University of Southern California Dissertations and Theses

Coexistence mechanisms for legacy and next generation wireless networks protocols

(USC Thesis Other)

Coexistence mechanisms for legacy and next generation wireless networks protocols

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content COEXISTENCE MECHANISMS FOR LEGACY AND NEXT GENERATION WIRELESS NETWORKS PROTOCOLS by Alex Chia-Chun Hsu A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Ful¯llment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (ELECTRICAL ENGINEERING) May 2008 Copyright 2008 Alex Chia-Chun Hsu Dedication This dissertation is dedicated to my parents, my aunts, and my wife. ii Acknowledgements I would like to thank Professor David S. L. Wei, Professor Norio Shiratori, and Professor Chung-Ju Chang. iii Table of Contents Dedication ii Acknowledgements iii List Of Tables vii List Of Figures viii Abstract xi Chapter 1: Introduction 1 1.1 Signi¯cance of the Research . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Review of Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.1 Coexistence Mechanism for Wi-Fi . . . . . . . . . . . . . . . . . . 4 1.2.2 Coexistence Mechanism for BT . . . . . . . . . . . . . . . . . . . . 5 1.2.3 Flexible Spectrum Allocation for Cognitive Radio. . . . . . . . . . 6 1.2.4 Decentralized MAC for Cognitive Radio . . . . . . . . . . . . . . . 7 1.3 Contributions of the Research . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.4 Outline of the Dissertation. . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Chapter 2: Research Background 14 2.1 IEEE 802.11 Wi-Fi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 2.1.1 Physical Layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 2.1.2 Ad-hoc and Infrastructure-based Networks . . . . . . . . . . . . . 16 2.1.3 Carrier Sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 2.1.4 Distributed Coordination Function (DCF) . . . . . . . . . . . . . . 20 2.1.5 Point Coordination Function (PCF) . . . . . . . . . . . . . . . . . 21 2.2 IEEE 802.15.1 Bluetooth. . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.2.1 Piconet and Scatternet . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.2.2 SCO and ACL Link . . . . . . . . . . . . . . . . . . . . . . . . . . 26 2.2.2.1 SCO Link . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 2.2.2.2 ACL Link . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 2.2.3 Packets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.3 The 2.4GHz ISM Band . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 2.4 Cognitive Radio. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 iv Chapter 3: Coexistence Mechanism Using Dynamic Fragmentation for Interference Mitigation between Wi-Fi and Bluetooth 35 3.1 Description of Dynamic Fragmentation Mechanism . . . . . . . . . . . . . 36 3.1.1 Interference Rate Analysis . . . . . . . . . . . . . . . . . . . . . . . 39 3.1.2 Transmission Time and Threshold Analysis . . . . . . . . . . . . . 41 3.1.3 Throughput Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 47 3.1.4 Delay Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 3.2 Computer Simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 3.2.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . 53 3.2.2 Simulation Results and Discussion . . . . . . . . . . . . . . . . . . 55 3.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 Chapter 4: Adaptive Hopset Frequency Hopping for Wireless Personal Area Networks in a Coexistence Environment 65 4.1 Adaptive Hopset Frequency Hopping (AHFH) . . . . . . . . . . . . . . . . 66 4.1.1 Etiquette Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 4.1.2 Carrier Sensing Mechanism . . . . . . . . . . . . . . . . . . . . . . 68 4.1.3 Description of AHFH Algorithm . . . . . . . . . . . . . . . . . . . 69 4.2 Theoretical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 4.2.1 System Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 4.2.2 Throughput Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 74 4.2.3 Occupancy Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 78 4.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 4.3.1 Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 4.3.2 Impact of Coe±cient ® . . . . . . . . . . . . . . . . . . . . . . . . 80 4.3.3 Impact of Carrier Sensing Mechanism . . . . . . . . . . . . . . . . 81 4.3.4 Performance Comparison under FD Interference Environment . . . 83 4.3.5 Performance Comparison under FD and FS Interference Environ- ments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 4.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Chapter 5: A Cognitive MAC Protocol Using Statistical Channel Alloca- tion for Wireless Ad-hoc Networks 90 5.1 Proposed SCA-MAC Protocol . . . . . . . . . . . . . . . . . . . . . . . . . 91 5.1.1 Overview of SCA-MAC Protocol . . . . . . . . . . . . . . . . . . . 91 5.1.1.1 Environment Sensing and Learning . . . . . . . . . . . . 92 5.1.1.2 CRTS/CCTS Exchange over Control Channel . . . . . . 93 5.1.1.3 DATA/ACK Transmission over Data Channels . . . . . . 95 5.1.2 Statistical Channel Allocation . . . . . . . . . . . . . . . . . . . . . 95 5.1.2.1 Optimum Operating Range . . . . . . . . . . . . . . . . . 95 5.1.2.2 Maximum Channel Aggregation . . . . . . . . . . . . . . 96 5.1.2.3 Closest Possible Opening . . . . . . . . . . . . . . . . . . 96 5.2 Successful Rate Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 5.3 Throughput Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 5.4 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 v 5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 Chapter 6: Dynamic Spectrum Access MAC for Wireless Ad-hoc Net- works 108 6.1 Role of DSA-MAC Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . 109 6.2 Analysis of Operation Range Assignment Strategies . . . . . . . . . . . . 112 6.2.1 System Model and Performance Metrics . . . . . . . . . . . . . . . 112 6.2.2 Non-Sharing (NS) Strategy . . . . . . . . . . . . . . . . . . . . . . 113 6.2.3 Full-Sharing (FS) Strategy . . . . . . . . . . . . . . . . . . . . . . 115 6.2.4 Partial-Sharing (PS) Strategy . . . . . . . . . . . . . . . . . . . . . 115 6.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 6.3.1 E®ects of Channel and Neighboring Node Numbers . . . . . . . . . 118 6.3.2 E®ects of Primary and Secondary Service Arrival Rates . . . . . . 119 6.3.3 Comparison of White Space Filling Rate . . . . . . . . . . . . . . . 121 6.3.4 Comparison of Slot Collision Probability . . . . . . . . . . . . . . . 121 6.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 Chapter 7: Conclusion and Future Work 125 7.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 7.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 Bibliography 131 vi List Of Tables 2.1 Parameters of HV packets.. . . . . . . . . . . . . . . . . . . . . . . . . . . 29 2.2 Parameters of DH packets.. . . . . . . . . . . . . . . . . . . . . . . . . . . 29 3.1 Parameters for Wi-Fi and BT systems. . . . . . . . . . . . . . . . . . . . . 55 4.1 Parameters of BT piconets. . . . . . . . . . . . . . . . . . . . . . . . . . . 68 5.1 Parameters of control and data channels. . . . . . . . . . . . . . . . . . . 103 vii List Of Figures 2.1 Transmission ranges and data rates for several wireless standards.. . . . . 15 2.2 The BSS infrastructure of a WLAN architecture. . . . . . . . . . . . . . . 18 2.3 Illustration of the hidden node and the exposed node problems. . . . . . . 19 2.4 Illustration of the distributed coordination function. . . . . . . . . . . . . 21 2.5 Illustration of a point coordination function. . . . . . . . . . . . . . . . . . 23 2.6 Illustration of the piconet and the scatternet. . . . . . . . . . . . . . . . . 26 2.7 Illustration of the operation of DH 1/3/5 packets.. . . . . . . . . . . . . . 30 2.8 Illustration of frequency occupancy of Wi-Fi and BT in 2.4GHz ISM band. 31 2.9 Illustration of spectrum holes. . . . . . . . . . . . . . . . . . . . . . . . . . 33 3.1 Successful transmission: (a) legacy 802.11 with no fragmentation (b) two fragments with no retransmission; Retransmission: (c) DF-I for mobile WLAN (d) DF-II for static WLAN . . . . . . . . . . . . . . . . . . . . . . 38 3.2 Illustration of the Wi-Fi packet transmission and the BT time slot. . . . . 39 3.3 Throughput as a function of PER for three schemes. . . . . . . . . . . . . 51 3.4 The Wi-Fi PER as a function of the BT tra±c load.(Validation of Eq. 3.2) 56 3.5 Throughput improvement for Wi-Fi with DF-I and DF-II. . . . . . . . . . 57 3.6 Throughput of the Wi-Fi network, which coexists with two SCO links on the BT piconet. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 viii 3.7 Throughput of the Wi-Fi network in the presence of ACL link on the BT piconet, where the background Wi-Fi tra±c is given by ¿ Wi¡Fi =0:3. . . 59 3.8 Throughput of the Wi-Fi network in the presence of ACL link on the BT piconet, where the background Wi-Fi tra±c is given by ¿ Wi¡Fi =0:7. . . 60 3.9 Throughput of the BT piconet with ACL tra±c ¿ BT =0:7, which coexists with Wi-Fi. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 3.10 ThroughputoftheWi-FinetworkinthepresenceofACLlinksonmultiple BT piconets, where the background Wi-Fi tra±c is given by ¿ Wi¡Fi =0:7. 62 3.11 The average Wi-Fi delay as a function of the BT tra±c load, where the background Wi-Fi tra±c is given by ¿ Wi¡Fi =0:5. . . . . . . . . . . . . . 63 4.1 Illustration of the channel occupancy of (a) the legacy frequency hopping system,(b)theorthogonalhopsethoppingsystemand(c)theAHFHsystem. 67 4.2 Illustration of the carrier sensing mechanism: (a) the original FH and (b) the modi¯ed FH with the carrier sensing mechanism. . . . . . . . . . . . . 69 4.3 Illustration of the necessary opening [¹s]: (a) the original DH1/3 packets and (b) the DH1/3 packets with carrier sensing.. . . . . . . . . . . . . . . 76 4.4 Comparison of throughput for FH, OH and AHFH with di®erent ® values. 81 4.5 Comparison of occupancy for FH, OH and AHFH with di®erent ® values. 82 4.6 Illustration of the impact of the carrier sensing mechanism on throughput. 83 4.7 Illustrationofthroughputvs. numberofpiconetsunderfrequencydynamic interference. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 4.8 Comparison of frequency occupancy as a function of number of piconets under an environment with FD interference. . . . . . . . . . . . . . . . . . 86 4.9 Throughput performance comparison as a function of the number of pi- conets in an environment with frequency dynamic and frequency static interference (¿ Wi¡Fi =0:7 in all 22 channels). . . . . . . . . . . . . . . . . 87 4.10 Occupancyperformancecomparisonasafunctionofthenumberofpiconets under frequency dynamic and frequency static interference with ¿ Wi¡Fi = 0:7 in all 22 channels. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 5.1 An illustrative example of the SCA-MAC protocol. . . . . . . . . . . . . . 92 ix 5.2 AnalyticalandsimulatedresultsofSuccessfulRate®vs. OperatingRange r with (n;m)=(2;1) at ¿ =0:5. . . . . . . . . . . . . . . . . . . . . . . . 105 5.3 Analytical and simulated results of Throughput ½ vs. Operating Range r with (n;m)=(2;1) at ¿ =0:5. . . . . . . . . . . . . . . . . . . . . . . . . 105 5.4 Analytical and simulated results of Successful Rate ® vs. Channel Aggre- gation m with (n;r)=(2;50) at ¿ =0:5. . . . . . . . . . . . . . . . . . . . 106 5.5 AnalyticalandsimulatedresultsofThroughput½vs. ChannelAggregation m with (n;r)=(2;50) at ¿ =0:5.. . . . . . . . . . . . . . . . . . . . . . . 106 6.1 AslottedmultichannelmodeloftheDSAsystem. Whenthereisacollision between transmissions A and B, scenario (1) shows the possible outcome of a time domain backo® (where transmissions are randomly delayed to a later time slot on the same channel) while scenario (2) shows the possible outcomeofafrequencydomainbacko®(wheretransmissionsarerandomly distributed to other channels). . . . . . . . . . . . . . . . . . . . . . . . . 110 6.2 Comparison of throughput performance of four ORA strategies as a func- tion of the channel number. . . . . . . . . . . . . . . . . . . . . . . . . . . 119 6.3 Comparison of throughput performance of four ORA strategies as a func- tion of the neighboring node number. . . . . . . . . . . . . . . . . . . . . . 120 6.4 Comparison of throughput performance of four ORA strategies as a func- tion of the packet arrival rate of the primary service. . . . . . . . . . . . . 121 6.5 Comparison of throughput performance of four ORA strategies as a func- tion of the packet arrival rate of the DSA node. . . . . . . . . . . . . . . . 122 6.6 Performancecomparisonofthewhitespace¯llingrateoffourORAstrate- gies as a function of the packet arrival rate of the primary service. . . . . 123 6.7 Performancecomparisonofthewhitespace¯llingrateoffourORAstrate- gies as a function of the packet arrival rate of the DSA node. . . . . . . . 123 6.8 PerformancecomparisonoftheslotcollisionprobabilityoffourORAstrate- gies as a function of the packet arrival rate of the primary service. . . . . 124 6.9 PerformancecomparisonoftheslotcollisionprobabilityoffourORAstrate- gies as a function of the packet arrival rate of the DSA node. . . . . . . . 124 x Abstract E®ective usage of unlicensed bands (UB) has received a lot of attention due to its poten- tial in ubiquitous computing and networking. One key issue in e®ective UB usage is the coexistence among devices of homogeneous or heterogeneous systems, e.g. wireless local area networks (WLAN) and wireless personal area networks (WPAN). To resolve the coexistence problem, we need to understand the interaction between concurrent trans- missions in overlapping frequency bands. Although some basic interference resolving process is mentioned in the standards, further performance improvement can be achieved by careful system analysis and parameter selection. In this research, we analyze the co- existence problem that these systems face and devise coexistence mechanisms to enhance performance. We develop a suitable analytical model that can accurately model the interference phenomena between WLANs and WPANs. Then, we propose non-collaborative solutions for WLAN and WPAN, respectively. For WLAN, we propose a dynamic fragmentation (DF)mechanismtooptimizethepacketlengthsuchthatWi-Fideviceshavebetterchance to avoid the interference caused by Bluetooth piconets. Both theoretical analysis and simulation results con¯rm that DF can signi¯cantly improve the performance of Wi-Fi in terms of throughput and transmission delay. xi For WPAN, we propose an adaptive hopset frequency hopping (AHFH) mechanism to avoid the interference from coexisting WLAN as well as the self-interference from collocatingBluetoothpiconets. WithAHFH,apiconetadjustsitshopsettominimizethe interferenceadaptively. ResultsshowthatAHFHimprovesperformanceandadaptability to the dynamic changes in the environment. Stale and ine±cient current spectrum regulation and the crowded ISM band makes °exible \cognitive radio (CR)" the logical next step. Along this research direction, we propose a statistic channel allocation MAC (SCA-MAC) and a dynamic spectrum access MAC (DSA-MAC). Both protocols can exploit the \spectrum hole" on unlicensed bands and minimize their interference on incumbent users and collision with other CR nodes. SCA-MAC avoids interference with primary services through statistics, and improves spectrum e±ciency. DSA-MAC further investigates strategies of the operation range assignment. Three strategies, namely, non-sharing (NS), partial sharing (PS) and full sharing (FS), were discussed and compared under various scenarios. It is shown that FS gives the best throughput. xii Chapter 1 Introduction 1.1 Signi¯cance of the Research In US and many other countries, the unlicensed band is expected to be populated by various wireless devices soon. Most of these devices will be used for wireless local area networking (WLAN) (e.g., the wireless ¯delity technology, or Wi-Fi) [1] and wireless personal area networking (WPAN) (e.g., the bluetooth technology, or BT) [16]. Since WLAN and WPAN are not competing but complementary technologies, the Wi-Fi and the BT devices are expected to be placed side by side while sharing the same frequency band. Due to the rapid integration of wireless communication technologies, collocations of BT and Wi-Fi devices, especially for hand-held or mobile computing devices, e.g., cellphone,PDA,laptop,etc.,arebecomingarealitynowadays. Whentheirradiocoverage areas overlap with each other, interference is likely to occur [23], [10]. Wi-Fi is the prevailing WLAN technology. Its medium access control (MAC) layer is designed to cope with the contention with other Wi-Fi devices, but it does not cooperate wellwithBTpiconets. TheperformanceoftheCSMA/CAmechanismadoptedbyWi-Fi 1 deteriorates under the coexisting scenario. In other words, BT devices operating in the proximity of Wi-Fi devices may gravely a®ect the performance of Wi-Fi devices, and vice versa. There is another interference problem associated with BT. That is, BT adopts the frequency hopping spread spectrum (FHSS) mechanism, which does not perform carrier sensing. While hopping blindly in the unlicensed band, a piconet interferes with neighboring piconets or other non-WPAN entities. In the near future, new wireless technologies such as ZigBee, UWB, WiMax, etc., are expected to join the pair and make the ISM band even more crowded. Overlapping coverageofwirelessnetworksinthesharedfrequencybandwillbecommonintheubiqui- tous communication environment. To uphold the quality of communication, all wireless technologies should have an e±cient mechanism to access the unlicensed band as well as a proactive mechanism to detect and mitigate the interference caused by other types of coexisting wireless devices. Thesuccessofunlicensedspectrumapplicationsrevealsthedrawbackofcurrentstatic spectrum allocation policies. The fact that the demand for spectrum is not uniform across time, frequency and space leaves a signi¯cant amount of \white space" (unused space) in the licensed radio spectrum. FCC sponsored studies [6] show that over 70% of the allocated spectrum is not in use for a long time even in a crowded area where the spectrum usage is intensive. On the other hand, the remaining portion of the unlicensed spectrumisbeingexhaustedbyemergingwirelessservicesandapplications,leadingtothe so-called spectrum scarcity problem. One solution to this problem is to give unlicensed- spectrum users an opportunity to use the white space in the licensed spectrum in such a way that the interference to licensed users is kept below an acceptable level. This is 2 called dynamic spectrum access (DSA) [19], [5] and can be realized by cognitive radio (CR) techniques [18], [11]. Fortoday'swirelesstechnologies, theperformancebottleneckcomesfrominterference caused by coexistence. Besides isolated solutions developed for individual wireless net- works, cognitive radio (CR) techniques have the potential to alleviate the bottleneck of coexistence. Workingwithetiquetterules, theymayrevolutionizethecurrentpartitioned spectrum access policy and enable more e±cient usage by exploiting \spectrum holes". 1.2 Review of Previous Work Recognizing the importance of coexistence and the need of studying the impact of inter- ference on the throughput of these devices, IEEE has created a coexistence task group called IEEE P802.15.2 [17]. The IEEE 802.15 TG2 has de¯ned two classes of coexistence mechanisms; namely, collaborative coexistence and non-collaborative coexistence. Even though such a classi¯cation is mainly developed for BT mechanisms, the principle can be applied to other coexisting scenarios. The collaborative mechanism works only when the information exchange is possible among coexisting heterogeneous networks. Examples of collaborative coexistence mechanisms of Wi-Fi and BT are META (MAC Enhanced Temporal Algorithm) or TDMA (Time Division Multiple Access) schemes [17]. Collab- orative mechanisms are feasible only when the two systems are installed on the same device and controlled by a centralized controller, e.g., a common driver. In contrast, the non-collaborative mechanism does not have such a constraint, and it can operate under 3 a broader scenario of practicality. With non-collaborative coexistence mechanisms, each device simply takes its own maneuver to reduce the interference. WewouldliketoreviewrelatedworkonWi-Fi,BTandcognitiveradiointhissection. Throughoutthechapter,weuseBTandWPANinterchangeablysinceBTistheprevailing technology for WPAN. The same applies to Wi-Fi and WLAN. 1.2.1 Coexistence Mechanism for Wi-Fi Wi-Fi adopts the direct sequence spread spectrum (DSSS) technology. It is more vulner- able to the interference caused by BT devices due to its longer data packets and lack of frequency agility. Thus, it is in great need of some coexistence mechanism. A schedul- ing mechanism called Voice-OverLap Avoidance (V-OLA) [3] has been proposed to avoid the interference from the BT voice tra±c by squeezing Wi-Fi transmission into the idle time between consecutive BT voice packets. The mechanism used by V-OLA is either to delay the transmission or to shorten the packet to ¯t into the idle interval. It works only when BT devices have a predictable access pattern, e.g., voice tra±c. Otherwise, it re- quirescoexistingBTdevicestoruntheprocessofData-OverLapAvoidance(D-OLA)[3]. Consequently, it is not a pure non-collaborative solution. Packet fragmentation [17] has been proposed as an e®ective way to mitigate inter- ference. A model was presented in [14] to ¯nd the optimal fragmentation by solving di®erential equations. In the coexistence environment of BT and Wi-Fi, Wi-Fi generally experiences more packet loss. Due to in°exibility of Wi-Fi spectrum access, few solutions have been proposed so far, not to mention a good solution. 4 1.2.2 Coexistence Mechanism for BT With a rapid frequency hopping mechanism and a broader operating frequency band, BT can avoid interference imposed by Wi-Fi devices more e®ectively. Besides charac- terizing interference caused by WLAN [13], [9], non-collaborative mechanisms have been developed to enhance the interference mitigation capability of BT. These solutions in- cludeadaptivefrequencyhopping(AFH)[17],[8],Bluetoothinterferenceawarescheduling (BIAS)[7],data-overlapavoidance(D-OLA)[3]anddynamicadaptivefrequencyhopping (DAFH) [29]. These mechanisms control the hopset to avoid overlapping in frequency. The basic idea is to distinguish good channels from bad ones and then keep the hopping sequence stay in good channels more frequently than bad ones. Besidesavoidingthefrequency-static(FS)interferencefromaWLANoramicrowave, the coexistence mechanism for a BT piconet should consider the frequency-dynamic in- terference from other collocating piconets. The self-interference experienced by BT is of an unpredictable and ephemeral nature. Being an ad-hoc network, a BT piconet cannot assumeanyinfrastructuretomanagethespectrumforcollocatedentities. Hence, WPAN should adapt its spectrum usage to the present interference pattern by applying a certain coexistence mechanism. Solutions proposed for self-interference often overlap with those for mitigating infer- ence from Wi-Fi, e.g. AFH [8], [17]. Coexistence mechanisms for BT need to address bothproblemssincetheycouldhappenconcurrently. Hopset(orhoppingset)adaptation is the essence of these mechanisms. The orthogonal hop set partitioning (OHSP) [20] prede¯nes ¯ve hopsets within the complete hopset allocated to BT. Each piconet chooses 5 randomly one of the hopsets without any procedure for adaptive hopset selection. Here, thenumberoforthogonalhopsetsissmallandsoisthenumberofcollocatedpiconetsthat can be accommodated by this scheme to be free from interference. The frequency rolling (FR)scheme[28]adoptsadaptivehopsetselectionanditcanaccommodatealargernum- ber of collocated piconets to be interference-free. However, it does not inherently posses the ability to deal with another FS-interferer. The dynamic adaptive frequency hopping (DAFH) method [29] deals both kinds of interference simultaneously. When experiencing high packet loss, it divides the complete hopset into two equal parts and chooses the half with less interference (i.e., with lower PER) as its new hopset. An etiquette rule is embedded to withhold constant occupancy with or without a coexistence mechanism. However, the ¯xed bisection behavior and the hopset bundle may hold back piconets to allocate their tra±c according to the real interference distribution. Moreover, the lack of multiple time-slot capability could signi¯cantly impact the performance. 1.2.3 Flexible Spectrum Allocation for Cognitive Radio The concept of a full cognitive and adaptive software de¯ned radio, which is also called Mitola radio [26], has been discussed for years. For example, it has been addressed in the European DRiVE project [19], where dynamic spectrum allocation (DSA) is proposed by collecting statistics of spectral usage to improve performance by time and space de- pendent spectrum sharing with coexisting wireless networks. If a certain portion of the licensed spectrum is not used for a while at some locality, the device in that area can usethespectrumwithoutmutualdisturbance,thusimprovingperformanceandspectrum e±ciency. IntheDARPAXGprogram[5], aschemecalledopportunisticspectrumaccess 6 (OSA) was developed to exploit the instantaneous availability of busy open spectrum, e.g., the ISM unlicensed band [21]. It allows the secondary wireless service to identify available spectrum resources and communicate opportunistically while guaranteeing the levelofinterferenceperceivedbytheprimarywirelessserviceisunderacertainthreshold. For example, o®-line applications of sensor networks within the WLAN coverage. OSA not only eliminates the white space (or spectrum hole) of occupied spectrum but also searches unused spectra for its usage. These two alternatives are complementary and could be combined to ful¯ll the goal of cognitive radio. 