ENHANCEMENT OF WI-FI COMMUNICATION SYSTEMS THROUGH SYMBOL SHAPING AND INTERFERENCE MITIGATION TANIM MOHAMMED TAHER

Transcription

1 ENHANCEMENT OF WI-FI COMMUNICATION SYSTEMS THROUGH SYMBOL SHAPING AND INTERFERENCE MITIGATION BY TANIM MOHAMMED TAHER Submitted in partial fulfillment of the requirements for the degree of Master of Science in Electrical Engineering in the Graduate College of the Illinois Institute of Technology Approved Advisor Chicago, Illinois December 2007

2

3 ACKNOWLEDGMENT I am grateful to my advisors Dr. Joseph L. LoCicero, Professor at the Electrical and Computer Engineering (ECE) department at the Illinois Institute of Technology (IIT), and Dr. Donald R. Ucci, Associate Professor at the ECE department at IIT, who have guided me through all the research work. They encouraged me to push the limits of my abilities and were always there to answer my questions when I met with any obstacles during my research. I would like to give a special thanks to my colleague Dr. Ayham Z. Albanna who served as my mentor. He also was involved in this research project with regards to the development of a Microwave Oven signal model. I would also like to thank Matthew J. Misurac, who served as my assistant in the Microwave Oven interference mitigation project, and the Barker symbol shaping project and who always asked the right tough questions that helped in our research path. I also thank my colleagues Roger Bacchus, Ghaith Assaf and John T MacDonald. My parents, Dr. Mohammed Abu Taher and Dr. Rowshan Ara Begum deserve the greatest appreciation for guiding me all these years, for all the love, care and support they provided me, and for encouraging me to always strive higher. I also thank my loving sister, Dr. Tania Taher. Finally, I thank God Almighty for all his blessings and for making any of this possible. iii

4 TABLE OF CONTENTS Page ACKNOWLEDGEMENT iii LIST OF TABLES vi LIST OF FIGURES vii LIST OF NOMENCLATURE xii LIST OF SYMBOLS xiv ABSTRACT xvii CHAPTER 1. INTRODUCTION IEEE Communication Systems Spectral Mask and IEEE Channels Wireless Interference Inter-Symbol Interference Barker Code and IEEE Mbps Signal IEEE CCK 5.5 Mbps Signal ComBlock Devices Research Methodology PULSE SHAPING FOR IEEE 1 MBPS BARKER SPREAD SIGNAL Pulse Shaping Methodology Sinusoidal Pulse Shaping Logarithmic Pulse Shaping Sinc m Pulse Shaping BER Measurements Comparison of Barker Symbol Shaped Systems BUFFERED PULSE SHAPED BARKER SPREAD SYSTEMS Rational for using Buffer Buffering 2 Bits Buffering of 3 Bits PULSE SHAPING FOR IEEE 5.5 MBPS CCK SIGNAL CCK Pulse Shaping Methodology Sinusoidal Pulse Shaping Sinc m Pulse Shaping BER Measurements iv

5 CHAPTER Page 5. EXPERIMENTAL STUDY OF MICROWAVE OVEN SIGNAL Main Features of MWO Signal FM Signal Amplitude Variation Transients MWO PSD MODEL OF MICROWAVE OVEN SIGNAL Necessity of MWO Model Model # Model #1 Simulation Drawbacks of Model # Model # Model #2 Simulation Model #2 Experimental Emulation MICROWAVE OVEN SIGNAL INTERFERENCE MITIGATION FOR IEEE SYSTEMS Interference Mitigation Technique Circuit Design and Description Experimental Setup BER Studies CONCLUSION Pulse Shaping for 1 Mbps Signal Pulse Shaping for 5.5 Mbps Signal MWO Signal Study MWO Signal Modeling MWO Interference Mitigation Future Work APPENDIX INTERFERENCE SPECTROGRAMS BIBLIOGRAPHY v

6 LIST OF TABLES Table Page 1.1 Differential QPSK encoding table used in CCK transmission Mbps CCK encoding table Simulated BER for Barker (no buffer) Experimental BER for Barker (no buffer) Comparison of PSD sideband attenuations Comparison of total power in each band ISI after filtering operation Simulated BER measurements for 2 bits buffered Barker spread system Symbol mapping table for 3 bits buffered Barker spread system Comparison of PSD sideband attenuations (3 bits buffered) Simulated BER measurements for 3 bits buffered Barker spread system Four possible vectors C Comparison of PSD sideband attenuations (unfiltered 5.5 Mbps data) Comparison of total power in each band ISI after filtering operation for CCK symbol Shaping BER for Case 1 (Wi-Fi at 2.46 GHz without interference mitigation) BER for Case 2 (Wi-Fi at 2.46 GHz with interference mitigation) BER for Case 3 (Wi-Fi at GHz without interference mitigation) BER for Case 4 (Wi-Fi at GHz with interference mitigation).. 98 vi

7 LIST OF FIGURES Figure Page 1.1 FCC Spectral Mask IEEE channels Symbol sequence without ISI ISI distortion of a symbol sequence Auto-correlation plot of 11-chip Barker sequence PSD of Barker spread 1 Mbps signal PSD of 5.5 Mbps CCK signal ComBlock transmitter system ComBlock receiver system Block Diagram of ComBlock transmitter system Block Diagram of ComBlock receiver system Sinusoidally shaped pulse plot Experimental Sinusoidally shaped pulse plot Analytic PSD of sinusoidal pulse shape Simulated PSD of sinusoidal pulse shape Experimentally obtained PSD of sinusoidal pulse shape Auto-correlation plot of sinusoidal pulse shape Logarithmically shaped pulse plot Simulated PSD of logarithmic pulse shape Experimentally obtained PSD of logarithmic pulse shape Auto-correlation plot of logarithmic pulse shape vii

8 Figure Page 2.11 Sinc shaped Barker pulse plot Auto-correlation plot of sinc pulse shape Simulated PSD of sinc Barker pulse shape Experimentally obtained PSD of sinc pulse shape MATLAB simulation methodology for each pulse shape Experimental methodology used for each pulse shape BER vs SNR study for pulse shaped Barker spread systems State transition diagram for 2 bits buffered line code Block Diagram of 2 bit buffered Barker spread system State symbols for sinusoidal pulse shaping State symbols for sinc m pulse shaping State symbols for logarithmic pulse shaping Simulated PSD of sinusoidal pulse shaping (2 bits buffered) Simulated PSD of sinc m pulse shaping (2 bits buffered) Simulated PSD of logarithmic pulse shaping (2 bits buffered) Experimental PSD of sinusoidal pulse shaping (2 bits buffered) Experimental PSD of logarithmic pulse shaping (2 bits buffered) Experimental PSD of rectangular pulse shaping (2 bits buffered) Simulated PSD of rectangular pulse shaping (2 bits buffered) Cross-correlation plots for 4 states (3 bits buffered) Symbols for the 8 states 3 bits buffered system State transition diagram for 3 bits buffered system viii

9 Figure Page 3.16 Simulated PSD of 3 bits buffered pulse shaped system Experimental PSD of 3 bits buffered pulse shaped system PSD of 5.5 Mbps CCK spread signal without symbol shaping Unmodified rectangular CCK CCK symbols shaped by sinusoidal pulse shaping Simulated PSD of sinusoidally shaped 5.5 Mbps CCK spread signal Experimental PSD of sinusoidally shaped 5.5 Mbps CCK spread signal CCK symbols shaped by sinc m pulse shaping Simulated PSD of sinc m shaped 5.5 Mbps CCK spread signal Experimental PSD of sinc m shaped 5.5 Mbps CCK spread signal Simulated BER vs SNR measurements for CCK symbol shaping Composite Experimental PSD plots for CCK symbol shaping Spectrogram showing key features of MWO signal Clean spectrogram of MWO signal Time domain envelope of MWO signal Transient locations of MWO signal shown in ZSM Experimental spectrogram of MWO showing transients MWO signal generation process PSD of experimental MWO #1 signal PSD of experimental MWO #2 signal PSD of experimental MWO #3 signal PSD of experimental MWO #4 signal ix

10 Figure Page 6.1 Qualitative representation of MWO signal Simulated PSD of MWO based on model #1 (1 MHz range) Simulated spectrogram of MWO based on model #1 (1 MHz range) Simulated PSD of MWO based on model #1 (100 khz range) Simulated spectrogram of MWO based on model #1 (100 khz range) Experimental spectrogram of an older MWO Remodeling the transients Qualitative representation of MWO signal model # Spectrogram of simulated MWO signal (model #2) Simulated PSD of MWO signal (model #2) Spectrogram of emulated MWO #2 signal PSD of emulated MWO signal measured by spectrum analyzer Experimental PSD of actual MWO Data transmission using channel Spectrogram of MWO signal & interference mitigation Block diagram for MWO interference mitigation system Photograph of Interference mitigation circuit Cognitive Radio Citcuit diagram made in PSpice Case 1: No interference mitigation BER study (Wi-Fi at 2.46 GHz) Case 2: Interference Mitigation (Wi-Fi at 2.46 GHz) Case 3: No interference mitigation BER study (Wi-Fi at GHz) Case 4: Interference mitigation BER study (Wi-Fi at GHz). 97 x

11 Figure Page A.1 Spectrogram A.2 Spectrogram A.3 Spectrogram xi

12 LIST OF NOMENCLATURE Abbreviation AC ADC AM AP AWGN BER BPSK CA CCK CSMA DAC FCC FM IEEE I iid IIR IIT ISI ISM Mbps Term Alternating Current Analog to Digital Converter Amplitude Modulation Access Point Additive White Gaussian Noise Bit Error Rate Binary Phase Shift Keying Collision Avoidance Complementary Code Keying Carrier Sense Multiple Access Digital to Analog Converter Federal Communications Commission Frequency Modulation Institute of Electrical and Electronics Engineers In phase independent and identically distributed Infinite Impulse Response Illinois Institute of Technology Inter-Symbol Interference Industrial, Scientific and Medical Mega Bits Per Second xii

13 mse Msps MWO OFDM PBCC PSD Q QPSK RF SIR SNR Wi-Fi WIL WiNCom ZSM mean squared error Mega Symbols Per Second Microwave Oven Orthogonal Frequency Division Multiplexing Packet Binary Convolutional Coding Power Spectral Density Quadrature phase Quadrature Phase Shift Keying Radio frequency Signal-to-Interference Ratio Signal to Noise Ratio Wireless-Fidelity Wireless Interference Laboratory Wireless Networking and Interference Research Center Zero Span Measurements xiii

14 LIST OF SYMBOLS Symbol B B m π Definition vector containing Barker sequence modified Barker sequence for 3 bits buffered system constant pi j the complex number 1 ƒ ω ƒ c T T c T (subscript) t n n c x(n) σ x 2 θ frequency variable in Hz frequency variable in radians signal carrier frequency time period of periodic signal time period of 1 chip time duration constants time variable discrete time interval where time t=nt number of chips binary input data at time interval nt variance of data x(n) arbitrary phase angle k, i, m, s arbitrary indices for numbering, summation or power k 0 to k i t 0 to t i d 0 to d i c 0 to c i constants time constants with indices 0 to i data bits chip sequence xiv

15 α n differential phase for the n th data symbol w(t) phase at time t v b( t) y(t) S y (f) p(t) P(f) s 0 to s i p si (t) P si (f) p s( t) p L (t) p c (t) modulated analog transmission signal using Barker spread analog baseband signal PSD of y(t) general pulse train PSD of p(t) indices for numbering pulse waveforms p si (t) pulse train with an index si PSD of p si (t) sinusoidally shaped Barker waveform Logarthmically shaped Barker waveform Barker waveform shaped by sinc function b, b (subscript) constants determining time response and bandwidth of sinc pulse b(t) a(t) s(n) v c (t) a I a Q x(n, k) y(n, k) x reversed continuous time Barker pulse continuous time Barker pulse positivie/negative sign of pulses for n th bit analog transmitted signal using CCK (at modulated frequency) In phase part of CCK signal Quadrature part of CCK signal I phase component of k th chip for the n th bit Q phase component of k th chip for the n th bit I phase CCK vectors xv

16 y c C v(t) x(t) s(t) c(t) f ac f 1, f (subscript) β Q phase CCK vectors All possible CCK vectors CCK vectors where C= c analog signal model for MWO signal at modulated frequency Amplitude function of MWO signal Frequency sweeping FM signal for MWO model ON cycle waveform for one period of MWO signal model AC line Frequency Carrier frequencies FM modulation index for MWO signal A, A (subscript) Amplitude constants t (subscript) E(f n ) E O N λ n F c y T (t) time delays for shifting pulses MWO signal s transient power at frequency f n amplitude scale factor for MWO transient power total number of sinc pulses in MWO model random variable for shifting sinc pulses in MWO model random variable for MWO s carrier frequency offset threshold detector output for interference mitigation xvi

17 ABSTRACT This dissertation presents two methods to improve the performance of existing Wireless Fidelity (Wi-Fi) networks. One method is symbol shaping, while the other is interference mitigation via the use of cognitive radio. The Federal Communications Commission mandates a spectral mask that governs the spectral signatures and bandwidth of all IEEE communication systems. Filters are necessary to achieve this spectral mask but they introduce Inter-Symbol Interference (ISI) that degrades the performance of the communications system. Symbol shaping is a technique that shapes the transmitted digital information symbol to lower the sideband amplitudes in the Power Spectral Density of the system. This method was applied to the IEEE Barker spread 1 Mbps signal, and the Complementary Code Keying 5.5 Mbps signals. The goal was to approximately achieve the spectral mask requirements so that a low order filter would satisfy the spectral mask requirement, thereby, lowering ISI and improving the wireless communication system. The resulting system was extensively studied experimentally and via simulation. The Microwave Oven (MWO) is a common appliance that interferes with IEEE Wi-Fi communication systems as it radiates in the same 2.4 GHz Industrial, Scientific, Medical band. The nature of the MWO radiated signal is studied in detail and an analytical model is developed that captures the key aspects of the radiation. This knowledge is used to develop an interference mitigation technique using cognitive radio that successfully mitigates MWO interference on Wi-Fi communications. This is implemented and studied experimentally. This cognitive radio system is used to improve Wi-Fi communications as it allows the mitigation of MWO interference. xvii

