USF Libraries
USF Digital Collections

Blind signal detection and identification over the 2.4GHz ISM band for cognitive radio

MISSING IMAGE

Material Information

Title:
Blind signal detection and identification over the 2.4GHz ISM band for cognitive radio
Physical Description:
Book
Language:
English
Creator:
Zakaria, Omar
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla
Publication Date:

Subjects

Subjects / Keywords:
OFDM Estimation
IEEE 802.11
Spectrum Sensing
Joint Time Frequency Analysis
Cyclostationarity Features
Dissertations, Academic -- Electrical Engineering -- Masters -- USF   ( lcsh )
Genre:
non-fiction   ( marcgt )

Notes

Abstract:
ABSTRACT: 'It is not a lack of spectrum. It is an issue of efficient use of the available spectrum"--conclusions of the FCC Spectrum Policy Task Force. There is growing interest towards providing broadband communication with high bit rates and throughput, especially in the ISM band, as it was an ignition of innovation triggered by the FCC to provide, to some extent, a regulation-free band that anyone can use. But with such freedom comes the risk of interference and more responsibility to avoid causing it. Therefore, the need for accurate interference detection and identification, along with good blind detection capabilities are inevitable. Since cognitive radio is being adopted widely as more researchers consider it the ultimate solution for efficient spectrum sharing 1, it is reasonable to study the cognitive radio in the ISM band 2. Many indications show that the ISM band will have less regulation in the future, and some even predict that the ISM may be completely regulation free 3. In the dawn of cognitive radio, more knowledge about possible interfering signals should play a major role in determining optimal transmitter configurations. Since signal identification and interference will be the core concerns 4, 5, we will describe a novel approach for a cognitive radio spectrum sensing engine, which will be essential to design more efficient ISM band transceivers. In this thesis we propose a novel spectrum awareness engine to be integrated in the cognitive radios. Furthermore, the proposed engine is specialized for the ISM band, assuming that it can be one of the most challenging bands due to its free-to-use approach. It is shown that characterization of the interfering signals will help with overcoming their effects. This knowledge is invaluable to help choose the best configuration for the transceivers and will help to support the efforts of the coexistence attempts between wireless devices in such bands.
Thesis:
Thesis (M.S.E.E.)--University of South Florida, 2009.
Bibliography:
Includes bibliographical references.
System Details:
Mode of access: World Wide Web.
System Details:
System requirements: World Wide Web browser and PDF reader.
Statement of Responsibility:
by Omar Zakaria.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 142 pages.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 002063216
oclc - 556084833
usfldc doi - E14-SFE0002996
usfldc handle - e14.2996
System ID:
SFS0027313:00001


This item is only available as the following downloads:


Full Text

PAGE 1

Blind S ignal D etection and I dentification O ver the 2.4GHz ISM B and for Cognitive Radio by Omar Zakaria A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Electrical Engineering Department of Electrical Engineering College of Engineering University of South Florida Major Professor : Huseyin Arslan, Ph.D. Paris Wiley Ph.D. Arthur David Snider Ph.D. Date of Approval: May 11, 2009 Ke ywords: OFDM Estimation IEEE 802.11, S pectrum S ensing J oint T ime F requency A nalysis Cyclostationarity F eatures, F eature D etector, S pectrum A wareness Copyright 200 9 Omar Zakaria

PAGE 2

Dedication To my wife and parents.

PAGE 3

Acknowledgements First, I would like to sincerely thank my inspirer and my advisor, Dr. H useyin Arslan for his encouragement, guidance, and support It has been a great privilege to have the opportunity to work as a team and to be under his wing. Dr. Arslan, you did not only teach me engineering, you taught me how to be a better person, and for that I give you my greatest respect and acknowledgment. I also thank Dr. Arthur Snider and Dr. Paris Wiley for their support throughout my studies I am hon ored to have their guidance throughout my master program Other thanks go to Dr. Srinivas Katkoori and to Catherine Burton for giving me the first chance to be in USF, and for being the fir st ones who believed in me. To Ali Gorcin Mustafa Emin Sahin Hasari Celebi, Ahmed H i sham and Hisham Mahmoud : thank you for your technical help, and for your priceless friendship. Special thanks to my friends Sabih Guzelgoz, and Evren Terzi and the ent ire WCSP group. I also want to thank my parents and my uncle Faris for their support, love, and selfless dedication toward me. My deepest gratitude goes to my wife, Dana, for her love and all the sacrifices she has made for her support, vast patience, and steady encouragement. Dana, many times I felt that this is the end of the line, but you always managed to pull me back on the track and keep me going. For that I give you my heartfelt thanks.

PAGE 4

i Table of Con tents List of Tables iv List of Figures v List of Acronyms v iii Abstract x Chapter 1 Introduction 1 1.1 Organization of the T hesis 3 Chapter 2 Cognitive Radio Model 5 2.1 Introduction 5 2.2 Cognitive Radio History 7 2. 3 Spectrum Sensing in Cognitive R adio 1 0 2.3.1 Matched Filter 12 2.3.2 Energy Detector 14 2.3.3 Cyclostationarity Detector 1 7 2. 4 Proposed Model 18 2.4.1 The RF Front End 20 2.4.1.1 The Sampling and Data Conversion Challenge 20 2.4.2 The Energy Detector 22 2.4.3 The Feature Extraction 24 2.4.4 The Central Processing Unit (CPU) 24 2.4.5 Adaptive Transmitter 25 2. 5 Conclusion 25 Chapter 3 The ISM B and 26 3.1 Introduction 26 3.2 The Wireless Communication Systems 29 3.2.1 Spread Spectrum 29 3.2.2 Orthogonal Frequency Division Multiplexing (OFDM) 35 3.2.2.1 OFDM System Model 37 3.2.2.2 Cyclic Prefix in OFDM Symbol 39 3. 3 WiFi IEEE 802.11 Standards 41 3.3.1 WiFi IEEE 802.11b 42 3.3.2 WiFi IEEE 802.11g 47

PAGE 5

ii 3. 4 IEEE 802.15.1/2 Bluetooth 50 3. 5 IEEE 802.15.4 Zigbee Networks 51 3. 6 Microwave Ovens 53 3. 7 Cordless P hones 55 3. 8 Unknown Signals 56 3. 9 Conclusion 56 Chapter 4 Features Extraction 5 8 4.1 Introduction 58 4.2 Feature Detector 59 4.3 Bandwidth and Central Frequency Estimation 61 4.3.1 Heisenberg Gabor P rinciple 62 4. 4 Power R elated M etrics 65 4.4.1 CCDF 66 4. 5 Single C arrier v ersus M ultic arrier 68 4.5.1 Moments Based Test 69 4 6 Modulation O rder and Type of Single Carrier Signals 75 4. 7 OFDM Signals Parameters Estimation 76 4.7.1 OFDM Time Parameters Estimation 77 4.7.1.1 Useful Symbol Duration 78 4.7.1.2 Total Symbol Duration 80 4.7.1.3 Cyclic Prefix Duration 82 4.7.2 OFDM Frequency Domain Parameters 83 4. 8 Cyclostationarity F eatures 8 4 4.8.1 Introduction to Cyclostationarity 85 4.8.2 Cyclostat ionarity for Signal De tection 90 4.8.2.1 Symbol Rate Detection 9 1 4.8.2.2 Chip R ate E stimation 99 4. 9 Hopping Sequence 101 4.9.1 Joint T ime F requency A nalysis 101 4.9.2 Spectrogram 102 4. 10 Conclusion 10 4 Chapter 5 Decision Making Algorithm 10 5 5.1 Introduction 10 5 5. 2 The Proposed Framework 106 5. 3 The Decision Making 108 5.3.1 FCC Regulat ions for the ISM B and 110 5.3.2 Control and Execution 111 5.3.3 Fuzzy L ogic and Soft Decision A lgorithm 113 5. 4 Location and Time of Occurrence 119 5. 5 Conclusion 120 Chapter 6 Summary and Conclusions 12 1 6 1 Summary of Works and Contributions 121 6 2 Conclusions 1 23

PAGE 6

iii 6 3 Future Work 1 2 4 References 12 6

PAGE 7

iv List of Tables Table 3.1 The t hree main branches of the IEEE 802.11 standard 42 Table 3. 2 The 11 channels assigned by the FCC to the ISM band 43 Table 3. 3 IEEE 802.11b d ata r ate s pecification s 44 Table 3. 4 Possible Barker codes 44 Table 3. 5 T he generation of the CCK codes depending on the data bits 46 Table 3.6 Data r ate modes and m odulation for IEEE 802.11g standard 48 Table 3.7 Three d efined power categories for Bl uetooth transmission 51 Table 4. 1 Ideal values for 20 and 30 71 Table 4. 2 Ideal values for 20 and 30 7 2 Table 5.1 T he identifying features for each wireless standard 109 Table 5. 2 T he FL decision tree of E xample 1 115 Table 5. 3 T he FL decision tree of E xample 2 117 Table 5. 4 T he algorithm performance results of success rate detection 118

PAGE 8

v List of Figures Figure 2.1 : The frequency spectrum allocation in the US [32] 6 Figure 2.2 : Cognitive radio transceiver (courtesy of the author [43]) 10 Figure 2.3 : The detection probability of the energy detector in different SNR values 1 7 Figure 2.4 : Spectrum awareness engine 1 9 Figure 2.5 : Frequency domain energy detector 24 Figure 3 .1 : The ISM and U NII bands [7 ] 26 Figure 3.2 : The process of spreading the information spectrum 30 Figure 3.3 : Illustration showing the DSSS immunity to narrow band interference 31 Figure 3.4 : Basic DSSS communication system 33 Figure 3.5 : Frequency h opping spread spectrum basic transceiver 34 Figure 3.6 : S pectrum of hopping signal 34 Figure 3.7 : OFDM signal of six subcarriers 36 Figure 3. 8 : Typical OFDM s ystem 39 Figure 3.9 : Illustration of cyclic prefix extension 40 Figure 3.1 0 : I llustration of the condition for two codes to be complementary to each other 45 Figure 3.1 1 : A typical IEEE 802.11b receiver 47 Figure 3.1 2 : The spectrum of a microwave oven signal 54 Figure 3.1 3 : Microwave oven time domain signal 54 Figure 4.1 : The proposed model 60

PAGE 9

vi Figu re 4.2 : Bandwidth estimation error with SNR values 65 Figu re 4.3 : CCDF implementation 67 Figure 4.4 : CCDF curves for different modulation schemes 68 Figure 4.5 : Moment s test performance with different SNR values 74 Figure 4.6 : Moment s test for FS K signals with different orders in 1 0 t abs c hannel SNR=0 76 Figure 4. 7 : The s tructure of the OFDM s ymbols 78 Figure 4.8 : Useful symbol duration estimation algorithm results over different SNR values for 10 OFDM s ymbols with useful symbol duration of 512 samples 80 Figure 4.9 : The sliding window technique for estimating the total symbol duration 81 Figure 4.10 : The result of the sliding window correlation based algorithm when tested on the same OFDM symbols for different CP lengths 82 Fig ure 4.11 : T he success rate of the time parameters estimation 83 Figure 4.1 2 : The cyclostationary detector results for three different cyclostationary signals 88 Figure 4.13 : PSD of and 92 Figure 4.14 : PSD of and 93 Figure 4.15 : The pulse shaping process 94 Figure 4.16 : PSD of the pulse shaped signal ( ) 94 Figure 4.1 7 : PSD of 96 Figure 4.18 : Symbol rate estimation without and with using Welch periodogram 97 F igure 4.19 : Symbol rate estimation algorithm performances with respect to SNR 98 Figure 4. 20 : WLAN IEEE 802.11b DSSS signal when tested using the nonlinear algorithm 100 Figure 4.21 : Typical WLAN DSSS transmitter 100

PAGE 10

vii Figure 4.22 : Spectrogram represen tation of a B luetooth signal 103 Figure 5.1 : The spectrum awareness engine flow chart 107 Figure 5.2 : Feature detection and decision making flow chart 112 Figure 5.3 : T he algorithm response for a WLAN 802.11g input signa l 114 Figure 5.4 : T he algorithm response for a Bl uetooth input signa l 116

PAGE 11

viii List of Acronyms Federal Communications Commission (FCC) N ational Information I nfrastructure (NII) The I nternational T elecommunications U nion (ITU) Industrial S cientific and M edical (ISM) Spectrum Policy Task Force (SPTF) Central Processing Unit (CPU) Unlicensed NII (U NII) Pseudo Noise C ode (PN code) Direct Sequence Spread Spectrum (DSSS) Frequency Hopping Spread Spectrum (FHSS) Orthogonal Frequency Division Multiplexing (OFDM) Frequency D ivision M ultiplexing (FDM) Peak to Average Power Ratio (PAPR) Digital Audio Broadcasting (DAB) Wireless L ocal A rea N etworks (WLAN) Asymmetric Digital Subscriber Line (ADSL) Fast Fourier T ransforms (FFT) Discrete Fourier T ransforms (DFT) Phase Shift Keyin g (PSK) Quadra ture Amplitude Modulation (QAM)

PAGE 12

ix Inter C arrier Interference (ICI) Inter S ymbol interference (ISI) Cyclic Prefix (CP) Time Division Duplex (TDD) Microwave Oven (MWO) Complementary Cumulative Distribution Function (CCDF) M inimum M ean S quared E rror ( MMSE) Power S pectrum D ensity (PSD) Fuzzy Logic (FL)

PAGE 13

x Blind Signa l Detection and Identification O ver the 2.4GHz ISM Band for Cognitive Radio Omar Zakaria ABSTRACT -conclusions of the FCC Spectrum Policy Task Force There is growing interest towards providing broadband communication with high bit rates and throughput, especially in the ISM band as it was an ignition of innovation triggered by th e FCC to provide, to some extent, a regulation free band that anyone can use. But with such freedom comes the risk of interference and more responsibility to avoid causing it. Therefore, the need for accurate interference detection and identification, al ong with good blind detection capabilitie s are inevitable. Si nce cognitive radio is being ado pted widely as more researchers consider it the ultimate solution for efficient spectrum sharing [1] it is reasonable to study the cognitive radio in the ISM ban d [2]. Many indications show that the ISM band will have less regulation in the future, and some even predict that the ISM may be completely regulation free [3]. In the dawn of cognitive radio, more knowledge about possible interfering signals should play a major role in determining optimal transmitter configurations.

PAGE 14

xi Since signal identification and interference will be the core concerns [4] [5], we will describe a novel approach for a cognitive radio spectrum sensing engine, which will be essential to de sign more efficient ISM band transceivers. In this thesis we propose a novel spectrum awareness engine to be integrated in the cognitive radios Furthermore, the proposed engine is specialized for the ISM band, assuming that it can be one of the most cha llenging bands due to its free to use appro ach. It is shown that c haracterization of the interfering signal s will help with overcoming their effects. T his knowledge is invaluable to help choose the best confi guration for the transceivers and will help to support the efforts of the coexistence attempts between wireless devices in such band s

PAGE 15

1 Chapter 1 Introduction r eady ? T his was the message content of the first wireless transmission on May 13, 1897 by Marconi [16]. From that early time people bega n to realize the importance of wireless communication and the scarcity of the electromagnetic spectrum. Wireless networks in t he US can only operate in the band of frequencies allowed by the Federal Communications Commission (FCC), and must follow the rules regulating the way that spectrum can be used. The FCC regulations are designed to set usage rules, increase the spectrum res ource usage efficiency, and to prevent interference. Until 1985 a large portion of the spectrum in the US was leased t o individuals exclusively for particular service s such as cellular or TV broadcasting. At that time interference was not a large problem as long as the users stay ed within their assigned band of frequency spectrum. In 1985 the FCC put in place a creative plan by op ening an unlicensed band of 2.4 GHz for wireless networks. This band was regulated by the FCC P art 15 rules [1]. These rules al low new and existing technologies to share the same frequency band and try to coexist and operate together. The FCC explained that creativity and better spectrum efficiency usage would be the results from opening a shared portion of the spectrum for the un coordinated wireless devices.

PAGE 16

2 In 1995, Apple Company petitioned the FCC to create a new unlicensed 5GHz band called N ational I nformation I nfrastructure (NII). Differ ing from the 2.5GHz unlicensed band, the NII technologies rules restrict possible uses of the NII band to wireless networks that use wideband communications. The I nternational T elecommunication U nion (ITU) announced a number of bands for i ndustrial, scientific and medical (ISM) applications and services that are not restricted to any specific w ireless technologies. The ITU develops frequency assignments that are adopted by count r ies in all regions by international treaty [4]. From the early beginnings of the ISM band, it became one of the popula r destinations for wireless system manufacturers With the increasing d emand for the wireless networks, s and services that need high bit rate like video streaming, there was an increasing need for frequency spectrum resource availability, n ot to mention the importance of peaceful coexistence between wireless users. Therefore the FCC began to encourage innovation and creativity to enhance spectrum usage, and it began with the ISM band. FCC was open to new approaches and techniques to efficiently shar e the spectrum in the ISM band. One of the more promising techniques that were looked at with hope was the cognitive radio. Cognitive r adio has the ability to sense, adapt and learn t o overcome environment ch anges and possible interference [36] The biggest challenges with cognitive radio are the ability to identify the existence of the primary users and avoid interfering with them or with other cognitive radios. To have this ability cognitive radio need s to cons tantly sense the spectrum and i d entify possible

PAGE 17

3 wireless users and base d on the identification result m ake the appropriate decision to overcome their effect as interference. 1.1 Organization of the T hesis The main topics covered in this thesis can be summarized as follows: a Cognitive ra dio, models and applications (Chapter 2) b The ISM band and its players, descriptions and analysis (Chapter 3) c Wireless signals features, analysis and extractions (Chapter 4) d Smarter decision making in spectrum sensing (Chapter 5) The outlines of these chap ter s are as follows. In Chapter 2, the cognitive radio concept is provided with a brief historical look in to cognitive radio evolution over the last decade. A conventional model of cognitive radio transceivers [43] is described, and analyzed. A detailed description for variou s spectrum sensing techniques is provided, with an evaluation of each performance. A proposal for a spectrum awareness engine is described to be integrated in the cognitive radio transceivers model. A description of the first two stages (the RF front end and the energy detector) of the proposed model is provided. A n extensive study a bout the ISM band is provided in Chapter 3, a long with thorough analysis of the main wireless standards that are activ e in the ISM ban d A brief description of the main modulation schemes that are commonly used in the ISM band is also provided in this chapter.