1.2.4 Decentralized MAC for Cognitive Radio Besides research on the PHY layer, several decentralized MAC protocols have also been proposed for cognitive radio. A random channel access protocol based on a continuous- time Markov model was introduced in [32], which was used to model spectrum etiquette with and without queuing. By assuming coexisting devices with di®erent operating spec- tra, a MAC technique was proposed in [32] to pack all radio systems tightly in the frequency domain. By packing, we refer to eliminating ine±cient spectrum allocation and preserving spectrum °exibility for future access. An optimal DC-MAC protocol and a suboptimal greedy DC-MAC protocol along with an analytical framework were proposed in [36]. The framework includes three com- ponents: 1) a channel occupancy model to captures dynamics of channel availability; 2) a performance metric to evaluate strategies; and 3) an decision-making process to select a channel to sense and access. The optimal DC-MAC was optimized using the partially observable Markov decision process (POMDP), and the suboptimal solution of 7 lower complexity is derived using a greedy search. The decision on channel selection is made upon the slotted time basis. However, there is no guarantee that the channel is slotted in reality unless all CR devices are synchronized, which is an extremely di±cult and challenging job in an ad-hoc network. To ¯t into unused space, we need to study problems such as how long the spectrum hole is and how long the channel will remain idle, which explains the importance of the statistics of spectrum utilization. A tri-band protocol called the dynamic open spectrum sharing (DOSS) MAC was proposed in [24]. This protocol consists of the control band, the data band and the busy-tone band. It allows CR devices to negotiate in the control band ¯rst, and then send data through any continuous fraction of the data band. The hidden/exposed node problem was eliminated by raising the busy-tone signal in the busy-tone band. DOSS MAC is scalable and it results in e±cient and real-time spectrum allocation. However, multiple radio transceivers are needed for the tri-band design. There is also a concern on the interoperability issue with existing 802-family wireless devices. A cognitive MAC protocol based on OSA [5] was proposed in [21], and a testbed was built to characterize the relationship between the loading of secondary users and its impact on primary users. However, important issues such as the e®ect of secondary user spectrum utilization upon the primary user carrier sensing, the MAC protocol overhead, the secondary-to-secondary user interference, etc., were not addressed there. The XG Radio system was ¯eld tested in the Northern Virginia in 2006 [25]. The system used the dynamic spectrum access technology to determine the unused spectrum locally, andthenoperatedonthesechannelswithoutcausinginterferencetoexistingnon- cooperative (or primary) users. There were three major test criteria: i) to be harm-free 8 (little interference), ii) to be functional (forming and maintaining connected networks), and iii) to provideadditional value(e±cientspectral usage). Some details were neglected in this test, e.g. how a new node searches the spectrum to join an existing XG net- work. This ¯eld demonstration provides convincing results that the XG technology can provide robust networks with excellent QoS in challenging mobile radio frequency (RF) environments. 1.3 Contributions of the Research Webeginwithourstudyfromtheinterferencephenomenainthecoexistenceenvironment, speci¯cally between Wi-Fi and BT. Then, solutions are proposed for both Wi-Fi and BT to mitigate their mutual interference. Here, we focus on the non-collaborative solution on since it is more practical. With the knowledge of wireless technologies in the existing unlicensed band, we investigate the problem of interference avoidance for cognitive radio andproposeaMACprotocolforCRdevicesthatallowsharmlesscoexistencewithlicensed spectrum users. More details are given below. ² Establishment of accurate interference model PacketlossduetomutualinterferenceofWi-Finetworksdoesnotnecessarilyimply packetlossonBTpiconets, andviceversa. Theinteractioniscomplicated, whichis controlled by multiple factors. A simple yet accurate model is important to provide insights and can be used to validate the potential improvement of proposed coex- istence mechanisms analytically. We propose an analytical model to describe the 9 PERofWi-FiandBTnetworks, andthenusethemodeltocomputethePERchar- acteristics of Wi-Fi and BT networks, respectively. Simulation results of proposed algorithms are consistent with our theoretical analysis. ² Improving Wi-Fi performance with coexisting BT piconets Packet fragmentation is speci¯ed in the 802.11b standard and it can provide a non- collaborative mechanism in reducing the interference of BT piconets when Wi-Fi and BT networks are collocated. The packet error rate can be used to determine whether the performance is improved or degraded. To lower interference for re- transmission reduction by fragmentation, the tradeo® is the increase in the packet overhead. Based on our model, we develop an optimal solution which provides an immediateperformanceboostratherthangradualperformanceadjustment. Rather than determining the optimal fragmentation as done in previous work, we deter- mine the proper time for fragmentation that provides a more practical and e®ective solution. Our proposed mechanism, called dynamic fragmentation (DF), adjust the frag- mentation level based on the current level of packet error rate (PER). Thus, the mechanism is triggered on demand. Our mechanism has two versions: DF-I for mobile networks and DF-II for static networks. It is shown that our model can be employed to determine the right time for further fragmentation or return to the previous state of un-fragmentation by examining PER. Furthermore, with the pro- posedmechanism, 802.11MACcandistinguishwhetherafailedtransmissionisdue to interference caused by a BT device or collision caused by a Wi-Fi station. Thus, 10 we can take di®erent actions to enhance the throughput. Simulation results show that our DF mechanism can reduce the interference between Wi-Fi and BT, thus improving the performance of Wi-Fi signi¯cantly in throughput and transmission delay. Although there is only slight performance improvement at the BT side, no performance deterioration is observed. ² Improving BT performance under mutual and self-interference Besides static spectrum interference from Wi-Fi, a BT piconet also faces interfer- ence from other piconets. An e®ective coexistence mechanism should solve these two types of interference simultaneously. We propose an enhanced adaptive fre- quency hopping (EAFH) algorithm that relies on dynamic hopset adaptation. By applying EAFH, a piconet constantly adjusts its hopset to avoid mutual or self- interference. The algorithm relies only on the observation of experienced PER under the assumption that a piconet cannot di®erentiate whether the interference is caused by coexisting Wi-Fi or collocating piconets. There are two important fea- tures of EAFH. First, it can take full advantage of multiple time slot packets, e.g. DH3/5, to improve throughput. Second, it preserves the original hopset as much as possible. SinceEAFHdoesnothaveprede¯nedhopsets,whethertoinclude/exclude a frequency channel in a hopset should be considered individually. As a result, the tra±c load is distributed according to the interference level. 11 ² Protection of incumbent users while enabling opportunistic spectrum access Toaddressinteroperabilityandachievehighere±ciencyofspectrumutilization, we propose a CSMA/CA-based cognitive MAC protocol using statistical channel allo- cation for wireless ad hoc networks. It is called SCA-MAC. SCA-MAC allows CR devicestodoreal-timeopportunisticaccesstoanycontinuouspartofthespectrum, licensed or not. The CR device gains intelligence by sensing the environment and collecting the statistics of spectrum usage. Based on the statistics, the probability of successful transmission and the chance of interference to licensed users can be evaluated. Thus, SCA-MAC can e®ectively use the spectrum hole to improve spec- trum e±ciency with little deterioration on the performance of coexisting licensed users. ² Collision resolution via frequency-domain backo® and operation range selection Collision resolution is an important issue for DSA nodes as well as incumbent users in DSA. As a precaution, a DSA node can control the operation range to reduce the probability of collision with neighboring nodes in DSA-MAC. Three OR strate- gies; namely non-sharing (NS), partial sharing (PS) and full sharing (FS), are con- sidered. Besides strategies for collision prevention, we adopt a frequency-domain backo® scheme. The traditional time-domain backo® scheme results in additional transmission delay after the occurrence of a collision. A DSA wireless system pos- sesses a swath of spectrum that is divided into many channels. By exploiting the adequate frequency resource, the frequency-domain backo® scheme can shorten the 12 delay. It is shown that DSA-MAC with the frequency-domain backo® and the op- eration range selection strategies can yield higher throughput and shorter average delay. 1.4 Outline of the Dissertation Therestofthethesisisorganizedasfollows. Thebackgroundonunlicensedbandwireless networks (including 802.11 WLAN and 802.15 WPAN) and dynamic spectrum access is reviewed in Chapter 2. Chapter 3 introduces the Dynamic Fragmentation (DF) mecha- nismtomitigatetheinterferencefacedbyWi-Fidevices. Chapter4presentstheAdaptive HopsetFrequencyHopping(AHFH)mechanismtosolvethemutual-andself-interference encounteredbyBTpiconets. Acognitive-radioMACscheme, calledSCA-MAC,waspro- posed in Chapter 5. SCA-MAC uses a statistical approach to increase the successful channel access rate. Another cognitive-radio MAC scheme, called DSA-MAC, was stud- ied in Chapter 6. DSA-MAC enhances the system performance by adopting a operation range selection strategy. Finally, concluding remarks and future research directions are given in Chapter 7. 13 Chapter 2 Research Background Many wireless standards have been proposed for various applications. The comparison of severalwirelessprotocols,includingtheirtransmissionrangesanddateratesareshownin Fig. 2.1. Among them, IEEE 802.11 Wi-Fi and IEEE 802.15.1 Bluetooth (BT) are main objects of our current research. These two legacy wireless standards will be reviewed in this chapter. We will explain how they access the medium, especially their medium access control (MAC) algorithms in the data link layer. The performance of a wireless protocol primarily depends on the e±ciency of its MAC algorithm due to the multiple access nature. Unlike wired communication systems, the bottleneck of wireless systems lies in the MAC layer. After the review of Wi-Fi and BT, we will go through the issues of coexistence in the unlicensed band. 2.1 IEEE 802.11 Wi-Fi TheIEEE802.11standard[1]isaspeci¯cationforWirelessLocalAreaNetworks(WLAN). IEEE 802.11 should appear as a traditional IEEE 802 LAN to higher layers (namely, lay- ers higher than logical link control (LLC)). This demands that the IEEE 802.11 network 14 Figure 2.1: Transmission ranges and data rates for several wireless standards. handles the issues of station (STA) mobility and wireless medium in the MAC sublayer. To meet the requirement of LLC on lower layers, it is essential for IEEE 802.11 to incor- porate functionality that is untraditional for MAC sub-layers. This, it is fair to say that the innovation of 802.11 lies in the MAC layers and it has some initiatives at the PHY layer. In the following, all components that connect to a wireless medium in a network are referred to as stations (STAs). All STAs are equipped with wireless network interface cards(WNICs). Stationsfallintooneoftwocategories: wirelessclientsandaccesspoints (AP). 15 2.1.1 Physical Layer The physical layers of IEEE 802.11 are fundamentally di®erent from wired media. The impact of wireless medium on 802.11 PHY design is listed below. 1. The medium has no absolute and observable boundaries, outside of which other 802.11 PHY transceivers are unable to receive data. 2. Transmission is unprotected from outside signals. 3. The medium is much less reliable than the wired one. 4. The network has a dynamic topology. 5. ThenetworkhasnofullconnectivitysothatsomeSTAsmaybehiddenfromothers. 6. The network has time-varying and asymmetric channels; namely, the forward path from STA A to STA B and the return path from STA B to STA A may not have the same channel conditions. Because of the limitation of the PHY layer, wireless LANs have to be built from basic building blocks to extend to a larger geographic distance. 2.1.2 Ad-hoc and Infrastructure-based Networks There are two WLAN architectures; namely, ad hoc or infrastructure-based networks. The smallest IEEE 802.11 LAN may consist of only two stations, say, two nearby note- books. They form an independent basic service set (BSS) together. This mode of opera- tion is possible when IEEE 802.11 stations are able to communicate directly. Since IEEE 16 802.11 LAN is often formed based on demand without pre-planning, this connection is referred to as an ad hoc network. The BSS infrastructure of a WLAN is shown in Fig. 2.2. An STA shall be associated to become a member of a BSS. These associations are dynamic based on distribution system service (DSS). The association between an STA and a BSS is dynamic (e.g. STAs turn on when it comes within the range and turn o® when it goes out of the range). DSS enables mobile device support by providing logical services to handle address-to- destination mapping and seamless integration of multiple BSSs. An access point (AP) is an STA that provides access to the mobile device with DSS. DSS and BSSs allow IEEE 802.11 to create a wireless network of an arbitrary size and complexity. This type of networks with an infrastructure is called the extended service set network in IEEE 802.11. TheWLANarchitecturehasanimpactonfunctionsoftheMAClayer. Therearetwo supporting MAC functions for two architectures. They are: the distributed coordination function (DCF) and the point coordination function (PCF). Before discussing DCF and PCF, we ¯rst explain how 802.11 handles collision detection. 2.1.3 Carrier Sensing Collision detection is relatively easy for wired systems but a major problem for wireless systems. The two main problems of wireless communications are the hidden node and the exposed node problems. The hidden node problem is illustrated in Fig. 2.3(a), where STAC cannot hear STAA sothat STAC maystart transmitting to STA BwhileSTAA istransmittingtoSTAB.Inthiscase,thereceiverisresponsibleforcollisionavoidance. A 17 Figure 2.2: The BSS infrastructure of a WLAN architecture. mechanism called virtual carrier sensing (VCS) is used in 802.11 to prevent the problem. The exposed node problem is shown in Fig. 2.3(b), where STA G is transmitting to STA H. STA F is within the coverage of STA G and overhears the transmission. Since STA F senses the busy medium, it would think that its attempted transmission might collide with the ongoing transmission and decide not to send. However, the transmission between STA E and STA F will not corrupt the existing transmission. Thus, the carrier sensing mechanism over-reacts, which hinders the transmission opportunity. The use of directional or smart antennas to limit the interference provides one solution. The other one is to relax the restriction of the VCS mechanism [34], [35] so as to allow more opportunities, which may result in more packet loss, too. 18 (a) Illustration of the hidden node problem (b) Illustrationoftheexposednodeprob- lem Figure 2.3: Illustration of the hidden node and the exposed node problems. Physical and virtual carrier sensing mechanisms are used to determine the medium state. A physical carrier sensing mechanism is provided by the PHY layer, which is called the clear channel assess (CCA). CCA measures channel energy and compares it to a threshold to determine whether the channel is idle or busy. However, physical sensing alone cannot guarantee the success of the transmission. A virtual carrier sensing mechanism is provided by MAC. This mechanism is referred to as the network allocation vector (NAV). NAV maintains a prediction of future tra±c on the medium based on the duration information announced in the Request-To-Send (RTS) and the Clear-To- Send (CTS) frames prior to the actual data exchange. The duration information is also available in MAC headers of all frames sent. The carrier sensing mechanism combines the NAV state and the transmitter status obtained by physical carrier sensing to determine the busy/idle state of the medium. The NAV may be thought of as a counter, which counts down to zero at a uniform rate. When the counter is zero, the medium is idle; otherwise, it is busy. The medium will be claimed to be busy whenever the STA is transmitting. The mechanism to set NAV using RTS/CTS in DCF and PCF will be described in the following subsections. 19 2.1.4 Distributed Coordination Function (DCF) ThefundamentalaccessmethodofIEEE802.11MACisDCF,whichisalsoknownasthe carriersensingmultipleaccesswithcollisionavoidance(CSMA/CA).DCFisimplemented inallSTAsforusewithinbothBSSandinfrastructure-basednetworkcon¯gurations. For aSTAtotransmit,it¯rstsensesthemediumtodetermineifanotherSTAistransmitting. If the medium is idle (with both physical and virtual carrier sensing), the transmission may proceed. The CSMA/CA distributed algorithm mandates that a gap of a minimum speci¯ed duration exists between contiguous frame sequences, e.g. various kinds of inter frame space (IFS). A transmitting STA shall ensure that the medium is idle for this required duration before attempting to transmit. If the medium is detected to be busy, the STA shalldeferuntiltheendofthecurrenttransmission. Afterdeferral,orpriortoattempting to transmit again immediately after a successful transmission, the STA shall select a random backo® interval and decrement the backo® interval counter while the medium is idle. Amechanismisusedtominimizecollisionfurthermore. Thatis, thetransmittingand receivingSTAexchangeshortcontrolframes,calledRTSandCTS,afterdeterminingthat themediumisidle,afteranydeferralsorbacko®s,andpriortodatatransmission. Details of CSMA/CA, deferrals, and backo®s are illustrated in Fig. 2.4. When encountering a packetloss,thebacko®windowisdoubledtoavoidcollision. Thesizeofabacko®window is doubled up to a system parameter CW MAX . After a successful transmission, the size of the backo® window is reset to CW MIN . 20 Figure 2.4: Illustration of the distributed coordination function. 2.1.5 Point Coordination Function (PCF) The IEEE 802.11 MAC also incorporates an optional access method called the point coordination function (PCF), which is only used for the infrastructure-based network con¯guration in the presence of an AP. This access method uses a point coordinator (PC), which operates at the access point of the BSS, to determine which STA currently has the right to transmit. The operation mainly relies on polling with PC performing the role of the polling master 1 . PCF uses a virtual carrier sensing mechanism aided by an access priority mecha- nism. PCF distributes information with Beacon management frames to gain control of the medium by setting the network allocation vector (NAV) in STAs. After receiving the 1 The PCF operation may require additional coordination, which is not included in the original stan- dard, to allow a more e±cient operation when multiple point-coordinated BSSs are operating in the same channel or the overlapping physical space 21 beacon, only STAs that need priority service stay active. In addition, all frame transmis- sions under PCF also use an inter frame space (IFS), which is assigned to a smaller value than the IFS for frames transmitted via DCF. The use of a smaller IFS implies that the point-coordinated tra±c should have a higher priority to access the medium than STAs in overlapping BSSs with the DCF access method. The access priority provided by a PCF may be utilized to create a contention-free period (CFP). CFP begins with a beacon from PC and ends with a contention-free- end (CF-End) control packet from PC as well. As shown in Fig. 2.5, PC polls other STAs by exchanging CF-Poll and CF-ACK during the CFP. These control packets can be piggybacked with normal DATA packets. PC controls frame transmissions of STAs to eliminate contention only for a limited period of time. DCF and PCF should coexist in a manner that permits both to operate concurrently within the same BSS. When a PC is operating in a BSS, the two access methods alternate, with a CFP followed by a contention period (CP). 2.2 IEEE 802.15.1 Bluetooth The proliferation of mobile computing devices, e.g. laptop, personal digital assistant (PDA)andwearablecomputer, hascreatedademandforwirelesspersonalareanetworks (WPAN). WPANs allow closely located devices to share information and resources with- out any cord. The BT wireless technology uses a short range radio link that has been optimizedforpower-conscious, battery-operated, small-size, lightweightpersonaldevices. 22 Figure 2.5: Illustration of a point coordination function. A BT-based WPAN supports both synchronous communication channels for telephony- grade voice communication and asynchronous communications channels for data commu- nications. These facilities enable a rich set of devices and applications to participate in a BT-based WPAN. For example, a cellular phone may use a circuit-switched channel to carry audio to and from a headset while using a packet-switched channel to exchange data with a notebook computer concurrently. A BT network is created in an ad-hoc manner whenever an application in a device desires to exchange data with matching applications in other devices. It may cease to exist when the involved applications have completed their tasks and no longer need to continue exchanging data. The BT network operates in the unlicensed 2.4 GHz ISM band. A fast frequency-hop (1600 hops/s) transceiver is used to combat interference and 23 fading in this band (i.e., reducing the probability that all transmission is destroyed by interference). A Gaussian-shaped, binary frequency shift keying (FSK) with a symbol rate of 1 Msymbols/s minimizes transceiver complexity. A slotted channel is used, which has a slot duration of 625 ¹s. A fast time division duplex (TDD) scheme is used to enable full duplex communications at higher layers. On the channel, the information is exchanged through packets. Each packet is transmitted on a di®erent frequency in the hopping sequence. A packet nominally covers a single slot, but can be extended up to either three or ¯ve slots. For data tra±c, an unidirectional (i.e., asymmetric) maximum of 723.2 kb/s is possible between two devices. A bidirectional 64 kb/s channel is used to support voice tra±c between two devices. The jitter for the voice tra±c is kept low using smaller transmission slots. Atthe¯rstglance, theoperationandobjectivesofIEEE802.15(WPAN)mayappear to resemble the operation and objectives of IEEE 802.11 (WLAN). Both of them allow a device to connect to its surrounding environment and exchange data over an unlicensed wireless link. However, WLANs have been designed and optimized for nomadic usage of portable computing devices (e.g., notebook computers). WPAN devices are truly designedformobileusage. Thetwotechnologiesdi®erinthefollowingthreefundamental ways. ² Power level and coverage In contrast with WLAN, WPAN trades the coverage for power consumption. A smallcoveragearea(about10m)allowsWPANtoreducepowerconsumption(typ- ically1mWoftransmitpower)andoperateinlowpowermodesforgreatportability. 24 Simple power-conscious and truly mobile personal devices can utilize the WPAN technology for data sharing. ² Transmission media control Devices that participate in a WPAN are designed for their personal appeal and functionality. They are not designed to be permanent members of an established networking infrastructure (even they may connect to it when it is necessary). A typical WPAN connection is set up on a demand base, and devices do not maintain a network-observable and network-controllable state. ² Network life span Unlike WLAN, the existence of a device does not mean the existence of a WPAN. ThelifespanofaWPANisalignedwiththedurationofanapplication. TheWPAN technology supports fast (i.e., in a few seconds) and ad-hoc connectivity without any pre-deployment, and connections created by mobile client devices in a WPAN are temporary in nature. 2.2.1 Piconet and Scatternet A piconet is a WPAN formed by a BT device serving as the master and one or more BT devices serving as slaves. A frequency-hopping channel based on the master's address de¯nes each piconet. All devices participating in communications in a given piconet are synchronized to the frequency-hopping channel of the piconet, using the master's clock. Slaves communicate only with their master in a point-to-point fashion under the control of the master. Master's transmissions may be point-to-point or point-to-multipoint. The 25 status of slave/mater is time variant, a slave device during one communication session can be a master in another. A scatternet is a collection of operational BT piconets overlapping in time and space. A BT device may participate in several piconets at the same time, thus allowing the possibility for the information to °ow beyond the coverage area of a single piconet. A device in a scatternet could be a slave in several piconets, but a master in only one of them. Fig. 2.6 illustrates how BT devices interconnect to form a piconet or a scatternet. Figure 2.6: Illustration of the piconet and the scatternet. 2.2.2 SCO and ACL Link The following two types of links can be established between the master and slaves: ² Synchronous Connection-Oriented (SCO) link; ² Asynchronous Connection-Less (ACL) links. The SCO link is a point-to-point link between a master and a single slave in the piconet. The master maintains the SCO link using reserved slots at regular intervals. The ACL 26 link is a point-to-multipoint link between the master and all the slaves participating in the piconet. In slots not reserved for SCO links, the master can establish an ACL link on a per-slot basis to any slave, including slaves already engaged in an SCO link. These two links are detailed below. 