually best hannel from its previous spetrum maps. After elaborating the RAC 2 E operational details, the following setion will give more insight into the performane behavior of the protool. IV. PERFORMANCE EVALUATION This setion provides performane evaluation for the previously introdued RAC 2 E protool. The first subsetion shows a numerial analysis alulating the probability that a ertain CR node will miss the ontrol hannel, already established from the other CR nodes along with some additional performane issues. The seond subsetion demonstrates a testbed implementation of RAC 2 E on a platform with four USRP2 nodes. A. Numerial analysis This subsetion elaborates the impat of the ontrol hannel duration to the RAC 2 E performane. The worst ase event upon whih the protool will be in outage is when a node that just beame ative and wants to establish ommuniation does not detet any ontrol message. The ritial point of RAC 2 E is the probability to miss all ontrol messages from all ative nodes (i.e. probability of missing ontrol hannel). Therefore, the numerial analysis is foused on minimizing the value of probability of missing ontrol hannel by appropriate seletion of ontrol hannel duration. The duration of the sensing interval T s is atually the 1848
4 needed time for performing spetrum sanning of all available hannels. In pratial RAC 2 E implementations, it is a hardware limitation stemming from the used equipment (e.g. USRP2). The ontrol hannel duration T is limited from the probability of missing ontrol hannel, i.e. that the nodes will not find eah other. This yields higher duration of T in order to minimize that probability. On the other hand, T should not have higher duration due to the dynami radio environment resulting in hanges in spetrum maps. Fig. 3. Arrival of the seond node The RAC 2 E targeted system and the alulation of the probability of missing the ontrol hannel refers to the ase when a ertain node is ativated in the network, denoted as first node on Figure 3. Then, a seond node an start being ative X seonds later. Fig. 3 summarizes all possible senarios. The observation is fixed on one yle (T s + T 1 ) and assumes the first node as a referent node. The moment a seond node appears in the network an be modeled as a random variable X with a uniform distribution. Aording to the protool definition, ontrol hannel messages (i.e. sensing report, onnetion request and onnetion reply) are transmitted at the beginnings and at the ends of eah T period. These messages are reeived from the seond node after some delay ΔT whih inludes the bakoff time of the used protool for medium aess (e.g. CSMA/CA), the propagation delay and the proessing delay. Probability that the seond node will not reeive any ontrol message of the two transmitted messages from the first node, whih is the probability of missing the ontrol hannel, an be alulated as a joint probability of the events: (X + T s > T s + ΔT) and (X + T s + T 2 < T s + T 1 + ΔT). The first event orresponds to the probability of not reeiving the first message at the beginning of the T interval, whereas the seond one to the probability of not reeiving the message at the end of the ontrol hannel interval. The observations of the above mentioned events are based on the situation when the ontrol hannels of both users are overlapped in the inspeted time limits (T s + T 1 ). Events falling outside this limit are analyzed in subsequent (T s + T 1 ) intervals not restriting the generality of the onlusions. The probability of missing the ontrol hannel for the ase with two users is alulated as: P = PX+ T > T+ΔT, X+ T+ T < T+ T +ΔT ) = PX ( >ΔT ) PX ( + T T <Δ ) = m ( s s s 2 s 1 2 1 T 3 ΔT ΔT 1 ( ln( T2 +ΔT ) ln( T1 +ΔT )) ( ln( T2 +ΔT ) ln( T1 +ΔT )) (1) 3 ΔT 6ΔT Hene the probability of ontrol hannel detetion from the seond user is: Pd = 1 P m (2) The alulations are made under the assumption of a uniform distribution of the random variable X in the time interval [0, T 1 + ΔT]. This interval is hosen aording to the time limits in whih the seond node is expeted. T 1 and T 2 are random variables uniformly distributed in the time interval [T 1, T 2 ], where ΔT = T 2 - T 1. The alulated result is valid for ΔT < ΔT. The analysis assumes only two CR nodes and alulates the probability of missing the already established ontrol hannel by the seond node. In ase of N ative nodes, the probability of missing the ontrol hannel i.e. that one node will not register any of the remaining (N - 1) nodes in the observing yle is simply the produt of the probabilities to not reeive any message from the first, from the seond up to the last observing node. This situation is the worst ase event whih orresponds to not reeiving any spetrum map, thus not being able to initiate or aept ommuniation with other nodes. Probabilities of missing and deteting the ontrol hannel when Nth node join the network are respetively given by equations (3) and (4): 1 P, = N m N P (3) m N 1 P d, N = 1 Pm (4) Fig. 4 and 5 present results from the elaborated numerial analysis. Fig. 4 onsiders the ase when the ontrol hannel duration is smaller than the sensing time duration, while Fig. 5 depits the opposite ase. The used value for T s is 2.6s and is hosen based on the hardware limitation imposed by the used equipment for the testbed demonstration in the following subsetion (i.e. it takes 0.2s to sense eah of the 13 WiFi hannels with USRP2). The probability of missing ontrol hannel messages is observed for 2, 4 and 6 network nodes for two possible values of ΔT. The first ΔT value is estimated to be around 20ms by negleting the CSMA/CA bakoff average duration and inluding the proessing and the propagation time. The seond one is around 100ms and orresponds to the ase when the CSMA/CA bakoff duration is inluded. The latter one is more realisti situation for large number of network nodes. It is evident from Fig. 4 and 5 that the probability of missing the ontrol hannel messages dereases as the ontrol hannel duration and the number of network nodes inrease. This indiates that RAC 2 E will operate suessfully in large networks (addressing the salability) unlike the protools listed in setion II that have smaller probability for meeting between the nodes as the number of nodes inrease (i.e. larger searhing/onnetion time). However, Fig. 4 and 5 also show that for larger bakoff values (i.e. higher ΔT) orresponding to the situation of larger number of nodes, the probability of missing ontrol messages inreases. Therefore, the effet of improved performanes for larger number of nodes is atually ompensated with the loss of performanes due to the inreased bakoff duration. As a result, only high values for ontrol hannel duration an guarantee suessful RAC 2 E 1849
5 operation. The ontrol hannel duration should be larger than T s, but smaller than 2T s, in order to allow preise monitoring of the spetrum ondition and diminish the probability that the CR nodes will not find eah other. For ontrol hannel searhing in the initialization phase, the reommended listening time should be about one average yle duration (T s + T ). The optimal operating point of the protool an be realized with dynami tuning of the ontrol hannel duration depending from the number of ative nodes. However, this optimization annot be expressed in losed analytial form and may be analyzed only by means of experimental results experienes. The inherent effets of possible CSMA/CA ollisions an also be inluded in the protool performane analysis. Namely, the overall probability of missing the ontrol hannel is the probability to not detet the hannel messages either due to protool operation or due to ollisions in the ontrol hannel. Sine these events are independent, the overall probability would be the sum of both probabilities. The first one an be alulated using eq. (3), whereas the seond one is the probability of CSMA/CA ollisions for appropriate number of nodes [13]. ertain hannel as a ontrol hannel (effetively, the hannel searhing time in the initialization phase is inreased under this ase). The numerial evaluation (Fig. 4 and Fig. 5) is done in MATLAB [14] and does not take into aount the radio propagation phenomena (e.g. fading, shadowing) whih is left for future work. However, the following setion will present a testbed implementation of RAC 2 E showing its operational potentials for pratial purposes. B. Testbed implementation This subsetion elaborates the atual implementation of the proposed RAC 2 E protool on a hardware platform in a real radio environment. The aim is to demonstrate a proof of onept for the proposed RAC 2 E protool. Fig. 6 depits the testing senario whih omprises four USRP2 [15] nodes utilizing RAC 2 E for ommuniation over a ontrol hannel. The appliation profile of the senario is that node 1 wants to establish data onnetion (video streaming) with node 2, while simultaneously node 3 wants to ommuniate with node 4. Fig. 6. RAC 2 E testbed platform Fig. 4. Probability of missing the ontrol hannel for Δ T < T s Fig. 5. Probability of missing the ontrol hannel for ΔT > T s In ase of ontrol hannel misdetetion, ertain node an prolaim some frequeny band as a ontrol hannel even if a ontrol hannel already exists in the network. This effet an be avoided by foring the nodes to searh the ontrol hannel over the whole frequeny band twie before delaring a One the ontrol hannel is established, the nodes start sending and olleting the spetrum maps on the mutually ommon ontrol hannel. The proess of reating loal spetrum map and exhanging spetrum maps among nodes is illustrated on Fig. 7 for node 1. After the exhange of the spetrum maps, node 1 tries to initiate the video streaming ommuniation to node 2 by sending a onnetion request message through the ontrol hannel by proposing to the node 2 hannel 6 as mutually the best hannel. Fig. 8 shows the onnetion reply message from node 2 and the atual hoie of the WiFi hannel to be used for data ommuniation. Similar