PAGE 18

4 In Chapter 4, a description of the ISM band spectrum sensing feature detector is proposed. We submit a list of w that are useful in the process of identifying them. Algorithms are proposed to extract and de tect each feature in the list, while maintaining the lowest computational complexity as possible. In Chapter 5, we demonstrate how different wirelesses standards may inherit similar features which may lead to confusion during the detection process A novel algorithm is proposed to utilize the extracted features before making the decision, along with a controlling algorithm to regula te the rest of the s s. The thes is concludes with Chapter 6, in which we summarize the thesis and discuss open research areas.

PAGE 19

5 Chapter 2 Cognitive Radio Model In this chapter we discuss the cognitive radio technologies and examine the ideology and the evolution ary history of the cognitive radios. We will choose one of the proposed architecture and try to design a realistic model to be integrated in the proposed cognitive radio architecture. 2 .1 Introduction With the i ncreased number of wireless devices and the number of users, the awareness of the frequency spectrum scarcity increased. From the early dawn of the wireless communications era, engineers realized the importance of utilizing the spectrum to increase the nu mber of users and provide better quality of service. A c loser look at the frequency spectrum allocation by the FCC shows that the spectrum is greatly underutilized [32]. Figure 2 .1 shows the current freq uency spectrum allocation in the US

PAGE 20

6 Fig 2 .1 T he freq uency spectrum allocation in the US [32] In Jun e 2002, t he Spectrum Policy Task Force (SPTF) was established to assist the FCC in the process of identifying and evaluating changes in spectrum policy to help increase the public benefits derived from the use of the radio spectrum [33]. The SPTF released a report in November 200 2 [ 34]. In this report, the SPTF demonstrate d that the current usage for the spectrum is no t very efficient and recommended rules and regulations for the efficient use of the radio spectrum and ways to improve the existing spectrum usage. C ognitive radio is being widely adapted as many researchers look to it as the ultimate solution for efficient spectrum sharing [35] [41]. Even though there is no formal definition of cognitive radio the concept is being addressed by researchers in various contexts as well as many efforts to standardi ze it [42]. The FCC attempts to define cognitive radio as, radio or system that senses its operatio nal electromagnetic

PAGE 21

7 environment and can dynamically and autonomously adjust its radio operating parameters to modify system operation, such as [to] maximize throughput, mitigate interference, facilitate interoperability, [and] 7] The main work s of this chapter are to: a Define cognitive radio and analyze its functionalities b Study one conceptual co gnitive radio architecture extensively c Propos e a novel and realistic design for the spectrum awareness engine in the mentioned archit ecture d Propose a combination of spectrum sensing algorithms to blindly identify primary signal s 2 .2 C ognitive Radio History Cognitive radio is a relatively new concept proposed by Joseph Mitola [35] in 1999. The concept aims to create a new smart generation of communication systems that dynamically interact with the environment in real time to modify its parameters such as band of operation, central frequency, waveforms, and the used modulation It aims to establish wireless systems with a state of awareness that will efficiently utilize the spectrum, with the ability to sense learn and adapt [36]. Cognitive radio provides a solution for the spectrum underutilization problem, through an opportunistic spect rum usage [37] [38] [40]. The main idea is to temporarily use the frequency channels that are currently not occupied by the licensed user (p rima ry) through cognitive radios

PAGE 22

8 (secondary) who are constantly looking for opportunities in the spectrum without disturbing the primary user. In 1999 where the cognitive radios interact with the outside world through: a Observation through the cognitive radio sensors b Orientation, to es tablish priorities c Planning, to develop the appropriate possible set of actions d Decision, to choose the best plan for the current set of factors e Action, to execute the decision that been taken f Learn ing T his function is a cross function between observing planning and deciding, to enable the cognitive radio to learn from the past in order to better plan in the fut ure. In 2005 a simplified understanding of the cognitive cycle was proposed by Haykin [36], where the focus is on three basic units : a Spectrum sensing unit, which mainly deal s with spectrum sensing analysis and white holes detection b Channel identification unit, which deals with channel estimation c Dynamic spectrum managements unit, to cognitively manage the spectrum resources

PAGE 23

9 The publisher explained that spectrum sensing and c hannel identification functionalities function is carried out by the transmitter. In 2008 a novel cognitive ra dio model was proposed [4 3]. This model describes a cognitive radio transceiver form that consists of mainly four engines : a Cognitive engine b Spectrum awareness engine c Location awareness engine d Environment awareness engine The author co nsidered the cognitive engine to be the main e ntity that control s and monitor s the other entities in the model in order to have goal driven and self directed tas k results. In the four engine model, all the information generated by the engines goes to the cognitive engine so that the proper system con figuration, for example, the proper waveform will be decided by the cognitive engine. The main responsibility of spectrum awareness engine is to handle any job related to the frequency spectrum usage and efficiency n ot to mention the most important role for this engine, the sensing part, where the success of the cognitive radio will greatly depend on its ability to detect unoccupied spectrum. Figure 2 .2 demonstrate s the cognitive system model we will adopt in our research

PAGE 24

10 Fig 2 .2 Cognitive radio transceiver ( c ourtesy of the author [43 ] ) 2 .3 Spectrum Sensing in Cognitive Radio To achieve the main goal of the cognitive radio which is utilizing spectrum usage, the system needs to continuously monitor the spectrum and identify any white spaces that may become available. A brief literature scan shows that there are three common techniques that can be used for spectrum sensing: a Matched filter b Energy detector c Cyclostationarity d etector

PAGE 25

11 Before we explain more about the three techniques, let u s assume the following hypothesis for detecting a signal: 0 : = = 0 1 1 ( 1 ) 1 = + = 0 1 1 ( 2 ) w here is the transmitted signal, and is the added white noise. T he white noise is usually modeled as a Gaussian zero mean distribution density [ ] ~ ( 0 2 ) ; therefore the spectral density of the noise is assumed to be 2 0 represents the null hypothesis and 1 represents the d etection hypothesis That means that equal zero in case of 0 The performance of the detection system can be characterized by two probabilistic measurements, the probability of detection and probability of false alarm The probability describes the probability of detecting the desired signal on the spectrum when the signal is truly present. Needless to say, we desire the largest probability. On the other hand, represents the probability that the test incorrectly decides that the signal exists when it does not. Therefore we try to minimize the f alse detection probabil ity value as much as we can. It i s important to point out that usually in detection systems, increasing the will increase the as well, and vice versa. Therefore it is important to find the optimum balance between these probabilities in any detection algorithm [50].

PAGE 26

12 2 .3.1 Matched Filter Matched filter is a filter that maximizes the signal to noise ratio. The main strength of this filter is that due to the coherenc y ; the filter does no t need a long time to achieve high processing gain [44]. In the case that the receiver has perfect knowledge of the transmitted signal, the matched filter will be the optimal detector [45]; in this case the optimal detector test stati stic will be [48]: = [ ] ( 3 ) will be used in the signal detection process, where H 1 = T 0 = T represents the absence of the signal. The value of threshold is critical as it impacts the desired detection and false alarm probabilities. The proo f is in the following a nalysis. As shown previously, is a jo intly Gaussian random variable. Since is a result of line a r operation of jointly G aussian random variables, consequently it i s Gaussian Therefore, if we define as the average power of the sampled signal [ 4 8 ], which is = 1 ( [ ] ) 2 ( 4 ) t hen :

PAGE 27

13 ~ ( 0 1 2 ) i n the case of 0 ( 5 ) a nd ~ ( 1 2 ) i n the case of 1 ( 6 ) So the = > 1 = 2 ( 7 ) In the same way = P T Y > 0 ( 8 ) = 2 ( 9 ) In [48] it is shown that the minimum number of samples needed for a successful detection is a function of the Signal to Noise R atio (SNR) and SNR = 2 2 Therefor e = 1 1 2 1 ( 10 )

PAGE 28

14 N = O ( SNR 1 ) ( 11 ) where the O notation represents the limiting behavior of the original number of samples function simplified to focus on its growth rate. Thus, 1/ SNR i s considered the lower bound on the number of sample s which is related to the sensing time. As we mentioned b efore, in the case that the receiver already has satisfactory knowledge of the transmitted signal, the matched filter will be the optimal detector. However this is usually not the case, as we often do no t have prior information about the signal. Also since th e cognitive radio will employ matched filter techniques to perform the detection, it will need a receiver design for each possible signal, mak ing it difficult to implement in real life [46]. 2 .3.2 Energy Detector Opposite to the matched filter method, the energy detector is used when there is no prior information about the signal I t also has low computational and implementation co mplexity. For all these reasons, it i s one of the common detectors [48] [49] [50] [51] [52]. In this detector the signal energy is compared to a predefined threshold to decide if the signal is present or absent. This threshold can be adjustable in an adaptive way depending on the noise variance and the channel [50] [75] Using the same assumption for the 0 and 1 in the previous sections, we know that the noise variance is 2 Since we do no t have prior information about the signal, we can model the samples of the signal [ ] as a Gaussian random process with variance of 2

PAGE 29

15 The detector tes t statistic will be : ( ) = ( [ ] ) 2 ( 12 ) will be used in the signal detection process, where H 1 = T 0 = T represents the absence of the signal. The value of threshold is critical as it impacts the desired detection and false alarm probabilities. Therefore the P D = > 1 13 = 2 2 ( 14 ) In the same way : = P T y > 0 ( 15 ) = 2 4 + 2 ( 16 )

PAGE 30

16 Closed form expressions for probability of detection under AWGN and fading ( Rayleigh, Nakagami, and Ricean) channels are derived. Average probability of detection for energy detector based sensing algorithms under log normal shadowing and Rayleigh fading channels is derived in [76]. Also it is proven that the minimum num ber of sampled required is = 1 1 2 2 ( 17 ) N = O ( SNR 2 ) ( 18 ) It is obvious for this kind of detector we need a higher number of samples in case of low SNR s compared to the matched filter detector. Some of the difficulties with the spectrum sensing based on the energy detector alone are: a The threshold value selection b The inability to distinguish interference from primary signal c Poor performance under low SN R values [74] The performan ce of the energy detector for 10 OFDM symbols in a Gaussian noise channel is shown in Figure 2 3.

PAGE 31

17 Fig 2 .3 The detection probability of the energy detector in different SNR values 2 .3.3 Cyclostationarity Detector Another technique used recently in research is the cyclostationary features searching in signals as a way to identify them. The cyclostationary theory was first introduced by G ardner [54] in his famous paper series about the exploitation of the cyclostationary features in random process es [54] [63]. Gardner tried to analyze the signals by extracting the hidden frequencies that exist in man mad e signals due to modulatio n, pulse shaping, shifting in frequency, sampling, repeated spreading codes an d a ny operation that may introduce a signal through the communication system. The theory explained that the communication processes that are applied on the original source signals introduce hidden frequencies (the author calls these cyclic frequencies) in the result signal. These frequencies can be det ected using a mathematical tool developed by Gardner which is the

PAGE 32

18 cyclic autocorrelatio n and the spectral correlation function. In the past the computational complexity was a large problem in the cyclostat ionarity analysis operations due to the nature of the estimation [64]. But with the development of FPGA s and microprocessors, this theory became popular is used in many proposed algorithms for signal detections [65] [70]. We will explain more about the c yclostationary analysis in Chapter 4 2 .4 Proposed Model Three detection techniques are used in this research for the purposes of spectrum sensing: energy detection, matched filtering, and cyclostationary feature detection. The three techniques are combined in the spectrum awareness engine design. Figure 2 4 describes the proposed s pectrum awareness engine. The proposed mod el consists mainly of five unit s: a RF front end s b Energy detector for initial stage channel sensing c Features extraction unit, for detailed detection and identification d Processing unit for decision making and controlling the engine component e Adaptive waveform generator of the transmitter (included for consistency )

PAGE 33

19 Fig 2 .4 Spectrum awareness engine As shown in [43] the spectrum awareness e ngine will pass the information to the cognitive engine and both the location awareness and the environment awareness engines. All these engines will cooperate to decide the best configuration for the current situation the cogniti ve radio is in. An example on how the spectrum awareness engine can cooperate with the location and/or environment engines can be that t he expected range information of the detected signal can be fed to the location awareness engine to participate in the decision making of the location, especially in the case of known wireless standards where usually the average range of the signal is predefined. As for the role of the spectrum engine information on the waveform configuration, it is important to know sign al features such as the duty cycle, the used hopping sequence, and the number of subcarriers in order t o design a signal that is robust against interference. In the following subsection we w ill briefly describe each unit of the spectrum awareness engine.

PAGE 34

20 2 .4.1 The RF Front End From the cognitive radio point of view, having an effective spectrum sensing ability requires cognitive radio to cover a large range of frequencies at the RF front end and t hen carry on the sampling process through a high speed ana log to digital (A/D) converter. This particular task became more possible after the development of sub sampling theorem and techniques. 2 .4.1.1 The Sampling and Data Conversion Challenge In cognitive radio applications RF signals need to be directly digitized by the cognitive radio RF front end [35] [40]. According to Nyquist, in order to successfully reconstruct a sampled signal, we must sample the signal at no less than twice the frequency of its highest frequency components Cognitive radio will deal with a wide range of frequencies, especially in the range of Gig a H ertz like the ISM band. This means that the ADC need s to sample the signals at much higher speeds tha n what current ADC s are capable of. To give an example, if we seek a signal in th e ISM band with a central frequency of 2.4GHz, we will need to sample it with a sampling frequency of at least 5GHz. Many techniques were developed to solve this sampling frequency problem in cognitive radios. One of these solutions is the sub sampling o [79]. This theory states that i f a band pass waveform has a s pectrum over the frequency band: f l < f < f h ( 19 ) a nd occupied bandwidth of

PAGE 35

21 B T = f h f l ( 20 ) t he signal may be reproduced from sample values if the sampling rate is f s = 2 B T ( 21 ) Thus instead of requiring an ADC with a sampling frequency at the Nyquist rate of at least 2 f h baseband sampling allows an ADC with a much lower sampling rate to do the same job. This leads t o much lower signal processing. After sampling the signal, measurements for detecti on of the primary user will be carried out [71]. We can safely say that one of the success ion factors for the cognitive radio will be the RF front end quality and flexibility to scan wideband in accurate and sensitive manners [46]. In the proposed algorithm, the band of interest will be selected down convert ed to the baseband, and sampled thr ough the wideband antennas with the help of adjustable band pass filters and the down converters. Signals can be found anywhere in the spectrum band of interest, which raise s the need for adjustable filters and local oscillators for the down conversion [78]. The dynamic range of the signal is an important factor in the cognitive radio RF front end to have suitable sensitivity for the low SNR signals. This is where the role of the A/D converter comes in, as it should be adaptive enough to cover a wide range of dynamic ranges.

PAGE 36

22 The output of the filter is sampled at Nyquist rate and N point FFT is applied to obtain the frequency domain samples which can be modeled as : ( ) = 0 + ( ) 1 = 1 , ( 22 ) w here X n the transmitted signal at the output of the FFT is W ( n ) is the white noise samples, and N is the used FFT size. Ma n y studies dea ling with the RF front end design and issues have been cond ucted [46] [71] [73] H owever, because it is not our focus in th is study, we will not consider them. 2 .4.2 The Energy Detector In this researc h we propose an energy detector as first stage sensing to help detect the presence of the signals before we process the sampled data and extract its features. This way we reduce the computational complexity of the whole process. After successfully receiving and sampl ing the band of in terest the blind signal detection process will begin in a form of energy detector to initially decide if there is a signal or just noise. The energy detector will also help in the decision process of whether the width of the band pass filter is sufficien t enough to capture the whole signal without losing any frequency domain information. Fine tuning to the correct central frequency and bandwidth of the presented signal will help in achieving some coherency in the detection. Furthermore, detecting the ba ndwidth of the signal will help to sample the filtered band at Nyquist rate.

PAGE 37

23 The energy detection is performed in the frequency domain. The magnitude square of the fast Fourier transform s (FFT) of the signal is calculated, and the output is compared to a predefined threshold to make the first judgment if a signal exist s or not. The processing gain in this method will be proportional to FFT size N and the averaging time T. Increase in t he size of FFT improves the frequency resolution whic h is helpful in detecting narro w band signals. Furthermore, if we reduce the averaging time it improves the SNR by reducing the noise power [44]. The energy estimation in the frequency domain can be described as: ( ) = ( ) 2 ( 23 ) w here ( ) represents the FFT output of the sampled spectrum and ( ) = 0 + ( ) 1 = 1 , ( 24 ) So the detection criteria will depend on the test equation, along with a predefined 1 = E 0 = E represents the absence of the signal. The impact of choosing the threshold on the detection performance wa s explained in S ection 2 .2. Figure 2 .5 demonstrate s the proposed energy detector design.

PAGE 38

24 Fig 2 .5 Frequency d omain energy detector 2 .4.3 The Feature Extraction When detecting energy in the band of interest which may i ndicate the presence of a signal, t he sampled signal will be passed to the features extractor to detect the main features that are present especially the bandwidth and central frequency so as t o fine tune the RF front end Also in this stage, detailed identification will be carried out based on the detecting features present in the sign al. This process will thoroughly be explained in C hapter 4 where we illustrate the features extraction methods and the cycl ostationarity detection method. 2 .4.4 The Central P rocessing Unit (CPU) This unit is responsible for the decision making process that is based on the parameters coming from the rest of the sensing and feature detection unit s. Also the CPU cont rols the rest of the engine unit s to optimize the spectrum awareness engine. T his stage will be explained in Chapter 5 where we describe the decision maki ng algorithm.