2.2.2.1 SCO Link The SCO link is a symmetric point-to-point link between the master and a speci¯c slave. It uses reserved slots and can be viewed as a circuit-switched connection between the master and the slave. It typically supports time-stringent information such as voice. The master can support up to three SCO links to the same or di®erent slaves. A slave can support up to three SCO links from the same master or two SCO links from two di®erent masters. SCO packets are never retransmitted. The master will send SCO packets at regular intervals, the so-called SCO interval denoted by T SCO (counted in slots), to the slave in the reserved master-to-slave slots. The SCO slave is always allowed torespondwithaSCOpacketinthefollowingslave-to-masterslotunlessadi®erentslave was addressed in the previous master-to-slave slot. If the SCO slave fails to decode the slave address in the packet header, it is still allowed to return an SCO packet in the reserved SCO slot. 2.2.2.2 ACL Link In slots not reserved for SCO links, the master can exchange packets with any slave on a per-slot basis. The ACL link provides a packet-switched connection between the masterandallactiveslavesparticipatinginthepiconet. OnlyasingleACLlinkcanexist 27 between a master and a slave. For most ACL packets, packet retransmission is used to assuredataintegrity. AslaveispermittedtoreturnanACLpacketintheslave-to-master slot if and only if it has been addressed in the preceding master-to-slave slot. If the slave fails to decode the slave address in the packet header, it is not allowed to transmit. ACL packets not addressed to a speci¯c slave are considered as broadcast packets and read by all slaves. If there is no data to be sent on the ACL link and no polling is required, no transmission will take place. 2.2.3 Packets Packets in a piconet are related to physical links. IEEE 802.15.1 de¯nes two physical links: the SCO link and the ACL link. For each of them, 12 di®erent packet types can be de¯ned, among them four control packets will be common to all link types. Here, we will review the most common and popular packet types, which will be used in our simulation. SCO packets are used in the SCO link. They do not include a cyclic redundancy check (CRC) and are never retransmitted. SCO packets are routed to the synchronous I/O (voice) port. SCO packets may carry an asynchronous (data) ¯eld as well as a synchronous (voice) ¯eld. They are typically used for 64 kb/s speech transmission. HV (High-quality Voice) packets are used for transmission of voice and transparent synchronous data. The voice packets are never retransmitted and need no CRC. There are three types of HV packets: HV1, HV2 and HV3. The parameters of the three HV packets are given in Table 2.1. For example, the most popular HV3 packet carries 3.75 ms of speech at a 64 kb/s rate and it has to be sent every six time slots (T SCO =6). 28 Packet Type T SCO Information Byte Max. Payload FEC CRC [bytes] [bytes] Encoding HV1 2 10 240 1/3 No HV3 4 20 240 2/3 No HV5 6 30 240 N/A No Table 2.1: Parameters of HV packets. ACLpacketsareusedinasynchronouslinks. Theinformationcarriedcanbeuserdata or control data. There are seven ACL packet types, six of which contain the CRC code andretransmissionisadoptedifnoacknowledgementofproperreceptionisreceived. The parameters of DH packet are listed in Table 2.2. The e±ciency of DH packets is shown in Fig. 2.7. We see that the DH5 packet has the longest payload and, therefore, highest e±ciency. There are more packet types other than HV and DH packets such as DM and DVpackets. Sincetheyarenotaspopular,wereferreaderstothestandarddocument[16] for more details. Packet Type Header Max. Payload FEC Encoding Max. Data Rate [bytes] [bytes] [Kbps] DH1 1 27 No 172.8 DH3 2 183 No 585.6 DH5 2 339 No 732.2 Table 2.2: Parameters of DH packets. 2.3 The 2.4GHz ISM Band Wi-Fi and BT devices both operate in the unlicensed 2.4GHz ISM band. Although there are many regulations that apply to operations of products in the 2.4GHz ISM band, the key requirements reviewed in this chapter are for the direct sequence spread spectrum (DSSS) of WI-Fi and the frequency hopping direct sequence (FHSS) for BT. The 2.4GHz 29 Figure 2.7: Illustration of the operation of DH 1/3/5 packets. ISMbandis83.5MHzwidewithalowerlimitof2.4GHzandanupperlimitof2.4835GHz. According to the FCC regulation, FHSS devices must hop over at least 75 channels and limit the maximum bandwidth of each hopping channel to 1MHz. BT devices hop over 79 frequencies that are 1MHz wide. Thus, a BT piconet may occupy 79MHz over time but only 1MHz at any speci¯c instance. Fig. 2.8 shows BT occupancy on the ISM band. Each Wi-Fi network maintains the same frequency usage over time and only uses a subset of the available bandwidth, i.e. 83.5MHz. The IEEE 802.11 standard de¯nes 11 possible channels, each of which is de¯ned by its central frequency. These central frequencies are spaced from one another by 5MHZ. Since any of adjacent channels are overlapping with each other, interference is expected if they are occupied concurrently. For this reason, collocated Wi-Fi networks typically operate on channels 1, 6 and 11 to prevent interference since there is no overlap between them. Fig. 2.8 shows occupancy of Wi-Fi on the ISM band. Unlike the regular pattern of slotted BT, a Wi-Fi packet transmission is typically followed by a short acknowledgement. If VCS is on, a pair of 30 Figure 2.8: Illustration of frequency occupancy of Wi-Fi and BT in 2.4GHz ISM band. RTS/CTS can also be observed. Whenever a BT packet and a Wi-Fi packet overlap in frequency and time, there is mutual interference and potential packet loss. Coexistence of Wi-Fi and BT networks is the main issue to be addressed in this research. 2.4 Cognitive Radio Most of the spectrum is allocated to various licensed services in the current regulation. Studies [6] show that over 70% of the allocated spectrum is not used at any location at any time. With little resource left and all good spectrum are occupied, obstruction on providing more access opportunity hampers the development of new wireless services. One solution to spectrum saving is open access, which operates in the open spectrum 31 or limited to low power underlay approach (e.g. UWB). Recognizing the trend, FCC has begun to modify the regulation to accommodate innovation, e.g. the IEEE 802.22 wireless regional area network (WRAN) that adopts cognitive radio (CR) functions for VHF/UHF TV spectrum sharing [18]. Cognitive radio [11] is "a radio system whose parameters are based on information in the environment external to the radio system". Such a wireless system with °exible parameters and opportunistic spectrum access is the strongest candidate for the next generation of wireless communication networks. Basically, a CR device monitors a swath of spectrum, including those occupied by the licensed services, and tries to identify the \white space" (or spectrum hole) as shown in Fig. 2.9. It is the idle period between consecutive accesses by licensed users. The CR technology exploits the opportunities withinthefrequencybandassignedtoalicensedserviceataspeci¯cgeographicallocation. In general, a CR device has to follow two basic principles. ² Limit interference to licensed users To borrow the licensed spectrum, a CR device has to guarantee the quality of services (QoS) of licensed users. Even some interference is unavoidable, it should be within the stipulated upper bound to preserve the service quality of licensed users. ² Coexistence with other CR devices The design has to coexist with other spectrum agile radios or with existing open spectrum systems whether they have coexistence mechanism or not [15]. Some 32 Figure 2.9: Illustration of spectrum holes. mutualetiquettesshouldbeestablishedtoregulatetheinteractionbetweendi®erent wireless protocols. Followingtheabovegeneralprinciples,threecriteriawereusedtomeasurethesuccess of a CR network in a recent ¯eld test [25]. ² The DSA network must do no harm to the incumbent service. When borrowing the spectrum from licensed services, DSA devices must guarantee tokeeptheexperienceofincumbentusersatthesamelevel. Topreservethequality ofservice(QoS)ofprimaryservices,DSAnodesshouldhavethecapabilitytodetect weaksignals from primaryservicesand leavethe overlappedchannel before causing 33 interference. Consequently, the channel abandonment time and the SINR of the CR system serve as performance measures accordingly. ² The DSA network must function properly. This requirement demands that a newly arrival DSA node can establish links with neighboring nodes. After abandoning a channel in use by primary services, con- nectedDSAnodesshouldbeabletore-establishconnectivitywithanotherchannel. Thus, the network join time for a newly coming CR node and the network re- establishment time for an existing CR network must be within some proper bound. AlthoughanidealCRnetworkprefersnopre-assignedfrequencyband,itisrealistic to have a centralized channel control scheme when the number of channels is large. ² The DSA network must add extra value. To maintain the e±ciency of a DSA network, we should allocate enough bandwidth totheDSAnetworkwithsu±cientspectrumholes. TheDSA-MACprotocolshould e±ciently utilize the given opportunity by keeping the white space ¯lling rate high. 34 Chapter 3 Coexistence Mechanism Using Dynamic Fragmentation for Interference Mitigation between Wi-Fi and Bluetooth A non-collaborative coexistence mechanism, called Dynamic Fragmentation (DF), for wireless-¯delity (Wi-Fi) and bluetooth (BT) systems based on dynamic packet fragmen- tationisproposedinthischapter. ThebasicideaistooptimizethepacketlengthofWi-Fi in the MAC layer such that the fragmented packet has a better chance to survive the interferencefromneighboringBTdevices. We¯rstdevelopananalyticalmodelthatspec- i¯es the information required by the Wi-Fi MAC layer to decide the best fragmentation strategy. Then, this model is extended to analyze the throughput and transmission delay of the Wi-Fi device. The analytical model is validated by computer simulation. Further- more, it is demonstrated by simulation results that the proposed coexistence mechanism improves the performance of Wi-Fi in throughput and transmission delay signi¯cantly while relatively smaller performance improvement is observed for BT. 35 3.1 Description of Dynamic Fragmentation Mechanism There are two fundamental issues that Wi-Fi has to cope with to mitigate interference in a coexistence environment. First, a Wi-Fi station cannot determine if a packet loss is due to collision or interference. The impotence of the PHY layer resolution on such an incidentlimitsthecapabilityofMACtoimprovethecoexistingperformance. Asaresult, Wi-Fi's collision avoidance mechanism (which is CSMA/CA) would treat all packet loss incidents in the same way, i.e. double the backo® window and retransmit the packet. This leads to the second issue: CSMA/CA is not e±cient in dealing with interference. Basically, CSMA/CA is designed to solve the tra±c congestion problem among Wi-Fi stations. With a longer backo® window, the tra±c is expected to average out over time and thus lower the collision rate. However, the interference rate will not be lowered by simply increasing the backo® time. Failing to lower the interference rate by increasing the backo® time, CSMA/CA is ine®ective and simply introduces unnecessary overhead. We would like to devise a new mechanism to address the problem of packet loss due to interference, which motivates the dynamic fragmentation mechanism. The basic idea is to develop an algorithm that adjusts the Wi-Fi packet length using theexistingfragmentationfunctionof802.11[1]andthelatestPERinformationtoreduce the probability of interference. In other words, legacy 802.11 MAC will be enhanced by the proposed DF algorithm so that it can handle the interference problem at run time. It is worthwhile to emphasize that the DF algorithm aims at reducing the interference rate but not the collision rate. The task of collision rate reduction is still on the shoulder of CSMA/CA. 36 There are two states in the proposed DF mechanism; namely, states 1 and 2. The entire communication payload is transmitted in one piece without fragmentation in state 1. Ontheotherhand, theentirecommunicationpayloadofasinglepacketisdividedinto ´fragments,whicharetransmittedsequentially,instate2. Thesystemcollectsthepacket errorrate(PER)informationperiodicallyina¯xedtimeinterval. Forstatetransition,we compare PER and threshold p. If the current system is in state 1 and the latest PER is higherthanp, thesystemtransitsfromstate1tostate2. Ifthecurrentsystemisinstate 2andthelatestPERislowerthanp,thenthesystemtransitsfromstate2backtostate1. For all other situations, the system remains in its original state without state transition. Determining a proper threshold value, p, is critical in the proposed DF algorithm. This can be achieved by comparing the fragmentation cost and the throughput gain, which will be elaborated later. Although the proposed DF mechanism employs two states only, to be generic, our analytical model is developed to model the transition from the state of n fragments to the state of ´n fragments such that it can be applied to other scenarios with more complicated fragmentation schemes as well. A basic fragmentation mechanism has been included in the 802.11 standard [1]. Al- thoughseldomactivated,manyWi-Fideviceshavetheabilitytodofragmentation. There- fore, only minor modi¯cation, i.e. some modi¯cations on the backo® strategy of the original mechanism, would be required to implement our DF mechanism. Fig. 3.1(a) shows the basic packet transmission of legacy 802.11 without fragmen- tation. It waits for a DIFS, keeps listening until the end of backo® window BW (or contention window) and then sends data in one piece. Finally, an ACK completes the 37 DATA hdr ACK DIFS Contention Window (a) DATA_1 ACK_1 (b) (c) SIFS DIFS SIFS DATA_2 ACK_2 SIFS SIFS DATA_1 ACK_1 DIFS SIFS DATA_1 ACK_1 SIFS DIFS Contention Window Contention Window Contention Window (d) DATA_1 ACK_1 DIFS SIFS DATA_1 ACK_1 SIFS Contention Window Figure 3.1: Successful transmission: (a) legacy 802.11 with no fragmentation (b) two fragments with no retransmission; Retransmission: (c) DF-I for mobile WLAN (d) DF-II for static WLAN transmission. Fig. 3.1(b) shows the packet transmission with the original 802.11 frag- mentation mechanism. In this example, the payload is divided into two fragments, i.e., DATA 1 and DATA 2. The ¯rst fragment is sent with the full contention mechanism. Then,afterSIFS,thesecondfragmentissentwithoutcontention. FragmentsareACKed separately. Fig. 3.1(c) shows what happens when some fragment su®ers from transmis- sion failure in the original fragmentation mechanism. Each of the failed fragments has to be retransmitted with the full contention mechanism. The bottom line is that the prior fragment has to be successfully received before any attempt of the next fragment. For 38 transmissions in the sequel, if it is not a retransmission, the contention mechanism could be saved. 3.1.1 Interference Rate Analysis To conduct the analysis, we need to develop an interference model ¯rst. Interference between Wi-Fi and BT occurs when transmissions of the two systems overlap both in frequency and time. Whenever the residual signal strength of one is higher than the SINRthresholdoftheother,itresultsinapacketloss. Dependingonthespatialrelation, transmission power, the carrier sensing threshold and channel conditions, there are three scenarios of transmission failure: BT packet loss, Wi-Fi packet loss, or both. Since the nature of collision due to BT/Wi-Fi interference is di®erent from that due to the contention of multiple Wi-Fi stations, we refer the packet loss caused by interference between BT and Wi-Fi as an interference incident and use collision exclusively for the packetlosscausedbythecontentionofWi-Fistationsthroughouttherestofthischapter. DATA hdr ACK BT time slot Wi-Fi packet transmission T w t T B T BA 22MHz 1MHz Figure 3.2: Illustration of the Wi-Fi packet transmission and the BT time slot. 39 Typically, a Wi-Fi packet has a length equal to that of several BT time slots. Let T W be the time duration of transmitting a Wi-Fi packet 1 , T B the BT time slot and T BA the active time within each BT time slot. Furthermore, we use t to denote the time interval between the beginning of a Wi-Fi packet and the beginning of the ¯rst overlapped BT time slot as shown in Fig. 3.2. The number of BT slots interfered 2 by a Wi-Fi transmission can be calculated as N = 8 > > > > > > < > > > > > > : d T W T B e¡1; if t¸T BA and t·d T W T B eT B ¡T W ; d T W T B e+1; if t<T BA and t>d T W T B eT B ¡T W ; d T W T B e; otherwise: (3.1) If the packet length in Eq. (3.1) is a random variable, we can replace T W and N by E[T W ] and E[N], respectively. The probability that a BT device hops on the frequencies that would interfere Wi-Fi transmission is denoted by P f 3 . Consider multiple collocated WPANs, where the number of WPANs is n BT . For the i-th piconet, the tra±c load (or piconet activity), denoted by G i , is the probability of a packet in a time slot. For a single WPAN and a single WLAN, we can express the PER of a Wi-Fi station as PER Wi¡Fi =1¡(1¡P f G) N ¼NP f G: (3.2) 1 TW contains both DATA and ACK and, if any portion of the duration is interfered, it will result in retransmission. 2 Since there is inactive time in a BT time slot, the number of interfered time slots is di®erent from the number of overlapped time slots de¯ned in Eq. (1) of [3]. 3 ThisprobabilityisafunctionoftheSINRratio,anditmayvarywiththedistanceandthetransmission power [12,31]. According to the interference scenario given in Sec. 3.2.1), we choose P f ¼ 22=79 under the assumption that WPANs and the WLAN are close enough such that there is interference as long as their packets are overlapped. 40 For multiple collocated WPANs, the PER becomes PER Wi¡Fi =1¡ n BT Y i=1 (1¡P f G i ) N ¼ n BT X i=1 NP f G i : (3.3) 3.1.2 Transmission Time and Threshold Analysis Following[14],forasuccessfultransmissionwithnofragmentationasshowninFig.3.1(a), the total time of a complete transmission can be written as DIFS+BW +T h +T DATA +SIFS+T ACK ; (3.4) where DIFS and SIFS are two kinds of inter-frame intervals, BW is the time of the backo® window (or contention window) and T h , T DATA and T ACK are the transmission timesfortheheader,DATAandACK ofapacket,respectively. BW isarandomnumber times the Wi-Fi time slot. IfthepacketisdividedintonfragmentsasshowninFig. 3.1(b),thetotaltransmission time with no retransmission can be expressed as DIFS+BW +T h + T DATA n +SIFS+T ACK +(n¡1)(SIFS+T h + T DATA n +SIFS+T ACK ) or DIFS¡SIFS+BW +n(SIFS+T h + T DATA n +SIFS+T ACK ) (3.5) 41 Note that T oh = T h +T ACK +2£SIFS is a ¯xed overhead for each fragment. It is easy to check that Eq. 3.4 is a special case of Eq. 3.5 when n = 1. Thus, the total transmission time, T s , for n fragments without any retransmission is equal to T s =DIFS¡SIFS+BW +n(T oh + T DATA n ) (3.6) If there is a fragment loss, a retransmission would take place such as the scenario in Fig. 3.1(c). Then, the time for a single retransmission is T r =DIFS¡SIFS+BW +T oh + T DATA n : (3.7) Finally, we could determine the total transmission time of a packet, which is divided into n fragments and su®ers a total R retransmissions. The total transmission time would be the successful transmission time of n fragments plus R times the single retransmission time, i.e. T n;R = DIFS¡SIFS+BW +n( T DATA n +T oh ) +R(DIFS¡SIFS+BW +T oh + T DATA n ) = (R+1)(DIFS¡SIFS)+ X 1+R BW i +(n+R)(T oh + T DATA n ) (3.8) Withtheaboveanalysis,wewouldliketodeterminethethresholdforstatetransition. In the proposed DF scheme, the decision on state transition depends on whether the transition could reduce the expected transmission time. If the total transmission time 42 can be reduced by a new state, the state transition will be conducted. From the state of n fragments to the next state, the number of fragments changes from n to n 0 (= ´n) and the retransmission number will change accordingly. We use R and R 0 to denote the numbers of retransmission before and after the state transition, respectively. Then, the condition for a transition to occur can be expressed as E[(R+1)(DIFS¡SIFS) + 1+R X i=1 BW i +(n+R)( T DATA n +T oh )]> E[(R 0 +1)(DIFS¡SIFS) + 1+R 0 X i=1 BW i +(n 0 +R 0 )( T DATA n 0 +T oh )] (3.9) After rearranging the terms, the inequality becomes (E[R]¡E[R 0 ])(DIFS¡SIFS)+(E[R]¡E[R 0 ]¡(´¡1)n)T oh +(E[R]¡ E[R 0 ] ´ ) T DATA n +E[ R X i=1 BW i ]¡E[ R 0 X i=1 BW i ]>0: (3.10) To calculate the threshold, we need to ¯nd all the expected values in Eq. 3.10. First, we have E[R]=E[ n X i=1 R i ]; (3.11) where R is the total number of retransmissions and E[R i ] is the expected number of retransmission for each fragment, and n times the value would be the expected value of the total number of retransmission. It is assumed that R i is geometrically distributed, 43 whose probability is in form of f R i (k) = p k (1¡p), where k is the retransmission count. Then, E[R] can be expressed as E[R]=n£E[R i ]= np 1¡p : (3.12) Next, we want to ¯nd E[R 0 ]. For terms in Eqs. 3.2 and (3.3), P f and G will remain the same after state transition, only N, which is a function of n, will change. Thus, we can obtain the ratio, PER Wi¡Fi PER 0 Wi¡Fi = p p 0 = N N 0 ; and then p 0 = N 0 N p: (3.13) Therefore, E[R 0 ]= n 0 p 0 1¡p 0 = n 0N 0 N p 1¡ N 0 N p = n 0 p N N 0 ¡p = ´np ·¡p ; (3.14) where ·= N N 0 and ´ = n 0 n . To ¯nd E[ P R i=1 BW i ] and E[ P R 0 i=1 BW i ], we have to calculate a few parameters. Let CW min and CW max be the minimum and maximum sizes of the backo® window (or contentionwindow), andconstantsaandbarede¯nedbyCW min =2 a ¡1andCW max = 2 b ¡1. For a fragment enters its kth retransmission, we have BW(k)2[0;1;2;:::;2 k+a ¡1]£T slot 44 so that the expected backo® window can be expressed as E[BW(k)]= 1 2 (2 k+a ¡1)£T slot : (3.15) For any fragment, E[BW(k)] is the average backo® of the kth retransmission. Since the expected value of total retransmissions of a packet is E[R], the expected value of total backo® of a fragment is P E[R i ] k=1 E[BW(k)]. For a packet divided into n equal length fragments, the total backo® of all fragments is E[ R X i=1 BW i ]=n£ E[R i ] X k=1 E[BW(k)]: Since the backo® window is doubled only up to the upper bound CW max , we consider the following two cases. ² Case A: E[R i ] · b¡a and the backo® window is not greater than the maximum contention window size. Then, we have E[ R X i=1 BW i ] = n£ E[R i ] X k=1 E[BW(k)]=n£ E[R i ] X k=1 1 2 (2 k+a ¡1)£T slot = 1 2 (2 a+1 (2 E[R i ] ¡1)¡E[R i ])£nT slot : 45 ² Case B: E[R i ] > b¡a and the backo® window reaches the maximum contention window size. Then, we get E[ R X i=1 BW i ] = n£ E[R i ] X k=1 E[BW(k)] = n£f b¡a X k=1 1 2 (2 a+k ¡1)+ E[R i ] X k=b¡a+1 1 2 (2 b ¡1)g£T slot = 1 2 (2 a+1 (2 b¡a ¡1)¡2 b (b¡a)+(2 b ¡1)E[R i ])£nT slot : Without loss of generality, we consider the case with E[R i ]·b¡a and E[R 0 i ]·b¡a. Then, the following expressions can be derived: E[ R X i=1 BW i ]= E[R i ] X k=1 1 2 (2 k+a ¡1)£nT slot = 1 2 (2 a+1 (2 E[R i ] ¡1)¡E[R i ])£nT slot ; (3.16) and E[ R 0 X i=1 BW i ]= E[R 0 i ] X k=1 1 2 (2 i+a ¡1)£´nT slot = 1 2 (2 a+1 (2 E[R 0 i ] ¡1)¡E[R 0 i ])£´nT slot : (3.17) Since all terms in Eq. 3.10 are now available, we could plug them in to calculate the threshold. 46 3.1.3 Throughput Analysis The throughput is the fraction of the total transmission time dedicated to payload trans- mission. Mathematically, we have throughput= T DATA E[(R+1)(DIFS¡SIFS)+ P 1+R i=1 BW i +(n+R)( T DATA n +T oh )] : (3.18) According to Eq. 3.18, one can either cut down the overhead cost or reduce the number of retransmissions to improve the throughput. There is a tradeo® between the number of retransmissions and the overhead cost in the fragmentation process. That is, the more a transmission is fragmented, the higher the overhead cost but the fewer the number of retransmissions. Whenthegainoftheretransmissionnumberreductioncannoto®setthe loss due to the increased overhead cost, fragmentation should not be performed. This explains the necessity of a careful analysis of the fragmentation cost so as to guarantee the performance improvement. We investigate the overhead cost and ¯nd an opportunity for further performance improvement below. The total overhead cost can be broken down into four parts: the header, the inter-frame space (IFS), the ACK message and the backo® window. The ¯rst three parts are ¯xed while the last one varies with PER. Furthermore, the overhead of the backo® window grows exponentially with PER. One way to cut down unnecessary overhead is to eliminate the backo® of fragment retransmissions incurred by interference. The backo® mechanism is designed mainly for reducingthe collisionchanceofretransmission. However, since it is observedthatbacko® 47 incurred by interference does not reduce the interference rate but introduce unnecessary overhead, this type of backo® should be avoided. We make the following two conjectures in order to enhance the resolution of collision and interference. Conjecture 1 If the ¯rst fragment in a sequence of multiple fragments is lost, it is most likely due to a collision. Under a simple static environment, i.e. no node mobility, collision happens only when two Wi-Fi stations randomly choose the same backo® window and they start their trans- mission at the same time. In other words, collision either happens from the beginning of a transmission or it does not happen at all. On the other hand, if an on-going BT transmissionisintheWi-FifrequencyandcloseenoughtointerfereaWi-Fitransmission, it would be detected 4 by the Wi-Fi device, which triggers backo® on the Wi-Fi transmis- sion. This phenomenon of backo® triggered by the BT transmission was shown in recent work, e.g. [37]. Although there is still small possibility that the ¯rst fragment lost is due to interference, the ¯rst fragment loss is due to collision in most time. Conjecture 2 If subsequent fragments encounter a transmission failure, it is most likely due to interference. For a normal Wi-Fi transmission in a simple static environment, there should be no collisioninthemiddleofatransmissionsession. Thescenarioisquitedi®erentwhenthere is a nearby BT device. Since BT has no carrier sensing mechanism, interference happens at any time during a Wi-Fi transmission except the ¯rst fragment for the aforementioned 4 The carrier sensing threshold is controllable and could be set to detect the harmful BT activity. 