18 1 CHAPTER 1 INTRODUCTION 1.1 IEEE Communication Systems Wireless digital data communications have become very popular and widespread over the last decade. This is in step with the growth of the Internet as well as the price reduction and widespread availability of portable devices like laptop computers and wireless routers. Initially, there were several competing technologies employed in wireless computer networks [TAN96] until the Institute of Electrical and Electronics Engineers (IEEE) undertook to develop standards. The IEEE protocols [IEE97] were formulated to standardize wireless computer networks. These protocols played a significant role in making wireless computer networks ubiquitous, as now a variety of devices based on different platforms could communicate wirelessly as long as they adhered to the IEEE standards. Additionally the popularity of the IEEE standards grew as the protocols evolved to include higher data rates. These rates permitted high bandwidth applications like video and image streaming, thereby making IEEE communications desirable for most business, residential, and college environments. The IEEE protocol has three common standards: IEEE a/b/g. The 2.4 GHz Industrial, Scientific and Medical (ISM) band has been allocated for the costfree operation b and g systems, while the a systems function in the 5 GHz band. The IEEE b [IEE99] permits data transmission at rates of up to 11 Mbps, while the IEEE a and g technologies support up to 54 Mbps data rate. In IEEE b

19 2 systems, data is transmitted at 1 and 2 Mbps by means of an 11 chip Barker spreading code. In the 1 Mbps case, Binary Phase Shift Keying (BPSK) is used, while Quadrature Phase Shift Keying (QPSK) [PRO94] is used to transmit the 2 Mbps signal. The 5.5 Mbps and 11 Mbps signals are transmitted by 8 chip Complementary Code Keying (CCK) codes, and QPSK modulation is used. The IEEE a and IEEE g systems use Orthogonal Frequency Division Multiplexing (OFDM) and Packet Binary Convolutional Coding (PBCC) to transmit at higher data rates. This research project aims to improve the performance of existing IEEE wireless systems by symbol shaping and interference mitigation. The layout of this thesis dissertation is as follows: chapters 2 and 3 apply symbol shaping to the 1 Mbps Barker spread signal, chapter 4 applies symbol shaping to the 5.5 Mbps signal, while chapters 5 to 7 involve microwave oven studies, modeling, and interference mitigation. 1.2 Spectral Mask & IEEE Channels All wireless communication systems have bandwidth and power regulations set by the Federal Communications Commission (FCC). These are the spectral mask requirements. The spectral mask regulations have been formulated to minimize the interference caused by the presence of wireless systems competing for the limited spectral region. The bandwidth limitations allow multiple users to share the spectrum using several Wireless-Fidelity (Wi-Fi) channels. The power limitation ensures that the interference caused by a wireless system is confined to a limited spatial space the size of which is proportional to the radiated power. For wireless computer networks adhering to the IEEE standards, the maximum allowable transmit power is 1 watt. The FCC spectral mask is shown in Figure 1.1. The mask limits most of the radiated Radio

20 3 Frequency (RF) power within a bandwidth of 22 MHz by requiring the transmitted signal Power Spectral Density (PSD) [PRO94] to be down by 30 db at 11 MHz from the carrier frequency, and down by 50 db at 22 MHz from the carrier frequency. In the 2.4 GHz ISM band, the IEEE b & g signals are allocated frequencies from 2401 to 2473 MHz in the United States [LEU03]. Outside of this range, the PSD of the signals is highly attenuated. This spectral space is further sub-divided into 11 channels as shown in Figure 1.2. Channel 1 has a center frequency of 2412 MHz and channel 11 is centered at 2462 MHz. Channels 2 through 10 are centered between channels 1 and 11 with adjacent channel spacing of 5 MHz each [MIS06]. However, each channel has a 22 MHz bandwidth. This means each IEEE channel overlaps with adjacent ones as seen in Figure 1.2. In practice, most wireless Access Points (AP) are assigned either channels 1, 6 or 11 as these channels are sufficiently far apart (25 MHz spacing) to avoid interference caused by spectral overlap [MAC07]. Figure 1.1. FCC spectral mask: The spectrum of an 11 Mbps rectangular pulse shaped signal is shown for comparison. (Source: IEEE standards)

21 4 Figure 1.2. IEEE b & g channel allocations in the US 1.3 Wireless Interference Officially, wireless interference is occurs when two or more Wi-Fi signals overlap with each other at the same spatial, spectral, and temporal locations. Interference causes degradation in the performance of the wireless system as the digital data is often corrupted due to the overlapping signals. Such interference necessitates retransmission, thereby data transmission. In addition, rates may need to be decreased. However, similar loss in data throughput performance occurs due to the listen-before-talk paradigm that is programmed into most wireless systems. This effect is often confused with wireless interference and, hence, warrants some explanation. Under normal conditions, all of the radios on a given channel share access to the airwaves by means of a listen-before-talk mechanism. The technical term for this mechanism is Carrier Sense Multiple Access/Collision Avoidance (CSMA/CA) [GOL05]. Basically, the radios listen to determine if another device is transmitting. If two or more physically close wireless data transmitters have PSDs that overlap, each transmitter will wait and transmit only if the channel is available, i.e., no other device is

22 5 transmitting in at that time. For example, suppose there are two such devices, namely, device A and device B. If device B wants to transmit data while device A is occupying the wireless channel, device B has to wait until the channel has been cleared by device A. Then, when device B takes over the channel, device A cannot transmit until the channel is again clear. So clearly, the users of both devices A and B experience degradation in the data communication performance, although wireless interference was avoided by means of the CSMA/CA algorithm. This is often misconstrued as wireless interference. Although the listen-before-talk mechanism helps prevent wireless interference, since data throughput is adversely affected by it, people often confuse its effect as wireless interference. Thus, the net effect and the cause are similar. The cause is that several devices are competing for the same spectral space. The effect is that the data throughput declines. However, there is one important difference. When wireless interference occurs, data for some or all the interfering devices is corrupted. This may make communication impossible. However, for the listen-before-talk mechanism, data is not corrupted as wireless interference does not occur. So communication is still possible, albeit, at a reduced throughput rate. Wireless interference is now the single most disruptive factor for wireless networks, in general. In a typical office building or multi-storied residential complex, there are many APs that provide the wireless infrastructure for different computer networks. The APs interfere as they compete for the limited number of IEEE channels. At best, the range of each Wi-Fi network is reduced due to interference. At worst, the problems caused by interference plus the listen-before-talk are so severe that the wireless connection between some APs and their subscribers can be lost. Figures

23 6 A.1, A.2 and A.3 in Appendix A are spectrograms [RAP02] that visualize the invisible phenomenon of interference. A spectrogram provides time, frequency and power information, and so the spectral and temporal overlap characteristics of interference are illustrated by the interference spectrograms. Considerable research effort has been invested in studies that attempt to mitigate interference or into the development of wireless communication systems that are more robust to its interference. This includes cognitive radio and the application of adaptive antennas [COM88]. In this research project, a wireless mitigation technique is developed that allows Wi-Fi systems to avoid wireless interference caused by MicroWave Ovens (MWOs) transmitting non-data carrying signals in the 2.4 GHz ISM band. 1.4 Inter-Symbol Interference Inter-symbol interference (ISI) is a form of data signal distortion where the previously transmitted symbols have an effect on the currently received symbol. In a communication system free from ISI, the energy from each received symbol is confined in the symbol interval. However, when ISI occurs, the time interval of each symbol is stretched, that causes the current symbol to receive some energy from preceding symbols. The energy from the previous symbols has a similar effect as noise, thus making the communication less reliable. As a result, the current symbol received has a greater probability of being decoded incorrectly. ISI occurs when the frequency response of the transmission channel is not flat and/or the channel s bandwidth is less than the bandwidth of the data signal. The result of transmission in such a channel is a filtering effect that introduces a filtering delay in each symbol. This delay elongates the symbol causing ISI. For the same reason, using

24 7 filters to adjust the PSD characteristics of a signal also leads to ISI. This phenomenon of ISI is clearly illustrated by Figures 1.3 and 1.4. Figure 1.3 shows a sequence of symbols before ISI distortion. Fig 1.4 shows what the symbol sequence may look like after ISI distortion. Notice how each symbol leaks into the following symbol. Figure 1.3. Ideal symbol sequence without ISI Figure 1.4. ISI distortion of a symbol sequence The PSD of the IEEE b signals fails to meet the FCC spectral mask requirements. Similar to the 11 Mbps rectangular pulse signal spectrum shown in Figure 1.1, the IEEE b signals have sidelobes that are not sufficiently attenuated. As a consequence, high order filters are required in the Wi-Fi transmitters to attenuate the sidelobes enough to satisfy the spectral mask. The resulting transmitted signal inherently suffers from ISI distortion. High ISI increases the Bit Error Rate (BER) of the communication system. The objective of this work is to achieve the spectral mask without using a high order filter, such that ISI is minimized.

25 8 ISI can be corrected either by increasing the signal s symbol duration or by pulse shaping. By increasing the symbol period, the time space between the data symbols is increased and thus the symbol stretching caused by ISI now has lesser effect on the data. However, increasing the symbol period means that the data transmission rate is lowered, and this is highly undesirable in today s high speed wireless networks. Symbol shaping, also called pulse shaping, involves changing the shape of the data symbols. Pulse shaping is done such that the new data signal possesses better PSD characteristics, such that the amplitude of the sidelobes is lowered. As a result, the filtering requirement necessary to meet the spectral mask is minimized, thus minimizing ISI. 1.5 Barker Code and IEEE Mbps Signal Generally speaking, a spreading code is a binary (uni-phase) or polyphase code that expands the bandwidth of a digital signal by modulating each data symbol. Each data symbol is transmitted by a sequence of n c chips, and the bandwidth is expanded by a factor of n c. This n c is called the processing gain. By spreading a data signal over a wider bandwidth, the signal becomes more resistant to narrowband interference. On the receiver side, the data sequence is de-spread accomplished by using the same spreading code to recover the original data. The 11 chip Barker spreading code expands the bandwidth of a regular 1 MHz bandwidth digital binary data signal by a factor of 11. The 1 Mbps data rate, however, of the digital signal remains unchanged. This factor of 11 processing gain of the Barker sequence makes the transmitted 1 Mbps wireless data signal highly robust to narrowband interference. The Barker chip sequence used in the standard is: B = [+1, 1,+1,+1, 1,+1,+1,+1, 1, 1, 1]

26 9 This sequence has good auto-correlation [PRO94] properties that make the code ideal for data transmission in wireless channels in indoor environments [SAL87]. In these type of environments, wireless signals experience multiple reflections from walls and furniture. Multipath fading [GOL05], therefore, is a significant problem that threatens to make the wireless channel error-prone. However, for a Barker spread signal, reflected and time-delayed multiple versions of the same signal are poorly correlated to each other. This is because the correlation detector [PRO94] gives a high value only if there is no time delay between the received symbol and the reference correlator symbol. The correlation value is low for any delayed multipath version of the same signal. This is demonstrated by a plot of the auto-correlation function of the 11 chip Barker sequence in Figure 1.5. The PSD of the Barker spread IEEE signal is plotted in Figure 1.6. The FCC spectral mask is shown by the dashed line. Clearly, it fails to meet the spectral mask requirement [LEE05]. Consequently high attenuation output filters are needed to satisfy the spectral mask requirement, which adds to the hardware cost and increases the BER by introducing ISI. Auto-correlation Time Delay (sec) x 10-6 Figure 1.5. Auto-correlation plot of 11-chip Barker sequence

27 10 Spectral Mask Figure 1.6. PSD of Barker spread 1 Mbps signal The IEEE Mbps signal is also spread by the same 11 chip Barker code as mentioned in Section 1.1. For the 2 Mbps case, QPSK is used. Two bits are taken at a time: one for the in (I) phase and the other for the quadrature (Q) phase. Both the I and Q phase bits are spread by corresponding I and Q phase Barker code waveforms. Due to interference, channel noise, and decreased signal strength with distance, it is quite common for the Wireless Local Area Networks (WLANs) to transmit at 1 or 2 Mbps between APs and laptops. Additionally, the Physical Layer Convergence Protocol (PLCP) packet [CON00] is transmitted using the 1 Mbps signal. So although Wi-Fi data rates have gone up tremendously, the 1 Mbps Barker spread signal still is crucial to IEEE systems. Thus if the 1 Mbps signal can be improved, this will cause an overall improvement of data transmission using IEEE Wi-Fi at any data rate. In this project, attempt is made to improve this 1 Mbps IEEE signal by pulse shaping to lower the signal s ISI such that it gives improved performance in BER versus Signal to Noise Ratio (SNR) studies.

28 IEEE CCK 5.5 Mbps Signal The 5.5 Mbps CCK signal uses an 8 chip polyphase spreading code to transmit 4 bits at a time. The symbol rate is Msps, and since the processing gain is 8, the baseband bandwidth of the CCK signal is 11 MHz. Unlike the real valued Barker spreading sequence however, the CCK spreading sequence is complex. This complex structure is what makes this spreading code a polyphase spreading code. In quadrature phase 8-chip CCK, there are possible code words, and sets of 64 that are nearly orthogonal. This is because it takes 16 bits to define each code vector (2 phases X 8 chips = 16 possible combinations). All the 64 nearly orthogonal code words are used in 11 Mbps CCK signal. However, for the 5.5 Mbps data rate signal, a subset of 4 of the 64 vectors having superior coding distance is used. In the 5.5 Mbps CCK signal, the incoming data is grouped into 4 bits nibbles where 2 of those bits select the spreading function out of the set of 4 while the remaining 2 bits modulate the symbol using QPSK. The 4 bit sequence is represented as [d 0, d 1, d 2, d 3 ]. Bits [d 0, d 1 ] select the differential carrier phase as specified in Table 1.1. Notice, that the symbol number affects this choice (even or odd). The 4 complex spreading sequences are shown in Table 1.2 and are selected by [d 2, d 3 ]. Table 1.1. Differential QPSK encoding table used in CCK transmission Dibit pattern (d 0, d 1 ) Even symbols Phase change Odd symbols Phase change 00 0 π 01 π/2 3 π/2 (-π/2) 11 π π/2 (-π/2) π/2