PAGE 39

25 2 .4.5 Adaptive Transmitter After identifying white space s in the spectrum detecting if there are any active signal(s), and reveal ing its properties, t he spectrum awareness engine should use the proper configuration for t he transmitter that provide s the best spectrum utilization and interference robustness Some of these configurations will use modulation schemes, duty cycle, hopping sequence, band of operation, bandwidth etc. By the spectrum awareness engine doing this and by cooperating with the rest of the cognitive radio engines, the best performance outcome is achieved. 2 .5 Conclusion In this chapter, we examine d the concept of the cognitive radio, and briefly described its history and previous work in cognitive radio research The cognitive radio is built on the principal of opportunit y to efficiently utilize the frequency spectrum. A creative model of cognitive radio architecture with location and env ironment awareness c ycles [43] wa s described. The importance of the spectrum awareness and spectrum sensing of the model was addressed and a brief analysis of the various spectrum sensing was conducted. In this chapter we proposed a novel design for spectrum awareness engin e and spectrum sensing algorithm that will be integrated with the cognitive radio architecture [43]. The RF front e nd and the energy detector unit design were also described.

PAGE 40

26 Chapter 3 The ISM B and In this chapter we will discuss the ISM band features and the FCC regulations for this band. We will discuss the main features of the active wireless standards in the ISM band. 3 .1 Introduction In the US the FCC defines the ISM and u nlicensed NII (U NII) bands as shown in Figure 3 1. The ISM bands are scattered in three different frequency bands, namely 900MHz, 2.4GHz, and 5.7GHz. U NII ba nds are mainly located in the 5 GHz segment of the frequency spectrum. Fig 3 .1 T he ISM and U NII bands [7] Those bands are license free where manufacturer s that build wireless devices operating in these bands are not required to buy the spectrum from FCC. However, there are some

PAGE 41

27 regulations concerning these bands and these are outlined in [1]. Each band has its own regulation and regulations may change fro m one to another. The 2.4 GHz band provides an attractive medium for many applications using the wireless technology that currently exists or may come up in the future. Different from other frequency bands where interference is avoided between wireless de vices through separation of operating frequencies, the ISM is a shared band which allows unlicensed wireless activities. Therefore coexistence between wireless devices is important to ensure perfo rmance. Operating in the 2.4GHz segment of the spectrum, the ISM band provides t he convenience of the license free band with worldwide availabil ity. Many wireless standards have been deployed to operate on the ISM band, such as wireless local area networks (WLAN), which is considered to be the largest wireless standard active in the ISM band. Also operating on the ISM band are the Bluetooth and Zigbee networks, some cordless phones, along with non standard wireless devices like microwave ovens. Coexistence between various wireless devices in the ISM b and was and still is the focus of much study and research To give an example about its importance, consider a wireless access point in a university library which provides the campus population with wireless access to the I nternet and the university data b ase. In the same library there are students using laptops and PDA s to access the I nternet, others using cellulars w ith s ome using Bluetooth headset s All these devices are using the same medium access ; specifically, the 2.4G Hz ISM band. Many possible sc enarios of interference between the w ireless devices can be envisioned in this specific example.

PAGE 42

28 Before we go further with this study, it is reasonable to first identify the standards and wireless technologies that are active in the ISM band, so that we c an study each separate standard and identify its key features. It is worth mentioning that our main concern will be the ISM 2.4GHz band; therefore, we will study the standards that are available in this band only. We presume (as many other studies in the literature do) that the major players in the ISM band can b e broken down to the following: a WiFi IEEE 802.11 standard b Bluetooth IEEE 802.15 s tandard c Cordless p hones d Zigbee n etworks IEEE 802.15.4 e Microwave f Unknown s ignals (prospective standards or potential secondary users) Before we explain the main features, properties, and di fferences of each standard, we will first explain some important modulation techniques that will play a major role in both the content of the standards, and the path of blind detection that we adopt in this research. The main work s of this chapter are to : a Analyze the main modulation schemes that are used in the ISM band b Study the ISM band wireless standards and active wireless devices extensively

PAGE 43

29 c Identify the physical layer fea tures in each wireless standard which can be used in the p rocess of blind identifications. 3.2 The Wireless Communication Systems Since the beginning of the wireless communication area, engineers competed to develop the best techniques to utilize spectrum usage and enhance spectrum management, in order to increase network capacity and achieve t he highest bit rate performance, b esides many other motivations like t he security, quality of service, etc. Communication systems evolve d over the decades from simple analog modu lation like the AM, FM and PM t o digital modulation like MSK, FSK, and PSK. With advances in integrated circuits and the development of the microprocessor, even more developed and complicated forms of modulations and wireless communication concepts beg a n to appear, all to support the overall performance of current communication systems, and to accommodate modern service demands and the rapidly increasing number of users. 3.2.1 Spread Spectrum Spread s pectrum is one of the popular digital communication schemes because of its various properties that makes it suitable for secure, multiple acce ss communication networks. The fact that it is hard to intercept or detect i s one reason why it was first used by the military [14]. Spreading spectrum may be defined as: a means of transmission in which the signal occupies a bandwidth in excess of the minimum necessary to send the information. The band spread is accomplished by means of a code which is independent of the data, and

PAGE 44

30 synchronized reception with the code at the receiver is used for de spreading, and This means that the occupied bandwid th of certain data is spread to a wider bandwidth, which will extend its power over a wider range at the same time. As shown in Figure 3.2, this is achieved by multiplying the signal with a higher freque ncy code sequence. The operation will spread the power spectrum density of the signal, reducing the effect of narrow band interference (both intentional and unintentional) which is one of the main features of the spread spe ctrum, as shown in Figure 3.3. Fig 3 .2 T he proc ess of spreading the information spectrum

PAGE 45

31 Fig 3 .3 Illustration showing the DSSS immunity to narrow band interference Other good properties of the spread spectrum are summarized below. a It has g ood tolerance towards narrow band interference and jammers. b It has h igh security due to the use of random codes which are known only to the transmitter and receiver. c It is suitable for multiple accessing, where more than one user shares the same bandwidth at the same time, such as has been deployed in the CDMA systems. Spread s pectrum can be classified into two main categories : Direct Sequence Spread Spectrum (DSSS) and Frequency H opping Spectrum (FHSS). The DSSS scheme uses a pseudo random sequence of positive and negative pulses at a very high repetition rate (chip rate) to spread the data bandwidth signal. The data signal is multiplied by the spreading code in the baseband stage and then up converts the signal to the required carrier frequency. The form of the spread signal at the output is given by:

PAGE 46

32 = ( ) cos ( + ) ( 25 ) where a(t) is a sequence of pulses used to spread the data, and d(t) is the digital data. At identical to the spreading signal applied at the transmitter. Figure 3.4 shows a basic system for a DSSS scheme. The spreading signal is called Pseudo Noise code (PN code). The PN sequences are high bit rate binary sequences, which exhibit randomness properties just like noise. The PN code rate is called the chipping rate (to dist inguish it from the information rate), so called because the code sequence applied to each bit results in chipping the original bit into smaller bits. The most important property of the PN sequence is its correlation properties. PN sequence should show n oise like correlation properties to the outsider, but the sequence is known to the two devices that are using it. The definition of randomness was studied by Golomb and requires three properties, which are described in [9]. Examples of the PN sequence ar e the M sequences, Gold codes and Kasami sequences.

PAGE 47

33 Fig 3. 4 Basic DSSS communication system On the other hand, i n FHSS transmission the random or PN sequence is used to change the carrier frequency in a random manner. This will cause spreading the data signal over a wide range of frequencies, yet no change to the original bandwidth of the data will occur. Instead, various portions of the data will be modulated and transmitted over different carrier frequencies. The or der and sequence of the carrier frequencies depends on the used PN sequence. The simplest frequency hopping form is given by: = b m cos 2 f m t P T b ( t m T b ) ( 26 ) where is the information sequence, is part of N frequencies chosen to be the random frequency sequence ; s o the data signals hop to a new frequency eve ry number of bits as shown in Figure 3.5. This way the information data is spread through frequency hopping. The time duration over which the data signal spends in each frequency is called

PAGE 48

34 the dwell time Figure 3.6 illustrates a simple frequency hopping communicatio n system. Fig 3 .5 Frequency hopping s pread spectrum basic transceiver F ig 3 .6 S pectrum of hopping signal

PAGE 49

35 3.2.2 Orthogonal Frequency Division Multiplexing (OFDM) Orthogonal Frequency Division Multiplexing (OFDM) is a multicarrier modulation scheme that provides efficient bandwidth utilization. OFDM is a mixture of special form of multicarrier modulation and special case of frequenc y division multiplexing (FDM) at the same time. Where the bandwidth itself is divided into independent subcarriers each subcarrier is modulated by a portion of the data after dividing the dat a in to parallel parts and then re multiplexed to create the OFDM carrier. At each subc arrier the data is modulated at a relatively low rate. This give s immunity against the delay spread of the channel. Ideally each subcarrier is narrow enough to face a flat fad ing channel. One way to intuiti vely look at the way OFDM works i s to use the analogy of making a shipment via truck. We have two options : we can either hire a big truck or four smaller trucks. Both methods carry the same amount of material (data) But in case of accident (interference) only 1/4 th the amount of material (data) in the entire shipment will suffer. This is exactly how the OFDM show s tolerance toward s interference ; in the case of interference, only some subcarrier s will get affected while the rest will not [10]. The main difference between the FDM and OFDM system is that OFDM does not use guard band to separate its subcarrier. On the contrary, OFDM allows some overlapping between the subcarrier without corrupting the data, through the orthogonality of the subcarriers which is the main concept of the OFDM. The subcarriers are chosen in such a way that there is no influence of other carriers in the detection of the information in a particular carrier when the orthogonality is maintained. Since the carriers are all

PAGE 50

36 sinusoidal waves, we know that the area under one full period of sinusoidal wave should equal zero. In the same way, if we multiply sinusoidal waves with different frequencies, the area under the product is zero if the sinusoi dal were orthogonal to each other as shown in F igure 3 .7. Fig 3 .7 OFDM signal of six subcarriers Although OFDM is relatively new concept, it has gained a great deal of att ention during the last decade a s it overcame many challenges, especially the one s associated with high bit rate communication, the main problems being frequency selectivity and time dispersion. OFDM is use d by many applications nowadays, including WLAN systems, Digital Audio Broadcasting (DAB) [11] and Terrestrial Digital Video Broad casting (DVB T) [12] in Europe, and in Asymmetric Digital Subscriber Line (ADSL) [13]. With all these powerful properties of the OFDM, it has its weak points, such as s ensitivity to frequency offsets caused by the mismatch between the transmitter and recei ver oscillator. This is a problem to the OFDM because it causes loss of orthogonality. Another unp rofitable problem is the large Peak to A verage Power Ratio (PAPR) of the OFDM signal, which requires high quality power amplifiers with large linear r anges. O ther problems include phase distortion, time varying channel and time

PAGE 51

37 synchronization which are not our main concern s in this research. To show the importance of the OFDM modulation and because it has a large role in the ISM band wireless standa rds we will describe in more detail the OFDM system and features in the following sections. 3.2.2.1 OFDM System Model The Discrete Fourier t ransform (DFT) of the discrete sequence y ( k ) with a length of N, Y k is defined as [13] Y k = y ( k ) e j 2 kn N N 1 k = 0 ( 27 ) a nd the Inverse Discrete Fourier transform (IDFT) is represented as y n = 1 N Y ( k ) e j 2 kn N N 1 k = 0 ( 28 ) As stated earlier, the OFDM system convert s the data stream from serial form to parallel blocks, each block with size of By using IDFT we obtain the OFDM signal. The t ime domain samples can be described as = { } ( 29 )

PAGE 52

38 = 1 ( ) 2 1 = 0 = 0 , 1 ( 30 ) w here X ( k ) is the symbol transmitted on the kth subcarrier and N is the number of subcarriers. The symbols are obtained from the data bits after being digitally modulated using one of the modulation schemes like Phase Shift Keying ( PSK), Quadrature Amplitude Modulation (QAM), etc. The symbols X k are considered a frequency domain signal and the samp les x n are considered the time domain of the signal. We have already stated that the most important fact about the OFDM is the orthogonality of the subcarriers. Only if we achieve orthogonality will we have no effects from the other subcarriers in the de tection of information at a particular subcarrier at the receiver. Otherwise loss of the orthogonality will cause inter carrier interference (ICI) Therefore, to maintain the orthogonality of the OFDM symbol the following should be achieved: 1 = ( 31 ) f is the subcarrier spacing, and Ts is the useful symbol duration. So if N point IDFT is used the total bandwidth of the OFDM signal will be W = N f ( 32 ) The time domain signal is then extended to avoid the inter symbol interference (ISI) between symbols. A typical OFDM system is shown in Figure 3 8

PAGE 53

39 Fig 3 8 T ypical OFDM s ystem 3 .2.2.2 Cyclic Prefix in OFDM S ymbol Passing signals thr ough a time dispersive channel may cause ISI and frequency selectivity i f the delay spread of the channel is greater than the symbol duration. Having I SI in the OFDM system can cause loss of orthogonality which may lead to an ICI problem. To overcome this problem a method introduced by Peled and Ruiz [15] proposed to cyclically extend the OFDM time signal by copying the last part of the OFDM time signal, called the cyclic prefix (CP), and replicat ing it at the front of the symbol during the transmission. This is th en remove d at the receiver side before demodulating the signal. One issue to be considered is that the CP length should be more than the delay spread to assure that the multipath components of the symbols will not interfere with the useful symbol to avoid the ISI, as shown in Figure 3 9 This way the CP will have three benefits:

PAGE 54

40 a It serve s like a guard to protect the symbols from ISI b It c an be used for synchronization and blind signal identification. c It will prevent the ICI because CP will convert the liner convolution with the channel i mpulse response in time, which cause s a scalar multiplication in the frequency domain resulting in preservation of the orthogonality. Fig 3 9 Illustration of cyclic prefix extension The main features and basics of t he OFDM system can be summarized by the following: a OFDM can achieve high bit rate with high delay spread tolerance.

PAGE 55

41 b OFDM system divides the data into lower bit rate parallel bit streams, and each parallel bit stream is modulated on an individual subcarrier out of N total number of subcarriers. c OFDM use s CP technique to avoid ISI and ICI. 3 .3 WiFi IEEE 802.11 S tandards The w ireless local area networks (WLAN) technologies appeared in the markets and beg a n to quickly increase the number of shipped equipment and the number of users thanks to rapid internet growth, busin esses data networks, and low co st integrated wireless radio designs. The first widely deployed wire less LAN solutions used the 2.4 GHz ban d since in the beginning this band was assigned for spr ead spectrum technologies [4]. Individual and large businesses widely ado pt ed IEEE 802.11 wireless network access points and client devices. IEEE 802.11 standard has three main branches: a IEEE 802.11a which works in the 5 GHz band b IEE E 802.11b, which wo rks in the 2.4 GHz band c IEE E 802.11g, which works in the 2.4 GHz band Table 3. 1 gives a quick glanc and main features.

PAGE 56

42 Table 3.1 The three main branches of the IEEE 802.11 standard Protocol Release Date Operation Feq. Data R ate (max) Modulation Technique Range (Radius Indoor) Range (Radius Outdoor) 802.11a 1999 5 GHz 54 Mbit/s OFDM 35 Meters 120 Meters 802.11b 1999 2.4 GHz 11 Mbit/s DSSS 38 Meters 140 Meters 802.11g 2003 2.4 GHz 54 Mbit/s OFDM 35 Meters 140 Meters Since our only concern is the blind detection in the 2.4GHz band, we will not deal with the IEEE 802.11a standard, not to mention that this standard has a lot of similarities with the IEEE 802.11b standard except in the band of operation. Also worth mention ing is that we will only focus on the physical layer featu res and properties that concern us in our detecting algorithm. 3.3 .1 WiFi IEEE 802.11b The I EEE 802.11b operates in the 2.4 GHz band. The FCC assi gns 11 channels in the ISM band, as sh own in Table 3.2. F or this standard, each channel is 22 MHz bandwidth [21 ].

PAGE 57

43 Table 3. 2 The 11 channels assi gned by the FCC to the ISM band Channel Lower Frequency Center Frequency Upper Frequency 1 2.401 2.412 2.423 2 2.404 2.417 2.428 3 2.411 2.422 2.433 4 2.416 2.427 2.438 5 2.421 2.432 2.443 6 2.426 2.437 2.448 7 2.431 2.442 2.453 8 2.436 2.447 2.458 9 2.441 2.452 2.463 10 2.451 2.457 2.468 11 2.451 2.462 2.473 Only three of these channels are none overlapping : 1, 6, and 11. Th is standard uses the DSSS modulation scheme and has different data r ate modes which are 1Mbps, 2Mbps, 5. 5Mbps and 11Mbps. The used spreading codes in this standard are the Barker code sequences in the low data rate mode (1, 2 Mbps) and the Complementary Code Keying (CCK) in the high data rate mode. The rest of the main features are shown in Table 3. 3. These spreading codes are used because they have low autocorrelation properties, as explained earlier in this chapter.