48 phenomenon in [37]. Thus, a transmission failure in the subsequent fragments is likely due to interference. When the hidden node problem occurs, the RTS=CTS mechanism has to be adopted so that the combination of RTS=CTS would be the ¯rst fragment. Based on the same reason, RTS packet lost is mainly due to collision. Once the handshake of RTS=CTS is cleared, the transmission failure of subsequent fragments is likely due to interference. Since the performance a®ected by interference is of our main concern, we follow [3] by assuming that there is no hidden node problem and the e®ect of RTS=CTS is ig- nored to simplify the analysis. Nevertheless, the outcome remains una®ected even when RTS=CTS is taken into account in our simulation study. WithoutDF,CSMA/CAalonecannotjudgewhetherapacketlossisduetoacollision or an interference. With DF, we can associate some suitable interpretation with the transmission status of a certain fragment to prevent unnecessary backo® incurred by interference. In CSMA/CA with DF, if a station enters a fragmentation state but does not observe a reduction in PER or any transmission failure in the second fragment and beyond, it is likely that the high PER is caused by collision only. Thus, it should switch back to the non-fragmentation state. Since under such a scenario, fragmentation will fail to reduce PER and decrease the throughput by introducing some unnecessary overhead. Except for the ¯rst fragment, retransmission of subsequent fragments is most likely caused by interference so that the backo® window assigned by the collision avoidance mechanism for those subsequent fragments will simply introduce unnecessary overhead and is of no use in reducing the interference probability. Consequently, we should simply 49 bypassthemasshowninFig.3.1(d). Underthisnewprocedure,whenasenderencounters a transmission failure for the second fragment or its subsequent ones, it will retransmit thefragmentimmediatelyaftertheACKtimeoutwithoutwaitingforthebacko®window. This revised scheme is called DF-II. The original scheme, called DF-I, can be used for mobile networks while DF-II can improve the performance of static networks. ForDF-II,sincecollisionmayincurtheretransmissionofthe¯rstfragment,weshould keep the backo® window for the retransmission of the ¯rst fragment. Let r be the re- transmission count of the ¯rst fragment. Then, R¡r is the retransmission count for subsequent fragments which are free from the backo® procedure. With the analogy given in Eq. 3.8, we can derive the total transmission time for successful transmission of n fragments plus R retransmissions for DF-II as T n;R;II = (DIFS¡SIFS+BW +n( T DATA n +T oh )) +r(DIFS¡SIFS+BW + T DATA n +T oh ) +(R¡r)(DIFS¡SIFS+ T DATA n +T oh ) = (R+1)(DIFS¡SIFS)+ 1+r X i=1 BW i +(n+R)( T DATA n +T oh ) (3.19) The new threshold can be found by plugging new values to Eq. 3.10. By removing unnecessary backo® windows, we can lower the fragmentation cost so that the threshold of entering the fragmentation state is lowered. WiththesystemparametersgiveninTable4.1,wecanstudytherelationshipbetween throughput and PER under di®erent schemes and determine threshold p accordingly. Three cases are examined in Fig. 3.3. They are the legacy 802.11 without fragmentation, 50 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Wi−Fi PER Wi−Fi Throughput legacy 802.11 analytical DF−I analytical DF−II analytical legacy 802.11 DF−I DF−II Figure 3.3: Throughput as a function of PER for three schemes. DF-I and DF-II with ¯xed fragmentation (´ = 2). The dashed lines in Fig. 3.3 are ana- lytical throughput results while the solid lines are obtained by computer simulation. We seeaclosematchbetweenanalyticalandsimulatedresults. Furthermore, thethroughput performance favors the case of no fragmentation in lower PER values but the case of fragmentation in higher PER values. Also, DF-II provides higher throughput than DF-I. These results corroborate our analysis. The legacy 802.11 intersects with DF-I and DF-II at PER equal to 0:4 and 0:3, respectively. It means that we should choose them as threshold values. For example, if PER is less than p II = 0:3, DF-II should stay in state 1 (no fragmentation). Otherwise, DF-II should be in state 2. It is worthwhile to mention that these threshold values are sensitive to the ¯xed transmission overhead, T oh , which accounts for the majority of the 51 fragmentation cost, and the ratio of overlapped BT time slots in one Wi-Fi fragment of the current and the previous state, which is · in Eq. 3.14. 3.1.4 Delay Analysis When a Wi-Fi packet is sent without retransmission nor fragmentation, the transmission delay for such perfect transmission comes from the requirements of the standard, such as DIFS, the contention window, SIFS, time for the ACK message and headers. However, besides these factors, there is another type of delay caused by interference, i.e. time due to retransmissions. Here, transmission delay is de¯ned as the di®erence between the actual transmission time and the transmission time for the non-fragmented payload. It is basically the extra time caused by retransmission, the fragmentation overhead and the spacing time set by the standard. Mathematically, we have delay=n£T oh +(E[R]+1)£(DIFS¡SIFS+T oh + T DATA n )+ E[R] X i=1 BW i : (3.20) Since fragmentation decreases the retransmission penalty, the proposed DF mecha- nism can decrease the delay caused by retransmissions. DF makes the state transition, i.e. performs fragmentation, only when the expected transmission time is less than that of the non-fragmentation case, thereby reducing the transmission delay. In other words, as far as the delay is concerned, DF in general outperforms scheduling-based algorithms, e.g. [3], in terms of packet transmission delay, which will be veri¯ed by computer simula- tion in the next section. 52 3.2 Computer Simulation 3.2.1 Experimental Setup Our computer simulation environment consists of one WLAN network and several pi- conets in proximity. In the open space environment, the path loss of a signal is usually modeledasthetwo-way groundmodel [30]. Thatis,thereceivingsignalpowerisinversely proportional to the fourth power of the separation distance. The SNR THRESHOLD of aWi-Fideviceissetto10[33]. SincethetransmissionpowerofaWi-Fideviceis40times of a BT node in Table 4.1, most severe interference happens when the distance between a Wi-Fi sender and its receiver is greater than 4 q 40 10 : =1:4 times of the distance between a BT sender to the receiving Wi-Fi receiver. This scenario, similar to that in [23], [10], is common in o±ces, households, airports, etc. The device separation distance in our sim- ulation is intentionally chosen to capture such interference scenario. That is, the Wi-Fi and the BT networks are so close, whenever a BT device operates in the Wi-Fi frequency band and there is a ongoing Wi-Fi transmission, both transmissions would fail. However, if BT transmits outside the Wi-Fi frequency band, concurrent transmission is possible. In other words, P f is exactly 22=79 in our simulation. Toreducethesimulationcomplexity,thefollowingassumptionshavebeenmadewith- outloss ofgenerality: (i) thepropagationdelayofWLANand WPANis neglected due to the short operational distance; (ii) WLAN is of low mobility, which is the most common scenario of Wi-Fi; (iii) piconets may be mobile, yet as compared with the wide coverage ofWi-Fi, itsmobilitywouldhavelittleimpactontheinterferencepattern; (iv)thepacket loss rate of Wi-Fi and BT due to noise is set to 0.1%. 53 The arrival rate of Wi-Fi packets is exponentially distributed. The Wi-Fi packet length is ¯xed in simulation 5 . The Wi-Fi and BT parameters used in simulation are summarized in Table 4.1. The header part in each Wi-Fi fragment consists of the PHY header (or called the preamble) and the MAC header. With these parameters, we can calculate the best ´, which is equal to two in our simulation, with respect to the assigned packet length. A higher ´ value, i.e. with more number of fragments, will not decrease · of Eq. 3.14, but just increase the overhead. Both SCO and ACL tra±c scenarios are simulated for BT. For the SCO link, the mostpopularHV3-typelinkisusedandapacketisgeneratedeverysixtimeslotsinboth directions. A BT slave can support up to three SCO links from the same master or two SCO links if the links are originated from di®erent masters. An SCO packet needs no ACK nor retransmission. If a BT slot is not reserved by SCO, the master could establish theACLlinkonperslotbasis. EachACLlinkpacketneedstobeACKedinthenexttime slot. IntheACLsimulation,theDH1-typelinkisusedsuchthatonedatapacketoccupies one BT time slot. The packet arrival rate of ACL is also exponentially distributed. With di®erent BT tra±cs, we consider three scenarios: ² Wi-Fi runs on legacy 802.11 without any fragmentation; ² Wi-Fi runs on dynamic fragmentation DF-I; ² Wi-Fi runs on dynamic fragmentation DF-II. Each plotted value shown in the ¯gures is the average result of at least 50 runs. 5 For a variable payload length, parameter max Fragment(Payload) length can be used to control the fragment length. 54 802.11 Wi-Fi Parameters Parameter Assigned Value Power 20 dBm Slot time 20 ¹s DIFS 50 ¹s SIFS 10 ¹s PHY header 192 bits MAC header 224 bits Payload 12000 bits ACK 112 bits CW min 32 CW max 1024 BlueTooth Parameters Power 4 dBm T B 625 ¹s T BA 366 ¹s Table 3.1: Parameters for Wi-Fi and BT systems. 3.2.2 Simulation Results and Discussion To verify our analysis, we ¯rst see that the state transition threshold obtained from the analytical model is close to that obtained by simulation as shown in Fig. 3.3. Based on Secs. 3.1.2 and 3.1.3, the threshold values for DF-I and DF-II are 0.38 and 0.31, respectively. They are also con¯rmed in Fig. 3.3. Furthermore, the relation between the PER of Wi-Fi and the BT tra±c load is shown in Fig. 3.4 which veri¯es Eq. 3.2. It is clear that simulation curves are consistent with our analytical prediction. These results demonstrate the accuracy of our analytical model, which can capture the interference phenomenon between Wi-Fi and BT well. Fig. 3.5 shows the throughput improvement of Wi-Fi with DF in the log scale. In our calculation, the Wi-Fi throughput is de¯ned in Eq. 3.18, which is the ratio of time dedicated to payload to the total transmission time, including the time spent on retrans- missions. When the PER of Wi-Fi is equal to 0.5 and 0.6, the throughput improvement 55 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 BT Traffic Load Wi−Fi PER legacy 802.11 analytical DF analytical legacy 802.11 DF−I DF−II Figure 3.4: The Wi-Fi PER as a function of the BT tra±c load.(Validation of Eq. 3.2) of DF-I is equal to 15% and 30% respectively. For DF-II, the throughput improvement becomes 28% and 56%, respectively. One noticeable trend is that the throughput im- provement grows exponentially with PER. On the other hand, one might think that we might have negative throughput improvement for low PER since the gain in reducing retransmissions caused by interference could be lower than the induced overhead. How- ever, since our algorithm is dynamically adjusted based on the PER level, no negative improvement would actually happen. That is, when PER is below the threshold, there will be no fragmentation. Fig. 3.6 shows the Wi-Fi throughput as a function of di®erent Wi-Fi tra±c loads when there are two SCO links in presence. Since our mechanism is neither collaborative nor scheduling, our mechanism will not have an extra bene¯t from the recursive nature 56 0.4 0.45 0.5 0.55 0.6 0.65 0.7 10 −2 10 −1 10 0 Wi−Fi PER Wi−Fi Throughput Improvement (LOG) DF−I analytical DF−II analytical DF−I DF−II Figure 3.5: Throughput improvement for Wi-Fi with DF-I and DF-II. of SCO tra±c. Two SCO links present a substantial tra±c load on the BT side. Thus, the DF mechanism outperforms legacy 802.11 even with a low Wi-Fi tra±c load. Simulation results of the Wi-Fi throughput as a function of the BT ACL tra±c load, with two background Wi-Fi tra±c loads, are shown in Figs. 3.7 and 3.8. We see a signif- icant improvement on the Wi-Fi throughput in Fig. 3.8 and the improvement increases exponentially with the BT tra±c load. The crossing points between DF-I/DF-II and legacy 802.11 represent appropriate thresholds. This threshold is a function of the back- ground Wi-Fi tra±c load. Intuitively, a higher background tra±c load will trigger the state transition earlier. Fig. 3.7 gives the result of the same simulation setup except for a lower background Wi-Fi tra±c load. We see a higher threshold, which is consistent with our intuition. 57 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.35 0.36 0.37 0.38 0.39 0.4 0.41 0.42 0.43 0.44 0.45 Wi−Fi Traffic Load Wi−Fi Throughput legacy 802.11 DF−I DF−II Figure 3.6: Throughput of the Wi-Fi network, which coexists with two SCO links on the BT piconet. Fig. 3.9 shows a slight improvement on the BT throughput. The curves of DF-I, DF-II and legacy 802.11 are close to each other when the Wi-Fi tra±c load is low. We see more signi¯cant performance improvement when the Wi-Fi tra±c load is su±ciently high. Even though the improvement is not impressive at the BT side, our scheme has no negativeimpactontheBTperformanceatanyevent. NotethattheproposedDFscheme can signi¯cantly decrease the cost of Wi-Fi retransmission, but cannot prevent any Wi- Fi transmission failure. For the same reason, it can not prevent any BT transmission failure. The small improvement at the BT side comes from less interference due to shorter retransmission induced by DF. Thus, the improvement is more substantial when the retransmission time takes up a signi¯cant portion of the total transmission time, which is always the case when the tra±c load is su±ciently high. 58 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 BT Traffic Load Wi−Fi Throughput legacy 802.11, τ Wi−Fi =0.3 DF−I, τ Wi−Fi =0.3 DF−II, τ Wi−Fi =0.3 Figure 3.7: Throughput of the Wi-Fi network in the presence of ACL link on the BT piconet, where the background Wi-Fi tra±c is given by ¿ Wi¡Fi =0:3. With two BT piconets that adopt ACL tra±c and coexist with a Wi-Fi network, we show simulation results of the Wi-Fi throughput with respect to the combined BT tra±c load in Fig. 3.10. We see that Figs. 3.8 and 3.10 are very similar. This implies that, regardless of the number of coexisting BT piconets, the degree of interference sensed by Wi-Fi is the cumulative contributions of all coexisting piconets, which is consistent with Eq. 3.3. Finally, Fig. 3.11 shows the average packet transmission delay of Wi-Fi as a function ofthe BTtra±c load. Inthe simulation, we¯rstcalculate the totaltransmission time for ¯lesof¯xedlength, e.g. 1000packets. Then,wesubtractthetimeneededtotransmitthe payload from the total transmission time to determine the total delay. Then, the total 59 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 BT Traffic Load Wi−Fi Throughput legacy 802.11, τ Wi−Fi =0.7 DF−I, τ Wi−Fi =0.7 DF−II, τ Wi−Fi =0.7 Figure 3.8: Throughput of the Wi-Fi network in the presence of ACL link on the BT piconet, where the background Wi-Fi tra±c is given by ¿ Wi¡Fi =0:7. delayis dividedbythenumberof packetsto gettheaveragedelayof a singlepacket. The improvementonpackettransmissiondelaycomes primarilyfromreduced retransmissions caused by interference. It is demonstrated by simulation results that DF can improve the transmission delay signi¯cantly. 3.3 Conclusions A non-collaborative mechanism, called Dynamic Fragmentation (DF), for improving the Wi-Fi performance in the presence of BT interference was proposed in this work. We ¯rst developed an analytical model to characterize the interference between Wi-Fi and BT. Then, we proposed DF to reinforce the coexistence ability of Wi-Fi networks. With 60 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.35 0.4 0.45 0.5 0.55 0.6 Wi−Fi Traffic Load BT Throughtput legacy 802.11, τ BT =0.7 DF−I, τ BT =0.7 DF−II, τ BT =0.7 Figure 3.9: Throughput of the BT piconet with ACL tra±c ¿ BT = 0:7, which coexists with Wi-Fi. DF, a Wi-Fi station can perform fragmentation dynamically to reduce interference and increase throughput. We also investigated the scenario of static networks and proposed an enhanced solution called DF-II to improve the performance furthermore. In addition to the throughput improvement, DF helps Wi-Fi di®erentiate between interference and collision. Thus, if transmission failures are mostly caused by collision at the ¯rst frag- ment, DF can recognize the di®erence and swiftly switch back to the non-fragmentation state toavoidunnecessary overhead. Thefragmentationmechanismhas already been de- scribed in the legacy 802.11 standard. Therefore, our proposed DF-I and DF-II solutions can be easily implemented. The derived analytical results were validated by simulation results. WiththePERlevelhigherthan0.6,wecouldgetmorethan56%improvementon throughput for static networks, and 30% throughput improvement for mobile networks. 61 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 Combined BT Traffic Load Wi−Fi Throughput legacy 802.11, τ Wi−Fi =0.7 DF−I, τ Wi−Fi =0.7 DF−II, τ Wi−Fi =0.7 Figure 3.10: Throughput of the Wi-Fi network in the presence of ACL links on multiple BT piconets, where the background Wi-Fi tra±c is given by ¿ Wi¡Fi =0:7. 62 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.5 1 1.5 2 2.5 3 BT Traffic Load Average Wi−Fi Packet Delay [ms] legacy 802.11, τ Wi−Fi =0.5 DF−I, τ Wi−Fi =0.5 DF−II, τ Wi−Fi =0.5 Figure 3.11: The average Wi-Fi delay as a function of the BT tra±c load, where the background Wi-Fi tra±c is given by ¿ Wi¡Fi =0:5. 63 The improvements grow exponentially with PER. Substantial improvement on the delay hasbeenobservedaswell. Constrainedbythenon-collaborativenature,theimprovement on BT networks is not as good as that on Wi-Fi, but still has slight improvement. The current DF mechanism divides the packet into fragments with equal length. It is interestingtodevelopanextendedmodelusingamorecomplicatedfragmentationscheme. Our analytical model can provide clues in the design of such an advanced fragmentation mechanism. Besides, owing to the explosive growth in the number of wireless devices, e®ortsonregulation,standards,etc.,areneededtoenhancethecoexistenceofalldi®erent kinds of wireless networks. This work provides important insights into the coexistence between Wi-Fi and BT networks, which can bene¯t future wireless technologies in the face of a crowded unlicensed spectrum. 64 Chapter 4 Adaptive Hopset Frequency Hopping for Wireless Personal Area Networks in a Coexistence Environment An enhanced frequency hopping (FH) mechanism, called the Adaptive Hopset Frequency Hopping (AHFH), is proposed in this chapter. AHFH aims to improve the performance of frequency hopping-based wireless personal area networks (WPANs). According to the packet error rate (PER) of each channel, AHFH trims the hopset of a piconet and then determines the proper packet length for active channels. The decision is made by each WPAN in a distributive manner with an objective to optimize the overall performance of piconets in a coexistence environment. An analytical model is developed to justify the occupancy and throughput performance of AHFH. Simulations are conducted under an environment that consists of some collocated Bluetooth (BT) piconets and a Wi-Fi networktovalidatetheanalysisandthesuperiorityofAHFH.Ascomparedwithexisting solutions, AHFH can provide much higher throughput while maintaining good channel occupancy. 65 4.1 Adaptive Hopset Frequency Hopping (AHFH) 4.1.1 Etiquette Rules Orthogonal hopset protocols reduce the hopset size to prevent frequency-dynamic (FD) interference and, consequently, the tra±c load in remaining channels increases signi¯- cantly. The interference to other unlicensed-band (UB) systems will increase, too. Low- power networks, e.g. 802.15.4/ZigBee, will face the most severe consequence. Although they are most likely imperceivable to BT, BT devices are still responsible to mitigate the potential starvation. To address this issue, etiquette rules are needed to guarantee fair sharing of unlicensed bands among di®erent wireless systems with various radio power and protocols. It is however a challenging task to develop good etiquette rules. Since a BT network does not know whether any of its channels is overlapped with anotherlow-powernetwork,thebeste®ortstrategyistodistributetheirtra±ctoavailable channels evenly, which is exactly what the original FH protocol does. In other words, the occupancy of any enhanced FH protocol should be close to the average occupancy as much as possible. Under the same occupancy level, it is desirable that the enhanced protocol should have higher performance without deteriorating the performance of the low-power system. Fig.4.1illustratestheoccupancydistributionoverallocatedchannelsofthreesystems. In all three sub¯gures, piconet ¼ 1 is fully loaded and the activity (or the tra±c load as indicated by the gray color) of ¼ 1 , denoted by G 1 , is equal to one. Besides ¼ 1 , there are three other collocated piconets of various tra±c load. Fig. 4.1(a) shows the occupancy distribution of piconets with legacy FH, where the total activity of collocating piconets is 66 (a) Occupancy Channel 1 M 1 1/M (b) 1 N/M M/N (c) 1 Figure 4.1: Illustration of the channel occupancy of (a) the legacy frequency hopping system, (b) the orthogonal hopset hopping system and (c) the AHFH system. evenly distributed among all channels. Thus, the occupancy is the same for all channels and the average occupancy at any channel is G 1 M = 1 M . Legacy FH has the best load balancing performance, which results in the least impact to low-power systems running under the shadow of BT. Fig. 4.1(b) illustrates the occupancy of an orthogonal hopset (OH) hopping system (e.g. DAFH [29]) that distributes collocated piconets to di®erent orthogonal hopsets. Each subset is occupied by one piconet exclusively. Since activities of di®erent piconets vary, the occupancy of orthogonal hopsets varies as well. The performance of the low- power system can be severely a®ected in the worst case. ThethirdscenarioisdepictedinFig.4.1(c),wheretheoccupancyofdi®erentchannels maynotbeequaltotheaverageoccupancyasshowninFig.4.1(a). However,itiscloserto theaveragethantheorthogonalhopsetsystem. TheproposedAHFHschemecanachieve this performance by adjusting the packet length and the hopset together. Although the occupancy of each channel varies, it does not deviate too much from the average. 67 4.1.2 Carrier Sensing Mechanism Both the hopset size and the packet length can be adjusted with the AHFH protocol. Generally speaking, the transmission a multiple-slot packet is more e±cient. As shown in Table 4.1, long DH3 packets can provide much higher throughput than DH1 pack- ets. However, long packets su®er some side e®ects. That is, both the PER and the retransmission cost tend to increase. A simple carrier sensing (CS) mechanism, e.g. the Listen-Before-Talk (LBT) in [37], can mitigate these unwanted e®ects. T slot T active Header Payload Throughput (625¹s) (¹s) (byte) (byte) DH1 1 366 1 27 0.56 DH3 3 1616 2 183 0.85 Table 4.1: Parameters of BT piconets. For any BT transmission, there is at least 250¹s before the next transmission. This turn-around time is reserved for the Rx/Tx transition and the synchronization of the next channel frequency. This time interval is long enough to allow carrier sensing before the next transmission. With the CS mechanism, each piconet can reach the next channel 50¹s earlier, which is called the carrier sensing window. If the selected channel is busy or becomes busy during the interval, the piconet will cancel the transmission attempt, skip the current channel and hop to another channel for the next chance. In the original FH scheme as given in Fig. 4.2(a), the interference will result in two corrupted packets and retransmissions. In the modi¯ed FH with the CS mechanism as given in Fig. 4.2(b), piconet 2 senses the ongoing transmission, hops to channel j and attemptstotransmitinthenexttimeslot. Ifthecarriersensingmechanismrevealsaclear 68 Piconet 1 Piconet 2 Channel i (a) Piconet 1 Piconet 2 Channel i (b) Channel j Carrier Sensing Window BT time slot Figure 4.2: Illustration of the carrier sensing mechanism: (a) the original FH and (b) the modi¯ed FH with the carrier sensing mechanism. channel j, piconet 2 begins its transmission. This example shows how the CS mechanism prevents the interference and the expensive retransmission of long DH3 packets. 4.1.3 Description of AHFH Algorithm In this section, we present an enhanced FH protocol called the adaptive hopset frequency hopping (AHFH) scheme. AHFH uses packet length control and hopset adaptation to mitigate the interference. Exceptforchannelsthatoverlapwithfrequency-static(FS)interferersrepresentedby group S, remaining channels can be classi¯ed into three groups. We use A to represent the group using 3-slot packets 1 , B the group using single-slot packet, and C represents 1 Weconsideronly3-slotpacketsasthemulti-slotpacketstosimplythediscussioninourcurrentstudy. However, it can be extended to include 5-slot packets, in which the fourth group will be needed. 69 the group of inactive channels. The sizes of these groups are denoted byjAj,jBj andjCj, and they add up to the total channel number, denoted as M. The AHFH mechanism is triggered by PER. Each piconet monitors not only the overall PER denoted by PER 2 but also the individual PER of each channel, denoted by per(m)forchannelm. ParameterPERprovidesanestimateonthenumberofcollocated piconets denoted by N 3 . N along with another parameter ® determines the number of channels to be assigned to the multiple-slot packet group A. The AHFH decision is conducted on per channel basis. The detailed mechanism of channel assignment is described below. ² Mechanism to mitigate FD interference When a piconet recognizes that the interference comes from other collocated pi- conets, it triggers the adaptation of the hopset size and the packet length. Among allchannels,AHFHrandomlyassigns®£N channelstogroupA. Atthesametime, it randomly deactivates 2jAj channels from the hopset to group C (jCj = 2jAj). Other channels remain in group B (jBj=M¡3jAj). After the initial assignment, AHFH keeps supervising the PER of each channel,i.e. per, and periodically update the member of each group. When a channel in group A hasper that is much higher than PER, AHFH swaps it with a channel with lower per in group C or B. Swap- ping between groups B and C is also possible. This channel swapping process can correct the possible misalignment in the initial assignment and re-arrange group members according to latest observations. 