29 12 Table Mbps CCK encoding table d 2, d 3 c 1 c 2 c 3 c 4 c 5 c 6 c 7 c j 1 1j -1 1j 1-1j j -1-1j 1 1j 1-1j j 1-1j -1-1j 1 1j j -1 1j 1-1j 1 1j 1 When combined, the resulting transmitted analog signal, v c (t) is represent able as ( 2π f t + α + w( n, + θ ) vc ( t) = 2 cos c n k), (1.1) where α n is the differential phase for the n th data symbol (from Table 1.1), w( n, k) is the phase of the n th data symbol s k th chip determined from Table 1.2, θ is an arbitrary phase angle, and f c is the carrier frequency of the signal [ALB06]. Cross coupling distortion occurs in M-ary phase shift keying for M > 2, where the I and Q phase signal energies leak into each other. This commonly occurs in wireless multipath environments because multipath signal components arriving with a delay are shifted in phase. Thus, the later arriving multipath with a phase shift corrupts the I and Q information from the primary signal ray. CCK that is used in the b standard is quite resistant to multipath distortion in the form of cross coupling. This is because the information in CCK is encoded directly onto complex chips, which cannot be crosscouple corrupted by multipath. Thus 5.5 Mbps CCK signals are often utilized by APs when channel conditions prohibit the use less robust higher data rate signals. The PSD of the 5.5 Mbps CCK signal is shown in Figure 1.7. It is interesting to note that this PSD has a simple sinc (sin x / x) shape that matches with the PSD of a 11 Mbps BPSK rectangular pulse shaped binary signal. The spectral mask is not achieved by the unfiltered 5.5 Mbps CCK signal and this necessitates high order filtering. Again, the filter introduces ISI that degrades the BER versus SNR performance of the

30 13 Wi-Fi system. Here also, the application of pulse shaping to reduce the ISI is a valid option. Power in dbm PSD plot Spectral Mask Frequency in Hz x 10 7 Figure 1.7. PSD of 5.5 Mbps CCK signal 1.7 ComBlock Devices ComBlocks [COM06] are modular communication chipsets that can be connected as blocks to construct transmitters and receivers. Each module performs a specific task, for example, modulation at 2.4 GHz. Modules are also swappable. In a transmit configuration, for example, the 2.4 GHz modulator can be replaced by a 900 MHz one, and the transmitter will function in a different frequency band. A transmitter was constructed using five ComBlock chipsets connected in series: a computer interface module, an arbitrary waveform generator, a high-speed baseband Digital to Analog Converter (DAC), a modulator operating in the 2.4 GHz band, and an amplifier. The transmitter is shown in Figure 1.8. Using this set up, any waveform with a maximum double sideband bandwidth of 40 MHz can be transmitted. In order to

31 14 transmit the waveform, a digital version of the waveform is generated in a computer using MATLAB software and this is saved in a data file. The data file is uploaded via the computer interface module into the arbitrary waveform generator. The generator outputs the digital waveform at up to 40 Msps, and the DAC converts this to a baseband analog waveform consisting of I and Q phases. The 2.4 GHz modulator mixes this baseband signal up to the ISM band, and the amplifier amplifies this for the transmit antenna. A receiver was also constructed using ComBlock modules. The receiver, displayed in Figure 1.9, consists of three chipsets. A 2.4 GHz receiver antenna receives the Wi-Fi signal. At the RF end, one chipset demodulates the RF signal and mixes the 2.4 GHz received signal to baseband. The same chipset then samples the baseband analog signal s I and Q phases and digitizes the received waveform using a high-speed Analog to Digital Converter (ADC). The digitized waveform is stored in the second module, which is a memory storage unit. The third module is a computer interface chip. The digital waveform is transferred from the memory storage unit to a computer via this module for analysis using MATLAB. Figure 1.8. ComBlock transmitter system

32 15 Figure 1.9. ComBlock receiver system Block diagrams of the transmitter and receiver are shown by Figures 1.10 and 1.11, respectively. The diagrams show the flow of information and summarize the chipsets operations. The blocks with dashed edges denote individual chipsets. Computer ComBlock USB chip Waveform storage DAC conversion I Q 20 MHz LPF I Q BPF ( MHz) 2.4 GHz Amplifier 2.4 GHz QPSK modulator Figure Block Diagram of ComBlock Transmitter Computer ComBlock LAN chip Waveform storage ADC conversion I Q 20 MHz Lowpass filter I Q BPF ( MHz) 2.4 GHz Amplifier 2.4 GHz QPSK demodulator Figure Block Diagram of ComBlock Receiver

33 16 In this research project, the ComBlock transmitter was used to experimentally transmit novel pulse shaped versions of the 1 Mbps signal. The receiver was used to capture the signal, demodulate the digital data, and obtain BER measurements to test the performance of the experimental Wi-Fi systems. The ComBlock transmitter was also used to emulate a MWO signal based on a model developed as part of this research. Additionally, an interference mitigation experiment was conducted where the transmitter and receiver both were used to test the efficacy of the interference mitigation system. Furthermore, the ComBlock receiver was used to obtain Spectrogram plots, which were instrumental in examining the phenomenon of wireless interference. Spectrograms were also useful in studying the signal characteristics of MWOs. 1.8 Research Methodology Research was conducted at the Wireless Interference Laboratory (WIL), which is a part of the Wireless Networking and Communications (WiNCom) Research Center at the Illinois Institute of Technology (IIT). Research undertaken at the WIL includes studies examining the impact of wireless interference on computer networks, IEEE signal characteristics and scope for signal improvement, characterization of wireless transmission devices, and interference mitigation. The research undertaken in this project falls under three of these categories: improvement of IEEE signals, characterization of a MWO, and mitigation of MWO interference. To ensure the validity of the research results, a three-pronged approach was used in addressing all the problems. Each problem was examined analytically, experimentally and via simulation. The results obtained using the different approaches were compared to obtain veritable conclusions. For example, when a sinusoidal pulse shaping function was

34 17 used for 1 Mbps Barker spread code, an analytical expression for the PSD was obtained and plotted. This was then compared to the simulated and experimentally obtained PSD plots. The three plots were in agreement and, thus, provided concrete support for the results. The three-pronged approach was also useful in detecting and weeding out errors and mistakes during the course of research. For example, when the experimentally generated PSD for a buffered and pulse shaped 1 Mbps signal diverged from the simulated PSD, an error in the simulation process was discovered and subsequently corrected. However, due to high levels of complexity for some problems, the threepronged approach could not always be used. A case in point: effort was made to find the analytical expression for the PSD of an MWO, but there was little success in this regard. MATLAB software [MAT07] was used throughout the research project. PSpice software was used to design the interference mitigation circuit and simple logic chips were used for its construction. For experimental work, ComBlocks were used. A spectrum analyzer was used for important measurements. Several measurement devices including oscilloscopes, voltmeters, etc. were used during the research.

35 18 CHAPTER 2 PULSE SHAPING FOR IEEE 1 MBPS BARKER SPREAD SIGNAL 2.1 Pulse Shaping Methodology For the case of the 1 Mbps Barker spread signal without any pulse shaping, each data bit is transmitted as a sequence of 11 Barker chips with chip interval, T c = T / 11, where T is the bit duration (1 µs). The chips are unit amplitude rectangular pulses, and the PSD for the unfiltered data signal modulated by this Barker wave shape does not satisfy the FCC spectral mask. Figure 1.6 shows that the second lobe must be filtered by at least 17 db and the third lobe by at least 32 db to be below the mask. In our simulation studies, a fifth order Butterworth filter with a cutoff frequency of 9.5 MHz was required to satisfy the mask requirements. This introduced considerable ISI. The Barker waveform was modified by smoothing the rectangular pulse shapes in the original Barker symbol. The modified pulse shapes still adhered to the general Barker sequence and maintained good autocorrelation properties. Several smoothing functions were applied to the Barker waveform of which three shapes that provided best results were examined thoroughly. Sinusoidal, logarithmic and a sinc m functions were used to smooth individual chips in these three cases. Regardless of the exact form of the Barker symbol shape, the baseband data signal can be represented as: n= y ( t) = x( n) p( t nt ), (2.1) where x(n) { 1, 1} is random binary data, that is independent and identically distributed, and p(t) is the pulse shaped Barker waveform. The signal, y(t), is a zeromean cyclostationary [PRO94] random process with PSD given as (2.2):

36 19 2 σ x 2 Sy ( f) = P( f), (2.2) T where σ 2 x is the variance of x(n) and P(f) is the Fourier transform of the pulse shaped Barker symbol, p(t). For each symbol shape, the resulting communication system was analyzed via simulation and experimentation. The PSD was of key interest in order to observe spectral improvements. The amount of sideband attenuation achieved was observed to check the degree to which the spectral mask was satisfied. However, the spectral mask could not be completely satisfied by symbol shaping alone and filtering was still required. However, for the novel signals, only low order filters are necessary to achieve the spectral mask. Thus, ISI is considerably reduced. This theoretically should translate to better system performance in BER versus SNR studies. In order to investigate this theory further, more simulations and experiments were conducted. An auto-correlation plot of each pulse shape was obtained and compared with that of the unmodified Barker waveform. A good similarity between the two implies that the new signal should have good multipath robustness. Beyond the auto-correlation plot itself, in order to further validate the performance of the communication systems, BER simulation studies were performed for each of pulse shaped Barker waveforms in MATLAB. In these studies, random binary data was Barker spread using each symbol shape to obtain a simulated transmit signal. A minimum order Infinite Impulse Response (IIR) filter [PRO96] was used to filter the baseband information signal such that the PSD satisfied the spectral mask without introducing excess ISI. This signal was applied to an Additive White Gaussian Noise (AWGN) channel and the SNR of the output noisy signal was recorded. After transmission through this simulated AWGN channel, a correlation

37 20 detector [PRO94] was used to decode the received bits. The decoded bits were compared to the transmitted bits to obtain a BER value for the channel at the recorded SNR level. Keeping true to the three-pronged approach described in Section 1.8, the communication systems with the various symbol shapes were emulated by the ComBlock transmitter and receiver. Experimentation was done to cross-check the simulation results. A Rohde & Schwarz spectrum analyzer (model no. FSP 38) was used to obtain the experimental PSD. Experimental BER studies were conducted where Barker spread data signals (using the novel pulse shapes) were transmitted over the air. The data signals were captured by the ComBlock receiver, the signals were demodulated and the received bit streams were decoded to obtain BER measurements. The ComBlock receiver, however, did not do the BER analysis. In each experimental run, the digitized waveform captured by the ComBlock was downloaded to a computer where a MATLAB program analyzed the waveform to decode the received bits using a correlation detector and an experimental BER value was obtained. For the experimental communication system, however, a 1 Mbps data rate could not be used. Due to hardware speed limitations, a 4 MHz chip-rate was used corresponding to a bit-rate of 363 kbps with the 11 chip Barker code. Using 4 Mchips/second, the main-lobe bandwidth in the experimental PSD is expected to be 4 MHz. The baseband modulated signal was viewed with a 400 MHz oscilloscope to provide a temporal domain representation. All these experimental plots were compared to the simulated and analytically expected results as based on the three-pronged research approach.

38 21 The implementation of complex pulse shaping in a transceiver system is feasible, where the system hardware can be constructed inexpensively by replacing the high-speed DAC with a discrete-time analog storage device. Such a device stores the modulator s signal level values as analog voltages. During each bit interval, the analog voltages will be output at discrete sub-time intervals to construct the complete pulse shape. Typically, a communication system requires a small finite set of pulse shapes. Thus, only a limited number of the analog storage cells are needed, thereby eliminating the need for complex digital logic circuits and DACs. Recently, a topic of promising research has been the integration of such analog waveform generators with digital communication systems [CHA05]. 2.2 Sinusoidal Pulse Shaping Figure 2.1. Plot of Sinusoidally shaped Barker waveform Sinusoidal pulse shaping was employed to shape sequences of Barker chips. The resulting wave shape is shown in Figure 2.1, where the energy per symbol is equal to that

39 22 of the original Barker wave shape. The sinusoidal pulse shape can be expressed analytically as where the compound parts are given as s S K k= 1 4 p () t = 2 p () t, (2.3) p () t = sin(2 πt/2 T ), 5T t 3T S1 C C C p () t = cos(2 πt/4 T ), 3T t T S2 C C C p () t = sin(2 πt/2 T ), T t 0 S3 C C p () t = sin(2 πt/6 T ), 0 t 6T S4 C C. (2.4) The waveform based on these equations was used to spread random digital data in a MATLAB program. A digitally represented waveform was obtained and the data file was uploaded to the ComBlock transmitter. The ComBlock s arbitrary waveform generator generated the analog waveform using this data. The experimental analog waveform was examined with an oscilloscope as is shown in Figure 2.2. It matches with Figure 2.1. Figure 2.2. Oscilloscope plot of sinusoidally shaped Barker waveform

40 23 The PSD of the sinusoidally shaped Barker waveform lends itself to an analytic study. Using rectangularly windowed and shifted sinusoids, as defined in (2.4), the Fourier spectrum of p s( t) is found to be P ( f ) = π T s c k 2 2 jk f Tc ( k / 2) cos( kπ ft c ) e 2 2 = 1 ( kπ ftc ) ( π / 2) ( 1) j (4k 2)sin((4 k 2) f Tc ) e j 2 2 ((4k 2) π f Tc ) π k (10 2k ) f T c. (2.5) Using (2.2) and (2.7) the PSD can be computed. The analytic PSD is displayed in Figure 2.3, and the MATLAB simulated PSD (with 5000 bits) is plotted in Figure 2.4. The agreement is excellent and we observe that there is an 11 db attenuation improvement over the rectangular Barker PSD. To satisfy the FCC mask requirement a simple second order Butterworth filter with a 9.5 MHz cutoff frequency is needed. An experimentally measured PSD using a sinusoidally shaped Barker waveform, modulated at GHz, is given in Figure 2.5, along with the FCC spectral mask (dashed lines). The analytic and simulated PSD, in Figs. 2.3 and 2.4, respectively, match very closely with the experimental PSD. Note that the experimentally emulated PSD in Figure 2.5 has a mask with transitions at 4 MHz and 8 MHz away from the carrier frequency since the chip rate is 4 MHz. With equal energy per shaped symbol, the peak of the autocorrelation function should be the same for the sinusoidal and rectangular pulse shaped Barker waveforms. Plotted in Figure 2.6 is the autocorrelation function of the rectangular Barker waveform (dashed line), and the sinusoidally shaped Barker waveform (solid line). The Barker code s autocorrelation property dictates that the autocorrelation function is bounded by

41 24 one-eleventh of its peak for time shifts of 1 chip or more. This property is largely preserved with the sinusoidally shaped Barker waveform, and is strictly preserved for time shifts of 3 chips or more. Consequently, the shaped Barker communication system should be robust to multipath distortion and noise. The mean square error (mse) between the two auto-correlation plots in Figure 2.6 is For comparison purposes, the power of the rectangular Barker s auto-correlation waveform is Analytical PSD of Sinusoidally Pulse shaped Barker waveform Power in dbm Frequency (Hz) x 10 7 Figure 2.3. Analytic PSD of sinusoidal pulse shaped Barker waveform.