PAGE 58

4 4 Table 3. 3 IEEE 802.11b d ata r ate s pecifications Data Rate Code Length Modulation Symbol Rate Bits/Symbol 1 Mbps 11 (Barker Code) BPSK 1 MSps 1 2 Mbps 11 (Barker Code) QPSK 1 MSps 2 5.5 Mbps 8 (CCK) QPSK 1.375 MSps 4 11 Mbps 8 (CCK) QPSK 1.375 MSps 8 Barker sequences codes consist of sequence s of +1s and 1s. The Barker code lengths that are used in the DSSS modulation are 11 and 13. Table 3.4 shows the possible B arker codes Table 3. 4 Possible Barker codes Length Codes 2 +1 1 +1 +1 3 +1 +1 1 4 +1 1 +1 +1 +1 1 1 1 5 +1 +1 +1 1 +1 7 +1 +1 +1 1 1 +1 1 11 +1 +1 +1 1 1 1 +1 1 1 +1 1 13 +1 +1 +1 +1 +1 1 1 +1 +1 1 +1 1 +1

PAGE 59

45 On the other hand, CCK code was first proposed by Golay [19]. Binary c omplementary codes are a subset of CCKs. These c odes are pairs of finite code sequence s with the same length. T he condition for two codes to be considered as complementary of each other is that the summation of the auto correlation func tions of each code should yield zero except for zero lag as shown in Figure 3. 1 0 It must be mentioned that the codes used in 802.11b are not real but complex (i.e. poly phase). (a) Autoco rrelation of Code 1 (b) Autocorrelation of Code 2 (c) Summation of the two autocorrelation Fig 3. 1 0 I llustration of the condition for two codes to be complementary to each other

PAGE 60

46 In 802.11b, CCK codes are generat ed using the formula: = 0 , 7 = ( 1 + 2 + 3 + 4 1 + 3 + 4 1 + 2 + 4 1 + 4 1 + 2 + 3 1 + 3 1 + 2 1 ) 33 In 11Mbps and 5.5 Mbps data rate modes, data bits are split into chips each having 8 and 4 bits respectively. Those chips are u sed to generate the spreading CCK code. In the case of 11 Mbps 6 out of 8 bits are used to determine the phase values and the remaining two are used to modulate the signal in QPSK by exploiting the common phase term in each code element. While in 5.5Mbps mode, 2 out of 4 bits are used for code generation and the remaining two are used for QPSK modulation Therefore, the poss ible number of CCK codes for 11 Mbps is ( 2 6 ) whereas it is ( 2 2 ) for 5.5 Mbps. Depending on the data bits the phases 1 , 4 are map ped in Table 3. 5 Table 3. 5 T he generation of the CCK codes depending on the data bits DIBIT(d i+1 ,d i ) Phase 00 0 01 10 11

PAGE 61

47 It i s only reasonable to have a correlation based receiver to detect the IEEE 802.11b standard, and this is what happens in reality. At the receiver the signal is correlated with every possible codeword. Figure 3. 1 1 demonstrates a typical diagram for a IEEE 802.11b receiver [20]. Fig 3.1 1 A typical IEEE 802.11b receiver To sum up the properties of the IEEE 802.11b standard: a It operates in the 2.4 GHz frequency range b It has 11 c hannel s assigned to it in the US, occupying a bandwidth of 22MHz. c Bit rate modes are 1Mbps, 2Mbps, 5.5Mbps and 11Mbps. d It employs Direct Sequence Spectrum Spreading (DSSS) e The lower data rates use Barker s equence s whereas the h igh data rates use Complementary Code Keying (CCK) 3 .3.2 WiFi IEEE 802.11g The I EEE 802.11g operates in the 2.4GHz band. T he FCC assigns 11 channels to it in the ISM ban d, and each channel is 22 MHz bandwidth [21]. The used channels are shown

PAGE 62

48 in Table 3. 2 Only three of these channels are none overlapping : 1, 6, and 11. Data r ate modes and modulation order are shown in Table 3. 6. Table 3. 6 Data rate modes and modulatio n for the IEEE 802.11g sta ndard Data Rate (Mb/s) Modulation Coding rate Coded bits/ subcarrier Coded bits/Symbol Data Bits/Symbol 6 BPSK 1/2 1 48 24 9 BPSK 3 / 4 1 48 36 12 QPSK 1/2 2 69 48 18 QPSK 3 / 4 2 69 72 24 16 QAM 1/2 4 192 96 36 16 QAM 3 / 4 4 192 144 48 64 QAM 2/3 6 288 192 54 64 QAM 3/4 6 288 216 This standard uses the OFDM modulation which makes it more effective in a multipath environment than the IEEE 802.11b standard. The number of subcarriers is 64, out of which 11 subcarriers at the end of both sides of the spectrum are set to zero for spectrum shaping reasons and to suppress the sideloops at the end of the OFDM spectrum to minimize the ICI. These sh ut off subcarriers will work as a guard bands at both ends of the spectrum. One subcarrier at zero frequency is set to zero as well to help the D/A and A/D converters and to get rid of the DC offset. Leaving 52 active subcarriers, four of these subcarriers are BPSK modulated pilot tones used for channel estimation. The

PAGE 63

49 subcarrier spacing is 312.5 KHz The total OFDM symb ol is 4 s; the useful symbol duration is 3.2s, and the CP rate in this standard is 1/4. Due to the total symbol duration th e symb ol rate of this standard is 250 KHz Due to the use of OFDM system, the PAPR is usually high, and it is vulnerable to Doppler spread. One interesting f eature in the IEEE 802.11g standard is that it supports higher data rates using the OFDM, and the low rates using CCK/Barker as well to ensure backward compatibility with existing IEEE 802.11b equipment. To sum up the main featu res of the IEEE 802.11g standard: a It operates in the 2.4 GHz frequency range b It has 11 c hannel s assigned to it in the US, occupying a bandwidth of 22MHz. c It has high bit r ate modes d It u ses OFDM modulation e It h as a useful symbol duration of 3.2 s, and a whole symbol duration of 4s f It h as subcarriers numbering 64, and a subcarrier spacing of 312.5KHz g It i s spectrally efficient h It i s more effective in a multi path environment (ISI) i It i s capable when narrow band interference is present j It h as a high PAPR.

PAGE 64

50 3 .4 IEEE 801.15.1/2 Bluetooth Bluetooth technology was first developed by Ericsson in 1994. This standard ope rates on the 2.4GHz bandwidth It is considered a short range (up to 10 meters) wireless personal area network (WPAN). It became very popular from the beginning of its development for its various applications and the services that can b e provided through it, from cellphone headsets to laptop applications and many others. Bluetooth standard uses a mixture of Time D ivision Duplex (TDD) and FHSS transmission mode over 79 channels with 1MHz spacing wit hin the range of 2.400 2.4835 GHz assigned to this sta ndard by the FCC. The central frequencies are chosen f rom the following equation [22]: = 2402 + = 0 , 78 ( 34 ) There are t wo data rate modes : the basic data rate with symbol rate of 1Mbp s and the enhanced data rate with symbol rate of 2Mbps/3Mbps. The signal hops from one channel to another with a rate of 1600 times per second. The hopping sequence is derived using a pseudo random sequence det ermined by the master device in the network and broadcasted to the slave devices. Transmission time is divided in to 625s time slots. One packet of transmission can take from one up to five time slots [23]. Two hopping modes in the Bluetooth are availab le. The basic is where the device use s a fix ed hopping list regardless of the channel status And t he adaptive frequency hopping (AFH) incorporate s interference identification to update the hopping list and exclude any

PAGE 65

51 channel that contains interference source. There are three defined power categories for the Bluetooth transmission, listed below and illustrated in Table 3.7. Table 3 .7 Three defined power categories for Bluetooth transmission Power Class Max. Output Power Nominal Output Power Min. Output Power Distance 1 100 mW (20 dBm) N/A 1mW (0 dBm) 100m 2 2.5 mW (4 dBm) 1 mW (0 dBm) 0.25 mW ( 6 dBm) 20m 3 1 mW (0 dBm) N/A N/A 10, To sum up the main features of the Bluetooth: a It operates in the 2.4 GHz frequency range. b It has 79 c hannel s with 1MHz separation occupying a bandwidth of 1MHz c It has two bit r ate modes d It uses FHSS and TDD. e Its t ime slot length is 625s, and the transmission can use up to five time slots f It has r esistance to interference, especially with the AFH mode g It ha s t hree power transmission modes 3. 5 IEEE 802 .15.4 Zigbee N etworks Zigbee is p art of the WPAN family that operates in ISM band and has the features of being small, low maintenance, and low power. It i s used for communication applications that require lo w data rate, a secure network, and low power consumption. This standard

PAGE 66

52 covers a transmission range up to 75 meters [26]. In the 2.4GHz ISM band, Zigbee has 16 defined channels with 5MHz bandwidth each. The central frequency of each channel is calculate d as : = 2405 + 5 11 = 11 12 , 26 ( 35 ) The bit rate offered is 250Kbps, with a symbol rate of 62.5Ksps. The modulation scheme used in this standard is the DSSS with a chip rate of 2000Kcps [25]. According to the standard specifications [24], the transmitter power is 0.5mW ( 3dBm). One of the main advantages of the Zigbee is the low duty cycle communicatio n with less than 10ppm duty cycle. L owering the duty cycle minimizes the power consumption thus increasing battery life. Transmission intervals may range as follows [27] : 15 36 2 0 14 ( 36 ) To sum up the main fea tures of the Zigbee that are useful for our purposes: a It operates in the 2.4 GHz frequency range. b It has 16 c hannel s with 1MHz separation occupying a bandwidth of 5MHz c Its b it r ate is 250Kbps d It us e s DSSS, with chip rate 2000Kcps. e Its t ime s lot length can be between 15 36 up to 251.65824 seconds f It has a low duty cycle (< 50% ).

PAGE 67

53 g It has l ow transmission power 3dBm h Its range is up to 75m 3 6 Microwave Ovens Microwave o vens (MWO) are the perfect example of non intentional interference of transmitters in th e 2.4GHz ISM band. Although m icrowave ovens were not meant to transmit electromagnetic waves, they usually leak th ese waves during operation in scattered power all ove r the ISM ban d. This phenomenon causes a non intentional interference and disturbs the other devices operating in the same band. Many studies have addressed the microwave signal model and its interference effects [29] [30] [31]. Using these studies as reference as well as examining a real microwave recorded signal we noticed that t he spect rum in microwave ovens has a distinguished shape ( see Figure 2. 1 2 ) and an occupied bandwidth of 20 MHz, where most of the energy is concentrated in 15MHz bandwidth. T he time domain signal is transmitted as bursts during the positive cycle of the standard electri c power lines frequency [29] W hen the positive cycle voltage exceeds some threshold, two bursts appear ( these bursts ar e referred to in the literature as the transient parts ) One s tarts at the beginning of the ON cycle, and the other one at the en d of the ON cycle of the microwave. The width of each transient part is ~ 1 The microwave signal in the ON mode is somehow similar to frequency modulation (FM) signals. The frequency sweep of the FM signal in the microwave has a duration close to half of the time period duration of the electricity power line, so in the US it is between 5 7ms. There are chan ging power levels during the frequency sweep of the ON period. These changes in the power level are expressed as an Amplitude

PAGE 68

54 Modulated (AM). So, the frequency sweeping part of the microwave signal is modeled as a combined AM FM signal waveform [31]. Fig ure 3. 1 3 shows a time domain m icrowave signal with th e two bursts that represent the transient parts of the ON period marked as A and B. Fig 3. 1 2 T he spectrum of a m icrowave oven signal Fig 3. 1 3 Microwave oven time domain signal

PAGE 69

55 Studies have shown that a microwave signal can be best modeled by the foll owing mathematical formula [30]: = cos 2 + sin 2 < 0 5 ( 37 ) w here = ( 2 ) and is the sweep tim e. To sum up the main features of the microwave oven standard that are useful for our present study: a It is in the 2.4 GHz band. b It has n o predefined channels or central frequencies. c It has p eriodic transmission, occupying a bandwidth of 20MHz d It transmits in bursts (transient parts) synchronized to electric power lines cycle. e It has a d istinguished power spectrum shape f Its s ignal is modeled as an AM and FM signal. g Its t ransient part width ~ 1 the AM FM part duration is 5 7ms. h Its d uty cycle is close to 50% 3 7 Cordless Phone s Cordless telephones have been o ne of the most popular technologies in the telecommunication market for a while now. Currently, t here ar e many types of cordless phones, depending on the band of operation. Since we are concerned with the ISM band, we will focus only on the types that operate on the 2.4GHz rang e The f irst noticeable

PAGE 70

56 feature is that cordless phones do n o t follow a specific protocol or standard. Each manufacturer defines its own devi and RF front end specifications. Most cordless phones that work in the 2.4GHz rang e use FHSS or DSSS. The devices that use DSSS have 8 16 channels of a bandwidth between 5MHz or 10MHz compared to the Bluetooth with its 1MHz bandwidth 79 channels. The bit rates for cordles s phones are less than 100kbps. 3 8 Unknown Signals ISM band technology would no t be an area of innovation without expecting to have unknown signals every now and then such as n ew prospective standards, cognitive radio secondary user s, and new unintentional interference This class is random and uncertain, yet it has to follow the FCC regulation in the ISM band. This fact can help us to form some idea s about what we may face. Therefore, we add unknown signals to our study as wel l to be pre pared for any future situation. 3 9 Conclusion The ISM band is a license free band, where wireless activities share the same spectrum with very limited regulations. Due to this fact it is now one of the attractive bands for manufacturer s, and many wireless standards are operating in this band. We have described the variety of wireless technologies that are working in this band, and we demonstrate d how important coexistence is for all these wireless activities to operate safely in this part of the spectrum.

PAGE 71

57 In this chapter we looked closely at the modulation schemes and communication systems that can exist in the ISM band, and demonstrate d how each system tries to utilize the spectrum and h ow they handle interference. We also thoroughly examined ea ch of the wireless standards and activities that may operate in the ISM band, and we identif ied the main physical layer features and properties of each wireless standard. Different wireless standards can utilize the same features or modulation techniques.

PAGE 72

58 Chapter 4 Features Extraction In this chapter we describe the features extractions stage and explain the algorithms used to extract each feature. A comprehensive list of features that can be used to detect the presence of the ISM band technologies is discussed and analyzed. 4.1 Introduction As we explained in C hapter 2 the cognitive rad io should have the capability to blind ly identify interference and try to mitigate its effects. T his capability will be executed in the spectrum awareness engine. In C hapter 2 a novel design for the spectrum awareness engine was proposed Descriptions of the RF front end and the energy detection components were given Studies show that energy detection alone is not sufficient to have an accura te idea about the available spectrum or the interference [74] [44] [81] [82]. Therefore we propose a feature detector stage to deeply explore the captured to try to identify them. Many studies in the literature examined the various features of the wireless signals. Some even proposed methods to extract these features. Some examples include : a In [50] the bandwidth is estimated through the use of FFT operation b In [84] 4th order cumulants test is used to extract the used carrier systems

PAGE 73

59 c In [102] m oment s test is applied to reveal the carrier system d In [103], a cosine modulated bank filter is used to blindly identify the multicarrier modulation e In [108] autocorrelation function is used t o estimate the OFDM time parameters f In [110] cyclostationarity is deployed to estimate the OFDM frequency domain parameters As it is shown, different approaches are proposed to extract different types of features. In this research we define the possible features that can be targeted and propose a comprehensive algorithms to extract each one of these features with the appropriate approach. The main work s in this chapter are to : a I dentify the possible PHY layer features that can help in the detection process. b S tudy the cyclostationarity theory and the cyclostationarity detector. c B uild algorithms to detect each of the proposed features. d P ropose a feature detector design that consists of handful of algorithms to detect the various features. 4.2 Feature Detector After applying the energy detection and making the first decision about signal presence, we try to extract as much information as possible from the captured signal These signal features can be detected using a single approach or multiple approaches such as

PAGE 74

60 autocorrelation based test, cyclostationarity based parameters extraction, and joint time frequency analysis. These algorithms can be used together for extracting the different features that may present in the signal Figure 4.1 illustrates the proposed design Fig 4.1 The proposed model First we define the p hysical layer features and characteristics that can be used to identify signals and interferences as below: a Power related: SNR, Peak to Average Power Ratio (PAPR) b Time o f occurrence (statistical observation over a period of time) c Frequency domain related features: central frequency, bandwidth (OBW, 3dB BW) d Duty cycle e Statistical characteristics: mean, variance, CCDF, moments (2 nd autocorrelation function properties f Cyclostationary feature

PAGE 75

61 g Distinguishing between single carrier or multicarrier h Single carrier: digital modulation, DSSS, FHSS i M ulticarrier parameters: time (symbol duration, CP du ration) and frequency (subcarrier spacing, number of subcarriers ) j Modulation type and order k Chip rates l Symbol rates m Hopping sequence n FCC regulation A comprehensive algorithm is proposed to extract each one these features. One thing to point out is that we took in consideration the computational complexity in the design of each algorithm 4.3 Bandwidth and Central Frequency E stimation T he wireless standards usually utilize predefined bandwidth s (depending on the data rate). The bandwidth of a detected signal is estimated and the bandwidth value is used in the process of identificat ion [50] [93]. The same applies to the central frequency of operation, as different wireless standards use different predefined central frequencies. Even in frequency hopping spread spectrum, there are certain predefined centra l frequencies the devices will operate on, as we observed in the Bluetooth case. There are some proposed ways in the literature for bandwidth and central frequency estimation. For instance in [94] wavelets decomposition is used to calculat e the

PAGE 76

62 bandwid th of the signal. I n [ 95] the author uses the Welch periodo gram to calculate the average power spe ctrum and find out its length, t hen detect s the two endpoi nts of the signal spectrum, calculate s the distance between these points and find s the bandwidth. I n [50], FFT is applied on the time domain signal, and a threshold is defined to decide which frequency bins are occupied to calculate the start and the end of the signal bandwidth. In our proposed algorithm, right after the energy detection stage, we need to check if th ere is one signal or more than one signal i n the spectrum, and to make sure that we captured all of the signals For this purpose we calculate the power spectrum density of the signal (PSD), and pass the PSD to an edge detector algorit hm to make sure that we have only one signal in the sampled spectrum To estimate the bandwidth and central frequency, we use the time frequency Heisenberg Gabor inequality concept. 4.3.1 Heisenberg Gabor P rinciple In many cases, the time frequency resol ution of a signal is restricted to the Heisenberg Gabor inequality Signals can be characterized in both time and frequency domains at the same time by considering their mean localization and dispersions in each of the mentioned domains. If we have: ( ) 2 ( 38 )

PAGE 77

63 a nd ( ) 2 ( 39 ) representing the probability distribution of the signal in both time and frequency domain respectively we can calculate the mean and the standard deviation as: = 1 ( ) 2 ( 40 ) = 1 ( ) 2 ( 41 ) 2 = 4 ( ) 2 ( ) 2 ( 42 ) 2 = 4 ( ) 2 ( ) 2 ( 43 ) is the energy of the signal and assumed to be finite: = ( ) 2 < ( 44 )

PAGE 78

64 Since we can calculate the power spectrum density of the signal ( ) we define the following: = 1 ( ) 2 ( 45 ) = 2 ( ) 2 ( ) 2 ( 46 ) where is the central point of the power distribution, hence the central frequency and is the frequency spreading around the ce nter point, hence the bandwidth. Then the Heisenberg Gabor inequality is: BT 1 ( 47 ) The main feature of this method of estimation is the simplicity of computation. The power spectr um density of sampled signals is easily calculated thanks to the simplicity of the current FFT circuitry allowing just two equations to give us a good estimat e for the bandwidth and the central frequency. Also t his method is independent of the SNR value, which means that we do no t need an SNR estimator. Figure 4.2 illustrates the performance of the proposed bandwidth estimation algorithm against various SNR v alues. As we can see, the algorithm gives relatively low error rate in low SNR values. The signal used in this evaluation is OFDM signal, 10 symbols, FFT size 512, CP 1/8.