2 This overall PER is calculated by those channels with FD interference only. 3 Since N is not known to a piconet, it is estimated via PER£M. 70 ² Mechanism to mitigate FS interference When a piconet senses that some channels constantly su®er severe interference, i.e. per >0:5, it recognizes that there is a FS interferer nearby and temporarily remove those channels from its hopset for the next update cycle (T). These channels are assigned to group S and are excluded from the channel swap mechanism. After being inactive for the while cycle, members of group S can be activated again to detect the absence of the FS interferer. Parameter ® controls the orthogonality of the hopset. Its selection depends on the number of coexisting piconets. When ® is larger, the AHFH mechanism shrinks the hopset more and the outcome would be similar to that of the OH protocol. On the other hand, smaller ® preserves the hopset and make the outcome more close to FH. With di®erent packet lengths in di®erent channels, AHFH makes FD interference more predictable in channels with multiple-slot piconets. Thus, a piconet can avoid severely interfered channels (i.e. channels already occupied by other piconets using multiple-slot packets) and improve the chance of successful transmission. To update the active/inactive status of a channel, a protocol has to exchange the control packet explicitly, i.e. the master of a piconet has to inform all slaves. Since the BT technology allows at most 7 slaves, a total of 14 time slots (a packet plus an ACK per slave) is required to synchronize the hopset information. The packet length update information requires little overhead, since it can be implicit and carried by the packet header. The hopset status can be updated at the beginning of a cycle, and the BT master can usetheabovemechanismforimmediateremedy. Forexample, ifamasterwantstoskipa 71 newlyfoundbadchannel,theeasiestwayistodelaythetransmissiontothenextchannel. The master can also use a look-ahead bu®er to avoid delay. For example, if a newly found bad channel is on the horizon, the master can use a multiple-slot packet to skip the bad channel, which is similar to the D-OLA mechanism in [3]. Controlling the packet length can relieve the pressure of keeping the hopset information up-to-date. However, if the hopset is truly out-of-date, the packet length mechanism may incur substantial overhead, e.g. the master may need to delay a transmission frequently. Then, an explicit synchronization mechanism may be a better choice. 4.2 Theoretical Analysis We develop an analytical model to analyze throughput and occupancy under the coex- istence environment, where BT devices can use any packet length adjustment or hopset manipulation technique, in this section. Our model provides insights into FD and FS interference phenomena, and captures essential properties of a coexistence environment so that throughput and frequency occupancy can be observed and calculated. 4.2.1 System Models Let M be the total channel number and N be the number of piconets. We use ¼ i to denote the i th piconet, where i = 1;2;¢¢¢ ;N, and m to denote the channel index. G i representsthetra±cload(oractivity)ofpiconet ¼ i . Intuitively, theincurredinterference of a piconet is proportional to G i . 72 The hopset of any piconet is a subset of the full set of spectrum allocated to WPAN. The hopset of ¼ i can be represented by the following utilization vector: u i =[u i (1); u i (2);¢¢¢ ; u i (M)]; (4.1) where u i (m) 2 f0;1g for m 2 f1;2;:::;Mg. The vector entry u i (m) = 0 means that frequency channel m is excluded from the hopset. Otherwise, channel m is active. The size of the hopset can be expressed as M i =ku i k= M X m=1 u i (m)·M: (4.2) Sinceapiconetchoosesachannelinapseudorandomfashionfromitshopsetwithanequal probability. The probability for piconet ¼ i to select an active channel m is p i (m) = 1 M i and the probability vector is p i =[p i (1); p i (2);¢¢¢ ; p i (M)]= 1 M i u i ; (4.3) which means that all channels that have nonzero utilization share the same probability of being selected. Although all channels have the same probability of being selected, a piconet will stay in channels with multiple-slot packets more often than those use single-slot packets. In 73 other words, the packet length parameter alters the probability that a channel is utilized. Let the average packet length of ¼ i in all channels be in form of l i =[l i (1); l i (2);¢¢¢ ; l i (M)]: (4.4) The assigned packet length for a piconet in a channel could be zero (channel excluded from the hopset), one (using single-slot packets, e.g. DH1) or three (using multiple-slot packets, e.g. DH3). Since the packet length of a channel is dynamically adjusted in AHFH,thepacketlengthisrepresentedbytheaveragevalueoveraperiodoftime. Thus, the average packet length l i (m) could be any real number between 0 and 3. Due to the in°uence of a variable packet length, we can ¯nd the e®ective utilization probability of channel m of ¼ i as ~ p i (m) = p i (m)l i (m) p i ¢l i ; (4.5) where p i ¢l i = M X m=1 p i (m)l i (m): ~ p i is the vector representation of the e®ective channel utilization probability. If all chan- nels share the same l i (m), ~ p i would be identical to p i . 4.2.2 Throughput Analysis Besides throughput improvement, AHFH should satisfy the etiquette rule that demands the occupancy of all channels to be close to the average as much as possible. Instead 74 of distributing activities evenly over all channels, AHFH distributes more activities to channels of lower utilization, fewer activities to those of higher utilization. As a result, theaggregatedoccupancyofcollocatedpiconetsineachindividualchannelisexpectedto be close to the average with small variation. Since all collocated piconets are expected to have the same size of the three groups, we argue that the fraction of piconets in di®erent groups in a channel is equal to the fraction of channels in di®erent groups of a piconet. Let ¹ be the ratio of N M . On a given channel, we would observe that among N piconets, ¹jAj piconets assign this channel to group A, ¹jBj piconets assign this channel to group B, and the remaining piconets exclude this channel from their hopset. We can analyze the throughput based on this observation accordingly. For a time slot in channel m, we ¯rst compute the probability that there is a trans- mission, which is denoted by P tr (m). Since it is the probability that at least one of N piconets intends to transmit in this channel using a single- or multiple-slot packet, we have P tr (m)=1¡ ¹jAj Y i=1 (1¡G i ~ p i (m)) ¹jBj Y j=1 (1¡G j ~ p j (m)): (4.6) If the time slot is not idle, it can be occupied by one or more piconets using single- or multiple-slot packets. We are interested in P A tr (m) which is the probability that at least one piconet using multiple-slot packets transmit at the time slot. It can be written as P A tr (m)=1¡ ¹jAj Y i=1 (1¡G i ~ p i (m)): (4.7) To successfully transmit a packet in a channel occupied by multiple piconets, the channel has to be idle for a certain period, which is called the necessary opening for a 75 DH1 732 DH1 DH3 DH3 3232 1616 366 DH1 DH3 1982 DH3 DH1 1982 DH1 416 DH3 1666 50 (a) (b) 50 Carrier Sensing Window Figure4.3: Illustrationofthenecessaryopening[¹s]: (a)theoriginalDH1/3packetsand (b) the DH1/3 packets with carrier sensing. packet or Á. Fig. 4.3 compares the necessary opening for a packet in two scenarios. Since BT transmission is slotted, we normalize the duration by time slots and denote a DH1 packet facing interference from a DH3 packet by Á 3!1 . As shown in Fig. 4.3(a), we have Á 1!1 = 732 625 =1:17; Á 1!3 =Á 3!1 = 1982 625 =3:17; Á 3!3 = 3232 625 =5:17: (4.8) 76 If the carrier sensing mechanism is used, the necessary opening is signi¯cantly reduced. As shown in Fig. 4.3(b), we obtain Á cs 1!1 =Á cs 1!3 =0:167; Á cs 3!1 =Á cs 3!3 =2:67: (4.9) The successful rate of a transmission using the single-slot packet, denoted by P B s (m), is the chance that only one piconet using single-slot packets transmits at the time slot, while all other piconets are idle. The successful rate of a transmission using multiple-slot packets, denoted by P A s (m), is the probability that only one piconet using multiple-slot packetstransmitsatthetimeslotwhileallothercollocatedpiconetsareidleforthewhole transmission period. They can be derived as P A s (m) = ¹jBj Y i=1 (1¡G i ~ p i (m)) Á 1!3 £ ¹jAj X j=1 fG j ~ p j (m) ¹jAj Y k=1;k6=j (1¡G k ~ p k (m)) Á 3!3 g (4.10) P B s (m) = ¹jAj Y i=1 (1¡G i ~ p i (m)) Á 3!1 £ ¹jBj X j=1 fG j ~ p j (m) ¹jBj Y k=1;k6=j (1¡G k ~ p k (m)) Á 1!1 g (4.11) ForBTdevices,thefailedtransmissionstilltakesthesameamountoftimeasthesuc- cessful transmission. Finally, the throughput of channel m is the time fraction dedicated to the payload during a transmission. It can be found by £(m)= P A s T A p +P B s T B p P A tr (3T slot )+(P tr ¡P A tr )T slot +T oh ; (4.12) 77 whereT slot isthedurationofaBTtimeslot,T oh istheoverheadrequiredforthecontinual hopset updates (including the overhead of retransmission and header), and T A p and T B p are transmission times of the payload for 3-slot and 1-slot packets, respectively. Eq. 4.12 expresses the throughput of a certain channel, which is also the overall throughput of collocated piconets since all channels are the same under AHFH. For a speci¯c piconet ¼ i , the throughput µ i is the inner product ~ p i ¢£, where £ is the vector presentation of throughput. 4.2.3 Occupancy Analysis The maximum aggregated activity level across all allocated frequency is called the fre- quency occupancy, which can be written as Occ= max m2M f N X i=1 G i ~ p i (m)g: (4.13) Generally speaking, Occ is proportional to the maximum interference that a device has in the worst case scenario. To give an example, the frequency occupancy of a device using the OH mechanism in Fig.4.1(b)is N M . TheoccupancyofDAFH[29], whichisaspecialcaseoftheOHscheme, can be written as Occ DAFH =max i2N f G i M i g= 2 dlog 2 Ne (max i2N fG i g) M ¼ N(max i2N fG i g) M : (4.14) 78 As mentioned in Sec. 4.1.1, FH distribute activities evenly to all channels, and its occu- pancy can be expressed as Occ FH = P N i=1 G i M (4.15) Finally, we have Occ AHFH ¼Occ FH (4.16) Based on the above discussion, we conclude that OH has higher occupancy than AHFH and, consequently, AHFH is more friendly than OH as far as coexistence is concerned. 4.3 Simulation Results 4.3.1 Simulation Setup Computer simulations are conducted with either the frequency-dynamic (FD) or the frequency-static(FS)interference. Weconsideracertainnumberofpiconetsatacrowded spot (e.g. airport lounge, conference room, etc.) with or without an FS interferer. In this simulation, interference is always mutually destructive, no matter whether it occurs among piconets or between piconets and the FS interferer. Table 4.1 excerpts the structuresandthroughputofDH1andDH3packetsfromIEEE802.15standard[16]. The much improved throughput of DH3 packets over DH1 packets contributes to the higher performance of AHFH over other mechanisms. To compute the occupancy, we record maximum utilization among all channels periodically, and then calculate the occupancy as an average of these maximal values. 79 We compare the performance of FH, OH and AHFH. Allocated channels in OH are partitioned into at most 5 orthogonal subsets. While the number of collocated piconets is not greater than ¯ve, piconets using OH are free from FD interference within their orthogonal subsets. However, once the number of collocated piconets exceeds ¯ve, more thanonepiconetmaybefoundinanorthogonalsubset. Then,thosesubsetswithmultiple piconets are no longer free from the FD interference. At the beginning of each run, the number of piconets is assigned and their activity levels, G i , are randomly selected. G i is uniformly distributed in interval [0:0;1:0]. After that, the simulation runs until reach 6,000,000 [slots] (around 60 minutes). The master of piconets updates the hopset of its slaves every 3,000 [slots]. In the following sections, each plotted value is the average of 30 trials. 4.3.2 Impact of Coe±cient ® Fig.4.4showsthethroughputofAHFHwithdi®erent®values. When®increases,AHFH assigns more channels to the multiple-slot group and excludes more channels from the hopset. Since more channels are assigned to the multiple-slot group, more multiple-slot transmission results in higher throughput. This explains the fact that, when the num- ber of collocated piconets is small, large ® could increase the throughput signi¯cantly. However, when the number of collocated piconets becomes larger, the chance of over- lapping multiple-slot channels of di®erent piconets becomes higher and the throughput deteriorates rapidly. Fig. 4.5 shows the occupancy of AHFH with various ®. When the number of piconets increases, the larger the ® value, the more rapidly the occupancy of AHFH increases. At 80 2 4 6 8 10 12 14 16 18 20 22 0.45 0.5 0.55 Number of piconets Throughput Throughput vs. Number of Piconets with various α AH (α=0.6) AH (α=0.8) AH (α=1.0) AH (α=1.2) AH (α=1.4) Figure 4.4: Comparison of throughput for FH, OH and AHFH with di®erent ® values. N =14, the occupancy of ®=1:2 and 1.4 exceed the occupancy of OH, which shows the severe adverse e®ect of opportunistic misalignment of multiple-slot channels for a large value of N. The occupancy behavior of AHFH is similar to that of OH with larger ® and to FH with small ®. To improve throughput and reduce occupancy, we choose ® = 1 in our simulations in the following sections. 4.3.3 Impact of Carrier Sensing Mechanism Fig. 4.6 shows the impact of the carrier sensing mechanism. For a small number of col- located piconets, the chance of interference is small and the application of the carrier sensing mechanism actually worsens the throughput performance due to the extra over- headrequired(e.g. thecarriersensingwindow). Webegintoseethebene¯tofthecarrier 81 2 4 6 8 10 12 14 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Number of piconets Occupancy Occupancy vs. Number of Piconets with various α OH FH AH (α=0.6) AH (α=0.8) AH (α=1.0) AH (α=1.2) AH (α=1.4) Figure 4.5: Comparison of occupancy for FH, OH and AHFH with di®erent ® values. sensingmechanismafterthenumberofpiconetsexceeds11. Byconsideringtheoperation range in our simulation, we do not use the carrier sensing mechanism for the FD and the FS interference scenarios. The interference increases as the number of piconets increases, and the performance gap between carrier-sensing-enhanced AHFH and basic AHFH be- comes more signi¯cant. The result suggests that the number of piconets a coexistence environment can sustain could be doubled with a proper designed carrier sensing mech- anism. This also reveals the necessity of any kind of carrier sensing or listen-before-talk mechanism for WPAN intended to operate in an heavy interfered environment. 82 5 10 15 20 25 30 35 40 0.3 0.35 0.4 0.45 0.5 0.55 0.6 Throughput vs. Number of Piconets Number of piconets Throughput AH AH with CS Figure 4.6: Illustration of the impact of the carrier sensing mechanism on throughput. 4.3.4 Performance Comparison under FD Interference Environment Fig. 4.7 shows the throughput as a function of the number of collocated piconets under an environment with FD interference only. The performance of original FH is severely degradedasthenumberofcollocatedpiconetsincreases. Thisisduetothefactthateach piconet uses all the allocated channels so that the chance that piconets will experience interference increases almost linearly with the number of piconets. OH has better perfor- mance than FH. We see that the degradation of OH starts to appear when the number of piconets reaches 5, which is the number of orthogonal subsets. Once all orthogonal sub- sets are occupied, a newly arriving piconet has to share an orthogonal subset with others and FD interference would occur in those shared orthogonal hopsets. AHFH adapts its 83 hopset so as to create a partially orthogonal hopset along with the adoption of highly e±cient multiple-slot packets for transmission. 2 4 6 8 10 12 14 0.51 0.52 0.53 0.54 0.55 0.56 0.57 0.58 0.59 Throughput vs. Number of Piconets Number of piconets Throughput AH theorectical AH OH FH Figure 4.7: Illustration of throughput vs. number of piconets under frequency dynamic interference. For a small number of collocated piconets, the throughput actually increases as the number of piconets goes up. This is because that, as the number of piconets increases, more channels are assigned to highly e±cient multiple-slot packets. Besides, the chance of overlapping multiple-slot channels is small since the number of collocated piconets is small. After the throughput reaches the peak value at a certain number of collocating piconets, i.e. ¯ve for AHFH in Fig. 4.7, interference due to overlapping multiple-slot channels among piconets begins to take a role. This °exibility gives AHFH an edge over benchmark mechanisms. The performance gain of AHFH comes from using highly e±cientmulti-slotpacketsonchannelswithlittleFDinterference. Finally,thetheoretical 84 curve of AHFH is plotted for comparison. It is calculated by ignoring the overhead of dynamic channel swapping before reaching the steady state (optimal alignment among piconets) and the overhead of continual hopset updates (i.e. setting T oh in Eq. 4.12 to 0). These overheads contribute to the gap between theoretical and realistic results. Fig.4.8showstherelationbetweenthefrequencyoccupancyandthenumberofcollo- cated piconets. We compare FH, OH and AHFH in an environment with FD interference only. FH has the best occupancy performance because it uses all the available channels. AHFH is designed to improve throughput while following the etiquette rules by keeping the occupancy pattern similar to FH. The increased occupancy is caused by the scenario that multiple piconets choose to transmit multiple-slot packets in the same channel. The channel swap process is used to compensate such an occurrence. OH does not average the tra±c load across orthogonal subsets and may result in locally concentrated occu- pancy. Thus, uneven occupancy is common with OH and it has the worst occupancy performance. 4.3.5 PerformanceComparisonunderFDandFSInterferenceEnvironments Fig. 4.9 shows the throughput of AHFH in an environment with both FD and FS inter- ference. The FS interferer chosen is the IEEE 802.11b DSSS WLAN. The Wi-Fi network occupies exactly 22 channels and transmitted ¯xed length packets. A single Wi-Fi trans- missiontimeisaround3BTtimelotslong. Theactivitylevelissetto0.7. Inotherwords, those22channelsareseizedbytheFSinterfererfor70%ofthetime. Inoursimulation,it is assumed that the Wi-Fi transmission will not backo® by the BT transmission. AHFH, AFH, FH, and OH mechanisms are compared. The linear degradation of FH is similar 85 2 4 6 8 10 12 14 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Number of piconets Occupancy Occupancy vs. Number of Piconets OH AH FH Figure4.8: Comparisonoffrequencyoccupancyasafunctionofnumberofpiconetsunder an environment with FD interference. to the curve in the simulation with FD interference only. AFH signi¯cantly improves the performanceofFH.Theperformancegapisachievedbyturningo®thoseheavyinterfered channels. The bene¯t of preventing packet losses and retransmissions on Wi-Fi channels outweighs the loss from the available bandwidth decrement. Both AHFH and OH in- herit the simple channel turn-o® technique. For OH, the number of orthogonal subsets decreases from 5 to 4 since one subsets is completely taken by the FS interferer. The piconetsallocatedtoremainingorthogonalsubsetshavelessFDinterferenceascompared with AFH or FH. Finally, AHFH further improves the throughput of AFH by adopting multiple-slot packets and partially orthogonal hopsets. Fig.4.10showstheoccupancyofdi®erentmechanismsunderFDandFSinterference. When calculating the occupancy, the Wi-Fi channels are exempted. The trend of FH 86 2 4 6 8 10 12 14 0.4 0.45 0.5 0.55 0.6 0.65 Throughput vs. Number of Piconets Number of piconets Throughput AH OH AFH FH Figure 4.9: Throughput performance comparison as a function of the number of piconets in an environment with frequency dynamic and frequency static interference (¿ Wi¡Fi = 0:7 in all 22 channels). is similar to that of OH in Fig. 4.8 except that there are 22 heavy interfered channels now. Like OH, the occupancy of AFH increases linearly with respect to the number of piconets. The increased occupancy of AFH over FH comes from the dwindled hopset. A smaller hopset makes the FD interference higher in remaining channels. The occupancy of AHFH is bounded between OH and FH at all time. AHFH improves the occupancy performance of OH by avoiding hopset over-shrinking. In additional to the partially orthogonal hopset, AHFH also improves the throughput using multiple-slot packets. OH has the worst occupancy performance. Under FS interference, the reduced orthogonal hopset number makes the occupancy increase more quickly than the case of only FD interference is considered. From all simulation results, we conclude that AHFH can 87 e±ciently mitigate the self-interference among piconets, while avoiding hopset's over- shrinking to maintain the etiquette rules by respecting other potential wireless systems in the same UB. The study also reveals the impact of coe±cient ®. The best value of ® is determined by the number of collocated piconets. The carrier sensing result suggests thatasmallcarriersensingoverheadcanimprovethroughputsigni¯cantlyeveninasever interference environment. 2 4 6 8 10 12 14 0 0.05 0.1 0.15 0.2 0.25 Number of piconets Occupancy Occupancy vs. Number of Piconets OH AH AFH FH Figure 4.10: Occupancy performance comparison as a function of the number of piconets under frequency dynamic and frequency static interference with ¿ Wi¡Fi = 0:7 in all 22 channels. 4.4 Conclusion TheproblemofFDinterferenceamongBTpiconetsintheUBusingthefrequencyhopping mechanism was discussed. We proposed Adaptive Hopset Frequency Hopping (AHFH) 88 mechanism to mitigate the self-interference and to preserve the fairness of UB. AHFH can manipulate the hopset size and the packet length simultaneously to enhance the coexistence ability of BT devices. An analytical model was developed to understand the performance of AHFH. We can increase its throughput (or decrease PER) while maintainingalmosttheaverageoccupancyacrossallocatedfrequencychannels. Computer simulations were conducted to demonstrate the superior performance of AHFH in an environment with frequency-dynamic (FD) and frequency-static (FS) interference. Onlythe3-slotpacket(DH3)wasconsideredforthemultiple-slottransmissioninthis work. Itwasshownthat,withproperdesign,wecanavoidtheincreasedPERthatusually comes with the long packet. From the BT speci¯cation [16], the 5-slot packet (DH5) can provide an even higher data rate. The impact of the coe±cient ® was discussed in Section 4.3.2. It is possible to assign ® dynamically to improve the performance. In the future, we will investigate this possibility to further enhance the performance of AHFH. 89 Chapter 5 A Cognitive MAC Protocol Using Statistical Channel Allocation for Wireless Ad-hoc Networks The MAC protocol of a cognitive radio (CR) device is supposed to enable the device to dynamically access unused or under-utilized spectrum without (or with minimal) inter- ference to primary users. To ful¯ll such a goal, we propose a cognitive MAC protocol using statistical channel allocation and call it SCA-MAC in this work. SCA-MAC is a CSMA/CA-based protocol, which exploits statistics of spectral usage for decision mak- ing on channel access. For each transmission, the sender negotiates with the receiver on transmission parameters through the control channel. A model is developed for CR devices to evaluate the successful rate of the transmission. A CR device should pass the threshold of the successful transmission rate via negotiation before it can begin a valid transmission on data channels. The operating range and channel aggregation are two control parameters introduced to maintain the MAC performance. To validate our ideas, we conducted theoretical analysis and simulations to show that SCA-MAC does improve the throughput performance and guarantees the interference to incumbents to be bounded by a predetermined acceptable rate. The proposed MAC protocol does not 90 need a centralized controller, as the negotiation between the sender and the receiver is performed using the CSMA/CA-based algorithm. 5.1 Proposed SCA-MAC Protocol Thereareseveraldesiredfeaturesforane±cientcognitiveMACprotocol. First,itshould be able to predict future spectrum usage based on statistics of local spectrum utilization up to the current time instance. To implement this feature, a CR device should monitor the spectrum usage continually to maintain an accurate view of spectrum utilization. Second, it can bundle several continuous idle channels from a wide spectrum hole to speed up data transmission. Third, it should be a distributed algorithm so as to be employed in ad-hoc networks. The proposed SCA-MAC protocol is designed to possess the above three properties. It is a CSMA/CA-based protocol so that it is a distributed algorithm by nature. It is also designed to allow channel aggregation. To control the in°uence on the QoS of primaryusers,SCA-MACcanevaluateitsimpactinrealtime,whichmeansitcanpredict the successful rate based on the incipient packet length and collected statistics to make decision among alternative choices. 5.1.1 Overview of SCA-MAC Protocol The proposed SCA-MAC protocol consists of three major operations: (1) environment sensing and learning, (2) CRTS/CCTS exchange over the control channel, (3) DATA transmission and ACK over data channels. These operations are detailed below, and an example is shown in Fig. 5.1. 91 Control channel Data channels DIFS CW CRTS CCTS CA DATA ACK Figure 5.1: An illustrative example of the SCA-MAC protocol. 