42 25-30 Baseband Transmitted signal PSD Power in dbm Frequency in Hz x 10 7 Figure 2.4. Simulated PSD of sinusoidal pulse shaped Barker waveform. Figure 2.5. Experimental PSD of sinusoidal pulse shaped Barker spread system emulated at a 4MHz chip rate.

43 26 15 Time Auto Correlation Auto-correlation Time (s) x 10-6 Figure 2.6. Auto-correlation function of sine-shaped Barker 2.3 Logarithmic Pulse Shaping A logarithmic smoothing function was used for the leading and trailing transitions of the Barker chip sequence. The general form of this function was p L t) = k + k log( k t + ), (2.6) ( t0 where k 0, k 1, k 2, and t 0 are constants. Plotted in Figure 2.7 is the logarithmically shaped Barker symbol and the (original) rectangularly shaped Barker waveform. The amplitude of the log-shaped waveform has been adjusted so that the energy per symbol is the same in both cases. A closed form analytic expression cannot be obtained for the PSD of the logarithmically shaped Barker symbol. Its PSD was found with a MATLAB simulation using the Welch PSD [PRO96]. This PSD result is plotted in Figure 2.8 using 5,000 data bits. This PSD shows an improvement in spectral characteristics, where the sidebands are attenuated 8 db more than the rectangular Barker PSD shown in Figure 1.6. A third

44 27 order Butterworth filter with a 9.5 MHz cutoff frequency is needed to satisfy the spectral mask, compared to a fifth order filter when pulse shaping is not used. The single sideband PSD of the experimentally emulated logarithmically pulse shaped Barker signal is shown in Figure 2.9. The auto-correlation plot of the logarithmically shaped pulse is shown in Figure 2.10 and compared to the ideal, the mse value is only Figure 2.7. Logarithmic and rectangular shaped Barker symbol with equal energies. -30 Baseband Transmitted signal PSD Power in dbm Frequency in Hz x 10 7 Figure 2.8. Simulated PSD of the logarithmic Barker waveform.

45 28 Figure 2.9. Experimentally emulated PSD of the logarithmic Barker waveform. 15 Time Auto Correlation Auto-correlation Time (s) x 10-6 Figure Auto-correlation function of log-shaped Barker 2.4 Sinc m Pulse Shaping A third shaping method was studied employing sinc-functions of the form [ sinc( bt + t ] m p = ), (2.7) c ( t) A 1

46 29 where 0 < m < 1. The best PSD, relative to the FCC mask, was obtained with two values for m, where m 1 = 0.55 for the 1-chip segments, and m 2 = 0.83 for the 2- and 3-chip segments, e.g., Simulation was used to find these optimal m 1 and m 2 values for the best spectral characteristics. The plot of this pulse shape is shown in Figure Its auto-correlation function is plotted in Figure 2.12 and the mse compared to that of rectangularly shaped Barker s auto-correlation is As with the log-shaped Barker symbol, the sinc-function shaped symbol does not lend itself to a closed form analytic expression for the PSD. The simulated PSD is plotted in Figure 2.13, while Figure 2.14 shows the experimentally obtained PSD. Observe that the first sideband is attenuated by 12 db compared to the rectangular Barker pulse, but there is also less influence on the third sideband. A second order Butterworth filter with 9.5 MHz cutoff frequency was adequate to meet the spectral mask in this case 1.5 The sinc m shaped waveform p(t) in Volts Time (s) x 10-7 Figure Plot of sinc-function shaped Barker waveform.

47 30 15 Time Auto Correlation 10 Auto-correlation Time (s) x 10-6 Figure Auto-correlation of sinc-function shaped Barker waveform. -30 Baseband Transmitted signal PSD Power in dbm Frequency in Hz x 10 7 Figure Simulated PSD of sinc-function shaped Barker waveform.

48 31 Figure Experimental PSD of sinc-function shaped Barker waveform. 2.5 BER Measurements As previously mentioned, BER simulation studies were performed for each of the four Barker pulse shaped waveforms in MATLAB at various SNR levels. A synchronous demodulator and correlation detector were used in all cases. The BER simulation methodology has been described in words in Section 2.1. Figure 2.15 provides a visual summary. The BER study results are shown in Table 2.1 using 50,000 random test bits each time. The unmodified rectangular Barker waveform signal was also used in the BER study as the control case. This table also indicates the filter order used to achieve the FCC spectral mask requirement after pulse shaping. With low order filters, ISI is minimized; but ISI will increase as the filter order grows. High noise levels were chosen

49 32 to obtain meaningful BER results with the available computing resources. A BER vs SNR plot is shown in Figure 2.17 showing the performance of the various pulse shaped systems. This plot shows that the sinusoidally shaped system performs about 0.5 db better than its unmodified rectangular shaped counterpart. MATLAB Simulation Methodology used for each Pulse Shape 1) Design Pulse Shape adhering to Barker Sequence. 2) Examine its Autocorrelation properties. Generate random bit sequence and spread each bit by pulse shape to obtain data waveform Obtain the PSD of data waveform using the Welch method. Add Additive White Gaussian Noise (AWGN). Examine Bit Error Rate Use Correlator to decode the received bits Use Correlator to obtain timing information from the received signal Figure Diagram illustrating BER simulation study Experimentation Methodology used for each Pulse Shape Design Pulse Shape adhering to Barker Sequence in Matlab. Examine Bit Error Rate Generate random bit sequence and spread each bit by pulse shape to obtain data waveform Use Correlator to decode the received bits Upload the data waveform to the Comblock transmitter. Transmit over the Air. Comblock receiver captures the received data waveform for computer download. Use Correlator to obtain timing information Figure Diagram illustrating experimental BER study

50 33 Figure BER vs SNR study for pulse shaped Barker spread systems In Section 2.1, the experimental BER method used to test each communication system was described. Figure 2.16 illustrates this diagrammatically. The experimental BER results, obtained using the experimental ComBlock test bed, are shown in Table 2.2. Table Simulated BER measurements for Barker pulse shape (no buffer). Pulse Shape Used Filter Order Bit Error Rate at SNR levels: 11.5 db 11 db 10 db Rectangular E E E-04 Logarithmic E E E-04 Sinusoidal E E E-04 Sinc-function E E E-04 Table 2.2. Experimental BER measurements at receiver-to-transmitter distance of 1 meter for Barker pulse shape (no buffer). Pulse Shape Used Rectangular Logarithmic Sinusoidal Sinc-function Experimental BER 9.99E E E E-03

51 Comparison of Barker Symbol Shaped Systems In general, the sinusoidally shaped system performed the best both in simulation and experimental emulation. This system experienced little ISI due to the low order filter used. The rectangular Barker system required the highest order filter and thus experienced more ISI that degraded the performance. The improvements in the unfiltered spectral characteristics are summarized in Table 2.3, where the rectangular pulse shape forms the unmodified control system with no symbol shaping. The table shows the attenuation in peak lobe powers. Additionally, Table 2.4 shows the total amount of sideband energy leakage in the PSD of the various signals. For lower sideband powers, there is lesser interference caused to nearby Wi-Fi channels. Table Comparison of PSD peak sideband attenuations Pulse Shape Used Second Lobe drop Third Lobe drop Rectangular 13.1 db 17.2 db Logarithmic 18.2 db 20.8 db Sinusoidal 24.0 db 35.9 db Sinc-function 24.7 db 30.5 db Pulse Shape Table Comparison of total power in each band Main Lobe Power % Second Lobe power % Third Lobe power % Second Lobe drop (db) Third Lobe drop (db) Rectangular Logarithmic Sinusoidal Sinc-function Table 2.5 quantifies the amount of ISI occurring in each system, where the metric is the amount of energy in one symbol that leaks into the next symbol interval. We notice that the rectangular system has almost 5% energy leakage due to ISI compared to only

52 35 0.2% for the sinusoidal case. The value of symbol shaping versus output filtering to satisfy the FCC spectral mask has thus been firmly established [TAH07]. Pulse Shape Table ISI after filtering operation Power within Bit Interval (%) Power leakage outside Bit Interval (%) Rectangular Logarithmic Sinusoidal Sinc-function 99 1 At the end of the chapter, it is worth mentioning that the pulse shaped systems functioned as novel Wi-Fi setups that performed better than the common IEEE Mbps system in several respects. These were better spectral characteristics, lower order filter requirement, and improved BER performance. Again, the best performance was obtained with sinusoidal pulse shaping. These conclusions were verified by matching results obtained by analytic, simulation, and experimental studies. One principle goal of this research project is to improve IEEE Wi-Fi systems. This goal has been partly achieved as it has been shown that the IEEE b 1 Mbps signal can be improved by applying the results of this pulse shaping study.

53 36 CHAPTER 3 BUFFERED PULSE SHAPED BARKER SPREAD SYSTEMS 3.1 Rational for using Buffer The research in Chapter 2 showed that the sinusoidal Barker waveform shaping was able to reduce the PSD considerably compared to the rectangular shaped Barker symbol. To achieve FCC mask compliance, however, a second order output filter was still needed. The dominant feature of the PSD of the original Barker waveform comes from the abrupt transitions after just one bit interval, that is, after 1 μs. This feature is also seen with the shaped Barker symbols, and is illustrated in the experimentally recorded oscilloscope plot in Figure 2.2, where sinusoidal shaping has been used. The sudden discontinuity in the time domain raises the power of the higher frequency components in the spectral domain. Thus, to eliminate the need for an output filter to satisfy the FCC spectral mask, it is necessary to eliminate these discontinuities. It is possible to introduce special line codes for the Barker waveform that will not have the discontinuity seen in Figure 2.2. The line code would examine the current data bit and the next bit. If a chip transition from +1 to 1 (or vice versa) is about to occur, the line code alters the pulse shape of the next Barker spread waveform in such a way to guarantee a smooth transition from the current bit to the next bit. Thus, the discontinuities would be removed resulting in better spectral characteristics relative to the FCC spectral mask. In this case it is necessary to buffer two or three bits before the appropriate pulse shaped Barker symbol is output for information transmission.

54 37 Two line codes were studied to test this hypothesis. One line code buffered 2 bits, while the other dealt with 3 bits at a time. The two methods described here combine both line coding and pulse shaping in an effort to attain the spectral mask. The results are detailed herein. 3.2 Buffering 2 Bits In this system, we use two Barker waveforms to form a line code: the original Barker waveform, and its time reversed version. Section develops this signal, mathematically, where two bits are buffered at a time. Section shows all the possible symbols Mathematical Foundation of 2 Bit Buffered Barker System. Let x(n) be binary random iid data from a set {-1,1}, with zero mean and unity variance. A waveform, a(t), is defined for 0 t < T, that is a symbol pulse based on the Barker sequence B ( ). Another waveform, b(t), is defined for the interval 0 t < T, that is a symbol pulse based on the reversed Barker pulse ( ), such that b(t) = a(t-t). (3.1) The +1 bit is to be represented by the symbol a(t), and the bit -1 is assigned the symbol b(t). Continuing with the definitions, s(n) is defined as the sign (+ or ) for the information symbol, y(t) at time interval nt. Combining all, during time nt < t < (n+1)t: y(t) = s(n) a(t-nt), (3.2a) if x(n) = 1. However, if x(n) = -1, then y(t) = s(n) b(t-nt). (3.2b)

55 38 Since s(n) can take on two possible values, there are a total of four possible symbol states. These states, namely 1 through 4 are given as follows: State 1 = +a(t-nt) State 2 = a(t-nt) State 3 = +b(t-nt) State 4 = b(t-nt). (3.3) The line code permits state transitions such that no discontinuities occur at the end of a bit interval. The line code thus selects the appropriate bit symbol, a(t) or b(t), and the sign s(n) in order to avoid discontinuities of the type shown in Figure 2.2. The state transition diagram for the line code is shown in Figure 3.1. x(n) = +1 State 1 State State 3 State 4 1 Figure 3.1. State transition diagram for 2 bit buffered Barker system In this line code, (3.4) is used to generate the sign value, s(n): s(n) = x(n) x(n 1) s(n 1). (3.4) Taking the initial condition, x( 1).s( 1) = 1, (3.4) can be further simplified for n 0, to: s(n) = (-1) n x(n). (3.5a) For a different initial condition, x(-1) s(-1) = +1, and we obtain (3.5b). s(n) = (-1) n+1 s(n). (3.5b)

56 39 With these definitions and equation (3.5a), the block diagram of the system is constructed. Figure 3.2 shows the structure that is used to generate the information signal based on the 2 bit buffered line code. Figure 3.2. Block Diagram of 2 bit buffered Barker spread system In Figure 3.2, the sign block, s(n), is selected by the line code according to (3.5), and this selection helps remove the discontinuities of the type shown in Figure 2.2. The system in Figure 3.2 works as follows: 1. If x(n) = +1, the top path produces a waveform y(t)= s(n) a(t-nt). In this case, the lower path produces a zero since +1 1=0. 2. If x(n) = -1, the lower path produces a waveform y(t)= s(n) b(t-nt). In this case, the top path produces a zero since 1+1=0. Thus, over the time interval nt t <(n+1)t, y(t) is expressed as y(t) = 0.5[x(n) + 1] s(n) a(t-nt) 0.5[x(n) 1] s(n) b(t-nt). (3.6) From (3.5a), s(n) = ( 1) n x(n). Thus, (3.6) simplifies to: y(t) = 0.5( 1) n {[1+x(n)] a(t-nt) [1 x(n)] b(t-nt)}. (3.7)

57 40 From (3.1), we know that b(t) = a(t-t). Thus, further simplifying, over the interval time nt t <(n+1)t, y(t) = 0.5 (-1) n {[1+x(n)] a(t-nt) [1 x(n)] a(-t + T + nt)}. (3.8) Finally, for all t 0, based on (3.8) and factoring in the initial conditions (3.5), we get y( t) = n= 0 0.5( 1) n+ 1 {[ 1+ x( n) ] a( t nt) [ 1 x( n) ] a( t + T + nt) } x( 1) s( 1). (3.9) Note that (3.9) completely defines the line coding performed using the 2 bit line code system. Close examination of (3.9) reveals that only one pulse shaped Barker waveform is required to be designed, i.e., a(t). This simplifies the effort involved in pulse shaping. Indeed, the methods for symbol design used for this pulse shaped line code are similar to the pulse shaping effort where no line code is used. For the systems in Chapter 2, only one pulse shape waveform is required to be made each time, analogous to this 2 bit buffered system. Figure 3.1 and (3.9) also reveal an interesting property of the line code: strictly speaking, there is no real need to have a buffer register as y(nt) really just depends on x(n). However, the principle for the line code is based on buffering two bits, hence, the nomenclature of 2 bit buffered Barker system Symbols Shapes Tested. According to Figure 3.1, it is clear that 4 symbols are necessary, one for each state. However, we noted that all four pulse shaped symbols can be obtained from just one pulse shape. This is obvious from (3.1) and the state equations (3.3). Once the symbol for a(t) has been designed, it corresponds to the state 1 symbol. State 2 can be obtained simply as a(t). States 3 and 4 can be obtained by time reversing the symbols for states 1 and 2 respectively. So the key is the shaping function used for the waveform a(t).