PAGE 79

65 It is w orth mentionin g that in the case of real recorded data we neglect some samples at the beginning and at the end of the signal spectrum to take the roll off factor of the filter in to consideration and compensate for the drop in magnitude at both ends due to the 4.4 Power Related Metrics The signal power and the SNR of the received signal can be a useful tool to provide an idea about the identity of the signal. For example in Zigbee networks the power transmission is low according t o the Zigbee networks specifications ( 3dBm). In Bluetooth there are three power transmission modes and each power mode has a specific tr ansmission distan ce; meaning that in the c ase of Bluetooth technology the transmission Fig 4.2 Bandwidth estimation error with SNR values

PAGE 80

66 power can indicate the effec tiv e distance of the device. And that is why power metrics are calculated in the proposed algorithm and feed into the decision making part of the algorithm. 4.4.1 CCDF The move t o 3G systems and the adoption of OFDM modulations is pushing signals to have higher peak to average power ratios. Current OFDM based communication systems combine subcarriers, resulting in a peak to average. This signal characteristic can be an identifying feature for the OFDM based sy stems, especially if we have prior knowledge about the primary signal statistics [97]. Here the Power Complementary Cumulative Distribution Function (CCDF) curves come into the picture as they provide critical information about the peak to average power b ehavior of the signal. The CCDF plot describes how much time the signal spends at or above a given power level [96]. To explain how t o construct the CCDF curves, let u s consider a signal power level with time representation, as in F igure 4.3a. The signa l in the mentioned form is difficult to quantify due to its randomness. In order to get some useful power information from the signal, we can statistically describe the power levels with respect to the average power in the signal. Figure 4.3b represents a specific power level above the average. We calculate the percentage of the time the signal spends at or above each power level which represents the probability for that particular power level, as in Figure4.3c. Then the CCDF can be defined as the powe r levels with respect to the average versus their probability. With the prior knowledge of the expected signal statistics and the channel, CCDF can help with the blind identification of the signals, especially the multicarrier based ones.

PAGE 81

67 ( a) The signal p ower level in time (b) Define average power level ( c) CCDF Curve Fig 4.3 CCDF implementation Figure 4.4 illustrate s the algorithm results for different types of modulations

PAGE 82

68 Fig 4.4 CCDF curves for different modulation schemes 4 .5 Single Carrier versus Multic arrier ISM band contain s different types of wireless standards as explained in Chapter 3. Some standards ado pt the multicarrier approach like the OFDM based WLAN, and some take the single carrier approach, like co rdless phones. Knowing this we identify the importance of detecting t carrier system, n ot only to participate in the process of the decision making of the blind detection but also to reduce the computationa l complexity of the decision making. There are two methods in the literature to discriminate the single carrier and multicarrier systems. Those are the 4th order cumulants test and the moments test. In the cumulant based test, since OFDM signals has Gaussian distribution or close to Gaussian a time domain statistical test for Gaussianity is applied on the signals [98], to detect if the signals are using multicarrier transmission. This approach was used for the

PAGE 83

69 first time by Akmouch e in 1999 [ 84 ]. According to the cumulants test, cumulants of order k > 3, which are generalizations of autocorrelation function, can be used to quantify departures from Gaussianity [98]. So if the data in hand (sampled signal) has a Gaussian distributio n, the k th order cumulants disappear for k > 3 w here the cumulants is defined as [98]: ( 1 , 1 ) = + 1 + 1 = ( 48 ) Some weak points were noticed in this m ethod of multicarrier test F or instance the test was SNR depende nt, and the a ccuracy of the results heavily a ffected in dispersive ch annels. For those reasons we ch ose not to use the cumulants based test. 4.5.1 Moments Based T est Moments test was first used as a modulation type and order identifier for single carrier systems by evaluating the summation results of power law elements [101]. Later on the test proposed to be used for the multicar rier signal identifications [100 ] [102 ]. To explain t he moment s test algorithm let us consider the baseband sampled signal model as: = + ( ) ( 49 ) where is the received signal, is the transmitted signal, and is the white Gaussian noise. The mixed moments of the received signal will be:

PAGE 84

70 + = ( ) ( ( ) ) ( 50 ) where th e denote d refers to the conjugation. Therefore we can form: 2 1 = = ( ) 2 ( 51 ) 4 2 = 2 2 = ( ) 4 ( 52 ) 6 3 = 3 3 = ( ) 6 ( 53 ) Furthermore, we define two parameters: 20 = 4 2 / 2 2 1 ( 54 ) 30 = 6 3 / 3 2 1 ( 55 ) The idea l values for 20 and 30 are shown in Table 4. 1

PAGE 85

71 Table 4. 1 Ideal values for 20 and 30 Since = + ( ) then: = 2 = 2 + 2 = + ( 56 ) The moments will be: 2 1 = = + ( 57 ) 4 2 = 2 2 = 2 2 + 4 + 2 2 ( 58 ) 6 3 = 3 3 = 3 3 + 9 2 2 + 18 2 + 6 3 ( 59 ) 20 30 64 QAM 1.378 2.21 32 QAM 1.306 1.88 16 QAM 1.312 1.93 MPSK 1 1 MFSK 1 1 OFDM 2 6

PAGE 86

72 So the new parameters for are: 20 = 4 2 2 2 1 = ( ( ) 4 ) 2 ( ( ) 4 ) = 20 ( / ) 2 + 4 / + 2 ( / ) 2 + 2 / + 1 ( 60 ) and: 30 = 6 3 3 2 1 = 6 3 2 = 30 ( / ) 3 + 9 20 ( / ) 2 + 18 / + 6 ( / ) 3 + 3 ( / ) 2 + 3 / + 1 ( 61 ) where 20 30 are the parameters 20 and 20 scaled by 2, and 6 respectively [102] and can be chosen from the Table 4. 2 Table 4. 2 Ideal values for 20 and 30 20 30 64 QAM 1.3782 2.216 32 QAM 1.3062 1.886 16 QAM 1.3122 1.936 MPSK 12 16 MFSK 12 16 OFDM 22 66

PAGE 87

73 Let us remember that = 10 10 ( / ) ( 62 ) From equation ( 60 ) we can derive the SNR equation: = 20 2 4 + 20 20 2 20 2 20 20 20 ( 63 ) Up to this point we need m 20 or the modulation type in order to estimate the SNR The algorithm proposes to use the SNR e stimation and modulation charac t e rization in [104] jointl y to detect the multicarrier signals as follows: a Calculate the moments 2 1 4 2 6 3 of the sampled signal. b Calculate 20 and 30 c Assume that the modulations that can be detected are A=B+C, either single carrier modulations B={ = 2 4 8 = 2 4 8 = 16 32 64 } or multicarrier modulation C={OFDM} d Assume a particular modulation is received in the sampled signal, the corresponding ( ) 20 and ( ) 30 can be obtained from T able 4. 2 e E stimate the ( ) through the estimation equation ( 63 ) using 20 and 20 values. f C alculate the estimation value of 30 we call it 30 using the ( ) 20 30 values in ( 61 ) equation.

PAGE 88

74 g R epe a t steps 4 6 and calcu la te the 30 for all possible modulations in {A} h C alculate the estimation error 30 30 for each modulation in {A} i Final ly, select t he modulation used in the rec e ived signal based on the minimum mean squared error(MMSE) criterion : = 30 30 2 ( 64 ) The algorithm is tested for different SNR values, and through 10 sample spaced uniformly distributed channel taps channel, t o evaluate the performance. Figure 4. 5 illustrates the outcome of the algorithm for different modulation schemes, when the ( ) 20 and ( ) 30 parameters are set to be OFDM signal [100]. We can clearly see that we have the minimum (MMSE) values when the signal is OFDM based, which indicate s that the tested signal was OFDM (a) 10 t abs c hannel SNR=10 (b) 10 tabs c hannel, SNR=15 Fig 4.5 Moments test performance with different SNR values

PAGE 89

75 (c) 10 t abs channel, SNR= 0 Fig 4. 5 (Continued) 4.6 Modulation O rder and Type of Single Carrier S ignals The same algorithm that is described in 4.5.1 is used to identify the digital modulation order. The same (MMSE) argument will hold when the algorithm is applied with different modulations parameters and the values that reflect the least MMSE value in ( 64 ) will represent the modulation used in the received signal. Despite the simplicity of the moment s test, it has been proven that it can be misleading when used to identify the digital modu lation orders [105], especially when the received signal has FSK modulation. Figure 4. 6 illustrates the results of the moment s test algorithm when the transmitted signal is FSK, showing that the test gives incons iste nt results. We explain how to overcome this problem by using our fuzzy logic like decision making process in Chapter 5.

PAGE 90

76 Fig 4. 6 Moment s test for FSK signals with different or ders in 10 tabs c hannel SNR=0 4.7 OFDM Signals Parameters E stimation There are many proposed algorithms to blindly estimate the OFDM parameters in both time domain and frequency domain [10 7 ] [10 8 ] [109 ] [1 10 ]. In [108] autocorrelation is performed and the total symbol duration is estimated through the distance between the correlation peaks. The cyclic prefix (CP) length is estimated through joint time frequency transform. In [109] the useful symbol duration is calculated through au tocorrelation based algorithm, w hile the tota l symbol duration is calculated by finding the distance between consecutive peaks in cross correlation based algorithm. In [110] different approaches were taken, where the author estimate s the sampling frequency using the cyclostationarity introduc ed b y the signal oversampling, uses the result of the cyclostationarity to estim ate the number of subcarriers, a nd finally estimate s the CP length and the symbol duration through cyclostationarity based algorithm.

PAGE 91

77 After a de tailed search in the literature and testing the pro posed methods o n both simulated and real captured signals, w e narrowed down our approaches to the following The OFDM symbol duration will be calculated through an autocorrelation based algorithm. The total symbol duration will be calculated through a slicing cross corr elation algorithm with fix ed window length. CP duration will be calculated based on the total symbol duration and the useful symbol duration results. The subcarrier spacing is calculated from the useful symbol results, which will eventually lead to the c alculation of the number of subcarriers. 4.7.1 OFDM Time Parameter s E stimation Let us recall the OFDM signal model and symbol component that we explained in Chapter 3 OFDM system converts the serial data stream into parallel parts of size N and modulates these parts in to different subcarriers through the inverse discret e Fourier transform (IDFT).The t ime domain signal can be described as : x n = IDFT X k = 1 N X k e j 2 kn N N 1 k = 0 n = 0 , N 1 ( 65 ) w here X ( k ) is the symbol transmitted on the kth subcarrier and N is the number of subcarriers. The OFDM time signal is cyclically extended by copying the last part of the OFDM symbol, and replicat ing it at the front of the symbol during the transmission. Figure 4. 7 illustrate s OFDM symbol s tructure

PAGE 92

78 Fig 4. 7 The s tructure of the OFDM s ymbols where T s is the total symbol duration, T c is the cyclic prefix duration, and T u is the useful symbol duratio n. Let us also assume that the baseband received signal over multipath channel is: r ( t ) = h l ( t ) s t l + w ( t ) l 1 l = 0 ( 66 ) w here s ( t ) is the OFDM signal, w ( t ) is the white Gaussian noise h l ( t ) is the path complex gain representation, with the path delay 1 and 1 is the sample spaced channel taps As shown, OFDM symbol will have cycli c reception which should cause correlation properties to exist between them in the OFDM symbol. We use this fact to develop algorithms to estimate the time parameters of the OFDM symbol. 4.7.1.1 Useful Symbol D uration After estimating the central frequency and the occupied bandwidth, the signal can be down converted and sampled And the autocorrelation function of the received signal r ( t ) can be defined as [109]:

PAGE 93

79 E r n n + = s 2 + w 2 = 0 s 2 e j 2 = N u ( 67 ) 0 other N u represents the useful symbol duration. Then the usefu l symbol duration will be: N u = ( ) ( ) = 1 2 , ( 68 ) W here N is the number of samples acqui red during the observation time, R Use(n) is the correlation function of the received signal with different correlation lags and en Use ( n ) is the power of data in each correlation window to nor malize the correlation results. So the peak site N u is the length of useful symbol in samples. This algorithm is robust against the frequency offset and phase offset [111] [112]. The performance of the algorithm is tested over different values of SNR and number taps multipath fading channel. Figure 4. 8 illustrates the acquired peak through the algorithm in diffe rent SNR values with a 15 sample spaced uniformly distributed multipath channel taps The r eason that the multipath does not overcome the useful symbol duration peak is that at the lag equal to the useful symbol duration, the CP of all symbols will correl a te at the same time, which create s a relatively high correlation power compared to the multipath components peaks.

PAGE 94

80 (a) 15 tabs c hannel SNR=1 ( b) 15 tabs c hannel, SNR=5 ( c) 15 tabs c hannel SNR=10 ( d) 15 tabs c hannel, SNR=20 Fig 4. 8 Useful symbol duration estimation algorithm results over different SNR values for 10 OFDM s ymbols with useful symbol duration of 512 samples 4.7.1.2 Total Symbol D uration The t otal symbol duration is estimated through the periodicity feature the OFDM symbol has due t o the CP [112]. An algorithm has been designed to search for the CP periodicity b y using a sliding correlation window with fixed window length e qual to the pos sible CP lengths and fix ed correlation length equal to the estimated useful symbol duration. To reduce the computational complexity we use our knowledge about the possible C P sizes

PAGE 95

81 in the wireless system standards which are 1/4, 1/8, 1/16, and 1/32 Figu re 4. 9 illustrates the mechanism of our algorithm. Fig 4. 9 The sliding window technique for estim ating the total symbol duration When using this method, consecutive peaks will be obtained. As we go closer to the actual CP length, we notice that the co nsecutive peaks become smoother. However this is not sufficient to be detected using MATLAB It was observed that the distance between neighboring consecutive peaks equal s the total symbol duration ( sy mbol duration + CP duration); therefore we mea sure t he distances between the midpoints of each two consecutive peaks and use a histogram to detect the most repeated value. This value will be equ al to the total symbol duration, therefore: 21 < 22 < 23 < ( 69 ) = 2 + 1 2 = 1 2 3 ( 70 ) = ( 71 )

PAGE 96

82 w here L 21 L 22 L 23 i s the midpoint of each consecutive peak in the correct sequence in which they appear, H p is the histogram fu nction of the distance between each two neighboring peaks, and N s is the total symbol duration estimation. Figure 4. 10 illustrates the consecutive peaks due to the sliding window algorithm with different values of CP. Fig 4. 10 T he result of the sliding window correlation based algorithm when tested on the same OFDM symbols for different CP lengths 4.7.1.3 Cyclic P ref ix D uration After estimating the useful symbol duration and the total symbol duration, the cyclic prefix wil l simply be the result of subtracting them both: N c = N s N u ( 72 ) The actual CP length

PAGE 97

83 Up to this point all the time paramete rs are estimated and detected, and what is left are the frequency domain parameters. Figure 4.11 shows the success rate of our algorithm for different SNR values. 4.7.2 OFDM Frequency D omain P arameters It was shown in Chapter 3 that it is important for the OFDM symbol to sustain the subcarrie r orthogonality. In order to do that this condition should apply: 1 T u = f ( 73 ) w here f is the subcarrier spacing. Thus, if N point IDFT is used the total bandwidth of the OFDM signal will be: Fig 4.11 T he success rate of the time parameters estimation

PAGE 98

84 W = N f ( 74 ) w here W is the total bandwidth of the OFDM signal and N is the FFT size. Assuming that the received O FDM si gnal sustains its orthogonality and since the useful symbol duration is known at this stage, we simply calculate the subcarrier spacing through equation ( 72 ) Furthermore, since the total bandwidth is known, the number of subcarrier can be calculated as well: N = W f ( 75 ) As it has been illustrated, no prior information is required in all th e proposed estimation algorithm, and no synchronization is needed. 4. 8 Cyclostationarity F eatures Cyclostationarity feature detection is one of the most popular methods for blind signal detection and identification [46] [65] [70] [1 26 ]. Many researchers look at the c yclostationarity as the answer to many blind identification and spectrum sensing problems. In this section we try t o explain the c ycl ostationarity and its features so that we may incorporate it into our signal identification algorithms. Much of the next section is part of the unfinished work of Dr. Arthur Snider [115].

PAGE 99

85 4. 8 .1 Introduction to C yclostationarity The c yclostationary theory was first introduce d by Gardner [54] in his paper series about the exploitation of the c ycl ostationary features in random processes [54] [63]. In [115] the c yclostationary process is described as a stationary random process (signal) that has been engineered and modified by a periodic operation, such as amplitude modulating (AM) the signal, freq uency shifting the signal, sampling the signal, or filtering the signal. For the purpose of illustration, let us look at some examples of cyclostationary processes. Assuming ( ) is the stationary random signal, then: = ( 76 ) a nd 1 2 = ( 1 2 ) ( 77 ) If we multiply ( ) with a periodic function like: a Frequency shifting, then: = b AM then : = cos ( + ) c Sampling then : = ( ) = All these operations will result in a c yclostationary signal and will introduce frequencies that were not in the original stationary process. The cyclostationary analysis

PAGE 100

86 in [54] is a method to detect these artificial frequencies that was introduced to the stationary p rocess for engineering purpose; the goal being to design a tool that will detect the hidden frequencies and ignore the original frequencies in the signal: = 0 Hidden frequencies 0 = Hidden frequencies ( 78 ) where [ ] is the proposed detector. In [54] and later in [115] this detector was developed in a form of a math ematical tool, t hat is: 1 ( ) / 2 / 2 1 ( 79 ) To examine the effect of this tool we apply it on the three examples of the cyclo stationary processes we mentioned earlier: a Frequency shifting a stationary signal : = ( 80 ) lim 1 ( ) / 2 / 2 = lim 1 / 2 / 2 ( 81 ) 1 representation for the c yclostationary detector [54].

PAGE 101

87 = = 0 ( 82 ) T he detector picks up the of the carrier frequency shifter. b AM: = cos ( + ) = ( + ) + ( + ) 2 ( 83 ) lim 1 ( ) / 2 / 2 = lim 1 ( + ) 2 + ( + ) 2 / 2 / 2 = 2 = 2 = 0 ( 84 ) T he detector picks up the frequencies of the amplitude modulator. c Sampling : = = = 1 2 = ( 85 ) lim 1 ( ) / 2 / 2 = lim 1 1 2 = / 2 / 2

PAGE 102

88 = = 2 0 86 So the detector will pick up the harmonics of the sampling frequency Figure 4.1 2 illustrates the different cyclostationary detector results for the cyclostationary processes examples. ( a) Frequency shift stationary signal ( b) AM signal (c) Sampled signal Fig 4.1 2 The cyclostationary dete ctor results for three di fferent cyclostationary signals One drawback the detector has occurs when the st ationary process (signal) has zero means.

PAGE 103

89 In this case: = 0 ( 87 ) T he detector will not work as planned. To overcome this problem we pass the signal through the nonlinear operation like a quadratic to force the signal mean to be nonzero. For instance, passing a zero mean stationary signal through a square law operation will change its mean to nonzero. In [54] the author proposes to multiply the signal with a conjugated shifted version of itself as a nonlinear operation to avoid the zero mean signal case. So if: = ( 88 ) t hen = + 2 2 ( 89 ) and the detector will be: lim 1 + 2 2 = 0 / 2 / 2 ( 90 )

PAGE 104

90 w here = 2 This final form of the detector is called the cyclic autocorrelation function (CAF). For a signal with finite samples to represent it, the cyclic autoco rrelation will be estimated by : = + 2 91 I n the process a spectral correlation function (SCF) is defined to simplify the detector function in some cases. The SCF for a sampled signal will be : = = ( ) 2 92 Using these tools we can examine the cyclostationarity features of signals. It is shown that all modulated signals contain cyclostationary features [ 54] [61] [88] [90] [107]. S o it i s only reasonable to examine the possible features in the received signal and to try to map the detected features to the possible standards. 4. 8 .2 Cyclostationarity for S ignal D etection As stated earlier, the cyclostationary approach is one of the most common methods for blind detection and spectrum s ensing. In our algorithm we use the cyclostationarity features to detect two main hidden periodicities in signals. Those are the symbol rate of t he single carrie r based signals and the chip rate of the direct spread spectrum (DSSS) signals.

PAGE 105

91 4. 8 .2.1 Symbol R ate Detection Assume that is a zero mea n stationary random process then: = 0 93 a nd = ( ) 94 a t h as defined autocorrelation and power spectrum density (PSD) where: = ( ( ) ) 95 Let be the amplitude modulation of at 0 carrier frequency : = ( ) cos 2 0 96 t hen the power spectrum density of is: ( ) = 1 4 + 0 + 1 4 0 97 Figure 4.1 3 illustrate s the PSD of both signals.

PAGE 106

92 Fig 4.1 3 PSD of and The carrier frequency f 0 in the AM signal is hidden periodicity due to the modulation operation. If we pass the signal though a quadratic operation like square law operation as suggested by [54] the result will be: = ( ) 2 = ( ) 2 cos ( 2 0 ) 2 98 = 1 2 + ( ) cos ( 2 ( 2 0 ) ) 99 w here : = ( ) 2 = K + c ( t ) 100 = 2 > 0 101

PAGE 107

93 Since has non zero mean, this will result in spectral line in the PSD at the zero frequency a nd two spectral lines components in the PSD of as shown in Figure 4.1 4 ( ) = 1 4 + + + 2 0 + 2 0 + 1 4 + 2 0 + 1 4 2 0 102 Fig 4.1 4 PSD of and Therefore the quadratic operation reveals the hidden periodicity of the AM signal. In digital communication systems ( ) is sampled and the pulses ( if pulse shaped ) are transmitted through a pulse shape d filter to prepare the signal and make it more suitable to be t ransmitted through the channel, as illustrated in F igure 4. 1 5

PAGE 108

94 Fig 4.1 5 T he pulse shaping process To examine the effect of the pulse shaping l et us assume that: = 103 w here is zero mean data, is the pulse shaping filter and is the symbol rate. Then the PSD of will be: ( ) = 1 ( ) 2 104 Figure 4.1 6 I llustrate s the PSD mentioned above. Fig 4.1 6 PSD of the pulse shaped signal ( )

PAGE 109

95 Again there are no spectral lines in the PSD, but the symbol period will cause a built in periodicity in the signal that we can look for. If we pass the signal through quadratic transformation of the square law, we have: = 2 = 105 w here = ( ) 2 = K + c ( ) 106 = ( ) 2 107 = ( ) 2 > 0 108 Now the squared signal has a positive mean, so its PSD will have spectral line components at each As illustrated in F igure 4.1 7 the PSD representation will be: ( ) = 1 ( ) 2 + 109

PAGE 110

96 Fig 4.1 7 PSD of As we show that spectral lines will appear at each period of the symbol rate, we use this feature and apply it on our received signal to detect the symbols rates as follow s We use the effect of nonlinear operations on the pulse shaped sig nal to detect the symbol rate pass the received signal through a square law operation, and calculate the power spectrum representation: = 2 110 ( ) = ( 2 ) 111 If we apply this operation on received signals to test the cyclostationary features, we are able to detect a peak that correspond s to the sy mbol rate of the digitally modulated signals as shown in F igure 4.1 8 a. Furthermore, [114] and [116] recommend to detect the symbol rate feature using the Welch periodogram [117]. A cyclostationarity detector is developed using Welch periodogram to detec t the symbol rate of the digital modulation type signals. The resul t of the algorithm is shown in F igure 4.1 8 b.

PAGE 111

97 ( a ) N onlinearity based algorithm to detect the symbol rate ( b) Welch based algorithm to detect the symbol rate Fig 4.1 8 S ymbol rate estimation without and with using Welch period o gram We notice that there is a dominant peak when the detector frequency is equal to the symbol rate. Furthermore there is another peak at frequency zero. To isolate the zero frequency peak we designed th e algorithm to search between 0 25 to make sure that we will pick up only the peak correspond ing to 1 We picked up this range based on the following analysis.

PAGE 112

98 In pulse shaped signal case s, the filter bandwidth and rol l of f factor impact the occupied bandwidth of the signal as follow s : 1 2 ( 1 + ) 112 w here is the roll off factor of the puls e shaping filter is the symbol rate, and is the signal bandwidth. Knowing that < 1 it is obvious that the bandwidth of the signal is larger than the symbol rate. Figure 4.19 illustrate s the symbol estimation perfo rmance with respect to the SNR in 5 Ta bs AWGN channel. Fig 4.19 Symbol rate estimation algorithm performances with respect to SNR

PAGE 113

99 4. 8 .2.2 Chip Rate Es timation Applying the same algorithm described in the previous section on the DSSS signals will result in re vealing the chip rate features. As we explained in Chapter 3 each bit of duration is spread in to a sequence of chips. Therefore : = 113 w here is the chip duration and is the spreading sequence length. The algorithm reveal s the chip rate as well as the symbol rate of the origin al data before spreading. As illustrated in F igure 4. 20 the used signal is WLAN signal IEEE 802.11b. The chip rate of this signal is 11Mcps, and the symbol rate is 1Mbps. We observe discrete spectrum lines with symbol rate int ervals, as well as a peak at 11 MHz that correspond s to the chip rate of the system Figure 4. 2 1 illustrate s a typical WLAN 802.11b transmitter.

PAGE 114

100 ( a) Full spectrum of the algorithm output ( b) Zoomed spectrum Fig 4. 20 WLAN IEEE 802.11b DSSS signal when tested using the nonlinear algorithm Fig 4. 2 1 T ypical WLAN DSSS transmitter These discrete spectral lines can help with the detection of DSSS signals by searching for the ir existence Also, we estimate the symbol period and the chip width by calculating the space of the discrete spectrum lines at the same time. A similar approach is used in [118] and those results matched our algorithm results.

PAGE 115

101 4. 9 Hopping S equence In frequency hopp ing spread spectrum (FHSS), a hopping sequence is deployed to spread the signal over a wide range of frequencies to avoid interference. The data stream is divided and transmitted over different central frequencies after modulating each part with Gaussian frequency shift keying (GFSK). The knowledge of the used hopping sequence is crucial to demodulate the received signal at the receiver. In this research we propose a method to detect the hopping sequence as a unique feature of each FHSS standard. 4.9 .1 Joint T ime F requency A nalysis The main benefits of the joint time frequency (JTF) are to give us the temporal spectrum components of the signal. Using JTF analysis will help us reveal the behavior of the signal in both time and frequency at the same time. This information is particularly important in case of frequency hopping signals, where both time information and frequency information will be n eeded t o analyze the hopping sequence. There are a handful of studies and approaches about the JTF analysis in t he literature [119] [120]. Some use the short time Fourier transform (STFT), others use wavelets transform approach, while Gabor expansion is deployed by other researchers. In this research we adapted the STFT approach to conduct our JTF analysis. I n STFT we simply divide the signal to short period s of time through a sliding windowing technique, and the Fourier transom of each windowed part o f the signal is calculated resulting in a two dimensional characterization of the signal. The STFT representation of a given signal is: = 1 14

PAGE 116

102 w here is the signal to be analyzed, and is the window function. As it i s shown in the equation, the result is a complex function that describes the phase and magnitude of the signal in both time and frequency domain. One thing that should be emphasized is that the tradeoff between time domain and frequency domain resolution is associated with the window selection [113]. Decreasing the window size will result in a better resolution in the time domain information because the length of the signal will be shorter, but the frequency domain resolution wills decrease. In general practices the window is chosen to be either Gaussian or Hanning windows. 4.9 .2 Spectrogram The spectrogram is one of the common applications of the STFT. The horizontal axis represent s time domain, while the vertical axis represent s the frequency domain. A third dimension is expressed in the spectrogram using color coding to describe the magnitude of the signal at a certain frequency and time point. The spect rogram is calculated through the STFT, and the representation of the spectrogram is: = ( ) 2 115 In the digital world we get the spectr ogram for a sampled signal through breaking the signal samples in to overlapped chunks. Then each chunk is passed through a Fourier transform operation to get the signal frequency representatio n and the spectrum magnitude. A measurement of magnitude versu s frequency for each time instant is

PAGE 117

103 performed, and the time plot is put side to side to construct the three dimension al image. Figure 4.2 2 illustrate s the designed spectrogram. A s pectrogram algorithm is performed on the received FHSS signal to reveal the time and frequency information. The three dimensional matrix is then analyzed to find the central frequency of each h op with the time of occurrence. This way the hopping sequence will be detected Fig 4. 2 2 Spectrogram representation of a Bluetooth signal There are many features that may be obvious or hidden in wireless communication signals. Identifying these features will be the success factor for any blind detection algorithm. In this chapter we define d the possible physical layer features that can participate in the process of identifying an unknown signal. Bandwidth and central frequency is estimated through a novel approach algorithm. Power related measurements of the signal are calculated. Moment s test based algorithm is designed to detect

PAGE 118

104 multicarrier signals. A comprehens ive OFDM parameter estimation has been proposed to estimate both time and frequency parameters. An introduction to the cyclostationarity is given, and an illustration of the c yclostationarity features detector was described. Symbol rate estimation is done throug h the nonlinearity Welch periodo gram approach, a s well as the DSSS chiprate and symbol rate estimation. An introduction to the joint time frequency analysis is given, along with a comprehensive JTF based algorithm that is proposed to detect the FHSS and the used hopping sequence. 4 10 Conclusion There are many features that may be obvious or hidden in wireless communication signals. Identifying these features will be the success factor for any blind detection algorithm. In this chapter we defined the possible physical layer features that can participate in the process of identifying an unknown signal Bandwidth and central frequency is estimated through a novel approach algorithm. Power related measurements of the signal are calculated. Moments test based algorithm is designed to detect multicarrier signals. A comprehensive OF DM parameter estimati on is proposed to estimate both time and frequency parameters. An introduction to the cyclostationarity is given, and illustration to the cyclostationarity features detector was described. Symbol rate estimation is done throug h the nonlinearity Welch per iodo gram approach, a s well as the DSSS chiprate and symbol rate estimation. An introduction to the joint time frequency analysis is given, along with a comprehensive JTF based algorithm that is proposed to detect the FHSS and the used hopping sequence.

PAGE 119

105 Chapter 5 Decision M aking A lgorithm In this chapter we descr ibe the decision making process that follow s the features extraction stage. We propose novel ISM band blind signal identifi cation algorithms which utilize all of the possible detected features before making a final judgment. 5.1 Introduction Many algorithms were proposed for blind signal identification but many of these studies target a specific type of signal or one wireless standard. For instance in [50] the proposed algorithm focuses only on the energy detection, bandwidth, and central frequency to make the judgment. I n [83] a threshold for the short time Fourier transform is s et to identify the DSSS signals. I n [85] bandwidth and ene rgy level of the signal is u sed to identify the signal type. I n [46] [65] [70] [86] [90] cyclostationarity is used to classify the signals. In [84] 4th order cumulants test is used to identify multicarrier systems, as well as many other algorithms that can be found where only a certain number of features are incorporated for the purpose of identification. This way may be sufficient enough to use for the band of interest where only certain licensed operators may be present. In the ISM band on the other hand, there ar e many standards that operate at the same time, with no license needed This makes signal detection and identification more complicate d and the uncertainties are larger. Therefore to build up reliable blind signal identification in the ISM band, we need to make sure that as many eventualities as

PAGE 120

106 possible are covered. This is mainly because in the ISM band many known and unknown prosp ective wireless standards can appear due to the license free quality of the ISM band. In this research we are trying to integrate all the possible features detection methods, and collect as much knowledge as possible about the signal. O nly then will we use the data we have collected about the received signal to make the judgment. The main work s in this chapter are to : a Propose a framework for the central processing unit in the spectrum awareness engine. b Integrate the entire feature extraction al gorithms in one controlled unit. c Utilize the detected features in a novel fuzzy logic like decis ion making mechanism. d Analyze the FCC regulation for the ISM and integrate the features rule in to the proposed algorithm. 5.2 The Proposed F ramework The final piece in our spectrum awareness engine will be the control and logic unit that will regulate the rest of and uti lize the incoming and outgoing information. Let us examine the proposed spectrum awareness engine flo w chart that is illustrated i n F igure 5.1. The band of interest will be chosen by the transmission upon request The RF front end wi ll sample the band of interest and pass the sampled data to the energy detection unit. The energy detector will identify the occupancy of the channel. If the channel is occupied the sampled signal will be passed to the feature detector to

PAGE 121

107 extrac t the features and information After searching for all the possible features a comprehensive control and decision unit will u tilize all the information and m ake the prop nature, will either initiate the proper transmitter configuratio n to overcome the interference or mark the band of interest as occupied and request a change of band. Fig 5.1 T he spectrum awareness engine flow chart

PAGE 122

108 The performance of the controlling and decision algorithm will define the overall effectiveness of our cog nitive radio performance. This is what makes it a very important part of our spectrum awareness engine. Th e following sections describe the decision making flow based on the detected features. 5.3 The D e cision M aking By now we can safely say that wireless standards have overl apping features and techniques. This means that although each wireles s standard is unique, there exist c ommon phy sical layer features which can be found between different wireless standards. Therefore, making a decision about a det ected signal is not as straight forward as it may appear, especially if we keep in mind that the ISM band can be the band of operation for many wireless sta ndards. From this point of understanding, we propose a novel approach to utilize the detected features while making the final judgment. In Chapter 3 we thoroughly investigate d each possible technology that may appear in the ISM band, an d we analyze d their mai n features and characteristics. Usi ng what we learned, an identification table was proposed. We cross linked each standard with the features that may identify it. Table 5.1 described the mapping of the defined features with each wireless standard.

PAGE 123

109 Table 5.1 T he identifying features for each wireless standard As demonstrated in Table 5.1, some features can be present in more than one wireless standard. This means t hat there is no precise answer; rather, approximations are m ore appropriate and hence fuzzy logic reasoning i s well suited to the situation. For this reason we ad apt a fuzzy logic like approach and soft decision making.

PAGE 124

110 5.3.1 FCC R egulations for the ISM B and Before we continue describing the proposed decision making algorithm, we need to descr ibe one last piece of the puzzle, the FCC regulations for the ISM band. We mentioned before that the ISM is a license free band, and any wireless device can be active in it. Although the band is license free it is not regulation free. The FCC regulate s the usage of the ISM band, and these regulations should be followed by any wireless device operat ing in it. For our spectrum awareness engine, th is is one of the best reference features. Because they are mandatory regulations, no one can by pass them. Thus, it is important to study the FCC regulations in the IS M band and to try to understand them and use these regulations for the benefit of our blind identifications In P art 15 Section 15.247 of T itle 47 of the Cod e of Federal Regulat ions (CFR) [1] [121], the FCC put up rules for the frequency hopping systems that operate in the ISM band, more precisely the 2.4G ISM band. We summarize th e points that deal with the 2.4GHz band that we also deem to be useful to our a lgorithms as 2 : a F requenc y hopping systems should have hopping channels frequencies with a minimum separation of 25 K Hz or the 20dB bandwidth of the hopping channel, whichever is greater. b The system shall hop to channel frequencies that are selected at the system hopping rate from a pseudorandomly ordered list of hopping frequencies. c Each frequency must be used equally on the average by each transmitter. 2 These regulations are located in Part 15 of the FCC rules (47 CFR 15.247).

PAGE 125

111 d Frequen cy hopping systems shall use at least 15 hopping frequencies. e The ma ximum 20 dB bandwi dth of the hopping channel is 1 MHz. f The average time of occupancy on any frequency shall not be greater than 0.4 seconds within a 30 second period. We believe that the most important rules are the fifth and sixth, as they state that any hopping sequence should hav e a bandwidth of no more than 1MHz. This allow s us to decrease the computationa l complexity in our algori thm because it means that we do not need to check if the signal is frequency hopping if its bandwidth is mo re than 1 MHz. Also it indicates that to che ck a frequency hopping sequence we need to observe the signal for at least 0.4 seconds. 5.3.2 Control and Execution The process begin s by first estimating the bandwidth and the central frequency of the signal. If the bandwidth is less than 2 MHz, we can safely assume that the signal might be a FHSS. In this case the signal will be passed to the joint time frequency analysis unit to check if the signal is FHSS and to extract the hopping sequence. If the signal bandwidth is larger than 2MHz, we can safely assume that the signal is not FHSS; therefore, we can overcome the joint time frequency analysis. P ower related measurements are conducted, as well as the duty cycle information by passing the signal through a burst detector. The signal will be tested for single carrier or multicarrier schemes. If the carrier test indicates that the signal is multicarrier, the

PAGE 126

112 OFDM parameter estimation will be applied to extract the time and frequency parameters of the signal. Otherwise, the moment s test and the nonli nearity based algorithm is executed to determine the modulation scheme. The outcome will be the modulation type and order identification or the confirmation that DSSS exist s in the signal. In case of a DSSS signal we extract the chip rate of the signal through the nonlinearity and cyclostationarity test. The symbol rate is estimated through the nonlinearity test for all the single carrier signals and is reported to the decision unit as well. At the end of this flow we have the features p arameters tha t we described in Chapter 4, and these are fed into our decision unit. Figure 5.2 illustrate s the feature detect or and the decision making unit work flow. Fig 5.2 Feature detection and decision making flow chart

PAGE 127

113 5.3.3 Fuzzy Logic and Soft Decision A lgorithm The concept of fuzzy l ogic was introduced by Lofti Zadeh, a professor at the University of California Berk e ley [92]. The author presented the concept not as a control method but as a technique of utilizing data by allowing partial set membership rather than crisp se t membership or non membership. Zadeh reasoned that people do not require precise numerical information input, and yet they are capable of highly adaptive control We adapt the same approach in the proposed algorithm. Instead of taking the path of hard decision and precise answer in the cogni tive radio, we believe that it is more reasonable to report a soft decision and probabilities about the present signal. Ther efore, we propose the following method to make the decision. A weight is given to each detected feature, and t hen according to the developed T able 5.1, the weight of the detected feature is transferred to the prospective wireless standards that match the feature within its sta ndard characterization (we discussed this characterization in Chapter 3 ) By the end of mapping all the detected features to the possible wireless technologies, the algorithm will calculate the total weight of each wireless technolog y, and a probability of the presence of each wireless standard will be reported as a soft decision of the current detected s ignal. This way we will take into consideration all the features present instead of dropping some features when making the decision The biggest benefi t of this is seen in the case of unknown signals or new standards, where all the detected signal c haracteristics will be taken in to account when setting up the transceiver configurations. For more illustrations, let us take the follow ing examples. In E xample 1 a WLAN standard IEEE 802.11g signal is passed to the

PAGE 128

114 features extraction and decision units. After extracting the features we have the following outputs: a Bandwidth = 20MHz b Fc = 2.417 GHz c No FHSS analysis is required since the bandwidth is >2 d Carrier test indicate s multicarrier based signal e OFDM time parameters estimation results: T s ~ 4 s T c ~ 0 8 s T u ~ 3 2 s f OFDM frequency parameters estimation results: f ~ 312 5 KHz N ~ 64 g CCDF curves indicate high PAPR The algorith m response is illustrated in F igure 5.3 Fig 5.3 T he algorithm response for a WLAN 802.11g input signal The decision tree that will be created in the FL decision unit is illustrated in Table 5.2

PAGE 129

115 Table 5. 2 T he FL decision tree of E xample 1 The final decision indicates that the biggest possibility is that t he detected signal is wireless LAN IEEE 802.11g. Another example is as follows. Example 2 represents a Bluetooth signal that is passed to the features e xtraction and the decision unit s. After extr acting the features we have the following outputs: a Bandwidth ~ 1 MHz b Fc = 2.4 06 GHz c FHSS analysis is required since the bandwidth is < 2

PAGE 130

116 d FHSS test positive e Hopping sequence The alg orithm response is illustrated in F igure 5. 4 Fig 5. 4 T he algorithm resp onse for a Bluetooth input signal The decision tree that will be created in the FL decision unit is illustrated in Table 5.3

PAGE 131

117 Table 5. 3 T he FL decision tree of E xample 2 The results indicate highest probability for the Bluetooth standard. Moreover the Bluetooth version 2 has slightly higher probability due to its match with the detected hopping sequence. As we see the more we know about the features and statistics of the prospective standards that we may encounter, the better our algorit hm performance will be. R ecall that the FSK pro blem we encountered in C hapter 4 is an example of the effect of the prior knowledge of standard specifications. The algorithm in the test for carrier system shows some inconsistence when tested for the FSK modulation. At some points

PAGE 132

118 the algorithm gave a result of OFDM signal while the signal w as FSK. If we were making our judgment only based on the result of the carrier system test, we may reach the wrong conclusion and think that the signal is OFDM. But using all the features to make the judgment we see that even though the algorithm give s a positive answer for the OFDM, the bandwidth does not support this decision since it indica te s an FHSS signal with a hopping sequence. So the final result will have larger probability to back it up that the signal is FHSS. The algorithm performance for different w ireless standard is calculated. T able 5.4 illustrate s the success rate of the alg orithm blind detection in different SNR environments. Table 5 .4 T he algorithm performance results of success rate detection

PAGE 133

119 5.4 Location and Time of Occurrence Some additional features we can use for information are the l ocation and time of occurrence. It was shown in [39] [40] [82] [122] and [ 123] that the location information can serve in cognitive cycle improvement. Furthermore in [40] [123] [125] it was explained how Bayesian theory can be used to incorporate t he past experience in the future decision making. Since the cognitive radio will monitor the spectrum continuously the history of the decision making and the spectrum usage information over time is valuable to the learning ability of the cognitive radio. We did not implement an algorithm for this particular purpose, but we will describe the general outlines for such algorithm for the sake of consistency in the aim of this research. The Bayesian theory state s that a relationship can be established betwe en an event and the prior knowledge about it. This means that it relates the conditional probability of an event given a certain observation. This theorem is considered as a model of learning, which makes it a perfect fit in the cognitive radio applicati on. The Bayesian theorem is expressed as: = ( ) ( ) 116 where r epresents a specific hypothesis, ( ) is the prior probability of H that was inferred before new evidence, E is the conditional probability of seeing the evidence if the hypothesis happens is true ( ) is the marginal probability of ( the a priori probability of witnessing the new evidence under all possible hypotheses ), and

PAGE 134

120 is the posterior probability of given It i s shown in the described fo rmula how the prior information (the history) is incorporated to decide the current probability. 5. 5 Conclusion In this chapter a no vel framework f or the central processing unit wa s developed. We demonstrate d full utilization of all the extracted features before making the decision. We briefly explain ed fuzzy logic and integrated i t into our algorithm. We explained how t he FCC rules are a common ground for the entire possible wi reless standard in the ISM band and that every device in the band should follow these rules These rul es were explained and analyzed. We pointed out that s ome of the rules can be used as features to indicate standards, so we integrated those rules in to the proposed algorithm to minimize the computati onal complexity. Some examples we re given to dem onstrate how the algorithm behave s in different situations. And finally we briefly explain ed the importance of the time of occurrence and the history of occurrence in the learning process of the cognitive radio.

PAGE 135

121 Chapter 6 Summa ry and C onclusions 6.1 Summary of Works and C ontributions This research deals with the cognitive radio implementations issues in the 2.4GHz ISM band and the possibility of coexistence between c ognitive radios and the pre exis ting wireless stan dards that are active in the band. In this thesis we proposed a new and realistic design to the spectrum awareness engine to be integrated with the model proposed in [43]. Furthermore, we designed the spectrum awareness engine to be compatible with the ISM band. The contributions and implementations of this thesis can be summarized as follows : a Cognitive r ad io concepts and pro posed models We defined and analyzed cognitive radio and its functionalitie s. We demonstrate the i mportance of the spectrum awareness and the continued sensing abilities in the cognitive radio performance. We identified the proposed models for cognitive radios and explained the common cognition cycle We proposed a novel design for the spectrum awareness engine which is both realistic and can be implemented with the current circuitry capabilities, with the help of the software defined radio. We described t he weaknesses and problems associate d with the two comm on choices for spectrum sensing: the energy detector and matched filter detector.

PAGE 136

122 Finally we proposed to use the energy detector only as a pre s tage in order to avoid its weaknesses. b The industr ial s cientific and medical band We studied the ISM band characteristics and regulations extensively. We analyzed the wireless standards that may operate in the ISM band We identified the main features of the wireless standards especially the p hysical layer features that can be used in the process of blind identifications. c Spectrum awareness engine We showed that there are many features which can be used to indicate wireless standards that are not utilized. W e propose d a list of features to be used in our b lind identification design. S ince each feature requires a special way of processing to extract it, we proposed appropriate algorithms to detect the feature s. We attempted to increase the performance and accuracy of each algorithm taking into consideratio n the computational complexity in the process of the design. We proposed a n ISM band feature detector design and integrated t he implemented algori thms for each individual feature into one feature detector. d Decision making and process controlling We demons trate d how some fe atures can exist in more than one standard and how t his may cause confusion in the process of decision making. We developed a work flow for the central controlling unit, to help organize the work of all the unit s together to give the hig hest performance. We proposed a proper decision makin g method to utilize all the possible det ected features and observations in order to avoid conflict which may be due to t he detection of common features or

PAGE 137

123 the outside impairments that the signal can su ffer from and that may distort some features. 6. 2 Conclusions The ISM band is one of the most popular destinations for wireless standards f or many reasons one of which is the fact that it is a license free band and opens to a ny wireless device. Although it is a free to use kind of band, there are regulations and rules to be followed, and in the US those rules are designed by t he FCC to e nsure fairness and innovatio n by the wireless devices. P eaceful coexistence betw een the wireless standards is important in the ISM band, a nd r ecently an increasing concern has been given t o this issue d ue to the fact that the numbers of users in the ISM band are increasing rapidly, which leads to many interference issues. A certainty is that c haracterizing the s ignals will help overcom e their interference effects, with the cognitive radio as the ultimate solution. The cognitive radio is one very promising technology. Day by day with the increasing developments in microprocessors and the software designed radio the cognitive radio is getting more attention and raises hopes Many models and work flows have been proposed for the cognitive radio, and more attempts should be done toward converting these models to realistic circuitry based models. The most important capability of the cognitive radio is the spectrum awareness performance since the spectrum is the most valuable wireless resource.

PAGE 138

124 Wir eless standards have many features some common and some different In this thesis, a novel design has been pr oposed to utilize all these features t o blindly identify the signals in order to evaluate how to overcome their effects. The decision making method will have a big impact on the spectrum awareness perf ormance of the cognitive radio, e s pecially in a band l ike the ISM where every device can operate. Fuzzy logic and a soft decision approach should be considered in the cognitive radio functionalities since flexibility shoul characteristics. 6. 3 Future Work The ISM band has become very popular and the sanctuary of many wireless standards. We expect that this rapid growth will cont inue, which mean only one thing -more congest ion and more interference. That i s why we believe that more research about the ISM band cognitive radio should take place. In this research we study the identifications of the wireless signals assuming that there is no interference, and only the signal of interest is present. For this reason, the next step in this research will be to study the effect of interference s on the identification process and develop methods to isolate the interfering signals during the identification process. Since there are some indications that the ISM band will have fewer regulations, another future work c an be the study of the features extraction in a total regulation free ISM band. Another poss ible open research area is to study the effect of channel impairments that the signal may suffer from and the effect of these impairments on the detection performan ce

PAGE 139

125 and features clearance Furthermore, methods can be developed to overcome the channel effect during the detection and identification process.

PAGE 140

126 References Telecommunication, Chapter I, Part15 Volume 1, 2007. [Online]. Available: http://www.access.gpo.gov/nara/cfr/waisidx_07/47cfr15_07.html [Accessed April 1, 2009]. [2] vol. 6, no. 4, pp. 13 18, August 1999. Technical Note Document FPG 2006 307.2 January 2008. [Online] Available: http://www.cisco.com/en /US/solutions/collateral/ns340/ns394/ns348/ns736/net_impleme ntation_white_paper0900aecd80554f8b.pdf [Accessed April 1, 2009] 2003. [Online] Available: http://www.super g.com/collateral/Atheros_Regulatory_whitepaper.pdf [Accessed January 11, 2009]. FPG 2006 328.3 Janu ary 2008. [Online] Available: http://www.cisco.com/en/US/solutions/collaterall/ns340/ns394/ns348/ns736/net_impleme ntat ion_white_paper0900aecd805eb8a5.pdf [Accessed April 2009]

PAGE 141

127 http://www.cisco.com/en/US/ products/ps9393/ index html [ Access ed April 2009]. [7] Morrow, Rober t, Wireless Network Coexistence. McGraw Hill Professional, 2004 [8] R. Pickholtz, D. Schilling, L. Milstein Theory of Spread Spectrum Communications -A Tutorial, IEEE Transactions on Communications vol. 30, no. 5, pp. 8 55 884, May 1982 [9] A.J. Viterbi, CDMA: Principles o f Spread Spectrum Communication. Addison Wesley, 1995. 22, 2002. [Online]. Available: http://www.complextoreal.com/chapters/ofdm2.pdf [Accessed April 2009]. [11] Radio broadcasting systems; Digital Audio Broadcasting (DAB) to mobile, portable and fixed receivers, ETSI European Telecommunications Standards Institute Std. EN 300 401, Rev. 1.3.3, May 2001. [12] Digital Video Broadcasting (DVB); Framing structure, channel coding and modulation for digital terrest rial television, ETSI European Telecommunications Standards Institute Std. EN 300 744, Rev. 1.4.1, J an. 2001. [13] Asymmetric Digital Subscriber Line (ADSL), ANSI American National Standards Institute Std. T1.413, 1995. [14] A. M. Wik, A. L. Lindblad, Novel LPI concept using filtered spreading codes IEEE Mil i tary Communications Conference McLean, Oct. 1996, vol. 1, pp. 90 94

PAGE 142

128 [15] A. Peled, A. Ruiz Frequency domain data transmission using reduced compu IEEE International Conference on Acoustics Speech and Signal Processing vol. 5 pp. 964 967, Apr 1980 [16] Wiki http://en.wikipedia.org/wiki/ Guglielmo Marconi#cite_note 12. [Accessed March 12, 2009]. http://en .wikipedia.org/wiki/802.11 #cite_note 0. [Accessed March 12, 2009]. A Wolfram Web Resource, 2009. [Online]. Available: http://math world.wolfram.com/BarkerCode.html [Accessed March 12, 2009] [19 ] M. Golay, IEEE Transactions on Information Theory, vol. 7, no. 2, pp. 82 87, April 1961 2000. [Online]. Available: http://www.eetasia.com/ARTICLES/2001MAY/2001MAY25_NTEK _DSP_AN.PDF [Accessed February 12, 2009]. 2007: Wireless LAN Medium Access [Online]. Available: http://standard s.ieee.org/getieee802/802.11.html [Accessed March 11, 2009].

PAGE 143

129 [Online]. Available: [http://www.bluetooth.org/foundry/adopters/document/Bluetooth_Core_Specification_v1 .2. [Ac cessed March 11, 2009] Bluetooth.org, 2009. [Online]. Available: http://bluetooth.com/Bluetooth/Technology /Works/Core_ Specification_v21__EDR.htm. [Accessed March 15, 2009]. [24] J. T. Adams, An int IEEE Aerospace Conference, Jul. 2007, pp. 8 [25] S.C. Ergen ZigBee/IEEE 802.15.4 Summary, September 2004. [Online] Available: http://www.sinemergen.com/zigbee.pdf [Accessed March 11, 2009]. http://en.wikipedia.org/wiki/ZigBee [ Accessed February 11, 2009]. [27] Y. Yamao, S. Takagishi, Time Shift Grouping Access in IEEE 802.15.4 MAC Beacon Mode for Layered IEEE Consumer Communications and Networking Conference pp. 338 342, Jan. 2008 [28] T. M. Taher, M. J. Misur ac, J. L. LoCicero, D. R. Microwave Oven Signal Interference Mitigation F or Wi IEEE Consumer Communications and Networking Conference, pp. 67 68, Jan. 2008 Proc. of IEEE PIMRC, vol. 3, 1997, pp. 1221 1227.

PAGE 144

130 [30] T.M. Taher, A. Z. Al Uninten tional Wi Fi Interference Device IEEE Military Commu nications Conference, Oct. 2006, pp. 1 7. [31 ] T. M. Taher, M. J. Misurac, J. L. LoCicero, D. R. Ucci M icrowave Oven Signal IEEE Wireless Communic ations and Networking Co nference, pp.1235 1238, April 2008 http://www.ntia.doc gov/ osmhome/ Allochrt.html. [Accessed February 2009]. [Online]. Available: http://www.fcc.gov/sptf/ [Accessed Feb ruary 2009]. [34] Federal Communications Commission, Rep. ET Docket no. 02 135, Nov. 2002. [35] J. Mi tola Cognitive radio for flexible mo IEEE International Workshop on Mobile Multimedia C omm unications pp. 3 10, Nov 1999 J ournal on Selected Areas in Communications vol. 23, no. 2, pp. 201 220, Feb. 2005. [37] S. Mangol d, Z. Zhong, K. Challapali, C. T. Chou, resource measurements f or opportunistic spectrum usage, in P roc IEEE Global Teleco mmunications Conference vol. 6, Dallas, TX, Dec. 2004, pp. 3467 3471.

PAGE 145

131 OFDM: Pushing Ultrawid eband Beyond Its Limit vi Journal of Communications and Networks Special Issue on Spectrum Resource Optimization, vol. 8, no. 2, pp. 151 157, June 2006. C ognitive Wireless Net IEEE Wireless Communications Magazine Special issue on Cognitive Wireless Net works, vol. 14, no. 4, pp. 6 13, Aug. 2007. [40] H. Arslan, Cognitive Radio, Software Defined Radio, and Adaptive Wireless Systems. Springer, June 2 007 [41] H. Cel ebi, I. Guvenc, On the Statistics of Channel Models f or UWB Ranging IEEE Sarnoff Symposium, Princeton, NJ, March 2006. [42] IEEE Standards Coordinating Committee 41 Dynamic Spectrum A ccess Networks [Online]. Availab le: http://www.ieeep1900.org awareness in cognitive Elsevier Computer Communications Special Issue on Advanced Location Based Services, vol. 3 1, no. 6, pp. 1114 1125, April 2008. [44] A. Sahai, N. Hoven, and R. Tandra, \ Some fundamental limits on cognitive radio," in Forty Second Allerton Conference on Communication, Control and Computing, September 2004. IEEE Trans action Information Theory, vol. 7, pp. 135 139, July 1961.

PAGE 146

132 for cognitive radios on Signals, Systems and Computers, vol. 1 Nov. 2004, pp. 772 776 Facilitating opportunities for flexible, efficient, and reliable spectrum use employing ET Docket No. 03 108, Dec 200 3 [Online]. Available: http://hraunfoss.fcc.gov/edocs_public/attachmatch/FCC 03 322A1.pdf [Accessed April 1, 2009]. [48] S. M. Kay, Fundamentals of statistical signal pr oc essing: Detection theory. Prentice Hall, 1998, vol. 2 Energy detection of in Proc IEEE, vol.55, no.4, pp. 523 531, April 1967 [50] T. Yucek, H. Arslan, Spectrum Characterization for Opportu nistic Cog nitive Radio IEEE Military Communications Conference Oct. 2006, pp 1 6 [51] S. Shankar, C. Cordeiro, and Symposium on New Frontiers in Dynam ic Spectrum Access Networks, Baltimore, Maryland, USA, Nov. 2005, pp. 160 169. for cognitive Eighth Asilomar Conference on Signals, Systems and Computers, vol. 1, Pacific Grove, California, USA, Nov. 2004, pp. 772 776.

PAGE 147

133 over fadi in Proc. IEEE International Conference on Commun ications ., vol. 5, Seattle, Washington, USA, May 2003, pp. 3575 3579. [54] W. Gardner, Exploitation of spectral redunda IEEE Si gnal Processing Magazine, vol. 8, no. 2, pp. 14 36, Apr 1991 [55] W. Gardner Spectral Correlation of Modulated S ign als: Part I -Analog IEEE Transactions on Communications vol. 35, no. 6, pp. 584 594, Jun 1987 [56] W. Gardner, W. B rown, Chih Spectral Correlation of Modulated Signa ls: Part II -IEEE Transactions on Communications, vol. 35, no. 6, pp. 595 601, Jun 1987 Signal interception: a unifying theoretical framework for feature detection ," IEEE Transactions on Communications, vol. 36, no. 8, pp. 897 906, Aug 1988 [58] W. Gardner, The Role of Spectral Correlation in Design and Perfor mance Analysis of Synchronizers ", IEEE Transactions on Communications, vol. 34, no. 11, pp. 1089 1095, Nov 1986 Signal interception: performance advanta ges of cyclic feature detectors IEEE Transactions on Communications, vol. 40, no. 1, pp. 149 159, Jan 1992 Robust feature de tection for signal interception IEEE Transactions on Communications vol. 42, no. 5, pp. 2165 2173, May 1994

PAGE 148

134 [61] W. Gardner, L Franks, Characterization of cyclosta tionary random signal processes IEEE Transactions on Information Theory vol. 21, no. 1, pp. 4 14, Jan 1975 [62] Representation and estimation of cyclostationary processes (Ph.D. Thesis abstr.) IEEE Transactions on Information Theory vol. 19, no. 3, pp. 376 376, May 1973 Meas urement of spectral correlation IEEE Transactions on Acoustics, Speech and Signal Processing vol. 34, no. 5, pp. 1111 1123, Oct 1986 [64] R. Roberts, Computationally efficient algorithm s for cyclic IEEE Signal Processing Magazine, IEEE, vol. 8, no. 2, pp. 38 49, Apr 1991 [65] N. Khambekar, L. Dong, spectrum sens ication and Networking Conf erence Hong Kong, Mar. 2007, pp. 38 42. c. IEEE Radio and Wireless Conference Atlanta, Georgia, USA, Sept. 2004, pp. 263 266. [67] Ghozzi Mohamed, Marx Francois, Dohler Mischa, Palicot Jacques, Cyclostatilonarilty Based Test for Detec 1st International Conference on Cognitive Radio Oriented Wireles s Ne tworks and Communications, pp. 1 5, June 2006 Spectrum Sensing Measurements of Pilot, Energ IEEE Military Communications Con ference, Oct. 2006, pp.1 7

PAGE 149

135 [69] Wenjie Ma, ShiMing Yang, Wu Ren ZhengHui Xue WeiMing Spectral correlation f Radio Science Con ference, Asia Pacific, pp. 197 200, Aug. 2004 Recognition Among OFDM Based Systems Utilizing Cyclosta tionarity Inducing Transmission 2nd IEEE International Symposium on New Frontiers in Dynamic Spectrum Access Networks April 2007, pp.516 523 [71] S. K. Mitra, Digital Signal Processing: A Computer Based Approach. New York USA: McGraw Hill, 2000. [ 72] Peter Kenington RF and Baseband Techni ques for Software Defined Radio Artech House Incorporated, 2005 [73] A. Poon, R. Brodersen, D. Degrees of freedom in spatial c IEEE Transaction s on Information Theory, vol. 51, Feb 2005. e in Proc. IEEE Int ernational Symposium on New Frontiers in Dynamic Spectrum Access Networks, Baltimore, Maryland, USA, Nov. 2005, pp. 151 159. [75] M. Olivieri, G. Ba allocation system with interference mitigation for teams of spectrally agile software def Symposium on New Frontiers in Dynamic Spectrum Access Networks Baltimore, Maryland, USA, Nov. 2005, pp. 170 179.

PAGE 150

136 in fading e Symposium on New Frontiers in Dynamic Spectrum Access Networks, Baltimore, Maryland, USA, Nov. 2005, pp. 131 136. [77] J. Lehtomaki J. Vartiainen, M. Juntti, with in Proc. IEEE Military Communications. Conference Washington, D.C., USA, Oct. 2006, pp. 1 7. layer filter in Proc. IEEE MTT S Inter. Microwave Symposium Digest, vol. 3, Philadelphia, Pen nsylvania, USA, June 2003, pp. 1767 1770. 1997 [Online]. Available: http://www.ittc.ku.edu/RDRN/papers/thesis/killoy_thesis_slide_061999.pdf [Accessed: Jan.1.2009] [80] IEEE Transactions on Signal Processing vol.39, no.9, pp.1973 1984, Sep 1991 [81] A. So nnenschein, P. Fis Radiometric detection of spread spectrum sign als in noise of uncertain power IEEE Transactions on Aerospace and Electronic Systems, vol.28, no.3, pp.654 660, Jul 1992 [82] Yucek T. Arslan, H. thms for cognitive radio IEEE Communicat ions Surveys & Tutorials, v o l. 11, pp. 116 130 2009.

PAGE 151

137 A threshold detection meth od of DSSS signal based on STFT International Symposium on Microwave, Antenn a, Propagation and EMC Technologies for Wireless Communications vol. 2, pp. 879 882, Aug 2005 [84] Detection of multicarrier modulations using 4th in Proc. IEEE Military Communications Conference, vol. 1, Atlantic City, Ne w Jersey, USA, Oct./Nov. 1999, pp. 432 436. [85] Detect and avoid: an ultra wideband/WiMAX coexistence mechanism [T IEEE Communications Magazine vol. 45, no. 6, pp. 68 75, June 2007 [86] Kim Kyouwoong, I Akbar, K. Bae, Urn Jung sun, C. M. Spooner, J. H. Reed, Cyclostationary Approaches to Signal Detection and Cl assification in Cognitive Radio 2nd IEEE International Symposium on New Frontiers in Dynamic Spectrum Access Ne tworks April. 2007, pp. 212 215. [87] P. D. Sutton K. E. Nolan, Cyclostationary Signatures for Rendezvous in OFDM Based Dy 2nd IEEE International Symposium on New Frontiers in Dynamic Spectrum Access Networks April 2007, pp.220 231 Cyclostationary Feature Detector Experim 2nd IEEE International Symposium on New Frontiers in Dynamic Spectrum Access Networks April 2007 pp. 216 219. Transmitter induced cyclostationarity for blind IEEE Transactions on Signal Processing vol. 45, no. 7, pp. 1785 1794, Jul 1997

PAGE 152

138 [90] Chen Hou Shin Gao Wen Spectru m Sensing Using Cyclostationary Properties and A IEEE Global Tel ecomm unications Conference Nov. 2007, pp. 3133 3138 [91] A. Gorcin, J. Mitola, H. Arslan with a robust coar Spectrum Access (DSA) Special Issue of EURASIP Advances in Signal Processing Journal. [92] Lotfi Asker Zadeh Fuzzy Sets, Fuzzy Logic, and Fuzzy Systems: Selected Papers by Lotfi A. Zadeh World Scientific Publishing Company, May 1996 [93] J. Palicot and C. IEEE Commun ications Mag azine, vol. 41, no. 7, pp. 124 132, 2003, Jul 2003. [94] P. L An OF DM bandwidth estimation scheme for spectrum in Proc. Int ernational Wireless Communications, Networking and Mobile Computing Conf erence vol. 1, Maui, Hawaii USA, Sept. 2005, pp. 248 251. [95] Hai ying Zhang, Chao wei Yuan A Method for Blind Detection of OFDM Signal Based Eighth International Conference on Software Engineering, Artificial Intelligence, Networking, and Parallel/Distri buted Computing, vol. 2, pp.181 186, Aug. 2007 [96] A.P. We bster, J. Paviol, Liu Jiang ; H. Arslan, L.P. Dunleavy, Measurement based modeli IEEE Rad io and Wireless Conference pp. 403 406, Sept 2004

PAGE 153

139 [97] el at 5 GHz U elsevier, Computers and Electrical Engineering, vol. 30, no. 5, pp. 331 345, July 2004. [98] G. B. Giann domai n tests for Gaussianity and time Transactions on signal processing v ol. 35, no. 1, pp. 18 26, Jan 1990 [99] Xu Bin Yang Chenyang Mao Shiyi, A multi carrier detection algorithm for OFDM IEEE International Conference on Communications May 2003, vol. 5, pp. 3377 3381 [100] International wireless communications, networking and mobile computing Conference, vol. 1, Sept 2005, pp. 261 264. [101] Signal classifica IEEE Transactions on Communications vol. 40, no. 5, pp. 908 916, May 1992 [102] Wang Bin Blind Identification of O FDM Signal in Rayleigh Channels Fifth International Conference on Information, Communications an d Signal Processing, pp. 950 954. [103] B. Farhang no.12. pp. 2057 2070, Dec 2003.

PAGE 154

140 [104] Wei Dai, Youzheng Wang, and Jing Wang, Jo int power estimation and modulation classification using second wireless communications, networking and mobile computing Conference vol. 1, pp. 155 158, Mar. 2002. [105] Su Wei, J.A. Kosinski, Yu Ming Dual Use of Modulation Recognition techniques for Digital Communication Signals Systems, Applications and Technology C onference, Long Island May 2006, pp.1 6. classifier f Instrumentation and Measurement Technology Conference, Anchorage, AK, May 2002, pp. 957 962. [10 7 ] H. Li, Y. Bar classification and parameter Cognitive Radio Oriented Wireless Networks and Communications, Mykonos Island, Greece, June 2006, pp. 1 6. IEEE Thirt y Fourth Asilomar Conference on Signals, Systems and Computers, vol.1, Pacific Grove, California, USA, Nov. 2000, pp.142 146. [109] Liu Peng Li Bing bing Lu Zhao yang Feng A blind time parameters estimation scheme for OFDM in multi path chan nel International Conference on Wireless Communications, Networking and Mobile Computing vol. 1, pp. 242 247, Sept. 2005

PAGE 155

141 [110] Shi Miao, Y. Bar Ness, Blind OFDM Systems Parameters Estima tion for Software Defined Radio 2nd IEEE International Symposium on New Frontiers in Dynamic S pectrum Access Networks, April 2007, pp.119 122. [111] OFDM Signal Identification and Transmission Parameter Estimation fo IEEE Global Tele communications Conference, Nov. 2007, pp. 4056 4060. OFDM Blind Parameter Ide ntification in Cognitive Radios IEEE 16th International Symposium on Personal, Indoor and Mob ile Radio Communications, vol. 1, pp. 700 705, Sept 2005 in Proc. IEEE Radio and Wireless Symposium pp. 639 642, Jan 2008. A robust baud rate esti mator for noncooperative demo dulation in Proc. 21st Century Military Communications Conference vol. 2, Oct. 2000, pp.971 975. [115] A. D. Snider, I ntroduction to Random Processes. ( Manuscript in preparation) Cyclic correlation ba sed symbol rate estimation Thirty Third Asilomar Conference on Signals, Systems, and Computers vol.2, Oct. 1999, pp.1008 1012. The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging ov er short, modified periodograms IEEE Transactions on Audio and Electroacoustics vol. 15, no. 2, pp. 70 73, Jun 1967.

PAGE 156

142 [118] Zhanqi Dong, Hanying Hu, The Detection, Symbol Period and Chip Width Estimation of DSSS Signals Based on Delay Multiply, Co rrelation and Spectrum Proc. of the International MultiConference of Engineers and computer scientists, Honk Kong, China March. 2007, pp. 1198 1201 lkernel time fre quency r IEEE Transactions on Sig nal Processing, vol 43, no. 10, pp. 2361 2371, Oct 1995 frequency actions on Signal Processing, vo l. 4 0, no. 2, pp. 413 420, Feb 1992. [121] Matthew, 802.11 Wireless Networks: The Definitive Guide, Second Edition. O'Reilly Media, Incorporated, April 2005 Cognit IEEE Signal Processing Magazine, vol. 23, no. 1, pp. 30 40, Jan 2006 [123] Rajamani Ganesh Sastri L. Kota Kaveh Pahlavan Ramn Agusti, Emerging Location Aware Broadband Wireless Ad H oc Networks. Springer, 2005 [124] Adaptive Radar Inter national Radar Symposium, May 2006, pp.1 4. [125] S. Haykin, Adaptive signal processing. John Wiley and Sons, 200 6 [126 International Conference on Cognitive Radio Oriented Wireless Networks and Communications, Aug. 2007, pp.408 413.


xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader cam 2200349Ka 4500
controlfield tag 001 002063216
005 20100316115727.0
007 cr mnu|||uuuuu
008 100316s2009 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002996
035
(OCoLC)556084833
040
FHM
c FHM
d FHM
049
FHMM
090
TK145 (Online)
1 100
Zakaria, Omar.
0 245
Blind signal detection and identification over the 2.4GHz ISM band for cognitive radio
h [electronic resource] /
by Omar Zakaria.
260
[Tampa, Fla] :
b University of South Florida,
2009.
500
Title from PDF of title page.
Document formatted into pages; contains 142 pages.
502
Thesis (M.S.E.E.)--University of South Florida, 2009.
504
Includes bibliographical references.
516
Text (Electronic thesis) in PDF format.
3 520
ABSTRACT: 'It is not a lack of spectrum. It is an issue of efficient use of the available spectrum"--conclusions of the FCC Spectrum Policy Task Force. There is growing interest towards providing broadband communication with high bit rates and throughput, especially in the ISM band, as it was an ignition of innovation triggered by the FCC to provide, to some extent, a regulation-free band that anyone can use. But with such freedom comes the risk of interference and more responsibility to avoid causing it. Therefore, the need for accurate interference detection and identification, along with good blind detection capabilities are inevitable. Since cognitive radio is being adopted widely as more researchers consider it the ultimate solution for efficient spectrum sharing [1], it is reasonable to study the cognitive radio in the ISM band [2]. Many indications show that the ISM band will have less regulation in the future, and some even predict that the ISM may be completely regulation free [3]. In the dawn of cognitive radio, more knowledge about possible interfering signals should play a major role in determining optimal transmitter configurations. Since signal identification and interference will be the core concerns [4], [5], we will describe a novel approach for a cognitive radio spectrum sensing engine, which will be essential to design more efficient ISM band transceivers. In this thesis we propose a novel spectrum awareness engine to be integrated in the cognitive radios. Furthermore, the proposed engine is specialized for the ISM band, assuming that it can be one of the most challenging bands due to its free-to-use approach. It is shown that characterization of the interfering signals will help with overcoming their effects. This knowledge is invaluable to help choose the best configuration for the transceivers and will help to support the efforts of the coexistence attempts between wireless devices in such bands.
538
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
590
Advisor: Huseyin Arslan, Ph.D.
653
OFDM Estimation
IEEE 802.11
Spectrum Sensing
Joint Time Frequency Analysis
Cyclostationarity Features
690
Dissertations, Academic
z USF
x Electrical Engineering
Masters.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.2996