onlusions and figures are valid for the other nodes (i.e. 3 and 4) that want to start other video streaming in the demo. The parameters used in the proposed USRP2 demo senario are as follows. T s is hosen to be 2.6s, the bit rate on the ontrol hannel is 200kbps (using GMSK modulation), T period length is hosen randomly in the interval [T s, 2T s ] and the video streaming appliation being used is VLC media player. Finally, Fig. 9 provides a snapshot from a spetral analyzer that was used to prove the atual ooperative sensing info exhange and to demonstrate the funtionality of the proposed rendezvous approah. The deteted hannels in use from both pair of nodes are hannel 6 and hannel 8. Other spetrum emissions, besides the ones stemming from the demo s video 1850
6 streaming appliations, are aused by surrounding IEEE 802.11 aess points at the laboratory premises where the demo was set up. designed to support ontrol hannel provisioning for ooperative ognitive environments. The onduted performane evaluation shows that RAC 2 E has fast reation to dynami environmental hanges as well as satisfatory performanes in dense nodes areas. Also, the adaptive ontrol hannel duration makes RAC 2 E flexible to various network onditions. Future work will onentrate on performane testing under larger number of nodes, analytially and in the demo platform, protool performane analysis in different radio environmental ondition, ooperative network senarios, future protool design enhanements, applying speifi data fusion tehniques et. Fig. 7. Spetrum information at node 1 ACKNOWLEDGMENT Parts of this work were funded by the EC through the FP7 projets ARAGORN (216856) [16] and QUASAR (248303) [17]. The authors would like to thank everyone involved. Fig. 8. Data hannel agreement, from node 1 viewpoint Fig. 9. Spetrum analyzer snapshot of the established onnetions This setion gave a numerial and pratial evaluation of RAC 2 E. It showed that the protool is salable (more nodes yield better performanes) and is lightweight for implementation (on USRP2 nodes). V. CONCLUSIONS AND FUTURE WORK The paper introdued RAC 2 E, a novel rendezvous protool for dynami ontrol hannel establishment among asynhronous CR nodes in ooperative environments. The protool differs from previous work in terms of addressing salability and allowing asynhronous (i.e. more realisti) behavior of the CR nodes. Moreover, it is speifially REFERENCES [1] I. Akyildiz et al. NeXt generation/dynami spetrum aess/ognitive radio wireless networks: a survey, The International Journal of Computer and Teleommuniations Networking, 2006. [2] D. Pu, A. M. Wyglinski and M. MLernon, A Frequeny Randezvous Approah for Deentralized Dynami Spetrum Aess Networks, Crownom 2009, Hannover, Germany, June 2009. [3] B. Horine and D. Turgut, Link Rendezvous Protool for Cognitive Radio Networks, IEEE DySPAN 2007, Dublin, Ireland, April 2007 [4] B. Horine and D. Turgut Performane Analysis of Link Rendezvous Protool for Cognitive Radio Networks, CrownCom 2007, Orlando, FL, USA, July 31 Aug 3 2007. [5] C. Cormio and K.R. Chowdhury, Common ontrol hannel design for ognitive radio wireless ad ho networks using adaptive frequeny hopping, Ad Ho Network Journal, Otober 2009. [6] P. J. Jeong and M. Yoo, Resoure-aware Rendezvous Algorithm for Cognitive Radio Networks, 9 th International Conferene on Advaned Communiation Tehnology, February 2007. [7] O. Simeone, J. Gambini, U. Spagnolini and Y. Bar-Ness, Cooperation and ognitive radio, in Pro. IEEE CogNet Workshop, 2007, Glasgow, Sotland, June 2007, pp 6511-6515. [8] B. Zayen and A. Hayar, Cooperative Spetrum Sensing Tehnique Based on Sub Spae Analysis for Cognitive Radio Networks, IEEE GLOBECOM2007, Washington, USA, November 2007. [9] J. Unnikrishnan and V. V. Veeravalli, Cooperative Spetrum Sensing and Detetion for Cognitive Radio, IEEE GLOBECOM2007, Washington, USA, November 2007. [10] Z. Quan, S. Cui and A. H. Sayed, An Optimal Strategy for Cooperative Spetrum Sensing in Cognitive Radio Networks, IEEE GLOBECOM2007, Washington, USA, November 2007. [11] A. R. Biswasyz et al. Cooperative Shared Spetrum Sensing for Dynami Cognitive Radio Networks, IEEE ICC2009, Dresden, Germany, August 2009. [12] S. Atapattu, C. Tellambura, and H. Jiang, Relay Based Cooperative Spetrum Sensing in Cognitive Radio Networks, IEEE GLOBECOM2009, Hawai, USA, 30 November- 4 Deember 2009. [13] V. Atanasovski and L. Gavrilovska, Throughput and optimal behavior analysis of non-saturated ontending IEEE 802.11a systems in Rayleigh fading hannels, 17 th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communiations (PIMRC 06), Helsinki, Finland, September 11-14, 2006. [14] MATLAB. Information available at: www.matlab.om [15] Universal Software Radio Peripheral 2 (USRP2). Information available at: www.ettus.om. [16] EC FP7 projet ARAGORN. Information available at: http://www.itaragorn.eu. [17] EC FP7 projet QUASAR, Information available at: http://www.quasarspetrum.eu. 1851