58 41 Several functions were used to shape the chips for a(t). This included sinusoidal, logarithmic and sinc m functions. Figures 3.3 through 3.5 show the state symbol plots for each of these pulse shaping methods. It should be noted that using this system, there is increased complexity at the receiver side. Two correlation detectors are required: one to detect states 1 and 2, and another correlation detector to detect states 3 and 4. 1 Plot of bit +1; state 1 1 Plot of bit +1; state Time in s 1 x 10-6 Plot of bit -1; state Time in s x Time in s 1 x 10-6 Plot of bit -1; state 4; Time in s x 10-6 Figure 3.3. State symbols for sinusoidal pulse shaping (2 bits buffered)

59 42 1 Plot of bit +1; state 1 1 Plot of bit +1; state Time in s 1 x 10-6 Plot of bit -1; state Time in s x Time in s 1 x 10-6 Plot of bit -1; state 4; Time in s x 10-6 Figure 3.4. State symbols for sinc m pulse shaping (2 bits buffered) 2 Plot of bit +1; state 1 2 Plot of bit +1; state Time in s 1 x 10-6 Plot of bit -1; state Time in s 1 x 10-6 Plot of bit -1; state 4; Time in s x Time in s x 10-6 Figure 3.5. State symbols for logarithmic pulse shaping (2 bits buffered)

60 PSD Plots. A random bit stream was applied to the system described in Section using each of the pulse shaped state symbols shown in Section The simulated PSD was obtained in MATLAB. Initially, promising results were obtained. The spectral mask seemed to have been satisfied without using any filter according to the simulations results, where the maximum power in the sidelobes was much attenuated compared to the mainlobe power. However, the experimentally measured PSD varied from the simulation results, and the sideband attenuations were not that promising. It was then realized that the parameters for MATLAB s Welch function used to estimate the simulated PSD needed calibration. This was one situation where the three-pronged approach came in handy, whereby error in the simulation process was discovered and rectified as there was discrepancy between the simulated and experimental results. After this correction was made, the simulated and experimental PSDs matched. Unfortunately, these PSD results showed that the line code introduced many tones in the power spectrum that made the system unviable for Wi-Fi communications. The tones made it impossible for any of the pulse shaped systems designed using this line code (2 bits buffered) to meet the requirements of the spectral mask. For the various pulse shaping functions, the PSDs obtained using simulation are shown in Figures 3.6 to 3.8. The matching experimentally obtained PSDs are shown in Figures 3.9 to Note the presence of tones in all cases. The FCC spectral mask is not applicable here due to the presence of the tones. It is surmised that the tones are a result of the line coding process as opposed to the particular pulse shaping functions used. Even when no pulse shaping is used, the

61 44 tones are still observed as is shown in Figure 3.12, which is the PSD of the 2-bit buffered system when a(t) is chosen to equal the unmodified rectangular Barker waveform bit buffer signal PSD Power in dbm Frequency in Hz x 10 7 Figure 3.6. Simulated PSD of sinusoidal pulse shaping with 2 bits buffered system bit buffer signal PSD Power in dbm Frequency in Hz x 10 7 Figure 3.7. Simulated PSD of sinc m pulse shaping with 2 bits buffered system

62 bit buffer signal PSD Power in dbm Frequency in Hz x 10 7 Figure 3.8. Simulated PSD of logarithmic pulse shaping with 2 bits buffered system Figure 3.9. PSD of experimental sinusoidal pulse shaping with 2 bits buffered system

63 46 Figure PSD of experimental logarithmic pulse shaping Figure PSD of experimental rectangular pulse shaping with 2 bits buffered system

64 bit buffer signal PSD Power in dbm Frequency in Hz x 10 7 Figure Simulated PSD of rectangular 2 bits buffered Barker system BER Measurements. Although this symbol shaping research using 2 buffered bits is not practically viable, the performance of the system was still studied to see if it is viable in alternate communication systems. The system was simulated using the line code and the various pulse shapes. Data was transmitted using the simulated communication system over an AWGN channel at various SNR levels. Correlation detection was used in the simulated receiver side and a BER value obtained. No filters were used in the simulations, as the spectral mask could not be applied to these systems due to the presence of tones in the PSD plots. The results are shown in Table 3.1. The results are poorer than the systems described in Chapter 2. Therefore, this avenue of research was not met with adequate success. Nevertheless, knowledge was gained in alternative symbol shaping techniques and line coding.

65 48 Table Simulated BER measurements for 2 bits buffered Barker spread system. Pulse Shape Bit Error Rate at SNR levels: Used 4.5 db 4 db 3 db Rectangular 0.40E E E-04 Logarithmic 2.80E E E-04 Sinusoidal 2.96E E E-04 Sinc-function 2.46E E E Buffering 3 Bits In this system sinusoidal pulse shaping was used as the results from Chapter 2 and Section showed that sideband attenuation was best achieved by sinusoidal shaping. A line code was utilized that buffered 3 bits: the previously transmitted bit, the current bit to be transmitted, and the next bit. Based on this set of 3 bits, one of 8 possible symbol states is selected for transmission such that all discontinuities are eliminated. A modified Barker sequence was used for the symbol shaping, B m ( ), as this provides the smoothest possible bit possible transitions. Symbols 1 through 4 transmit the +1 bit and were based on +B m, while symbols 5 through 8 transmit the -1 bit and followed B m. The difference between symbol 1 and symbol 2/3/4 is in how the symbol s first 3 and last 3 chips are shaped. Thus, how the bit begins and ends is different such that there is a smooth transition. However, Barker sequence s autocorrelation function is no longer preserved as shown in Figure 3.13.

66 49 Cross correlation state1 to state Cross correlation state1 to state Cross correlation state1 to state Cross correlation state1 to state Figure Cross-correlation between state 1 and other states The principle idea behind this line code is that based on the previous and the next state symbols, the current symbol can be chosen such that there is no discontinuity at the beginning and end of the current bit interval. Thus, the discontinuities such as those seen in Figure 2.2 that are characteristic of the non-buffered Barker spread wireless signal are completely eliminated by this 3 bits buffered system. As a result, the sidebands in the PSD are expected to be attenuated since the high frequency components that arise as a result of any signal discontinuity are removed. The 8 state symbols used in this study are shown in Figure 3.14.

67 50 2 Plot of bit +1; state 1 2 Plot of bit +1; state Time in s 1 x 10-6 Plot of bit +1; state Time in s x 10-6 Plot of bit -1; state Time in s x 10-6 Plot of bit -1; state Time in s x Time in s 1 x 10-6 Plot of bit +1; state Time in s x 10-6 Plot of bit -1; state Time in s x 10-6 Plot of bit -1; state Time in s x 10-6 Figure Symbols for the 8 state 3 bits buffered system The 3 bit sequence used in the line code can be represented as d -1, d 0, d 1, where d -1 is the previous bit, d 0 is current bit to be sent, and d 1 is the next bit. Table 3.2 shows the symbol mapping employed by the line code based on this 3 bit sequence and Figure 3.15 shows the state transition diagram. Applying this map to Figure 3.14, it is possible to

68 51 check that there is always a smooth signal transition from the previous bit to the present bit, and then onto the next. Table Symbol mapping table for 3 bits buffered Barker spread system. Bit stream State symbol transmitted d -1 d 0 d State # State # State # State # State # State # State # State #8 11 State 1 10 State State 3 State State 4 10 State State 7 11 State Figure State transition diagram for 3 bits buffered Barker system

69 52 This line coded system using sinusoidal pulse shaping system was simulated to obtain the PSD shown in Figure For result verification, the experimentally generated PSD for this system is shown in Figure The PSD of this system shows the best spectral characteristics observed while dealing with Barker spread signals. The sidebands were more attenuated than any of the other systems investigated in this study. Table 3.3 shows the sideband attenuation achieved in comparison with the unmodified rectangular Barker spread signal. As a result, a simple second order filter with a cutoff frequency of 10 MHz is all that is needed to satisfy the spectral mask. Table Comparison of PSD sideband attenuations (unfiltered 1 Mbps data signals) Pulse Shape Used Second Lobe drop Third Lobe drop Rectangular (No Buffer) Sine-shaping (3 bits buffered) 13.1 db 17.2 db 26.0 db 38.7 db -30 Unfiltered signal PSD Power in dbm Frequency in Hz x 10 7 Figure PSD of 3 bits buffered system with sinusoidal shaping

70 53 Figure PSD of experimental 3 bits buffered system with sinusoidal shaping The performance of this 3 bits buffered system was studied to see its BER versus SNR performance. The system was simulated using random binary data, and the signal was transmitted over an AWGN channel at various SNR levels. Correlation detection using four correlators (states 1 to 4) was used in the simulated receiver side, and the BER value was recorded. The results are shown in Table 3.4. The results show improved performance compared to the control case (unmodified IEEE Mbps Barker spread system). However, compared to the systems described in Chapter 2, the improvement is less significant as the correlation function (Figure 3.13) is less than ideal. Table Simulated BER measurements for 3 bits buffered Barker spread system. Pulse Shape Used Rectangular (No Buffer) Filter Order Bit Error Rate at SNR levels: 11.5 db 11 db 10 db E E E-04 Sinusoidal E E E-04

71 54 CHAPTER 4 PULSE SHAPING FOR IEEE 5.5 MBPS CCK SIGNAL 4.1 CCK Pulse Shaping Methodology In Section 1.6, we observed how the CCK time domain signal can be represented. To recap, the transmitted signal, v c (t) can be represented as ( 2π f t + α + w( n, + θ ) vc ( t) = 2 cos c n k), (4.1) where α n is the differential phase for the n th data symbol (from Table 1.1), w( n, k) is the phase of the n th data symbol s k th chip determined from Table 1.2, θ is an arbitrary phase angle, and f c is the carrier frequency of the signal. The 5.5 Mbps b signal in (4.1) can be expressed as ( π θ) ( π θ) v () t = a cos 2 f t + w( n, k) + + a sin 2 f t + w( n, k) +, (4.2) c I c Q c where, a = 2cos( α ), (4.3a) I and, a = 2sin( α ). (4.3b) Q The implementation of the CCK spread signal uses α n {±π/4, ±3π/4} resulting in binary (±1) sequences for a I and a Q at the chip rate. The signal in (4.2) can be rewritten as n n ( π θ) ( π θ) v () t = x( n, k)cos 2 f t+ + y( n, k)sin 2 f t+, (4.4) c c c where, x( nk, ) ai cos ( wnk (, )) aqsin ( wnk (, )) = +, (4.5a) and, ynk (, ) ai sin ( wnk (, )) aqcos ( wnk (, )) = +, (4.5b) with the index n representing the symbol number and the index k representing its kth chip.

72 55 In this development, there are a total of 16 x vectors and 16 y vectors, corresponding to the 16 c vectors for the CCK code used in 5.5 Mbps signal. The group of 16 y vectors is exactly the same as the 16 x vectors but in different order. It is important to note that this set of 16 vectors contains 8 groups with 2 identical vectors each. Additionally, it was observed that for any vector C, there exists the negative of that vector in another group. Therefore, we see that there are only 4 possible vectors C. Table 4.1 shows the 4 possible vectors C. Table Four possible vectors C. Chip # vector vector vector vector Furthermore, we notice that vector 2 can be obtained by simply reversing vector 3 and vice versa. Similarly, vector 1 can be obtained by reversing vector 4. Thus, there are only two truly unique set of vectors in the 5.5 Mbps system, that is vectors 1 and 2. For pulse shaping using the 5.5 Mbps system, these two vectors 1 and 2 only need to be designed, thereby simplifying the effort considerably. Several shaping functions like sinusoidal and sinc m were used to shape the chips in vectors 1 and 2. The resulting pulse shaped symbols still adhered to general shape of the CCK code vectors. Then the two symbol vectors 1 and 2 were reversed to obtain the pulse shaped symbols for vectors 3 and 4, completing the symbol set for vector C. Then from C, the complete symbol set was obtained for all possible x and y vectors. For each type of pulse shaping that was performed for vectors 1 and 2 as described above, the communication system was simulated. Binary random data at a rate

73 56 of 5.5 Mbps was spread using the pulse shaped CCK symbols and the Welch method was used to obtain the simulated PSD. The PSDs of the pulse shaped CCK systems were compared to the PSD of the unmodified CCK system shown in Figure 1.7. Of particular interest was the improvement in sideband attenuation. The pulse shaped CCK signals showed marked improvement in spectral characteristics. The PSDs of the pulse shaped CCK signals were also examined experimentally. For the experimental PSD, however, a 5.5 Mbps data rate could not be used. Due to hardware speed limitations, a 4 MHz chip-rate was used corresponding to a bit-rate of 2 Mbps with the 8 chip CCK code. Using 4 Mchips/second, the main-lobe bandwidth in the experimental PSD comes out as 4 MHz. The experimental PSDs were compared to the simulated ones to check if the results were in agreement. The spectral mask could not be completely satisfied by symbol shaping alone and filtering was still required. However, for the novel signals, only low order filters are necessary to achieve the spectral mask and, thus, ISI is considerably reduced. In order to examine the performance of the communication systems, BER simulation studies were performed for each of the pulse shaped CCK systems in MATLAB. In these studies, random binary data was CCK spread using the shape symbol vectors to obtain a simulated transmit signal. A minimum order IIR filter was used to filter the baseband information signal such that the PSD satisfied the spectral mask without introducing excess ISI. AWGN noise was added to simulate the channel and the SNR was recorded. After transmission through this simulated AWGN channel, a synchronous correlation was used to decode the received bits. A total of eight correlators were needed: one each for the four possible I and Q phase vectors. The decoded bits

74 57 were compared to the transmitted bits to obtain a BER value for the channel at the recorded SNR level. For purposes of easy comparison, the PSD of the 5.5 Mbps signal spread using the unmodified rectangular chip CCK is repeated in Figure 4.1. A fourth order Butterworth filter with a 9.0 MHz cutoff frequency is required to make this signal meet the spectral mask requirement. The symbols representing the four vectors C are shown in Figure PSD plot -40 Power in dbm Frequency in Hz x 10 7 Figure 4.1. Simulated PSD of 5.5 Mbps CCK signal with no pulse shaping 1 Plot of vector 1 1 Plot of vector Time in s 6 x 10-7 Plot of vector Time in s 6 x 10-7 Plot of vector Time in s 6 x Time in s 6 x 10-7 Figure 4.2. Unmodified CCK symbols