5.1.1.1 Environment Sensing and Learning One way to learn about the environment is achieved by extensive spectrum sensing. For cognitive radio, it is not just ordinary spectrum sensing, but sensing with broadband capability and narrowband resolution. For example, a CR device should sense a swath of spectruminoneshotanddiscoverthedetailutilizationinformationofspectralpartitions. Such advanced spectrum sensing technology is crucial to the success of CR devices. It is feasible via a DSP technique called the cyclostationary feature detection [2]. An alter- native way to acquire such high resolution broadband utilization information is through successive partial sensing, i.e. one fraction of the spectrum after another randomly or sequentially. Naturally, the latter takes longer time to ¯nish spectrum sensing and the information may not be up-to-date. Sensing is performed continuously and periodically. The sensing period is prede¯ned but adjustable. Upon activating the protocol, a run length of the idle/busy period is maintained for each channel. When the idle duration is ended by a transmission incurred 92 by the primary user, the run length is recorded in a circular bu®er, e.g. a circular bu®er of size 1000 records the run length of the last 1000 idle periods. These spectrum holes are opportunities that CR devices can use. These records provide the statistics of each channelandhelpthedevicemakedecisiononintelligentchannelallocationwithbounded interference. 5.1.1.2 CRTS/CCTS Exchange over Control Channel Generallyspeaking,aCRprotocolattemptstoaligntransmissioninspectrumholesinside the accessible spectrum. When the channel access is opportunistic, we have to determine which channel a sender should use to communicate with the receiver in the ¯rst place, which demands a mechanism to initiate the transmission. Here, we introduce a control channel that provides a common channel for initiation hand-shaking. The access to this single control channel is implemented by the CSMA/CA mechanism so that our protocol is still a decentralized one. A careful design can resolve the control channel saturation problem [22]. WhenaCRdevicewantstoinitiateatransmission,itfollowsthestandardCSMA/CA protocol to access the control channel to negotiate with the receiver. To be more speci¯c, the sender listens to the control channel and waits until it becomes idle. Then, it waits for the channel to remain idle for another DIFS duration before it begins the countdown of the contention window (CW). If the channel is still idle after the contention window period, it transmits a Control-channel-Request-To-Send (CRTS) packet. Upon receiving CRTS, the receiver screens the potential transmission opportunities based on its own 93 statistics and the current channel status, and then replies with a Control-channel-Clear- To-Send packet (CCTS) which contains the information of the best opportunity. If there is a collision on CRTS or CCTS, the sender would repeat the negotiation process but double the contention window size. The channel allocation mechanism will be described in the next section. In short, the control packets carry the information of channel aggre- gationandpacketlength,whoseexpectedsuccessfulratemeetstheinterferencethreshold. Renegotiation is needed if no choice satis¯es the interference threshold. A parameter carried by CCTS has to be speci¯ed. It is the collision avoidance (CA) windowforthecomingtransmissiononthedatachannel. Thisbacko®windowisdesigned to reduce the probability of collision resulted by two transmission pairs that happen to select the same or overlapped data channelsfor transmission. Even though the basic idea is borrowed from CW in CSMA/CA, it is adopted by our protocol here because of its desired property. We assume that the number of neighboring nodes of the receiver is n, which can be obtained from CCTS. Then, the sender assigns CA according to N =jCAj= 8 > > > > > > > > > < > > > > > > > > > : 2 ;n=0; 2 n ;1·n<5; 32 ;n¸5: (5.1) because the probability of collision at the receiver highly depends on n. This CA window is only related to n at the receiver side, and will not change with respect to the count of retransmission. 94 5.1.1.3 DATA/ACK Transmission over Data Channels Once CRTS/CCTS have been successfully exchanged, the sender and the receiver will tune their transceivers to the agreed data channels. The sender begins the countdown of a counter randomly selected from the range of the CA window received via CCTS. If the channelisstillidleupontheendofcountdown,thesenderbeginstheDATAtransmission. If data are successfully received, an ACK will be sent by the receiver after SIFS. If some other node acquires the channel before the end of countdown, the sender has to go back to the control channel to renegotiate. The transmission is considered done after ACK is successfully received. If the transmission failed (e.g., no ACK received), the sender has to go back to the control channel for negotiation again. 5.1.2 Statistical Channel Allocation Toallocatechannelssuchthattheinterferencetotheprimaryserviceisboundedbelowan acceptablelevel,weshouldevaluatethesuccessfulrateofanytransmissionbeforeittakes place. Channel aggregation and the packet length a®ect the transmission successful rate. For cognitive radio, the total number of available channels and the number of potential combinations can be very large. Thus, we need to set some parameters and rules to lower the complexity. 5.1.2.1 Optimum Operating Range Although it is straightforward to calculate the successful rate of a single channel, the complexity to evaluate all available combinations could be signi¯cant for a wide range of operating spectrum. To reduce such a complexity, we introduce a parameter, r, called 95 the operating range to specify the proper spectrum range that a node would search for transmission opportunities. It is a parameter related to the level of availability of spec- trum holes. A CR device can dynamically change its operating range. If the spectrum is crowded and ¯nding a roomy enough spectrum hole is di±cult, the CR device may increase its operating range, and vice versa. To decrease the level of overlapped oper- ating ranges among neighboring nodes, we can use some algorithm based on device's MAC address to spread the central channels of devices over the whole spectrum. Such a technique achieves spectrum load balancing nicely, and largely prohibits neighboring nodes from selecting the best but the same opportunity, which causes unwanted collision or re-negotiation. 5.1.2.2 Maximum Channel Aggregation By transmitting over multiple channels simultaneously, we can decrease the transmission time and increase the successful rate. According to Shannon's channel capacity formula, channel capacity W is proportional to bandwidth B, i.e., W = Blog 2 (1+SNR). Thus, we can get m fold shrinkage on transmission time with m channels aggregated together. 5.1.2.3 Closest Possible Opening Anidlechannelisnormallythe¯rstchoice. However,ahighersuccessfulratemayrequire a CR device to wait for some channels to become idle. Intuitively, if several similar opportunities coexist, we prefer the opportunity that demands the shortest waiting time. As a result, we should judge each opportunity by its successful rate ® based on collected channel statistics, and employ a successful rate threshold, ® T , to bound the interference 96 to theprimary service to be under1¡® T . This boundguaranteesthat the interferenceis within the tolerance and will result in no noticeable impact on the QoS of primary users. 5.2 Successful Rate Prediction In this section, we evaluate the successful rate of a channel for the prediction purpose. It consists of two subproblems: the probability of successful channel allocation within the operatingrangeandtheprobabilitythatthespectrumholeonallocatedidlechannelscan accommodate the speci¯c incipient packet. They deal with the following two problems, respectively: 1) the probability of collision with another CR devices in an available chan- nel, and 2) the probability of interference to the primary service by studying the packet length and the spectrum hole duration. In the following analysis, we use r to denote the dynamic operating range, n the number of neighboring nodes, and m the number of data channels in channel aggregation forthetransmissionopportunityunderevaluation. Furthermore,¿ denotestheutilization of the primary service, ¿ c denotes the average utilization of neighboring nodes, and m denotes the average channel aggregation of neighboring nodes. ² Channel availability ® c Parameter ® c represents the probability of successful channel allocation within the operating range of the receiver. The expected number of idle channels is (1¡¿)r while the expected number of channels occupied by CR devices is ¿ c ¢n¢m. Then, 97 their ratio is the probability of collision with neighboring CR nodes. With the consideration of channel aggregation, we obtain ® c =(1¡ ¿ c ¢n¢m (1¡¿)r ) m : (5.2) ² Spectrum Hole su±ciency ® L Parameter ® L represents the probability of a speci¯c packet of length L can ¯t the spectrumholeofdurationT onchanneliwithstatisticsC i intermsoftransmission time. For example, channel C 1 has been idled for time t 0 , the probability that it will remain idle for another L period can be written as p(C 1 :T ¸t 0 +LjT ¸t 0 )= p(C 1 :T ¸t 0 +L) p(C 1 :T ¸t 0 ) : (5.3) Then, by considering channel aggregation of channel i to channel i+m¡1, we can get ® L = i+m¡1 Y j=i p(C j :T ¸t 0;j + L m jT ¸t 0;j ): (5.4) By combining Eqs. 5.2 and 5.4, the successful rate for a packet of length L to be transmitted on channels i to i+m¡1 can be written as ®([i;i+m¡1];L)=® c ¢® L : (5.5) 98 Forallidlechannelsandtheircombinationswithmaximumchannelaggregationm t within operating rage r, we can calculate their successful rates and then select the highest one as the prediction to the successful rate: ®(r;m t )= max 8m2[1;mt];8i²[0;r¡m] f®([i;i+m¡1];L)g: (5.6) The complexity of exhaustive search is r(r+1) 2 . However, by excluding channels that are currently occupied, the complexity is reduced to (1¡¿r)(2¡¿r) 2 . This is an upper bound since available channels may be separated. We see from Eqs. 5.2-5.6 that the successful rate highly depends on operating rage r, channel aggregation m and packet length L, which can be controlled by a CR device. Besides, the successful rate also depends on the number of neighboring nodes n and channel statistics C i . The predicted and simulated successful rates ® are plotted as a function of operating range r and channel aggregation m in Figs. 5.2 and 5.4, respectively, where all channels are assumed to be equally likely with the same statistical parameters. 5.3 Throughput Analysis To analyze the throughput, we ¯rst analyze the collision and the interference phenomena of the SCA-MAC protocol. Since the control channel occupies the spectrum without any primaryservice, thereisnointerference. AllCRTS/CCTSfailuresarecausedbycollision 99 among neighboring nodes and it will incur renegotiation over the control channel. The average negotiation time, T c , in the control channel can be calculated as T c =DIFS+CW £T slot +T CRTS +SIFS+T CCTS : (5.7) Weusep c todenotethecollisionprobabilityonthecontrolchannel. Then,E[R c ]= p c 1¡pc is theexpectednumberofrenegotiationduetosuchacollision. Renegotiationisalsoneeded if no agreement is reached after the exchange of CRTS and CCTS. This is denoted by R 0 c . The cases with the data channel are more complicated. Despite prediction, there is still a probability that a transmission is corrupted by an early primary service access, which causes mutual interference, or by some neighboring CR device that chose the same CA counter. Either case will trigger a renegotiation and a retransmission. The average time for a successful transmission, T s , on the data channel is T s =CA£T slot +T d +SIFS+T ACK : (5.8) The time required for a retransmission over the data channel is the same as that for a successful transmission, i.e. T r = T s . Unlike the CW on the control channel, CA does not change with the retransmission count. It only re°ects the local spectrum usage. With the predicted successful rate ®, the expected number of retransmission is equal to E[R d ]= 1¡® ® . 100 In addition to the above cases, there is another scenario. That is, when some CR device is in the middle of CA, another device acquires the idle channel ¯rst. Then, the current CR device will dismiss the countdown and renegotiate for another opportunity. Thus, a collision and the retransmission process can be avoided due to the use of the collision avoidance mechanism. We use p 0 to denote the data channel collision avoidance possibilityandE[l]theexpectednumberofwaitingtimeslotsbefore¯ndingthatachannel is occupied. Then, the expected number of renegotiation caused by collision avoidance is E[R ca ]= p 0 1¡p 0 ; where p 0 (k)= k N , k =0;1;2;¢¢¢ ;N¡1, and p 0 =E[p 0 ]= (N¡1) 2N , and E[l]= 1 N N¡1 X k=0 l(k)= n(N¡1)(N¡2) 6N ; where l(k) = n N P k¡1 i=0 i = nk(k¡1) 2N . Then, the expected transmission time on control channel T control and data channel T data for a packet are equal to E[T control ] = (1+E[R c ]+E[R 0 c ]+E[R ca ]+E[R d ])T c ; E[T data ] = E[R ca ]E[l]T slot +(1+E[R d ])T s ; respectively. With all parameters available, we can determine the throughput of the data channel as ½= T d E[R ca ]E[l]T slot +(1+E[R d ])T s : (5.9) 101 The predicted and simulated throughput values ½ are plotted as a function of operating range r and channel aggregation m in Figs. 5.3 and 5.5, respectively. 5.4 Simulation Results Our simulation environment consists of one primary service network and one cognitive radio network which runs SCA-MAC in proximity. The channel is equally divided into 100 subchannels. Instead of a speci¯c system, we implemented a general primary service whichhasfollowingproperties. Eachprimaryservicedevicetransmitsitspacketsoverthe channelwithoutchannelsensing. Thepacketarrivalrateofprimaryusersisexponentially distributed, and so is the packet length. Moreover, we assumed the number of primary usersissu±cientlylarge. Asaresult,theoverallaccesspatternisexpectedtobeinnormal distribution due to the central limit theorem and each subchannel would have similar statistics in terms of average idle/busy time. We also represented the user experience of primary user in term of packet error rate (PER). Although the ¯xed payload length is adoptedinsimulation,theactualtransmissiontimevariesduetothepossibilityofchannel aggregation. We set the successful rate threshold ® T to 0.9, which means it transmits only when the predicted successful rate is higher than the threshold. Otherwise, the CR device renegotiates for a better opportunity. This limits the expected interference to the primary service to be 1¡® T = 0:1. The utilization of primary service is chosen to be ¿ =0:5. Other system parameters are shown in Table 5.1. Although there may be unused gaps in the spectrum between di®erent primary ser- vices in reality, all subchannels are assumed to be occupied to test the extreme case in 102 Parameter Assigned Value PHY header 192 bits MAC header 224 bits Slot time 20 ¹s DIFS 50 ¹s SIFS 10 ¹s CRTS 160 bits CCTS 112 bits CW min 32 CW max 1024 CA Eq. 5.1 Payload 11000 bits ACK 112 bits ® T 0.9 subchannel # 100 ¿ 0.5 Table 5.1: Parameters of control and data channels. simulation. Since the collision between primary users is not our concern, we assume no collision among primary services. All CR devices adopt SCA-MAC and every device is a neighbor node to one another. It is assumed that there is always a packet to transmit for each CR device. Concurrent transmission of a CR device with primary service is possible as long as they are on di®erent subchannels. Otherwise, collision and interference would take place. For performance benchmark, we have implemented a simple cognitive MAC, which has no channel prediction and no guarantee on interference. It simply selects an idle channel (or channels) to transmit randomly. We show theoretical and simulation results of the successful rate and the throughput with respect to operating range r in Fig. 5.2 and with respect to channel aggregation m in Fig. 5.4. All simulation data plotted in Figs. 5.2 and 5.4 are the averaged results of at least 20 runs. 103 AlargeroperatingrangegivestheCRdevicehigher°exibilityonopportunityselection and a higher probability on ¯nding quali¯ed ones. Thus, ® and ½ increase as r increases for the simple MAC as shown in Fig. 5.2. The proposed SCA-MAC outperforms the simple MAC in all operating ranges. The performance gap becomes more obvious when the operating range is smaller. WeseefromFig.5.4that®and½decreaseasmincreasesforthesimpleMAC.Thiscan be explained as follows. In the simulation, we implemented a general primary user access pattern in which each subchannel has resembled statistics in terms of average idle/busy time. However, their idle periods do not begin and end at the same time. Thus, even higherchannelaggregationcouldshortenthetransmissiontimebyseveralfold,thebene¯t of channel aggregation does not o®set the potential retransmission overhead incurred by theunsynchronizedchannelaccess. Incontrast,ourSCA-MACprotocolalwaysmaintains the success rate at the desired level and the throughput at a level higher than that of generic MAC due to the prediction of spectrum opportunities. If a primary service has a high level of synchronized channel access, a higher improvement could be achieved. 104 10 15 20 25 30 35 40 45 50 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Operating Range r Successful Rate α Successful Rate α vs Operating Range r with n=2 m=1 SCA−MAC, α T =0.9 Theoretic SCA−MAC, α T =0.9 Generic MAC Theoretic generic MAC Figure 5.2: Analytical and simulated results of Successful Rate ® vs. Operating Range r with (n;m)=(2;1) at ¿ =0:5. 10 15 20 25 30 35 40 45 50 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Operating Range r Throughput ρ Throughput ρ vs Operating Range r with n=2 m=1 SCA−MAC, α T =0.9 Theoretic SCA−MAC, α T =0.9 Generic MAC Theoretic generic MAC Figure 5.3: Analytical and simulated results of Throughput ½ vs. Operating Range r with (n;m)=(2;1) at ¿ =0:5. 105 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Channel Aggregation m Successful Rate α Successful Rate α vs Channel Aggregation m with n=2 r=50 SCA−MAC, α T =0.9 Theoretic SCA−MAC, α T =0.9 Generic MAC Theoretic generic MAC Figure5.4: AnalyticalandsimulatedresultsofSuccessfulRate®vs. ChannelAggregation m with (n;r)=(2;50) at ¿ =0:5. 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Channel Aggregation m Throughput ρ Throughput ρ vs Channel Aggregation m with n=2 r=50 SCA−MAC, α T =0.9 Theoretic SCA−MAC, α T =0.9 Generic MAC Theoretic generic MAC Figure 5.5: Analytical and simulated results of Throughput ½ vs. Channel Aggregation m with (n;r)=(2;50) at ¿ =0:5. 106 5.5 Conclusion AcognitiveMACusingStatisticalChannelallocation,calledSCA-MAC,wasproposedin thiswork. Ananalyticalmodelwasdevelopedtoexplainitsperformance. Tofullyexploit the spectrum-time-space opportunity, we introduced couple controllable parameters, the operating range and the channel aggregation. Computer simulation was conducted to demonstratethesuperiorperformanceofSCA-MACoverthatofasimplecognitiveMAC. 107 Chapter 6 Dynamic Spectrum Access MAC for Wireless Ad-hoc Networks The dynamic spectrum access (DSA) mechanism enables the access to unused or under- utilizedspectrumwithoutcausinginterferencetoexistingincumbentservice. Oneimpor- tant issue to consider is collision resolution, which is demanded to avoid the interference not only with incumbent users but also between DSA nodes. In our design, a DSA node can control its operation range (OR) to reduce collisions with neighboring nodes. Three operation range assignment (ORA) strategies; namely, non-sharing (NS), partial sharing (PS) and full sharing (FS), will be discussed in this chapter. Furthermore, we examine the frequency-domain backo® scheme. After a collision, traditional time domain backo® schemes result in additional transmission delay. A DSA wireless system generally pos- sesses a swath of spectrum which is divided into many channels. By taking advantage of the rich frequency resource, the frequency-domain backo® scheme avoids unnecessary delay. To validate our design, a slotted CSMA model with the Poisson packet arrival rate is used to analyze the performance of the ORA strategies. Computer simulations are 108 conductedtoverifyourtheoreticalresults. Itisshownthatthereexistsatradeo®between throughput and the white-space ¯lling rate. While FS gives the best throughput, NS has the best white space ¯lling performance. 6.1 Role of DSA-MAC Protocol AlthoughtheremaybemanynodesinaDSAnetwork,atransmissioncouldonlybeinter- fered by a node's neighboring nodes. In our model, we de¯ne a node and its neighboring nodes as a network unit. When there is an on-going transmission between the node and one of its neighbors, no other neighboring nodes could transmit in the same channel. In otherwords, anodeanditsneighborssharethelocalfrequencyresource. Alargewireless network is formed by the union of many such basic units. For simplicity, we ignore the boundary e®ect of a ¯nite DSA network, and assume that all DSA nodes have the same number of neighboring nodes in our analysis. SincetheDSAsystemexploitslocallyunusedfrequencyresource, itgenerallyrequires a wider spectrum to secure su±cient transmission opportunity. Thus, a DSA wireless network is always a multichannel system. The number of channels a node could use, which is called the operation range of a node, is normally larger than the number of neighboring nodes it has. Here, we assume all channels are slotted, as illustrated in Fig. 6.1. A white slot in the ¯gure indicates that the slot is not used either by the primary service or DSA nodes so that it is idle. A gray slot means that only one pair of DSA nodes access the slot and the transmission is successful. A black slot represents a collision. A collision (or an interference) happens when two or more neighboring nodes' 109 transmissions overlap in time and in frequency, and the collision resolution mechanism of the DSA MAC is responsible to solve the problem. collision A B B A (1) (2) Figure 6.1: A slotted multichannel model of the DSA system. When there is a collision betweentransmissionsAandB,scenario(1)showsthepossibleoutcomeofatimedomain backo® (where transmissions are randomly delayed to a later time slot on the same channel) while scenario (2) shows the possible outcome of a frequency domain backo® (where transmissions are randomly distributed to other channels). Spectrum sensing is crucial to the success of DSA networks. Through continuous sensing, a DSA node learns about the environment and then adapts to it. Adaptation is achieved by the MAC protocol, which is responsible for avoidance of primary services and link establishment between neighboring nodes. For example, if a node senses that a primary service is extremely busy on a certain channel, the DSA-MAC protocol will exclude the channel from its operation range. Assigning channels exclusively can avoid collision among neighbors, yet leave fewer channels to exploit in the white space. The impact of di®erent operation range assignment (ORA) strategies will be discussed and analyzed in the next section. Besides reducing collision through ORA, DSA-MAC should resolve collisions e±- ciently after their occurrence. There are two kinds of collisions: with the primary service 110 andwithanotherDSAtransmission. Uponapacketarrival,aDSAnoderandomlyselects a channel from its operation range and listens to the channel. During this period, a DSA node should be intelligent enough to recognize any ongoing incumbent service. Since all DSA communications should do no harm to the primary service, a DSA node quits lis- tening to the channel once it detects any ongoing primary service. Then, the DSA node would try to transmit again at the next time slot. Although there is no packet loss under this scenario, we still call it the primary service collision. If the DSA node decides that the channel is clear of primary service, it begins its transmission. In our slotted model, a primary service always occupies the whole time slot. However, other neighboring nodes may also select the same channel to transmit. This incident is called the DSA collision, where all overlapped DSA packets are lost. Then, they would select a channel from their operation range and try to transmit at the next time slot again. Upon a collision, the transmission is postponed randomly to a later time slot in the same channel using the time-domain resolution scheme. This corresponds to scenario (1) in Fig. 6.1. Since the DSA system has multiple channels, we propose a frequency-domain collision resolution scheme that exploits the nature of multiple channels in resolving con- tention. Similar to the 1-persistent ALOHA, this scheme retransmits in the following slot except that it selects one from its operation range randomly, which is called the frequency-domain backo® scheme. This corresponds to scenario (2) in Fig. 6.1. As com- pared with the time-domain backo® scheme, the frequency-domain backo® scheme has less delay, which is referred to as \fast retrial" in OFDMA networks [4,27]. Although the two aforementioned collision types are fundamentally di®erent, they can be both well resolved using the same frequency-domain backo® scheme. 111 6.2 Analysis of Operation Range Assignment Strategies Although a DSA system has multiple channels, a node does not have to use all of them. The set of channels that a node would use is called the operation range (OR). The OR of any node is a subset of total available channels. Typically, OR is managed by DSA- MAC.Inthissection,wewillanalyzethreeoperationrangeassignment(ORA)strategies: non-sharing (NS), partial sharing (PS) and full sharing (FS). In fact, the NS and the FS schemes are special cases for the PS scheme. 6.2.1 System Model and Performance Metrics We adopt a ALOHA-like model for our theoretical analysis. The Poisson distributed packet arrival rate of each DSA node is represented as G i and the arrival rate of primary service in each channel is G p . There are total M available channels and each node has the same number of neighboring nodes, which is represented by n. The OR size of a DSA node is m. We use G t to represent the total packet arrival rate in a channel, which includes packets from both primary and secondary services. To evaluate these three strategies, we examine the following performance metrics: throughput, the collision probability and the white space ¯lling rate. ² Throughput The throughput of a node, denoted by S, is the product of its packet arrival rate and the probability that the medium is idle at a time slot. Thus, it is expressed as S =G i e ¡G t : (6.1) 112 ² Collision Probability Slot collision occurs when two or more primary or secondary packets arrive at the same time slot. Thus, the probability of slot collision Q could be represented as Q ns =1¡(1+G ns )e ¡G t : (6.2) ² White Space Filling Rate The chance for a time slot not being occupied by the primary service is e ¡G p . Similarly, the chance that it is not used by any service is e ¡G t . The white space is de¯ned as time slots not used by the primary service. Then, the white space ¯lling rate in a channel can be written as W =1¡ e ¡Gt e ¡G p ; (6.3) whichistheprobabilitythatawhitespaceisoccupiedbyasecondarytransmission. 