75 Sinusoidal Pulse Shaping Sinusoidal functions were used to shape the chips in the CCK vectors. The sinusoidally symbol shaped vectors are displayed in Figure 4.3. Comparing with Figure 4.2, notice that the pulse shaped CCK vector waveforms still follow the CCK vectors chip values. The simulated PSD obtained by this shaping method is shown in Figure 4.4. The experimentally obtained PSD is shown in Figure 4.5. An improvement of 10 db is obtained for attenuation of the first sideband, while the second sideband is attenuated by 15 db more than the system without any pulse shaping. As a consequence, the spectral mask is met by using a second order Butterworth filter with a 8.75 MHz cutoff frequency. Plot of vector 1 Plot of vector Time in s 6 x 10-7 Plot of vector Time in s x Time in s 6 x 10-7 Plot of vector Time in s x 10-7 Figure 4.3. CCK symbols shaped by sinusoidal functions

76 59-30 PSD plot -40 Power in dbm Frequency in Hz x 10 7 Figure 4.4. Simulated PSD for CCK symbols shaped by sinusoidal functions Figure 4.5. Experimentally obtained PSD for CCK symbols shaped by sinusoidal functions 4.3 Sinc m Pulse Shaping Functions of the form sinc m were used to shape the chips in the CCK vectors. The symbol shaped vectors designed in such away are displayed in Figure 4.6. The simulated

77 60 PSD obtained by this shaping method is shown in Figure 4.7, and the experimentally obtained PSD in Figure 4.8. An improvement of 7 db is obtained for attenuation of the first sideband, while the second sideband is attenuated by 7 db more than the system without any pulse shaping. Filtering by a third order lowpass filter (cutoff frequency 9.5 MHz) is necessary here also. With sinc m pulse shaping for the CCK spread 5.5 Mbps signal, the spectral improvements are not as good compared to the application of sinc m for shaping the Barker spread 1 Mbps signal. Nevertheless, spectral improvement is observed. Plot of vector 1 Plot of vector Time in s 6 x 10-7 Plot of vector Time in s x Time in s 6 x 10-7 Plot of vector Time in s x 10-7 Figure 4.6. CCK symbols shaped by sinc m functions

78 61-30 PSD plot -40 Power in dbm Frequency in Hz x 10 7 Figure 4.7. Simulated PSD for CCK symbols shaped by sinc m functions Figure 4.8. Experimentally obtained PSD for CCK symbols shaped by sinc m functions 4.4 BER Measurements The BER vs SNR simulation study for the CCK signals was mentioned in Section 4.1. Each of the pulse shaped systems, including the unmodified rectangularly

79 62 shaped CCK waveform was subjected to this study using 200,000 random bits in order to test the system performance. The results of the BER versus SNR studies are shown in Figure 4.9. The specifications for the Butterworth filter used to satisfy the spectral mask are included in the table. The filter order is indicative of the amount of ISI introduced, and the system performance is indicated by the BER values. Figure 4.9. Simulated BER vs SNR measurements for CCK symbol shaping The simulated BER results show that the sinc m function pulse shaped 5.5 Mbps CCK system has lesser ISI and performs about 1 db better (in terms of BER versus SNR) than the rectangular system. Compared to the Barker spread signal, there is somewhat greater improvement (about 0.5dB) in the system performance for the CCK spread signal through pulse shaping. The Barker signal is BPSK, but the CCK spread signal is QPSK. By filtering any signal, ISI occurs in both the I and Q phases. However, for the Barker spread system this is no issue as it is a BPSK signal and Q phase ISI has no effect. However, for the QPSK CCK signals, this effect has a greater impact leading to signal

80 63 distortion. Since filtering adversely affects CCK signals more, reducing the spectral mask filter s order provides increased gain in performance for CCK spread. The spectral improvements for using the pulse shaped CCK systems are summarized in Table 4.2. The table shows the attenuation in peak lobe powers. Additionally, Table 4.3 shows the total amount of sideband energy leakage in the PSD of the various signals. For lower sideband powers, there is lesser interference caused to nearby Wi-Fi channels. The composite PSD plot showing the PSD of the different CCK systems is shown in Figure Notice that the sinusoidal system performed the best both in terms of spectral improvement and in terms of the better system performance with respect to the BER vs SNR study. The rectangular unmodified 5.5 Mbps CCK signal performed worst in both aspects. Thus, the results show that the IEEE Mbps CCK signal can be considerably improved by symbol shaping. Table Comparison of PSD peak sideband drops (unfiltered 5.5 Mbps data signals) Pulse Shape Used Second Lobe drop Third Lobe drop Rectangular 13.3 db 17.4 db Sinusoidal 23.0 db 32.9 db Sinc-function 19.8 db 24.2 db Table Comparison of total power in each band (unfiltered 5.5 Mbps CCK signals) Pulse Shape Main Lobe Power % Second Lobe power % Third Lobe power % Second Lobe drop (db) Third Lobe drop (db) Rectangular Sinusoidal Sinc-function

81 64 Table 4.4 quantifies the amount of ISI occurring in each system, where the metric is the amount of energy in one symbol that leaks into the next symbol interval. We notice that the rectangular system has almost 5% energy leakage due to ISI compared to only 2.4% for the sinc case. Table ISI after filtering operation for CCK symvol shaping Pulse Shape Power within Bit Interval (%) Power leakage outside Bit Interval (%) Rectangular Sinusoidal Sinc-function Rectangular pulses Sinc m pulses Sinusoidal pulses Figure Experimentally obtained composite PSD plots for CCK symbol shaping

82 65 CHAPTER 5 EXPERIMENTAL STUDY OF MICROWAVE OVEN SIGNAL 5.1 Main Features of MWO Signal As mentioned in Chapter 1, the 2.4 GHz band is dominated by high-speed data communications and Wi-Fi. Access points, wireless laptops, Personal Digital Assistants, Bluetooth [GUI04] devices, and cordless phones [BAT01] all intentionally operate in this band for the purpose of communicating. On the other hand, various commercial devices not intended for Wi-Fi communications, such as microwave ovens and other residential and industrial products, also radiate in the 2.4 GHz band. The emitted electromagnetic RF signals they produce act as interference to Wi-Fi users. The composite interference from intentional Wi-Fi transceivers and unintentional emitters results in reduced network performance, and even connectivity loss. The microwave oven is one of the most common unintentional interference device [KAM97]. There are two types of MWOs: residential and commercial MWOs. The residential MWO contains a single magnetron that periodically turns on and off as the 60 Hz AC line voltage changes from positive to negative [GAW94]. Thus, the MWO signal goes through ON and OFF cycles characterized by the respective presence and absence of RF radiation. A commercial MWO has two magnetrons that operate 180 degrees out of phase such that energy is always radiated into the MWO cavity. RF energy leaking from the MWO cavity causes interference in the 2.4 GHz ISM band. In this research project, the interference signal from the residential MWO was studied in detail [TAH06].

83 66 In this chapter, an overview of MWO operation is provided, as well as experimental signal characteristics. In particular, we explore the frequency-sweeping phenomenon of the MWO signal, the envelope of the MWO signal in the time domain, and the transient signals that exist in the MWO signal but have been often overlooked in prior MWO studies. 5.2 FM Signal The residential MWO signal, in the ON mode, is similar to a Frequency Modulated (FM) signal [PRO94], with a fixed carrier frequency, and an instantaneous frequency that changes with time. The MWO center frequency varies with the manufacturer and model, but for the models tested, it was in the 2.45 GHz range. The MWO signal is repetitive in nature with a period of ms, which is the inverse of the 60 Hz frequency of the AC supply line powering the MWO. However, the frequencysweep in the MWO signal is less than half of the 60 Hz time period, typically 5-6 ms. This is shown in the spectrogram in Figure 5.1, where the sweep of the MWO signal is clearly seen. Figure 5.2 repeats the same spectrogram image but without any markings for visual clarity. This figure also shows transients before and after the frequency-sweep. The spectrogram is particularly useful in developing a model for MWO emissions because it experimentally reveals the characteristics of the frequency-sweeping and transient aspects of the MWO signal. All the spectrogram plots were obtained using the ComBlock receiver described in Section 1.7. The spectrogram s bandwidth is 20 MHz. This is because the digital sampling rate of the ComBlock is 40 MHz, and, according to the Nyquist criterion [PRO96], a maximum 20 MHz bandwidth of spectral information can be obtained from this digitally sampled data. MATLAB s spectrogram function

84 67 [MAT07] was used to obtain the plots from the data. The function compartmentalizes the signal versus time data, and for each such subset of data, it estimates the frequency spectrum using algorithms based on the fast Fourier transform [PRO96]. During the frequency-sweeping part of the ON cycle, the radiated signal does not behave like conventional FM where the power level is constant. However, the signal can be characterized as an FM signal with varying power levels. The latter property lends itself to an Amplitude Modulated (AM) mode [PRO94]. Thus, a combined AM-FM waveform will serve as a basis for the frequency-sweeping part of the signal [TAH06]. The approximate sinusoidal shape in Figure 5.1 represents the FM signal that sweeps the spectrum over 15 MHz for approximately one half of the 60 Hz AC cycle. A thorough investigation for the amplitude of the MWO signal is detailed in Section 5.3. AM-FM Signal A B Transients Figure 5.1. Spectrogram of MWO signal with key features labeled

85 68 Figure 5.2. Clean Spectrogram of MWO signal 5.3 Amplitude Variation The envelope of the MWO signal varies significantly during the ON cycle. To study the characteristics of the actual envelope of the MWO signal, measurements were carried out in the WIL. The Zero-Span Mode (ZSM) of a Rohde & Schwarz Spectrum Analyzer (model no. FSP 38) was used to capture the envelope of the RF MWO signal. The spectrum analyzer s Resolution Bandwidth (RBW) was set to 10 MHz and the center frequency to GHz. The time domain MWO signal captured by the spectrum analyzer is shown in Figure 5.3. Observe that the oven is on about half of the 60 Hz cycle. The amplitude of the MWO signal can be approximated by a sinusoidal waveform when the microwave oven is on. Careful observation of Figure 5.1 also gives support of this approximation. The increase in shading indicates that the power of the AM-FM signal increases during the ON cycle and then decreases as it approaches the OFF cycle; the power depends on the amplitude and, hence, the amplitude change is also observable in the spectrogram.

86 69 Figure 5.3. The envelope of the MWO signal over two 60 Hz cycles (3.33 ms/div) It is important to notice the transient signals at the beginning and end of each ON cycle in Figure 5.3. These transient signals, together with the frequency-sweeping signal, comprise the radiated MWO signal. The transient signals are studied in detail in Section Transients The transient part of the MWO was observed in Figures 5.1 to 5.3. In each period of the MWO signal, there are two transient signals, one occurring at the beginning and another occurring at the end of the ON cycle of the MWO. The characteristics of the transient signals in the time and frequency domains are further investigated here. Numerous ZSM measurements were taken to estimate the bandwidth of the transient signal as well as its duty cycle. The ZSM captures were obtained at different frequencies across the ISM band, using a narrow resolution bandwidth of 10 khz. If a periodic transient signal was detected at that ZSM center frequency, then its power and duty cycle were measured.

87 70 To synchronize all the ZSM captures of the transient signal at different frequencies, a 60 Hz line trigger was used. With this experimental setup, the zero-span captures of the transient signals at different frequencies are aligned. This is illustrated in Figure 5.4 where the periodic transient signals are at the same time locations, even though the capturing frequencies are different (2.46 GHz and 2.44 GHz). Observe that the width of these transient signals is approximately 1 ms at both the frequencies in Figure 5.4, with the turn-on transient slightly longer than the turn-off transient. Transients Turn-on Turn-off Turn-on Turn-off Transients Figure 5.4. Zero-span measurements at 2.46 GHz and 2.44 GHz over two 60 Hz cycles (3.33 ms/div) A programmed spectrum analyzer captured a series of ZSM measurements at uniformly spaced frequencies over the 85 MHz ISM band to estimate the bandwidth of the turn-on and turn-off transients. Measuring the periodic time-varying power signatures of the transient signal over the 2.4 GHz ISM band and combining all the zerospan captures, an experimental spectrogram was generated showing the transient signals. The contour plot of the spectrogram is shown in Figure 5.5 over one power cycle with the

88 71 low-level noise suppressed. The transient signals are broadband, extending over 60 MHz in bandwidth. Also, the power of the transient signals is concentrated at frequencies where the sweeping part of the MWO signal meets the transient part in the spectrogram plot (see points A and B in Figure 5.1). The spectrogram obtained using this method has a bandwidth of 80 MHz, that is, four times the ComBlock spectrogram s bandwidth. Hence, this ZSM based spectrogram method was utilized to study the very wideband MWO transient signals. The limitation of this ZSM based spectrogram method is that it can be used to obtain power versus frequency and time plots only for periodic signals. The advantage of this method is that large bandwidth spectrograms are obtainable. It is useful to understand why the transients occur in the MWO. The MWO magnetron needs a minimum threshold voltage (Volt A, Figure 5.6) to operate, i.e., to emit microwave energy. Since Volt A is positive, the time duration for the ON cycle is less than that of the OFF cycle. This is observable in Figures 5.1 to 5.3. Figure 5.5. An experimental spectrogram for transient signals over one 60 Hz cycle (2 transient signals)

89 72 Transients Threshold Volt B Threshold Volt A ON OFF Time (s) Figure 5.6. MWO signal generation process The minimum threshold voltage (Volt A) is inadequate for sustained operation of the magnetron. A second threshold (Volt B > Volt A) is required for the MWO to generate a frequency-sweeping signal. Between the two thresholds, the MWO emits wideband transient pulses. The transient areas are shown shaded in Figure 5.6. The threshold values and the transient times are manufacturer dependant, with a nominal transient duration of 1 ms. Obviously, transients are periodic and synchronized to the AC line signal. 5.5 MWO PSD Figure 5.7 shows the PSD of an actual MWO, experimentally measured in the WIL. The maximum power is concentrated at the higher frequencies, that is near the frequency-swept region shown in Figure 5.1 (points A and B). The power in the lower frequencies comes from the transients shown in the spectrogram of Figure 5.5 and is

90 73 25 db weaker in strength. Similar characteristics were observed for other MWOs (Figures 5.8 to 5.10) whose spectra were measured. Figure 5.7. Experimental PSD for MWO 1 (center 2.42 GHz, 12 MHz / division) Figure 5.8. Experimental PSD for MWO 2 (center 2.45 GHz, 10 MHz / division)

91 74 Figure 5.9. Experimental PSD for MWO 3 (center 2.45 GHz, 10 MHz / division) Figure Experimental PSD for MWO 4 (center 2.43 GHz, 10 MHz / division)

92 75 CHAPTER 6 MODEL OF MICROWAVE OVEN SIGNAL 6.1 Necessity of MWO Model An analytical model is highly useful in wireless network simulation studies. For example, simulations that study wireless network throughput and performance must account for RF interference from other radiating sources. If this interference is a microwave oven, a model of the device becomes necessary [TRA04]. In this chapter, two MWO analytical models have been developed. A good analytical model can be utilized in wireless network simulation as one of the wireless interferers operating in the simulated physical layer [JER92]. Also, a proper model allows better understanding of the RF signal from a MWO, which is important in understanding the nature of wireless interference caused by MWO and in developing interference mitigation techniques. Indeed, the model is used in Chapter 7 to develop an interference mitigation technique. 6.2 MWO Model #1 From the experimental data and analysis presented in Chapter 5, an analytical model was developed [TAH06]. The MWO signal can be expressed as the sum of two wideband transient signals and a frequency-swept signal during the ON cycle, and zero during the OFF cycle. The frequency-swept signal is modeled as an AM-FM signal. Based on the shapes of the MWO signals in Figures 5.1 and 5.3, the frequency-swept signal, s(t), is modeled as a sinusoidally modulated FM signal with a sinusoidally shaped amplitude, x(t). Here, both the modulations are at the 60 Hz line frequency.

93 76 The 1 ms (approximate) transient signal pulse was modeled as the sum of two sinc waveforms modulated at different carrier frequencies. The two sinc pulses also have different main lobe widths in the time domain and thus different bandwidths in the frequency domain. One sinc waveform has a wide spectral bandwidth to provide power across the entire ISM band, while the other sinc waveform has a narrower bandwidth with power concentrated in the frequency-swept band. The transient bandwidths are MHz and the main lobe of each sinc waveform is in the order of nano-seconds. Figure 6.1 shows a qualitative plot of the time domain locations of these signals for each ON cycle. The two transient signals are centered in each of these locations The frequency sweeping FM and AM modulated signal. The cosine shape shows the AM modulating envelope used. Time (ms) Figure 6.1. Qualitative representation of MWO signal model The complete MWO signal, v(t), can be expressed as the sum of ON cycle waveshapes, c(t), that is, vt () = ct ( nt), (6.1) n= where T = 1/f ac and f ac = 60 Hz. Using the structure shown in Figure 6.1 and the signal description above, the ON cycle wave-shape can be written as,

94 77 ct () = Apt ( + t; b)cos(2 π ft) 1 a A pt ( + t; b)cos(2 π ft) + st () 2 a A pt ( t; b)cos(2 π ft) 1 a A pt ( t; b)cos(2 π ft), 2 a 2 2 (6.2) where the pulse waveform, p(t), is, ptb (; ) = sinc( bt), t < 0.5 T p, (6.3) The power in the transient pulses is dictated by the amplitudes, A 1 and A 2, and the center of their spectra is determined by the carrier frequencies, f 1 and f 2. The time locations of the transient pulses are at ± t a and their duration is T p. The bandwidths of the two transients are determined by b 1 and b 2. The AM-FM signal, with sinusoidal modulation, can be written as, ( π β π ) st () = Axt ()cos 2 ft+ sin(2 f t), t < 0.5 T, (6.4) where the amplitude variation is given by, c ac s x() t = cos(2 π f t). (6.5) ac The power in s(t) is dictated by the amplitude A and the sweep time, T s. The peak frequency deviation is determined by the modulation index, β, while T s and β determine the frequency-swept band. The center frequency of the magnetron is given by f c. Using the model, any MWO signal can be represented by a total of 12 parameters. It is, of course, possible to refine the model with different pulse widths for the turn-on and turn-off transients, non-symmetric pulse locations, and other pulse shapes. However, the 12 parameter model, when simulated, provides reasonable agreement to experimental measurements as detailed in the next section.

95 MWO Model #1 Simulation The analytical model presented in the previous section was simulated using MATLAB. The simulations were carried out in the MHz and khz ranges for computational convenience. Our simulations have shown that this analytical model is scalable to all frequencies and bandwidths as the general characteristics of the PSD and the spectrogram are preserved. Figure 6.2 shows the Welch PSD estimate for one simulation run, and Figure 6.3 shows its spectrogram. Here, the FM carrier frequency was set to 1 MHz and the FM sweep bandwidth was set to 0.05 MHz. In this simulation, the transient bandwidths are each 0.05 MHz, and the transient carrier frequencies, f 1 and f 2, are chosen so that the combined transient spectra span a 0.1 MHz range. Frequency (MHz) Figure 6.2. Simulated PSD of the MWO signal (carrier frequency in 1 MHz range)

96 79 Frequency (MHz) Figure 6.3. Simulated spectrogram MWO signal (carrier frequency in 1 MHz range) In a second simulation, the FM carrier frequency was set to 100 khz, the sweep bandwidth was fixed at 10 khz and the total transient bandwidth was set at 20 khz. The PSD and the spectrogram are displayed in Figures 6.4 and 6.5, respectively. In this case, the lower power wide transient bandwidth was 20 khz and the higher power narrow transient bandwidth was 10 khz. Here, f 2 was set so that the narrow transient signal s spectrum overlapped with frequency-swept band. Frequency (khz) Figure 6.4. Simulated PSD of the MWO signal (carrier frequency in 100 khz range)

97 80 Frequency (khz) Figure 6.5. Simulated spectrogram of MWO signal (carrier frequency in 100 khz range) We see that the simulation results, for the analytical model introduced in Section 6.1, capture the main features of the actual MWO PSD. The maximum power is concentrated at the higher frequencies in the frequency-swept region. The power in the lower frequencies comes from the transients and is 25 db weaker in strength. This reduced PSD level is seen both in experimental and simulated results. The simulated spectrograms in Figures 6.4 and 6.5 compare quite well with the experimental spectrogram shown in Figure 5.1. However, greater similarity between the simulated model and experimental measurements from actual MWO devices is desired. 6.4 Drawbacks of Model #1 The model developed in Section 6.2 (model #1) has three major drawbacks. These limitations with model #1 necessitate modifications in order to obtain an analytical model that better captures the characteristics of MWO ovens in general. The first problem with model #1 is that for a bandwidth of 50 MHz, the transient durations come out to be in the order of nanoseconds as opposed to milliseconds. This is

98 81 because for the sinc pulse used in the model, the time domain duration of the main lobe of the pulse is inversely proportional to the signal s frequency domain bandwidth. Thus, for a signal bandwidth in the MHz range, the transient duration comes out only in the order of hundreds of nanoseconds. However, we know from Chapter 5 that the transients last for about a millisecond: a discrepancy of fourth order magnitude between the model and actual MWO signals. Second, the FM carrier frequency of a MWO is not constant but varies. This is dramatically illustrated by the spectrogram of an old 1980s MWO in Figure 6.6. Although the newer MWO devices do not have such highly fluctuating characteristics, the carrier frequencies are not stationery either. Careful observation of the spectrogram in Figure 5.2 reveals that the AM-FM signal carriers on adjacent ON cycles differ. Frequency (Hz) Time(s) Figure 6.6. Experimental spectrogram of an older MWO Finally, we notice from Figures 5.2 and 5.5 that the transient power PSD is not flat. However, model #1 treated the transient power PSD essentially as flat PSD, albeit

99 82 with two discrete power levels. From Figures 5.2 and 5.5, the varying transient power level can be approximated by means a bell curve, but with a short tail on the high frequency curve. A second analytical model, MWO model #2 was thus developed [TAH08a] in order to address these issues. This is described in the following sections. 6.5 MWO Model #2 In model #2, the transient duration problem was corrected and the carrier frequency from one ON cycle to the next was made random. The transients were formulated as a sum of sinc pulses modulated at uniformly spaced frequencies, where the sinc pulse power was a function of the frequency following a modified Rayleigh distribution shown by the left sub-plot in Figure 6.7. If in the frequency and power axes, this distribution (left sub-plot) is compared with the transient spectrogram (right sub-plot in Figure 6.8), satisfactory correlation is obtained. Transient Power vs frequency in model Normalized Amplitude Frequency (GHz) x 10 9 Time (ms) Frequency (GHz) Figure 6.7. Remodeling the transients. Left sub-plot shows the plot of function used to control the frequency domain power of the transients. Right sub-plot shows experimentally measured transient powers for an MWO.

100 83 Based on these three modifications, analytical model #2 of the MWO signal was developed as a derivative of the earlier model #1. During each period, the signal can be expressed as a sum of two transients and an AM-FM signal to represent the frequency swept signal. As in the previous model #1, the modeled AM-FM signal, s(t), consists of a sinusoidally modulated FM signal with a sinusoidally shaped amplitude, x(t). The AM and FM modulations are both sinusoidal in nature at the 60 Hz line frequency. The large bandwidth of the transient signals was modeled as the sum of sinc pulses modulated at different subcarrier frequencies. Figure 6.8 shows a qualitative plot of the time domain locations of these signals for each ON cycle. The two transient signals are centered in these locations T s T P The frequency swept AM-FM modulated signal. T P t d t d Time (ms) Figure 6.8. Qualitative representation of MWO #2 signal model The complete MWO signal, v(t), can be expressed as the sum of ON cycle waveshapes, c(t), that is, vt () = ct ( nt), (6.6) n= where T = 1/f ac and f ac = 60 Hz. Using the structure shown in Figure 6.8 and the signal description above, the ON cycle wave-shape can be written as

101 84 n= 1 ( ) ( ) ( π ) ct () = E f p t t cos 2 ft E( fn) p( t td ) cos( 2π fnt) (6.7) n= 1 + st (), where the transient pulse waveform is given by N N + + n d n ( λ ) pt () = sinc bt ( + ), t < 0.5 T, (6.8) with b a bandwidth parameter (usually 4 khz), T P the width of the transient pulse centered at ±t d, and λ n a random variable uniformly distributed over ± 0.5T p to provide a time offset for each sinc pulse in the transient signal summation. The transient signal is the sum of N sinc pulses modulated by subcarriers, f n, uniformly spaced from f 1 to f N. Here, f 1 and f N are the minimum and maximum values of f n, respectively, such that (N 1)b = f N f 1. The energy in each sinc pulse is determined by the function E( f n ). Several curve fitting functions were tested for E( f n ) but best results were obtained with a modified Rayleigh function [RAP02] defined as 2 h n 2 fn 2 h ( fn ) 2 f ( fn fn) E( fn) = EO e, (6.9) f where f = f f, (6.10) h N pk E O is an amplitude scale factor, and f pk is the subcarrier frequency with the maximum transient energy. The AM-FM signal, with sinusoidal modulation, can be written as where the amplitude variation is given by ( π β π ) st () = Axt ()cos 2 Ft+ sin(2 f t), t < 0.5 T; (6.11) c ac s p

102 85 x() t = cos(2 π f t), (6.12) and the power in s(t) is dictated by the amplitude, A, with the sweep time given by T s. The peak frequency deviation is determined by the modulation index, β. The carrier frequency of the AM-FM signal is a random variable, F c, that is uniformly distributed between frequencies f a and f b. During any given period, F c is fixed, but it varies from one ON cycle to the next. The operating range of F c, that is f b f a, is typically 5 MHz. Using the model, any MWO signal can be represented by appropriately choosing a set of 13 independent parameters. This model, when simulated and emulated, provides very good agreement to experimental measurements as detailed in the next section. ac 6.6 MWO Model #2 Simulation The model described in the previous section was studied by experimentation and via simulation to examine its accuracy. The model #2 described was simulated in MATLAB software. Simulations were performed in the megahertz range for computational convenience. Simulations at higher and lower frequency ranges have shown that the model is scalable to all frequencies and bandwidths without altering the general signal characteristics. Figure 6.9 shows a spectrogram obtained using the simulated model. Figure 6.10 shows the PSD computed over 100 cycles. The parameters were chosen such that the PSD in Figure 6.10 closely matched the characteristics of the MWO PSD shown in Figure 5.7. For computational feasibility, however, the MWO total bandwidth was limited in simulation to 1.5 MHz compared to the 60 MHz bandwidth of the experimental MWO in Figure 5.7.