6.2.2 Non-Sharing (NS) Strategy One ORA strategy is to allocate one part of the spectrum exclusively to a node. Since a collision happens when more than two transmissions overlap in time and frequency, a nodecanavoidacollisionusingthisstrategy. Ononehand,thenon-sharing(NS)strategy eliminates collisions among DSA users. On the other hand, the OR size of a node also shrinks, which means fewer channels to explore for a DSA node to transmit. To compute the total packet arrival rate, G t , at the ith time slot in a channel, we should consider the 113 numbers of new arrival packets and retransmissions due to collisions at the (i¡1)th time slot. If the arrival rate of new packets is G 0 , we can obtain G t (i)=G 0 +P c P m G t (i¡1); (6.4) where P c is the probability of collision and P m is the probability of the node would select the same channel to transmit again. In the steady state, we can express G t as G t = G 0 1¡P c P m : (6.5) In the NS strategy, MAC distributes the local frequency resource equally to neighboring nodes. Thus, the size of OR is m= M n . The arrival rate of new packets is G ns 0 =G p + G i m (6.6) Furthermore, we have P c =(1¡e ¡Gp )(1¡e ¡ G i m ) and P m = 1 m : (6.7) Finally, the total arrival rate of the NS strategy, G ns t =G ns , could be found via Eq. 6.5. 114 6.2.3 Full-Sharing (FS) Strategy With the full sharing (FS) strategy, each DSA node fully shares the local frequency resource with its neighboring nodes and the size of OR is m=M. Since the tra±c of all nodes is distributed to M channels, the arrival rate of new packet on a channel becomes G fs 0 =G p +n G i M : (6.8) There is a chance that a transmission may be interfered by neighboring nodes. The collision probability includes not only the backo® incurred by the primary service but also transmission collisions among neighboring DSA nodes. Thus, we have P c = (1¡e ¡G p )(1¡e ¡n G i M ) (6.9) + e ¡G p (1¡e ¡n G i M (1+n G i M )): Along with P m = 1 M , we could calculate G fs t =G fs via Eq. 6.5. 6.2.4 Partial-Sharing (PS) Strategy The partial sharing (PS) strategy allows a node to control the degree it wants to share the OR with its neighbors. Generally speaking, the OR size of such a node is between M n (non-sharing) and M (full sharing). Under the PS strategy, a channel may be reused by some of the neighbors. We use r to depict the reuse rate of a channel. Unlike ¯xed r in the previous two strategies 1 , the reuse factor may vary from a channel to another in the 1 Note that r =0 for the NS strategy and r =n for the FS strategy. 115 PSscenario. However, withasimpletrimmingalgorithm, wecanaveragethereusefactor of each channel and reduce it to the theoretical upper-bound with respect to its OR size, which is equal tod n£m M e. Then, we can calculate the averaged number of neighbors that a DSA node shares its channels; i.e., n 0 =(r¡1)+ b M ; (6.10) where b is the remainder of n£m divided by M. Since a channel is shared by n 0 nodes in average, the arrival rate of new packet on a channel becomes G ps 0 =G p +n 0 G i m : (6.11) With P c = (1¡e ¡G p )(1¡e ¡n 0 G i m ) (6.12) + e ¡G p (1¡e ¡n 0 G i m (1+n 0 G i m )); and P m = 1 m , we obtain G ps =G ps t = G ps 0 1¡P c P m : (6.13) It is straightforward to see that the FS and the NS strategies are just two extreme cases of the PS strategy. Comparing the analytical results of the three ORA strategies, we can obtain the following lemma. 116 Lemma 1 Given the same numbers of channels and neighboring nodes, and the same packet arrival rates of the primary service and DSA nodes, the throughput performance of the three ORA strategies is given by S FS ¸S PS ¸S NS ; while their white space ¯lling rates satisfy W NS ¸W PS ¸W FS : The above lemma is a direct consequence of the total arrival rates of three ORA strategies, which have the following order: G NS ¸G PS ¸G FS : Furthermore, with the PS strategy, the throughput of a node increases as the PR size increases. On the contrary, the white space ¯lling capability of a node decreases as the OR size increases. 6.3 Simulation Results In our simulation, the channel is slotted and the packet arrival process in each channel of the primary service and a DSA node is Poisson distributed. When a DSA node wants to access a channel, it ¯rst listens to the channel for a short period of time to decide whether there is an ongoing primary service. If there is one, it randomly picks up a 117 channel from its OR and tries again at the next time slot. If the channel is clear, it begins its transmission. Since the number of available channels is generally larger than thatofneighboringnodes, wedonotimplementanytime-domainbacko®beforetheDSA transmission. Only the frequency-domain backo® is implemented to resolve collisions among DSA nodes. The frequency-domain backo® of our DSA-MAC works like the 1- persistence ALOHA, it retransmits again at the next time slot in a di®erent channel. All channels in the OR have the same probability to be selected. We compare four cases in our simulations: FS, NS and two PS schemes. All DSA nodes adopt the same ORA strategy. We run the simulation for 1000 time slots with respect to each single trial, and each plotted curve in the ¯gure is the averaged result of at least 20 trials. 6.3.1 E®ects of Channel and Neighboring Node Numbers Fig. 6.2 shows the impact of channel numbers on the throughput. With there are more channels, each DSA node has more white space to access and a lower collision possibility with its neighbors. Thus, the throughput of all strategies improves when the channel number increases. Among the four strategies under comparison, FS gives the highest throughput,whichimpliesalargerchannelselectionpooldoeshelp. ThePSstrategywith a larger OR size has a higher channel reuse rate that also results in higher throughout which ranks the second. The NS strategy gives the lowest throughput. In other words, the gain from fewer collisions cannot compensate the loss due to a smaller OR size. Fig. 6.3 shows the impact of the number of neighboring nodes. With a ¯xed total channel number, increasing the neighboring node number implies increasing the tra±c 118 5 10 15 20 25 30 35 40 45 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Available channel number M Throughput S Throughput vs. channel number n=6 G p =1 G i =5 non−sharing full sharing partial sharing m=8 partial sharing m=12 Figure 6.2: Comparison of throughput performance of four ORA strategies as a function of the channel number. load in a DSA node. Then, the throughput decreases as the neighboring node number and the collision probability increase. In this setup, the FS strategy still has the best throughput, followed by the two PS strategies, and then the NS strategy. Figs. 6.2 and 6.3 reveal that a DSA network demands su±cient frequency resource (i.e. more channels) and a smaller number of neighboring nodes for better throughput. 6.3.2 E®ects of Primary and Secondary Service Arrival Rates Fig. 6.4 compares the throughput performance of four ORA strategies as a function of the primary service arrival rate under a ¯xed arrival rate at each DSA node. The rising arrival rate of the primary service increases the burden in each channel, which leaves less white space for the secondary service. When the tra±c load of the primary service is moderate, the performance gap between di®erent strategies is obvious. However, as the 119 2 4 6 8 10 12 14 0 0.2 0.4 0.6 0.8 1 1.2 1.4 Number of neighbor n Throughput S Throughput vs. neighbor node number M=36 G p =1 G i =5 non−sharing full sharing partial sharing m=8 partial sharing m=12 Figure 6.3: Comparison of throughput performance of four ORA strategies as a function of the neighboring node number. tra±c load aggravates, the performance gap shrinks. One way to cope with the heavy primary service tra±c is to allocate more spectrum to the DSA system. With a ¯xed primary service arrival rate, Fig. 6.5 compares the throughput perfor- manceoffourdi®erentORAstrategiesasafunctionofthepacketarrivalrateoftheDSA node. The FS strategy gives the best performance. Note that each throughput curve has a peak at some arrival rate, which is the optimal operation tra±c load for a DSA node under the e®ects of multiple parameters, i.e. the channel number, the neighboring node number and the primary service arrival rate. Thus, the ideal tra±c load is expected to shift to the right when the primary service arrival rate decreases, and vice versa. 120 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 Packet arrival rate of primary service G p Throughput S Throughput vs. arrival rate M=18 n=6 G i =5 non−sharing full sharing partial sharing m=8 r=2 partial sharing m=12 r=3 Figure 6.4: Comparison of throughput performance of four ORA strategies as a function of the packet arrival rate of the primary service. 6.3.3 Comparison of White Space Filling Rate Figs. 6.6 and 6.7 compare the white space ¯lling rate of di®erent ORA strategies. Unlike throughput, the NS strategy performs the best for this measure. With a smaller OR and, hence, fewer channels to operate on, the NS strategy has less °exibility but focus on ¯lling the white space in its OR. These two ¯gures also indicate that the tra±c load of the primary service has more impact on the white space ¯lling capability than that of DSA nodes. 6.3.4 Comparison of Slot Collision Probability Figs. 6.8 and 6.9 compare the slot collision probability of di®erent ORA strategies. By slotcollision,werefertonon-idleslotsthatareneitherusedbytheprimaryservicenorby any DSA transmission successfully. In other words, the slot is wasted by collision among 121 0 2 4 6 8 10 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 Packet arrival rate of each node G i Throughput S Throughput vs. arrival rate M=18 n=6 G p =1 non−sharing full sharing partial sharing m=8 r=2 partial sharing m=12 r=3 Figure 6.5: Comparison of throughput performance of four ORA strategies as a function of the packet arrival rate of the DSA node. primary or secondary services. The FS strategy has the best performance (i.e. the least slot collision probability) among the four ORA strategies. 6.4 Conclusion A DSA-MAC protocol that is capable of adapting the operation range and resolving collision via frequency-domain backo® was proposed in this chapter. It was shown by bothanalyticalandsimulationresultsthattheFSstrategyprovidesthebestperformance on throughput but the worst performance on the white space ¯lling rate. Thus, there exists a tradeo® between throughput and the white space ¯lling rate to be considered in the design of DSA networks. 122 1 1.5 2 2.5 3 3.5 4 4.5 5 0.82 0.84 0.86 0.88 0.9 0.92 0.94 0.96 0.98 1 Packet arrival rate of primary service G p White space fill factor White space fill factor vs. arrival rate M=18 n=6 G i =5 non−sharing full sharing partial sharing m=8 r=2 partial sharing m=12 r=3 Figure6.6: Performancecomparisonofthewhitespace¯llingrateoffourORAstrategies as a function of the packet arrival rate of the primary service. 0 2 4 6 8 10 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Packet arrival rate of each node G i White space fill factor White space fill factor vs. arrival rate M=18 n=6 G p =1 non−sharing full sharing partial sharing m=8 r=2 partial sharing m=12 r=3 Figure6.7: Performancecomparisonofthewhitespace¯llingrateoffourORAstrategies as a function of the packet arrival rate of the DSA node. 123 1 1.5 2 2.5 3 3.5 4 4.5 5 0.75 0.8 0.85 0.9 0.95 1 Packet arrival rate of primary service G p Probability of collision Collision probability vs. arrival rate M=18 n=6 G i =5 non−sharing full sharing partial sharing m=8 r=2 partial sharing m=12 r=3 Figure 6.8: Performance comparison of the slot collision probability of four ORA strate- gies as a function of the packet arrival rate of the primary service. 0 2 4 6 8 10 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Packet arrival rate of each node G i Probability of collision Collision probability vs. arrival rate M=18 n=6 G p =1 non−sharing full sharing partial sharing m=8 r=2 partial sharing m=12 r=3 Figure 6.9: Performance comparison of the slot collision probability of four ORA strate- gies as a function of the packet arrival rate of the DSA node. 124 Chapter 7 Conclusion and Future Work 7.1 Conclusion To improve the performance of unlicensed band technologies under a coexistence envi- ronment, we presented two non-collaborative mechanisms in this thesis. Respectively, they are the Dynamic Fragmentation (DF) scheme for Wi-Fi WLAN in Chapter 3 and theAdaptiveHopsetFrequencyHopping(AHFH)schemeforBluetoothWPANinChap- ter 4. These two mechanisms can enhance the coexistence performance of Wi-Fi and BT individually and collectively without any opposing e®ect to each other. They lift up performance of the legacy wireless systems under coexistence scenarios and provide a friendlier environment for upcoming new wireless technologies. DFimprovesthecoexistenceofWi-Fiinthecoexistenceenvironment. Webeganwith developingananalyticalmodeltocharacterizetheWi-FiinterferenceposedbyBluetooth. Basedonthismodel,weproposedtheDFschemetoenhancethecoexistenceabilityofWi- Fi networks. We also investigated the scenario that the wireless networks are stationary and presented an enhanced DF scheme to further improve the performance. Besides 125 the improvement on throughput, DF also empowers a Wi-Fi network to distinguish that packet loss is due to interference of Bluetooth or collision from other Wi-Fi nodes. If transmissionfailuresaremostlyfoundatthe¯rstfragmentofapacket,thosetransmission failures are largely caused by other Wi-Fi nodes according to our model. Therefore, the node(orAP)wouldrecognizethepatternandmakedecisiontostayattheun-fragmented stateandavoidtheunnecessaryoverhead. SinceDFfollowsthefragmentationframework asdescribedinthe802.11standard,ourproposedsolutioncanbeeasilyimplementedwith small modi¯cation. Analytical and simulation results were presented in Chapter 3. The AHFH scheme enhances the coexistence ability of BT networks. The spatial overlap among BT piconets, known as self-interference, is the main source of throughput degradation. AHFH addressed not only the self-interference arising from collocating BT piconetsbutalsotheinterferencefromWi-Fi. Althoughtoavoidthroughputdegradation is usually the ¯rst priority of coexistence mechanisms, it is not the sole target for AHFH. AHFH also considers the impact of each node's hopset adaptation to surrounding non- FH devices. This motivated us to include an etiquette rule in AHFH to uphold the coexisting spirit in the unlicensed band. In a nutshell, the proposed AHFH mechanism combines packet length selection and hopset adaptation to adjust the tra±c load on each channel according to the interference level (in term of PER) under the constraint that the tra±c load should satisfy the upper-bound of the maximum channel occupancy set by the etiquette rule. The results were detailed in Chapter 4. One of the most challenging tasks in developing networks with dynamic spectrum access (DSA) is in the design of cognitive medium access control (MAC), especially the decentralized cognitive MACs. A decentralized cognitive MAC using statistical channel 126 allocation, called SCA-MAC, was proposed in Chapter 5. From knowledge gained from the statistics, a CR node could adjust its operation range and channel aggregation to decrease packet loss rate as well as the interference to the primary service. We built a statistical model to predict the interference of a CR node. Detailed analysis and results were reported in Chapter 5. WiththeframeworkdescribedinChapter5, wefurtherinvestigatedtheimpactofthe channelselectionpolicyadoptedbyCRnodesonthesystemperformance. Weproposeda DSA-MACschemethatcanadjusttheoperationrangeadaptivelyandresolvethecollision through frequency-domain backo® in Chapter 6. Three operation range strategies, i.e., non-sharing(NS),partialsharing(PS)andfullsharing(FS),werestudied. Itiscon¯rmed by mathematical analysis and computer simulation that the FS strategy has the best throughput but the worst white-space ¯lling rate. Therefore, there is a tradeo® in the operation range assignment. A ALOHA-like model with the Poisson distributed packet arrivalrate isestablishedinanalysis. Computer simulationsundervariousscenarioswere conducted to show the in°uence of the operation range selection strategy. 7.2 Future Work Cognitivetechnologyisabletoutilizethescarcespectrumresourcemoree±ciently. Based on insights gained from analysis of Wi-Fi and BT, we believe that the cognitive network is the logical direction for future wireless technology. If we can understand the dynamics amongneighboringcognitiveradio(CR)devicesaswellasthebehaviorofprimaryservice 127 more, it is likely that we could do a better job on designing a cognitive MAC protocol. To extend the current research, we proposed following research problems. ² Collisions among neighboring CR devices result from the limited view of surround- ing spectrum usage. To further eliminate unwanted collision, a CR device needs to gain a view that goes beyond a single hop, which was done by exchanging CRTS/CCTS in SCA-MAC. Normally, 2 or 3-hops view of networks is su±cient, and a mechanism is required to disseminate the spectrum usage information. This can be done by all nodes individually and then broadcast through the control chan- nel(whichissimilartoroutinginformationexchange)orpiggybackedaspartofthe regular transmission on the data channel. ² The access pattern of incumbent service highly a®ects the prediction performance. If the access pattern is known to CR devices, interference would be less likely to occur and the CR spectrum partition can also be done more precisely to achieve better channel aggregation. Besides the access pattern, the QoS requirement of the primary service is another important factor of consideration. For example, we should consider PER and delay to re°ect user experience in the context of multimedia streaming. ² Thanks to the spectrum °exibility of software-de¯ned radio (SDR), CR networks canformmorecomplextopologiesthantraditionalwirelessnetworks. Besidesbuild- ing topology that only utilizes a single channel, we may build topology that runs on multiple channels, which vary hop to hop and time to time. The problem would become even more complicated when each CR device has more than one interface. 128 Thus, the number of possible topologies can be huge even for a mid-size CR net- work. It is a great challenge to form an optimal or near-optimal topology. Besides theconnectivityconcern, newtopologyneedstoconsiderinterferenceamongneigh- boring transmissions and with primary service as well. A new policy is needed to construct a topology to accomplish the following objectives: { Form a connected topology through intelligent assignment of radio interfaces to spectrum holes that are heterogeneous from node to node in CR networks. EventhoughtheCRnetworkmayseemonlypartiallyconnectedateachchan- nel, the superimposed graph from all channels manifests a fully connected network. { Assign links that not only maintain network connectivity but also maximize network capacity and minimize interference among neighboring nodes. { Select diverse channels on a routing path to prevent interference between ad- jacent hops in the path and improve spatial reuse and network capacity at the same time. AwelldesignedcognitiveradioMACalongwithtopologymaysolvetheuntractable mesh networking problem in traditional wireless networks and handle the spectrum scarcity problem more e±ciently. ² TheSCA-MACschemeproposedinChapter6providesthreeoperationrangeselec- tion strategies. In our analysis and simulation, all nodes adopt the same operation range. The homogeneous assignment may be reasonable when all nodes share the 129 same packet arrival rate and have the same number of neighboring node. How- ever, if nodes have di®erent tra±c loads and/or di®erent QoS requirements, we may assign an individual node a speci¯c operation range. Although the dynamics of the non-homogeneous operation range assignment is complicated, the increased °exibility may help CR networks leverage the real world environment better. 130 Bibliography [1] Wireless LAN medium access control (MAC) and physical layer (PHY) speci¯cation, ANSI/IEEE Std. 802.11, 1999. [2] D. Cabric, S. M. Mishra, and R. W. Brodersen, \Implementation issues in spectrum sensing for cognitive radios," in Proceedings of the 38th Asilomar Conference on Signals, Systems, Paci¯c Grove, CA, 2004. [3] C. Chiasserini, R. Rao, and D. di Elettronica, \Coexistence mechanisms for inter- ference mitigation between IEEE 802.11 WLANs and Bluetooth," INFOCOM 2002. Twenty-First Annual Joint Conference of the IEEE Computer and Communications Societies. Proceedings. IEEE, vol. 2, 2002. [4] Y.-J.Choi,S.Park,andS.Bahk,\MultichannelRadomAccessinOFDMAWireless Networks," IEEE JSAC, vol. 24, no. 3, March 2006. [5] DARPA. ATO the next generation program. [Online]. Available: http://www.darpa.mil/ato/programs/XG/index.htm [6] FCC Spectrum Policy Task Force, \Report of the spectrum e±ciency working group," Federal Communications Commision, Technical Report 02-155, November 2002. [7] N. Golmie, \Bluetooth dynamic scheduling and interference mitigation," ACM Mo- bile Networks, vol. 9, no. 1, Feb 2004. [8] N. Golmie, N. Chevrollier, and O. Rebala, \Bluetooth adaptive frequency hopping and scheduling," in Proc. of IEEE Military Communications Conference (MILCOM '03), vol. 5, October 2003, pp. 1138{1142. [9] ||, \Bluetooth and WLAN coexistence: challenges and solutions," IEEE Wireless Communications Magazine, Dec 2003. [10] N. Golmie, R. E. V. Dyck, A. Soltanian, A. Tonnerre, and O. Rebala, \Interference evaluationofBluetoothandIEEE802.11bsystems,"ACMWirelessNetworks,vol.9, pp. 202{211, 2003. [11] S.Haykin,\Cognitiveradio: brain-empoweredwirelesscommunications(InvitedPa- per)," IEEE JSAC, vol. 23, no. 2, February 2005. [12] I. Howitt, \WLAN and WPAN coexistence in UL band," Vehicular Technology, IEEE Transactions on, vol. 50, no. 4, pp. 1114{1124, 2001. 131 [13] ||, \Bluetooth performance in the presence of 802.11b WLAN," in IEEE Trans- actions on Vehicular Technology, vol. 51, no. 6, November 2002, pp. 1640{1651. [14] I. Howitt and F. Awad, \Optimizing IEEE 802.11b packet fragmentation in collo- cated Bluetooth interference," in IEEE Transactions on Communications, vol. 53, no. 6, Jun 2005, pp. 936{938. [15] A. C.-C. Hsu, D. S. L. Wei, and C.-C. J. Kuo, \Coexistence mechanism using dy- namic fragmentation for interference mitigation between Wi-Fi and Bluetooth," in MILCOM'06, Washington D.C., October 2006. [16] Wireless medium access control (MAC) and physical layer (PHY) speci¯cations for wireless personal area networks (WPANs), IEEE Std. 802.15.1, 2002. [17] IEEE 802.15 WG. IEEE 802.15 WPAN task group 2 (TG2). [Online]. Available: http://www.ieee802.org/15/pub/TG2-Coexistence-Mechanisms.html [18] IEEE 802.22 WG. IEEE 802.22 wireless regional area networks WRAN. [Online]. Available: http://standards.ieee.org/announcements/pr 80222.html [19] IST. Spectrum e±cient uni- and multicast services over dynamic multi- radio networks in vehicular environments IST-2001-35125. [Online]. Available: http://www.comnets.rwth-aachen.de/»o drive/index.html [20] Z. Jiang, V. Leung, and V. Wong, \Reducing collisions between Bluetooth piconets byorthogonalhopsetpartitioning,"inProc.IEEERadioandWirelessConf.(RAW- CON '03), August 2003. [21] S.Jones,N.Merheb,andI.-J.Wang,\Anexperimentforsensing-basedopportunistic spectrum access in CSMA/CA networks," in Proc. of IEEE Symposium on New Frontier in Dynamic Spectrum Access Networks, November 2005. [22] P.Kyasanur,J.Padhye,andP.Bahl,\Onthee±cacyofseparatingcontrolanddata into di®erent frequency bands," in IEEE Broadnets, October 2005. [23] J. Lansford, A. Stephens, and R. Nevo, \Wi-Fi (802.11b) and Bluetooth: enabling coexistence," IEEE Network, vol. 15, no. 5, pp. 20{27, Sep 2001. [24] L.Ma, X.Han, andC.-C.Shen, \DynamicopenspectrumsharingMACprotocolfor wirelessadhocnetworks,"inProc.ofIEEESymposiumonNewFrontierinDynamic Spectrum Access Networks, November 2005. [25] M. McHenry, E. Livsics, T. Nguyen, and N. Majumdar, \XG Dynamic Spectrum AccessFieldTestResults,"IEEEWirelessCommunicationsMagazine,vol.45,no.6, June 2007. [26] J.Mitola,\Cognitiveradiofor°exiblemobilemultimediacommunications,"in Proc. of IEEE Int. Workshop on Mobile Multimedia Comm., 1999, pp. 3{10. 132 [27] P. Pawelczak, R. Prasad, and R. Hekmat, \Opportunistic Spectrum Multichannel OFDMA," Communications, 2007. ICC'07. IEEE International Conference on, pp. 5439{5444, 2007. [28] P. Popovski, H. Yomo, S. Aprili, and R. Prasad, \Frequecny rolling: a cooperative frequency hopping for mutual inteferring WPANs," in Proc. ACM Mobihoc Conf., May 2004, pp. 199{209. [29] P. Popovski, H. Yomo, and R. Prasad, \Dynamic adaptive frequency hopping for mutually interfering wireless personal area networks," IEEE Transactions on Mobile Computing, vol. 5, no. 8, pp. 991{1003, August 2006. [30] T. Rappaport, Wireless Communications: Principles and Practice. IEEE Press Piscataway, NJ, USA, 1996. [31] A. Stranne, O. Edfors, B. Molin, A. TeliaSonera, and S. Malmo, \Energy-based in- terferenceanalysisofheterogeneouspacketradionetworks," Communications, IEEE Transactions on, vol. 54, no. 7, pp. 1299{1309, 2006. [32] Y. Xing, R. Chandramouli, and S. Mangold, \Dynamic spectrum access in open spectrum wireless networks," IEEE JSAC, vol. 24, no. 3, March 2006. [33] K. Xu, M. Gerla, and S. Bae, \E®ectiveness of RTS/CTS handshake in IEEE 802.11 based ad hoc networks," Ad Hoc Network Journal, vol. 1, no. 1, pp. 107{123, 2003. [34] ||, \How e®ective is the IEEE 802.11 RTS/CTS handshake in ad hoc networks?" in GlobeCom '02, 2002. [35] F. Ye, S. Yi, and B. Sikdar, \Improving spatial reuse of IEEE 802.11 based ad hoc networks," in GlobeCom '03, 2003. [36] Q.Zhao,L.Tong,andA.Swami,\Decentralizedcognitivemacfordynamicspectrum access," in Proc. of IEEE Symposium on New Frontier in Dynamic Spectrum Access Networks, November 2005. [37] B. Zhen, Y. Kim, and K. Jang, \The analysis of coexistence mechanisms of Blue- tooth,"VehicularTechnologyConference,2002.VTCSpring2002.IEEE55th,vol.1, 2002. 133