103 86 Figure 6.9. Spectrogram of simulated MWO signal Figure 6.10 Simulated PSD of MWO signal 6.7 Model #2 Experimental Emulation To verify the simulation studies and to further validate the model, the MWO model was emulated experimentally such that the model parameters matched with a different MWO whose PSD is shown in Figure This experimentation was done as

104 87 part of the three pronged approach to check if the simulation results could be verified independently by emulation. For this purpose, a ComBlock transmitter unit operating in the 2.4 GHz range was used to emulate the MWO signal based on the model equations. Figure 6.11 shows the experimentally emulated spectrogram. Figure 6.12 is the PSD of this emulated signal obtained with a spectrum analyzer. Due to experimental limitations, the emulated MWO model s bandwidth was limited to 1.5 MHz as opposed to 50 MHz for the actual MWO PSD in Figure The simulation and emulation studies show that the model is a good approximation to the MWO signal. Furthermore, they demonstrate that the model s parameters are readily adjustable to approximately match the characteristics of different MWOs. Figure Spectrogram of emulated MWO #2 signal

105 88 Figure PSD of emulated MWO signal measured by spectrum analyzer Figure Experimental PSD of actual MWO

106 89 CHAPTER 7 MICROWAVE OVEN SIGNAL INTERFERENCE MITIGATION FOR IEEE SYSTEMS 7.1 Interference Mitigation Technique While the CSMA protocol is effective in Wi-Fi collision avoidance, the MWO is oblivious to this type of interference avoidance. Hence, an alternate interference mitigation mechanism needs to be developed. In this section, a technique that allows Wi-Fi devices to avoid interference caused by MWO signals is outlined. From Chapter 5, it was seen that the frequency-swept part of the MWO signal spans a relatively narrow bandwidth (approximately 15 MHz). The transient signal bandwidths are much larger (60 MHz or more) and they occupy the entire ISM band. Due to this relatively large bandwidth, the MWO affects data communications in all IEEE channels [AVA02] [INT98] [ZHA05]. However, the transient bursts are periodic. Thus, if the transient time locations are known, then MWO interference can be avoided by simply stopping data transmission during those time intervals. Consider the case in which an IEEE signal is being transmitted at channel 1, centered at GHz, with a main-lobe bandwidth of 22 MHz. The MWO frequency-swept spectrum does not impinge on channel 1, so we can freely operate in this channel during the OFF cycle and during the time interval when the MWO emits the AM-FM signal. Since the transient signals exist for only 2 ms out of the ms period, then in theory, 88% of the time, MWO interference can be avoided in channel 1. Now, consider the case when another common IEEE channel is being used, that is, channel 11, which is centered at GHz. The MWO frequency-swept

107 90 signal and the transient pulses both interfere with this channel. However, during the OFF cycle, that is 50-60% of the time, the MWO signal interference can be avoided while transmitting in channel 11. For each IEEE channel and MWO spectral signature, an effective interference mitigation paradigm can be formulated. For MWO interference mitigation, the Wi-Fi transmitter must be synchronized with the AC line signal. For Wi-Fi devices with AC power, the synchronization is relatively easy to achieve. Once synchronized, the position of the transient pulses can be estimated from the zero crossings of the AC voltage, the average duration of the transient pulses, and the average frequency-sweep time. For the Wi-Fi devices that are batterypowered, synchronization can be done by using the 60 Hz periodic transient bursts of the MWO signal that are detectable throughout the ISM band. To implement MWO emission mitigation, the Wi-Fi device simply requires a detector that uses the signature of the MWO interference signal to identify when a MWO is operating. The MWO model #2 developed in Chapter 6 can be used as a reference MWO signature that is compared to receive RF signals to detect when MWO interference is present. The MWO interferer is present when there is a 60 Hz periodic signal in the ISM band, synchronized with the AC line voltage. If the Wi-Fi device is using channel 11, it can switch to a channel outside the frequency-swept band, like channel 1, or employ a mitigation mode where it only transmits during the OFF cycle duration. Any Wi-Fi device operating on channel 1 (or on other channels outside the frequency-swept region) can transmit data in the manner shown in Figure 7.1, i.e., it transmits at all times other than when the transients are present. For data transmission in IEEE channel 11 can be achieved by the scheme shown in Figure 7.2. As part of this research project, the

108 91 interference mitigation technique illustrated by Figure 7.2 was implemented [TAH08b] using a cognitive radio [GOL05] circuit and is described in this chapter. Threshold Volt B DATA DATA DATA DATA Threshold Volt A Time (ms) Figure 7.1. Data transmission using channel 1 (shaded areas are transient locations) Experimental MWO #1 Spectrogram Transmit Data Packets during OFF cycles Data Data D at a Transients Frequency Sweep/AM-FM Figure 7.2. Spectrogram of MWO signal & interference mitigation 7.2 Circuit Design and Description Figure 7.3 shows the block diagram for the interference mitigation circuit developed in this research project. The mitigation principle is based on Figure 7.2. For

109 92 successful interference mitigation, it is necessary to detect the presence of MWO interference signals and synchronize the data transmitter with the MWO s ON-OFF cycles. The signature of a radiating MWO signal is detected and a Wi Fi transceiver is controlled to communicate only during the OFF cycles. The function of each system block is described in the following paragraph. Baseband Converter Threshold Detector 60 Hz AC Line Reference Baseband Logic Circuit y T (t) Transient Detector Transmit Controller (50 / 100 %) Figure 7.3. Block diagram for MWO interference mitigation system The 2.4 GHz ISM band signal received by the antenna is down-converted by the baseband converter in Figure 7.3. The threshold detector senses any received signal above the background noise threshold. The transient detector compares the threshold detector output, y T (t), with the AC line reference signal. If the timing of y T (t) matches the expected MWO transient time location, then the cognitive radio records the detection of a transient. The expected transient time locations are 2 ms time durations before or after the zero voltage crossings of the sinusoidal AC line reference. If the transient detector records the presence of several transient pulses over consecutive AC line cycles, the cognitive radio circuit concludes that a MWO interference signal is present. This smart radio system ignores all Wi-Fi signals and only triggers when a MWO signal is present.

110 93 If a MWO signal is present, the transmit controller instructs the Wi-Fi transmitter to synchronize with the AC line cycle and operate only during the MWO OFF cycles. If the MWO signal is not detected, the Wi-Fi transmitter is instructed by the transmit controller to operate normally. A circuit was constructed that successfully implements the interference mitigation function [TAH08b]. The circuit was constructed and was tested to work properly. Figure 7.4 is a photograph of the baseband logic circuit. Figure 7.5 shows its PSpice [PSP07] circuit diagram. Figure 7.4. Photograph of Interference mitigation circuit (left) showing digital logic chips. The ComBlock transmitter (right) is being controlled by this circuit.

111 94 Figure 7.5. Baseband logic control circuit diagram made in PSpice

112 Experimental Setup An experimental Wi-Fi communication system was used to transmit and receive digital data in the presence of MWO interference. The Wi-Fi signal was transmitted by the ComBlock transmitter at a rate of 363 kbps with the 11 chip Barker spreading code. This data signal s bandwidth is 8 MHz. This signal was chosen because it is very similar to the 1 Mbps data rate IEEE signal that is used to transmit the physical layer convergence protocol [CON00] and often data for wireless local area networks. Thus, the results of this interference mitigation study applied to the 8 MHz Wi-Fi signal are well applicable to IEEE Wi-Fi systems in general. The data is transmitted in 128 bit packets by the ComBlock transmitter. The ComBlock receiver captures and decodes the data packets. The transmitted and received packets are compared to get the experimental BER. In all experiments, the receiver was placed in a position equidistant from the Wi-Fi transmitter and an interfering MWO. Three different MWOs were used in the BER study. Four experimental scenarios were tested for each MWO and the BER was recorded each time. Case 1 is shown by the spectrogram in Figure 7.6. Here, the Wi-Fi transmitter operates at 2.46 GHz without any interference mitigation. In this frequency range the AM-FM signal of the MWO exists and hence there is high interference. Case 2 is shown in Figure 7.7, where the interference is mitigated and the Wi-Fi transmitter frequency is still at 2.46 GHz. In Case 3 and Case 4, the Wi-Fi transmitter carrier frequency is at GHz, where there is less interference as only low duty-cycle MWO transients exist. Interference is not mitigated in Case 3 but it is mitigated in Case 4 by the cognitive radio system. The spectrograms for cases 3 and 4 are shown in Figures 7.8 and 7.9 respectively.

113 96 Figure 7.6. Case 1: No interference mitigation BER study (Wi-Fi at 2.46 GHz) Figure 7.7. Case 2: Interference mitigation (Wi-Fi at 2.46 GHz)

114 97 Frequency (Hz) Time(s) Figure 7.8. Case 3: No interference mitigation BER study (Wi-Fi at GHz) Figure 7.9. Case 4: interference mitigation BER study (Wi-Fi at GHz) 7.4 BER Studies Tables 7.1 through 7.4 show the experimentally recorded BERs for each of the scenarios described in Section 7.3. The results vary depending on the MWO used as the interference source acting on the experimental ComBlock wireless system.

115 98 Table 7.1. BER for Case 1 (Wi-Fi at 2.46 GHz without interference mitigation) MWO # Data Rate BER kbps kbps kbps Table 7.2. BER for Case 2 (Wi-Fi at 2.46 GHz with interference mitigation) MWO # Data Rate BER kbps kbps kbps Table 7.3. BER for Case 3 (Wi-Fi at GHz without interference mitigation) MWO # Data Rate BER kbps kbps kbps Table 7.4. BER for Case 4 (Wi-Fi at GHz with interference mitigation) MWO # Data Rate BER kbps kbps kbps Although the data rate drops to 50% in the interference mitigated case, the BER is minimized. This means that data packets will be reliably transmitted by a Wi-Fi device even when a MWO is operating. In the case where this interference mitigation is not used, the data rate remains at 100% but the BER is much higher, as shown in Table 7.1. This means that many data packets are likely to be dropped as a result of interference and the actual throughput may be much less than the mitigated case even though the data transmission rate is higher. At high BER and high packet drop rates, the Wi-Fi connection may be severed [AVA02]. The MWO interference mitigation technique solves this problem completely.

116 99 It should be noted that the Barker spread IEEE signal is the most resistant to interference and noise effects. For other IEEE signals the BER is likely to be higher in similar experimental settings. Also, the BER greatly depends on the relative received signal strengths of the data signal and the MWO signal, that is, the Signal-to- Interference Ratio (SIR) [PAU06]. Due to this effect, the BER varies considerably if the distances between the receiver, transmitter, and the MWO are changed. Therefore, Tables 7.1 to 7.4 are meant only for comparative purposes to demonstrate the performance of the experimental Wi-Fi system in the different scenarios, particularly with or without interference mitigation. Tables 7.1 and 7.3 also show that MWO interference significantly degrades the wireless communication system performance making interference mitigation valuable. Furthermore, this method is practically realizable on consumer access points and other Wi-Fi devices.

117 100 CHAPTER 8 CONCLUSION 8.1 Pulse Shaping for 1 Mbps Signal This research was described in Chapter 2. A new Barker spread modulation scheme was investigated that incorporated pulse shaping techniques with an 11-chip Barker code. Four pulse shapes were studied and their PSDs determined. In all cases the PSD was compared to the FCC spectral mask. The sinusoidally shaped Barker waveform required the least output filtering in order to satisfy the spectral mask. The BER performance was also studied. The pulse shaped systems performed better in several respects: better spectral characteristics, lower order filter requirement, and improved BER performance. The conclusions were verified by matching results obtained by analytic, simulation, and experimental studies. Two novel line coding techniques were employed in conjunction with pulse shaping as an effort to further boost performance. Both the two systems were tested via simulation and experimental study. One system employed the buffering 2 of bits at a time. However, the spectral characteristics of this line coded Barker spread system were observed to be poor. The second line code system utilized the buffering of 3 bits at a time. The spectral characteristics of this system were the best observed, such that the FCC spectral mask was nearly achieved without using any filtering. Simulated BER results for this system showed that its performance was better than the rectangular pulse Barker spread system with no buffering.

118 Pulse Shaping for 5.5 Mbps Signal This research direction was the result of logical progression of applying symbol shaping to the Barker spread 1 Mbps signal and onto the CCK spread 5.5 Mbps signal, and the work has been described in this dissertation in Chapter 4. A new CCK spread modulation scheme was investigated that incorporated pulse shaping techniques to modulate the four possible I and Q phase vectors. Several shapes were studied and their PSDs determined. In all cases the PSD was compared to the FCC spectral mask. The sinusoidally shaped CCK waveform required the least output filtering in order to satisfy the spectral mask. The BER performance was also studied by simulation. The pulse shaped systems performed better in several respects: better spectral characteristics, lower order filter requirement, and improved BER performance. The results of this research work can be used to improve the performance of the existing IEEE Mbps CCK based wireless signal. 8.3 MWO Signal Study Chapter 5 described the research undertaken here. The MWO signal was meticulously studied as part of this research project. In particular, key features of the MWO signal: the ON-OFF duty cycle, the AM-FM frequency sweeping nature of the signal, and the transients were thoroughly investigated. Prior MWO signal studies have often neglected the short duration transients, but our research has shown conclusively that the transients are important and have critical interference impacts on Wi-Fi communication on account of their high bandwidth.

119 102 As a result of this signal study, valuable insights were attained that made it possible to accurately model the MWO signal and develop an interference mitigation technique. The results of this experimental signal are valuable to engineers studying MWO leakage and its interference on Wi-Fi systems. 8.4 MWO Signal Modeling In this work, described in Chapter 6, analytical models were developed for the MWO signal based on its experimental characteristics. Model #1 was developed that expressed the major features of the MWO signal: ON-OFF duty cycle, the AM-FM frequency sweeping nature of the signal, and the transients. This model was simulated and compared with the actual experimental MWO measurements. Although there was some degree of correlation between the two, more accuracy was desired in the model. As a result, model #2 was formulated and was carefully designed to cover more details of an actual MWO. This analytical model was simulated and emulated. Emulation was done to support of the model. The emulation also sheds light on our three-pronged research approach. The spectrogram and PSD plots obtained by simulation and emulation matched extremely well with actual experimental plots of the MWO signal. Model #2 is more refined and can be utilized in wireless network simulation studies that aim to improve IEEE Wi-Fi transmission. Thus, the modeling research aims to improve Wi-Fi communications indirectly by aiding network simulation engineers in their quest to optimize wireless networks.

120 MWO Interference Mitigation The interference mitigation research presented in Chapter 7 provides the single biggest improvement to Wi-Fi communication. An interference mitigation technique was formally proposed and implemented. The implemented circuit performed fully in line with its expected function. Thus, interference on Wi-Fi systems due to MWOs can now be eliminated, thereby improving Wi-Fi communication system performance in certain interference rich environments. The interference mitigation technique was practically implemented and BER results were obtained via an experimental Wi-Fi communication system. Promising results were obtained, thereby validating the research results and effort. This novel system is practically realizable on consumer wireless APs in order to improve the performance of wireless computer networks based on AP infrastructure. 8.6 Future Work As part of ongoing and future work, pulse shaping to improve spectral characteristics will be extended to other IEEE signals; particularly the higher data rate signals transmitted using PBCC. It is this researcher s goal to study the pulse shaping systems via simulation and experimental emulation and, wherever possible, by analytical means. It has been shown that the existing IEEE Mbps data signal can be improved through symbol shaping. Although not investigated in this research project, it is expected that the same pulse shaping techniques are applicable to shape the I and Q

121 104 phase symbols of the QPSK IEEE Mbps data signal, and similar results are expected. For future work, this hypothesis needs to be tested. The RF signal and interference from a residential MWO has been thoroughly studied. In the future, we intend to look at RF leakage from the other types of MWOs: commercial and switching MWOs. As part of future work, interference mitigation based on the technique shown in Figure 7.1 will be practically implemented. Interference mitigation using cognitive radio has proved valuable for MWO interference mitigation. It is hoped that similar cognitive radio algorithms will be applied to mitigate interference on IEEE systems from other wireless devices like the cordless telephone.

122 105 APPENDIX INTERFERENCE SPECTROGRAMS

123 106 In this appendix several interference spectrograms are plotted. The figure titles describe which devices are interfering. MWO signal AP data packet Cordless phone s data packet ComBlock transmitter signal Figure A.1. Spectrogram 1: interference between MWO, AP, ComBlock transmitter, and DSSS cordless phone.

124 107 Cordless phone signal MWO signal ComBlock transmitter data packet Figure A.2. Spectrogram 2: interference between MWO, ComBlock transmitter, and DSSS cordless phone. Figure A.3. Spectrogram 3: interference between ComBlock transmitter, and DSSS cordless phone.