Abstract (if available)

Abstract Effective usage of unlicensed bands (UB) has received a lot of attention due to its potential in ubiquitous computing and networking. One key issue in effective UB usage is the coexistence among devices of homogeneous or heterogeneous systems, e.g. wireless local area networks (WLAN) and wireless personal area networks (WPAN). To resolve the coexistence problem, we need to understand the interaction between concurrent transmissions in overlapping frequency bands. Although some basic interference resolving process is mentioned in the standards, further performance improvement can be achieved by careful system analysis and parameter selection. In this research, we analyze the coexistence problem that these systems face and devise coexistence mechanisms to enhance performance.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

A framework for worst-case performance evaluation of MAC protocols for wireless adhoc networks

PDF

Transport layer rate control protocols for wireless sensor networks: from theory to practice

PDF

A protocol framework for attacker traceback in wireless multi-hop networks

PDF

Congestion control in multi-hop wireless networks

PDF

Reliable and power efficient protocols for space communication and wireless ad-hoc networks

PDF

Resource allocation in OFDM/OFDMA cellular networks: protocol design and performance analysis

PDF

Realistic modeling of wireless communication graphs for the design of efficient sensor network routing protocols

PDF

Adaptive resource management in distributed systems

PDF

Reliable languages and systems for sensor networks

PDF

Exploiting side information for link setup and maintenance in next generation wireless networks

PDF

Distributed edge and contour line detection for environmental monitoring with wireless sensor networks

PDF

Optimal resource allocation and cross-layer control in cognitive and cooperative wireless networks

PDF

Robust routing and energy management in wireless sensor networks

PDF

Multichannel data collection for throughput maximization in wireless sensor networks

PDF

Understanding the characteristics of Internet traffic dynamics in wired and wireless networks

PDF

Relative positioning, network formation, and routing in robotic wireless networks

PDF

On practical network optimization: convergence, finite buffers, and load balancing

PDF

Balancing security and performance of network request-response protocols

PDF

Fundamentals of two user-centric architectures for 5G: device-to-device communication and cache-aided interference management

PDF

Space-time codes and protocols for point-to-point and multi-hop wireless communications

Asset Metadata

Creator Hsu, Alex Chia-Chun (author)

Core Title Coexistence mechanisms for legacy and next generation wireless networks protocols

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Electrical Engineering

Publication Date 01/29/2008

Defense Date 11/13/2007

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Bluetooth 802.15,MAC protocol,OAI-PMH Harvest,Wi-Fi 802.11

Language English

Advisor Kuo, C. C. Jay (committee chair), Govindan, Ramesh (committee member), Neely, Michael J. (committee member)

Creator Email protoalex@hotmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m998

Unique identifier UC1169139

Identifier etd-Hsu-20080129 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-591411 (legacy record id),usctheses-m998 (legacy record id)

Legacy Identifier etd-Hsu-20080129.pdf

Dmrecord 591411

Document Type Dissertation

Rights Hsu, Alex Chia-Chun

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu