Performance Evaluation of Pattern Reconfigurable Antennas in MIMO Systems. Yu Zhou

Transcription

1 Performance Evaluation of Pattern Reconfigurable Antennas in MIMO Systems by Yu Zhou A thesis submitted in conformity with the requirements for the degree of Master of Applied Science The Edward S. Rogers Sr. Department of Electrical and Computer Engineering University of Toronto Copyright c 2012 by Yu Zhou

2 Abstract Performance Evaluation of Pattern Reconfigurable Antennas in MIMO Systems Yu Zhou Master of Applied Science The Edward S. Rogers Sr. Department of Electrical and Computer Engineering University of Toronto 2012 With the fast adoption of LTE and IEEE N, more devices are employing multiple antennas to boost the data rate and reliability of the communication link. Traditionally, fixed antennas are used in such devices. In recent years, reconfigurable antennas have been sought out to further boost the performance, which can adaptor to the changing wireless channel by altering their radiation characteristics, and maintain or exceed the performance of fixed antennas. This thesis studies the possibility of performance increase using pattern reconfigurable antennas as receivers. Their performance potential was first estimated using simulations, and then demonstrated using two electrically steerable passive array radiator (ESPAR) antennas against a pair of monopole antennas on a hardware bit error rate (BER) testbed. The former produces equal performance in BER with certain pattern combinations and excels in theoretical capacity with substantial lead making pattern reconfigurable antenna a potent option as receiver in MIMO-related applications. ii

3 to my parents, family and dearest iii

4 Acknowledgements I would like to take this opportunity and thank the people who have made this thesis possible. Firstly, I owe my deepest gratitude to my supervisor Dr. Sean Victor Hum who has generously shared his knowledge and time to guide me towards the completion of this thesis. His sage advice and insightful ideas have provided tremendous help in my research and it has been an honour to work under his supervision. I would like to extend my gratitude to Dr. Raviraj Adve, for his support and knowledgeable advice in digital communications during the course of my research. I would like to also thank my fellow graduate students for their support and friendship in the years of my graduate studies, and special thanks to Jonathan Lau, Krishna K. Kishor, Tony Liang, Neeraj Sood, and William Chou for spending their valuable times to provide aid and tips that have assisted my research progress. And last but not least, I would like to thank my family, especially my parents, and my girlfriend Amanda, for their tremendous and unconditioned support and encouragement, both financially and morally, in making all this possible. I will be forever indebted to them. iv

5 Contents 1 Introduction An Emerging Technology: MIMO Motivation and Proposal Thesis Goal Thesis Outline Understanding Wireless Communications and MIMO Types of Wireless Communication Systems SISO System MISO and SIMO Systems MIMO Systems Channel Capacity and BER Receive Diversity Techniques to Improve MIMO Performance Digital Versus RF Beamforming Simulations and System Performance Estimation Simulation Method Evaluation Method and Figure of Merit Simulations with Omni-directional Receiving Antennas Simulation Setup v

6 3.3.2 Simulation Results Simulations with Synthetic Pattern Reconfigurable Receiving Antennas Simulation Setup Simulation Results Reconfigurable Antennas for MIMO Motivation Common Reconfigurable Antennas for MIMO Selection Criteria for Performance Studies ESPAR Antenna Physical Design and Characteristics Monopole Simulation Single ESPAR Simulations Dual ESPAR Simulations Fabrication Evaluation Experimental Results Test Methodology Existing Evaluation Methods in Literature BER as a Method of Evaluation BER Testbed Testbed Signal Flow Hardware Realization for SIMO Practical Implementation Issues and Proposed Solutions Testbed Validation BER Results with ESPAR Antennas Channel Capacity Evaluation of SIMO/MIMO with ESPAR Antennas vi

7 5.4.1 SIMO Performance MIMO Performance Conclusion Research Summary and Achievements Final Remarks and Thesis Contributions Future Work A Figures with Complete Simulation Results 124 B Derivation Process of ESPAR Parameters 127 Bibliography 132 vii

8 List of Tables 2.1 Alamouti encoding scheme for two transmitting antennas ESPAR design parameters Simulated monopole heights and input impedances (Ω) Dual ESPAR setup test configurations List of ESPAR Test Configurations and S-parameter Measurements; Configurations are denoted from C1 to C List of BER Testbed Hardware Components viii

9 List of Figures 2.1 A SISO system. It only employs a single antenna at both the transmitter and receiver A SIMO system. It employs a single antenna at the transmitter and multiple antennas at the receiver A MISO system. It employs multiple antennas at the transmitter and a single antenna at the receiver A MIMO system. It employs multiple antennas at both transmitter and receiver Digital beamforming RF Beamforming The sampling region in simulation, with coordinates in metres Channel Capacity CCDF, SISO vs 2 2 MIMO A 2D planar array in the azimuth plane The floor plan of BA8177 showing the plane of coverage by the receiver patterns and the three principal patterns (in red, blue and green) tested Synthetic patterns of the 3 3 array for simulating directive antennas with the beam angles indicated in degrees Channel Capacity CCDF, SISO, at fixed transmitting SNR. Synthesized radiation patterns are introduced into the simulations, with the corresponding beam angle in degrees shown in parentheses in the legend ix

10 3.7 Channel Capacity CCDF, 2 2 MIMO, at fixed transmitting SNR. Synthesized radiation patterns are introduced into the simulations, with the corresponding MIMO element beam angles in degrees shown in parentheses in the legend Channel Capacity CCDF, SISO versus 2 2 MIMO, with only the best, worst and reference cases at fixed transmitting SNR are shown. Synthesized radiation patterns are introduced into the simulations, with the corresponding MIMO element beam angles in degrees shown in parentheses in the legend Channel Capacity CCDF, SISO versus selected 2 2 MIMO, at fixed receiving SNR. The H matrices are normalized to eliminate power disparity between configurations Degrees of freedom CCDF, 2 2 MIMO. This metric indicates the number of channels the system effectively utilizes Degrees of freedom CCDF, 2 2 MIMO. This metric indicates the number of channels the system effectively utilizes. The plot is zoomed in between 1 and 1.1 on the abscissa Diagram of an ESPAR antenna showing the corresponding design parameters. The driven element is highlighted in gray Top view of ESPAR structure generated and meshed using Gmsh Simulation model of the ESPAR in HFSS Simulated radiation pattern of ESPAR in HFSS, R sub = 0.75λ Gain patterns of HFSS simulated ESPAR antenna with 3 directors and 3 reflectors Simulation model of the dual ESPAR in HFSS x

11 4.7 Radiation pattern of coupled dual ESPAR antennas. Red dotted lines represent the ideal (uncoupled) patterns, blue solid lines represent the coupled patterns Radiation pattern of coupled dual ESPAR antennas in modified test case 6. The red dotted line represents the ideal (uncoupled) pattern, the blue solid line represents the coupled pattern of ESPAR no. 1, black solid line represents the coupled pattern of ESPAR no Equivalent circuit of a PIN diode under forward and reverse bias conditions. When forward biased it has a positive reactance; when reverse biased, the reactance becomes negative Loading and biasing scheme for the passive monopoles. The loads are on the top layer, while the DC biasing traces are on the bottom layer A photo of the fabricated duo-espar antennas on a single FR-4 substrate. The two RJ45 ports used to control the diode voltages are visible in the photo at the far edge of the substrate and pointing away from the camera Far-field radiation patterns of the dual-espar antennas Far-field radiation patterns of the dual-espar antennas Far-field radiation patterns of the dual-espar antennas Far-field radiation patterns of the dual-espar antennas Far-field radiation patterns of the dual-espar antennas Far-field radiation patterns of the dual-espar antennas Far-field radiation patterns of the dual-espar antennas Far-field radiation patterns of the dual-espar antennas Far-field radiation patterns of the dual-espar antennas TX Pipeline Configuration with Single Transmitting Antenna TX Pipeline Configuration with Dual Transmitting Antennas xi

12 5.3 Implementation of Alamouti coding at the transmitter in MISO/MIMO testbed RX Pipeline Configuration for Single Receiving Antenna RX Pipeline Configuration for Dual Receiving Antennas Implementation of Alamouti coding at the receiver in MIMO testbed SIMO TX Pipeline Configuration Part 1 consists functions implemented in Matlab script executed from a computer. The pipeline marked 1 and 2 continue to part 2 in Figure SIMO TX Pipeline Configuration Part 2 consists of hardware instruments which are used to realize and modulate the digital waveforms generated by Matlab script, fully controlled by Matlab script via imbedded Matlab toolbox and LAN interface Illustration of Assembled I/Q Demodulator RX Pipeline Configuration Part 1 consists of functions realized by hardware instruments and the pipeline continues in Figure RX Pipeline Configuration Part 2 continuing from Part 1 in 5.10, consists of post-processing steps implemented in software The hardware implementation of BER Testbed with full-blown equipments at its current development stage. Note the two green cables are the outputs from receiving antennas, and there are four channels going into the digital oscilloscope representing the I and Q channels of both antennas The positioner used as part of the testbed, manufactured by Nearfield Systems Inc. The positioner can cover a vertical plane of 1.6 m horizontally and 0.5 m vertically during testing A flowchart showing the implementation of the stitching technique during capturing stage for extending BER curves xii

13 5.15 A flowchart showing the implementation of the stitching technique during post-processing stage for extending BER curves A detailed example illustrating BER curves before and after applying the stitching technique Sanity check on the implementation of the BER testbed ESPAR antennas mounted on the positioner. Two CAT5 ethernet cables provides power and correct biasing voltage to the diodes on the antenna for creating steerable patterns Floorplan of BA8177 showing the antenna placements and setup. Red blocks indicate antennas; blue blocks indicate absorbers; pink blocks indicate benches; yellow blocks indicate positioner components. The transmitter is located at a fixed location near the entrance. The receivers are placed on the positioner of the testbed and are confined to movements in the azimuth plane in south-north directions. The paths between the TX and RX are NLOS Radiation patterns of C1 with respect to the channel Radiation patterns of C2 with respect to the channel Radiation patterns of C3 with respect to the channel Radiation patterns of C4 with respect to the channel Radiation patterns of C5 with respect to the channel Radiation patterns of C6 with respect to the channel Radiation patterns of C7 with respect to the channel Radiation patterns of C8 with respect to the channel Radiation patterns of C9 with respect to the channel Measured SIMO BER results comparing the ESPAR antennas to the monopole antennas. All results are shown xiii

14 5.30 Measured SIMO BER results comparing the ESPAR antennas to the monopole antennas. Only the reference results and the best and worst ESPAR results are shown. It is evident that the ESPARs can match or exceed the performance of monopoles, even when they are unmatched for optimal return loss Simulated SIMO BER results comparing the ESPAR antennas to the monopoles. The results show general agreement with the experimental results Simulated SIMO BER results comparing the ESPAR antennas to the monopoles. The power loss due to the impedance difference between the ESPAR antennas and monopoles have been fully compensated, eliminating any role of return losses in the comparison Selected simulated SIMO BER results comparing the ESPAR antennas to the monopoles. The power loss due to the impedance difference between the ESPAR antennas and monopoles have been fully compensated, eliminating any role of return losses in the comparison Theoretical SIMO capacity versus Transmitting SNR; the capacity is obtained using measured channel matrix of various receiver configurations Spatially averaged theoretical SIMO capacity versus transmitting SNR; the capacity is obtained using measured channel matrix of various receiver configurations. The power loss due to the impedance difference between the ESPAR antennas and monopoles have been fully compensated, eliminating any role of return losses in the comparison Spatially averaged theoretical SIMO capacity versus receiving SNR; the capacity is obtained using measured channel matrix of various receiver configurations xiv

15 5.37 Spatially averaged theoretical SIMO capacity versus receiving SNR; the capacity is obtained using measured channel matrix of various receiver configurations. It contains the same results presented in Figure 5.36 but within a narrower SNR range to make the curves easier to distinguish Maximum theoretical SIMO capacity versus receiving SNR; the capacity is obtained using measured channel matrix of various receiver configurations Maximum theoretical SIMO capacity versus receiving SNR; the capacity is obtained using measured channel matrix of various receiver configurations. Contains same results presented in Figure 5.38 but within a narrower SNR range to make the curves easier to distinguish Spatially averaged theoretical MIMO capacity versus transmitting SNR; the capacity is obtained using measured channel matrix of various receiver configurations Spatially averaged theoretical MIMO capacity versus transmitting SNR; the capacity is obtained using measured channel matrix of various receiver configurations. The impedance mismatch delta between the ESPAR antennas and monopoles have been fully compensated, eliminating any play of return losses in the comparison Spatially averaged theoretical MIMO capacity versus receiving SNR; the capacity is obtained using measured channel response of various receiver configurations. The curve with solid circle markers is the reference monopole configuration Spatially averaged theoretical MIMO capacity versus receiving SNR; the capacity is obtained using measured channel response of various receiver configurations. A more zoomed in version of Figure Maximum theoretical MIMO capacity versus receiving SNR; the capacity is obtained using measured channel matrix of various receiver configurations.116 xv

16 5.45 Maximum theoretical MIMO capacity versus receiving SNR; the capacity is obtained using measured channel matrix of various receiver configurations. It contains the same results presented in Figure 5.44 but within a narrower SNR range to make the curves easier to distinguish A.1 Channel Capacity CCDF, SISO versus 2 2 MIMO, at fixed transmitting SNR. Synthesized radiation patterns are introduced into the simulations, with the corresponding MIMO element beam angles in degrees shown in parentheses in the legend A.2 Channel Capacity CCDF, SISO versus 2 2 MIMO, at fixed receiving SNR. The H matrix is normalized to noise power to eliminate power disparity between configurations B.1 Normalized radiation patterns (E θ ) for R sub = 0.75λ and load configuration The ESPAR produces a main beam in the direction of ϕ = B.2 Normalized radiation patterns (E θ ) for R sub = 0.75λ and load configuration The ESPAR produces a main beam in the direction of ϕ = B.3 Normalized radiation patterns (E θ ) for R sub = 0.75λ and load configuration The ESPAR produces a main beam in the direction of ϕ = B.4 Normalized radiation patterns (E θ ) for R sub = 0.5λ and load configuration B.5 Normalized radiation patterns (E θ ) for R sub = 1λ and load configuration B.6 Best normalized radiation patterns (E θ ) at load configuration for substrate sizes of R sub = 0.5λ, R sub = 0.75λ, and R sub = 1λ xvi

17 Chapter 1 Introduction Wireless local area network (WLAN) access has become a norm in our modern society, but just fifteen years ago, such standards did not exist. In 1997, the first IEEE WLAN standard, , was introduced. However the data rate was feeble and it was never widely adapted, until the deployment of the subsequent b standard. Moving into the early twenty-first century, laptop computers and mobile computing have seen a tremendous increase in popularity. Convenience has taken a higher priority and people wanted to have access to the Internet without limiting their freedom by plugging an Ethernet cable into their laptops. The development of the IEEE standards have enabled wireless access to the Internet. The endless search for higher data rates saw the introductions of IEEE g in 2003 with 5 times the data rate of the earlier IEEE b, and finally the draft IEEE n standard in 2009, with up to a 150 Mbit/s data rate promising three times the performance of IEEE g. On a different front, Moore s Law has pushed the semiconductor industry to embed handheld mobile devices such as cellular phones with significant computing power under the hood while maintaining excellent battery rundown time. The widespread deployment of 3rd generation (3G) mobile telecommunications network coverage has granted people with high-speed access to the Internet using their mobile handheld devices that was once 1

18 Chapter 1. Introduction 2 a job only computers could accomplish, and the amount of information moving in the air is greater than ever before. Developments in the wireless communication industry have introduced significant improvements to the venerable 3G protocol, with the 4G technologies on the horizon. 1.1 An Emerging Technology: MIMO There is one technology that plays an important role in the IEEE n standard and the 4G technologies such as WiMAX and Long-Term Evolution (LTE) is the multipleinput and multiple-output (MIMO) system. The concept of using multiple antennas in wireless communication is dated back to the 1970s, and in 1996 Greg Raleigh and John M. Cioffi refined the concept and proposed co-located transmitting antennas at one transmitter for more effective throughput improvements [1]. Traditionally, two wireless systems only communicate through one single antenna on each side. This is a singleinput single-output system, or SISO system. Because of the unpredictable nature of the wireless channel between the two ends, the propagated signal may experience significant fading at the receiver and a severe drop in signal-to-noise ratio (SNR). A MIMO system tackles this problem by using multiple antennas at the transmitter and receiver. With the many antennas, they can establish multiple communication links between the transmitter and receiver, so that even if one link is in deep fade, the others may have acceptable SNR to maintain a critical level. 1.2 Motivation and Proposal Most existing commercial approaches, e.g. cellular phones and pagers, utilize fixed microstrip antennas such as monopoles and planar inverted F antennas (PIFA). Both of these two types of antennas produce directive but usually very broad patterns so that they can guarantee certain reception of incoming signals, and work fairly well under most

19 Chapter 1. Introduction 3 circumstances for mobile devices that are constantly in motion. However a major downside for having a fixed and broad antenna pattern is that these antennas possess relatively low directivity and cause problems in keeping the antennas optimally pointed for best performance. Omni-directional antennas may partially solve this problem, as they have a uniform gain in all directions. However the more omnidirectional the antenna pattern, the lower the gain, thus the less power the antenna can receive. When there exists a dominant angle of arrival (AOA), an omni-directional antenna can only get limited power due its low gain, while an antenna with higher directivity that is oriented to point in the direction of AOA, can receive considerably more power. Therefore when optimally oriented, a directive antenna may excel and deliver much more performance than what an omni-directional antenna could achieve. If the directive pattern is pointed in the direction of arrival (DOA), the higher antenna gain would greatly increase the received power. This concept of using directional antennas in multi-antenna applications is very simple, however despite this conceptual simplicity, there is a problem that needs to be addressed first. If the pattern is not pointing at the right AOA, the high directivity of the antenna would actually hurt performance as the gain in directions off the main beam are usually much lower, and could even worse than that of an omni-directional antenna. If the antenna can dynamically steer its main beam to always point at the DOA of the strongest signal, then it could potentially increase the probability of obtaining a respectable level of receiving SNR. This pattern agility can potentially increase the receive diversity of the system, when the patterns can adaptively choose the channel they wish to receive the signal from. Pattern reconfigurability can be achieved with conventional antenna array, however it not only occupies considerable physical space but also requires a complex feeding network and phase shifters for every antenna element in the array. Enter the pattern reconfigurable antenna. By integrating tunable electronic components into the antenna, it can obtain the ability to electrically alter the characteristics of

20 Chapter 1. Introduction 4 its radiation pattern on-the-fly. A carefully designed reconfigurable antenna can generate identical patterns that point towards different directions depending on the configurations of the tunable components. Comparing to antenna arrays, reconfigurable antennas are in general cheaper to make, have fewer components and lower volume, which make they more suitable for MIMO related applications. There have been reported studies suggesting increased performance in MIMO using pattern reconfigurable antennas. In 2007, Samsung proposed an antenna pattern selection system (APS) using multiple pattern reconfigurable antennas [2] and measured the channel capacity against equivalently numbered and spaced omni-directional antennas and shows capacity improvements. There are very little details provided on the antenna used and how the capacity measurements are conducted. In another published study, a MIMO reconfigurable antenna array in a 2x2 MIMO system is tested in an indoor nonline-of-sight (NLOS) environment with two distinct propagation paths and has demonstrated that by steering the antenna patterns, better channel capacity can be achieved with certain pattern combinations [3]. In this demonstration, the test environment is a two reflector NLOS setup and has a very deterministic channel. Though the results are well presented in a very analytical manner, the test scenario was limited in showing what is possible at one spatial location with a very specific channel. They study did not provide information on what capacity improvements are possible with reconfigurable antennas in a more realistic channel, which must take into statistical considerations such as spatial averaging. A more thorough study in this topic can greatly contribute to the area of MIMO research. 1.3 Thesis Goal The main focus of this thesis is on studying the impact of using pattern reconfigurable antennas on the performance of wireless systems with multiple receiving antennas. In

21 Chapter 1. Introduction 5 order to accomplish this, the thesis establishes several goals in a few stages that will lead to the final achievements. The first goal is to estimate the performance impact of using directive patterns in MIMO systems. This establishes a guideline for the subsequent analysis on the reconfigurable antennas and provides an initial assessment to our hypothesis that pattern reconfigurable antennas as receiving antennas can increase the performance of a MIMO system. The performance estimation should be as generalized as possible and applicable to all pattern reconfigurable antennas. The second goal is to design a pattern reconfigurable antenna for carrying out the experimental measurements. At least two such antennas are required to be used for evaluating the effectiveness of steerable patterns in an indoor environment. In order to generalize the experimental results and set this thesis further apart from the existing studies, the antenna should be able to demonstrate more flexibility and agility than most of the existing works and should have a better representation of the ability of an ideal pattern reconfigurable antenna. Available antenna designs for similar applications are studied in order to establish the criteria of the antenna characteristics for guiding our design. Finally, the third goal is to be able to experimentally carry out the measurements to confirm our earlier findings, and validate our idea with convincing results. A suitable method of evaluation is sought, and based on the decision, a corresponding test setup is developed. To gain more understanding into the ground of reconfigurable antennas in MIMO applications, the results are analyzed thoroughly to determine the roles of factors such as antenna gain and diversity order in changing the performance of a MIMO system and their contribution to the overall performance.

22 Chapter 1. Introduction Thesis Outline This thesis presents our progressive study on pattern reconfigurable antennas in MIMO applications and their potential benefits. The thesis is structured as follows. Chapter 2 reviews important concepts in wireless communications, and background on various types of systems including SISO, SIMO, MISO, and MIMO. It touches on the details of traditional diversity schemes for both transmitter and receiver, and introduces the concept of channel capacity which is considered to be one of the most important figures of merit when evaluating performance of a wireless system. Lastly, it categorizes methods that attempt to improve MIMO capacity such as by using more advanced diversity schemes, or by employing novel antenna designs. Chapter 3 states the motivation for running software-based channel simulations, and describes the simulation method, setup and procedures, and the final results. The simulation model is developed based on the real channel where the experiments are conducted, and the simulated MIMO capacity results with omni-directional receivers are compared to that with directive radiation patterns, showing what can be potentially achievable with reconfigurable antennas. Chapter 4 briefly introduces various types of common reconfigurable antennas designed for MIMO applications, and proceeds to discuss the requirements and criteria in searching for a suitable reconfigurable antenna design that can efficiently demonstrate the purpose of this study in the experiments. Finally, the electronically steerable passive array radiator (ESPAR) is introduced as the antenna chosen for carrying out the experiments. The design and fabrication procedures of the antenna are fully documented and the measured far-field patterns are presented. Chapter 5 starts off with discussions of advantages and disadvantages of each evaluation method for wireless communication systems. It states the usefulness in measuring the bit error rate (BER) as a valuable figure of merit for real-world performance along with the traditional but more theoretical channel capacity. A BER testbed for MIMO is

23 Chapter 1. Introduction 7 proposed in detail and its hardware implementation for SIMO testing is revealed. The experimental BER results, Monte Carlo simulations on BER and theoretical capacity evaluation using measured channel matrices are presented and compared to each other. The similarities and contradictions between each method is discussed and preliminary conclusions are drawn on the performance advantage of wireless systems employing ES- PAR antennas. Finally, Chapter 6 summarizes the thesis with conclusions on all major findings and discusses possible improvements and further studies as future work.

24 Chapter 2 Understanding Wireless Communications and MIMO In order to understand MIMO, the performance evaluation methods and results in the later chapters, some fundamental concepts surrounding wireless communication systems are reviewed in this chapter. We start with the types of wireless communication systems, the concepts of channel capacity, MIMO and receive diversity, and end with ways to improve MIMO performance. 2.1 Types of Wireless Communication Systems This section expands the discussion on types of wireless communication systems from the preceding chapter, including SISO, single-input and multiple-output (SIMO), multipleinput and single-output (MISO) and MIMO systems and their associated diversity schemes. The diversity schemes are methods for improving the reliability or performance of a communication system via the use of multiple communication channels. 8

25 Chapter 2. Understanding Wireless Communications and MIMO 9 TX 1 h11 1 RX Figure 2.1: A SISO system. It only employs a single antenna at both the transmitter and receiver SISO System A SISO system, which is the acronym for a single-input single-output system, is the most basic communication system to be discussed in this section. With a single antenna both in the transmitter and receiver, it is relatively straightforward to model and build a SISO system than the other types of wireless communication systems. Figure 2.1 illustrates a SISO wireless system. The channel, shown as the dashed line between the transmitter and receiver, can be presented by a complex quantity h 11 which is the transfer function between the output of the transmitter, to the input at the receiver. Even though a SISO system is cheaper to build with fewer components and does not require the use of any special coding schemes at the transmitter and at the receiver, it is the least reliable wireless system of all types mentioned here. In a typical environment where these systems operate, there is often no clear line-of-sight (LOS) channel between the transmitter and receiver. The transmitter sends the signal through its sole antenna, and due to multipath, the signal bounces off different objects and create multiple instances of signals that propagate to the receiver with varying amplitude, distortion, time delay and phase offsets. Each of the arriving signals may destructively interfere with one another potentially resulting in a deep-fade and severe penalty to the SNR at the receiver. When the only antenna in the receiver experiences a deep-fade, the system cannot be able to reliably determine the bits it has received due to the extremely poor SNR.

26 Chapter 2. Understanding Wireless Communications and MIMO 10 h11 1 TX 1 hn1... N RX Figure 2.2: A SIMO system. It employs a single antenna at the transmitter and multiple antennas at the receiver MISO and SIMO Systems A MISO or SIMO system employs one of the several wireless diversity schemes by using multiple antennas at the transmitter or receiver end respectively. Figure 2.2 and Figure 2.3 illustrates a wireless systems employing SIMO and MISO respectively. The channel of a SIMO system can be represented with a column vector h SIMO, h 11 h 21 h SIMO =, (2.1)... where h 11 to h n1 are the transfer functions between the transmitter and the n ports at the receiver respectively. Similarly, the channel of a MISO system can be represented h n1 with a similar row vector h MISO with elements h 11 to h 1n, h MISO = [h 11 h h 1n ], (2.2) where the elements are the transfer functions between the n ports of transmitter and the receiver respectively, The purpose of having two or more spatially separated antennas in the transmitter or receiver is to reduce the effect of multipath fading and combat the occurrence of poor SNR in one or more of the many transmitter-receiver links. With more antennas deployed, there is an increasing possibility that at least one of the other links still has a

27 ... Chapter 2. Understanding Wireless Communications and MIMO 11 1 h11 TX M h1m 1 RX Figure 2.3: A MISO system. It employs multiple antennas at the transmitter and a single antenna at the receiver. respectable SNR for the received signal to be correctly processed and interpreted. In order to take advantage of the multiple antennas, a MISO or SIMO system can employ one of the many types of diversity schemes at the transmitter or receiver. These diversity schemes help to achieve one single goal to improve the reliability of a wireless communication system. For a SIMO system, there are commonly three types of diversity combining techniques that can be used practically at the receiver. These techniques are namely the selection combining, maximal-ratio combining, and equal-gain combining [4]. Despite the recent emergence of multi-antenna technology, the diversity schemes have been around for more than five decades. The first of the three, selection combining (SC) or selection diversity, is a switched technique. At any given time, the system evaluates the signal strength at each receiving antenna terminal and picks the best one. SC is easy to implement, as it does not require the receiver to know the actual channel nor apply any additional treatment to the outputs. Therefore, results obtained with SC can only be as good as the best SISO performance from the receiver. Maximal ratio combining (MRC) technique differs from the selection diversity technique in that instead of discarding all the signals that are not equal to the highest SNR, it combines weighted versions of all the signals to yield the highest SNR in the output

28 Chapter 2. Understanding Wireless Communications and MIMO 12 signal. The weights are applied to the received signals from the antennas according to y(t) = w H (h SIMO x(t) + n) (2.3) where y is the combined output, x(t), h SIMO, and n are all N-element vectors, where N is the number of receiving antennas. x(t) represents the signals sent from transmitter to each of the receiving antennas, h SIMO is the SIMO channel vector, n is the noise, and y(t) is the final output at the end of the RF chain. w represents the weightings for each receiving antenna. The system achieves optimal SNR when w is linearly proportional to h, which is based on the assumption that the system has full knowledge of the channel h. Using this information, the output from an MRC combiner for a two-port SIMO system can be derived from Equation 2.3 to be y(t) = a{ h h 21 2 }x(t) + ah 11n 1 + ah 21n 2 (2.4) where w = ah, and ( ) denotes the complex conjugate. The optimized SNR is γ = N γ n (2.5) n=1 where γ n is the per-element SNR, and γ is final output SNR after being processed with MRC technique. It involves co-phasing of the signals in all receiving branches. With full knowledge of the channel, the phase offset of received signals can be estimated and corrected; if erroneously estimated or uncorrected, the phase offsets can lead to reduced SNR. MRC technique is an N-time improvement over the average SNR at each receiver element. Compared to the SC technique, MRC can provide additional SNR advantage over individual SISO outputs at the expense of implementation complexity. Equal-gain combining technique is very similar to MRC, except the weights are unity in all receiving channels, i.e. w = 1. The technique also requires the receiver to have knowledge about the channel in order to correct the phase offsets. The mean SNR is given by Eγ = [1 + (N 1) π ]Γ (2.6) 4

29 Chapter 2. Understanding Wireless Communications and MIMO 13 Table 2.1: Alamouti encoding scheme for two transmitting antennas antenna 1 antenna 2 time t s 0 s 1 time t+t s 1 s 0 where Γ is the average per element SNR. Despite its simpler implementation, it can achieve a comparable improvement to MRC in the final output SNR. For a MISO system, different transmit schemes can be implemented to exploit the transmit diversity and improve performance. Space-time block coding allows the encoded data to be split into m streams and sent through m independent transmitting antennas simultaneously. One common space-time block coding scheme is the Alamouti scheme [5]. Its encoding scheme for two transmitting antennas is presented in Table 2.1, where the operator denotes complex conjugation and T denotes the symbol duration. s 0 and s 1 are two adjacent bits, and the scheme uses two adjacent symbol periods to transmit the two symbols. The receiver receives the following signals at time t and t + T : r 0 = r(t) = h 0 s 0 + h 1 s 1 + n 0 (2.7) r 1 = r(t + T ) = h 0 s 1 + h 1 s 0 + n 1 (2.8) where h 0 and h 1 denote the channel response from antenna 1 and antenna 2 to the receiver, and n 0 and n 1 denote the noise at time t and t + T respectively. The combining scheme involves using a maximum likelihood detector: s 0 = h 0r 0 + h 1 r 1 (2.9) s 1 = h 1r 0 h 1 r 1 (2.10)

30 ... Chapter 2. Understanding Wireless Communications and MIMO 14 1 TX M h11 hn1 h1m hnm 1... N RX Figure 2.4: A MIMO system. It employs multiple antennas at both transmitter and receiver. Substituting Equations 2.7 and 2.7 into 2.9 and 2.10, and we obtain: s 0 = ( h h 1 2 )s 0 + h 0n 0 + h 1 n 1 (2.11) s 1 = ( h h 1 2 )s 1 + h 0 n 1 + h 1n 0 (2.12) The resulting combined signal in Equations 2.11 and 2.12 is equivalent to that obtained from MRC with two receiving antennas and therefore achieves a similar gain in SNR MIMO Systems MIMO, which is the acronym for multiple-input and multiple-output system, employs multiple antennas at both transmitter and receiver. A MIMO system can employ a transmit diversity scheme at the transmitter and a receive diversity at the receiver at the same time, allowing it to combine all the advantages offered by SIMO and MISO systems. Figure 2.4 depicts a MIMO system. The channel of a MIMO system can be described by a M N matrix where M is the number of receiving antennas, and N is the number of transmitting antennas. For example, a 2 2 H matrix is in the form of H MIMO = h 11 h 12. (2.13) h 21 h 22

31 Chapter 2. Understanding Wireless Communications and MIMO 15 In theory, MIMO systems produces the highest throughput when compared to the other types of wireless systems introduced previously in Section and Section 2.1.2, and is often measured as channel capacity. 2.2 Channel Capacity and BER Channel capacity is one of the most commonly used figures of merit in defining the performance of a wireless system. The term channel capacity was defined by Claude E. Shannon in developing information theory back in the 1940s [6]. It is defined as the upper limit of the amount of information that can be reliably transmitted over a wireless channel. In an additive white Gaussian noise (AWGN) channel, the Shannon capacity for a single-antenna scenario can be obtained with C = W log 2 (1 + P N 0 W ) (2.14) where W is the bandwidth, P is the average received power in Watts, N0 is the noise power spectral density in W/Hz, and P N 0 W represents the received SNR. This equation has a unit of bits/s and presents an ideal system that transmits without errors at the rate C with a given bandwidth W Hz. It represents the absolute best-case scenario in the given channel, and cannot be realized with any finite encoding process [7]. Further developing the equation allows us to compute the capacities using the H matrix. For instance, the capacity of a SISO system can be computed using C = log 2 ( 1 + h2 SISO P N 0 ) (2.15) where h SISO is the transfer function of the channel, P is the average transmitted power in Watts. This equation represents the capacity that can be achieved for every Hz of available bandwidth. It is often used to evaluate an ideal SISO system without involving the bandwidth constraints. Similarly, using a matrix form of Equation 2.15, we can

32 Chapter 2. Understanding Wireless Communications and MIMO 16 obtain the MIMO capacity: C = log 2 [det ( I + P )] (HRHH ) N 0 (2.16) where I is the identity matrix, H is the M N channel matrix, R is the covariance matrix, and ( ) H is the Hermitian operator. The P here is the average transmitted power per antenna, in Watts. The expression P N 0 in the equation represents the ratio of total transmitted power to the noise power present at the receiver and we define it as the transmitting SNR. The channel capacity as a figure of merit provides extremely useful guidance to designing and evaluating a wireless system. Another commonly used figure of merit is the bit-error-rate (BER). BER, also known as bit error probability, is the percentage of received bits of a data stream over a communication channel that have been altered due to noise, interference, distortion or bit synchronization errors. The BER can be calculated by dividing the number of bit errors by the total number of bits transmitted. Unlike the channel capacity, BER is a quantity that represents the realistic data rate of the system under a given channel and transmitting power level. By finding the BER at different transmitting SNR level, we can construct a curve of BER values versus their corresponding SNR that presents the probability of bit error with respect to transmitting power increase. BER can become a powerful indicator of the average realizable performance of a system by spatially averaging the BER results across multiple sampling locations to cover a variety of scenarios. 2.3 Receive Diversity While the channel capacity is a measure of the absolute maximum performance of the system and BER represents the realistic performance achievable under given constraints, receive diversity is a quantity that represents the relative performance of a multi-antenna system versus a SISO system. There are two important concepts in receive diversity. The first one is diversity order, which is the scale factor relating the performance improvement

33 Chapter 2. Understanding Wireless Communications and MIMO 17 with multiple receiving antennas over a single one as SNR increases. In BER measurements, the slope of a spatially averaged BER curve at high SNRs is the diversity order. The second concept is diversity gain, which is defined as the improvement in power required to achieve a certain performance criterion; and frequently, the measure of power is presented by the equivalent SNR. Another concept associated with receive diversity is the eigenvalue spread. The term eigenvalue here refers to the square of the respective singular values of the channel correlation matrix HH H. For a 2 2 MIMO system, the eigenvalue spread is defined as the ratio between the larger eigenvalue and the smaller eigenvalue. A more useful metric in measuring the eigenvalue spread is the number of degrees of freedom (DOF), which gives an indication of the number of effective channels that the system is able to utilize. The DOF is defined as the ratio between the sum of all eigenvalues and the maximum eigenvalue. For a 2 2 MIMO, the maximum DOF it can have is 2, since the system is capable of receiving at most two independent copies of the signal. For a 2 2 system, its DOF can be obtained by [8] DOF = {λ1, λ 2 } max{λ 1, λ 2 } (2.17) where λ 1 and λ 2 are the two associated eigenvalues. 2.4 Techniques to Improve MIMO Performance There have been extensive studies on the methods to increase MIMO capacity. The majority of these studies can categorized into three main groups. The first category is by using advanced diversity schemes. Traditional schemes such as selection combining and equal gain combining have been proven in the past to boost the performance and increase the reliability of a MIMO system. These techniques have a relatively low complexity and thus are very practical in implementation. Novel techniques with higher complexity have been exploited more recently to further increase the

34 Chapter 2. Understanding Wireless Communications and MIMO 18 capacity [9, 10]. For example, an adaptive switching technique based on switching between low-complexity transmission schemes is proposed to achieve better capacity over fixed transmission schemes [10]. Another category is the use of smart antennas. A typical smart antenna is an antenna array with advanced signal processing algorithms. It may use switching techniques found in diversity schemes with fixed antennas, or use beam steering techniques with reconfigurable antenna or adaptive arrays, or a combination of both. An example of such a smart antenna for MIMO uses a large number of receiving antennas with a passive weighting process and claims to have achieved same performance as a passive beamformer that has many more RF chains [11]. The third category involves innovative hardware implementations such as reconfigurable antennas. This approach aims at the same goal of improving MIMO performance from a different perspective. By improving link quality through the use of advanced radio frequency (RF) hardware, the instantaneous or average capacity can be improved. There are other techniques that also promise improved MIMO performance through different approaches from the above three categories. For example, a new modulation technique called space shift keying (SSK) has been studied [12] which claims to achieve higher capacity than the conventional phase modulation techniques. 2.5 Digital Versus RF Beamforming Beamforming has long been regarded as one method of increasing MIMO performance. It can be achieved in two ways: 1) through digital signal processing, and 2) through RF beamforming. Each method has its own advantages and disadvantages. Digital beamforming is a digital signal processing step that is performed after the RF chains of the receiver. Figure 2.5 illustrates the flow diagram of digital beamforming. A matrix W is applied digitally in a post-processing stage and produces a final output as

35 Chapter 2. Understanding Wireless Communications and MIMO 19 x H + n W ỹ Figure 2.5: Digital beamforming x H W + ỹ n Figure 2.6: RF Beamforming presented in ỹ = W H (Hx + n) (2.18) where x is the original waveform, H is the channel matrix, and n is noise. While it performs weighted sums on the received signal Hx, it also performs the same process on the noise, since the beamforming matrix is applied after the RF chains. In order to perform digital beamforming, digital processing of multiple copies of the received signals is required. This requires a separate RF chain for each receiving antenna element, and the cost for having many such receivers becomes expensive. Moreover, studies have shown that no improvement in capacity can be achieved when the beamforming is performed after the dominant noise source, therefore using digital beamforming will not improve MIMO performance[13]. The alternative to digital beamforming is RF beamforming depicted in Figure 2.6, where the beamforming happens before noise from the receiver chain distorts the signal. By using either a directive antenna or an RF beamforming network at the output before the signal gets converted to a digital one, the output ỹ can be represented by ỹ = (W H Hx + n) (2.19) With a directive antenna, the antenna s own radiation pattern automatically applies a

36 Chapter 2. Understanding Wireless Communications and MIMO 20 phase and amplitude adjustment to signals arriving from different angles of arrival, and the complex sum of these signals, which is the output of the antenna gets passed into the receiver chain. An RF network can also achieve similar results but requires additional RF hardware in the outputting network. Therefore the ability to perform beamforming on the antenna level requires the fewest RF chains as the combining is performed before going into the receiver, and this reduces the cost and number of components in the receiver significantly. Unlike digital beamforming, RF beamforming can potentially improve capacity if it is employed in receiver noise limited-systems (versus sky noise-limited systems) [14, 15, 13]. Therefore, to investigate the potential benefits of using directive patterns, pattern reconfigurable antennas are considered due to their flexibility in forming different antenna patterns. By changing the radiation pattern, a set of different beamforming weights is applied to the arriving signals, and analyzing the performance results with various patterns will help us determine the correct configurations that can increase MIMO performance.

37 Chapter 3 Simulations and System Performance Estimation In Section 2.5, the advantages of using RF beamforming and the potential benefit of using pattern reconfigurable antenna was discussed. Before experimentally measuring MIMO performance with pattern reconfigurable antennas, we would like to estimate the impact a non-uniform radiation pattern will make on the performance of a MIMO system in this chapter. Therefore computer aided simulations are performed, where two MIMO systems are simulated and compared against each other. One uses two omni-directional patterns, while the other uses directive patterns. 3.1 Simulation Method In order to provide performance guidance to the subsequent experimental results, the simulation simulates the environment where the physical experiments take place. The Antenna Research Lab at the University of Toronto, Room BA8177 in the Bahen Centre of Information Technology, is a good place for conducting this experiment. The inner walls and file cabinets inside the lab can can act as scatterers to enhance the multipath nature of the channel. 21

38 Chapter 3. Simulations and System Performance Estimation 22 Figure 3.1 shows the floor plan of BA8177 that is used to construct the simulation model. The lab has dimensions of 12.5 m by 8.23 m, and the corresponding coordinate system is labeled in the figure as well. The walls inside the lab are modeled as dielectric blocks with the parameters: ϵ r = 5, σ = S/m, and 0.15 m in thickness. The cabinets are modeled as perfect electric conductor (PEC) blocks. The harmonic simulation has a frequency of 2.28 GHz, which is also the frequency used in the experimental measurements. This test frequency is determined by careful measurements and examination of frequency spectrum inside the lab, and after consulting with the Table of Frequency Allocations published by the Industry Canada, where the 2.28 GHz band appears to be vacant. The simulations are performed on SEMCAD X from SPEAG, a readily available EM simulation software package. Using the finite-difference time-domain (FDTD) method, it can simulate complex three-dimensional geometries. FDTD can be very advantageous in solving broadband simulations as it is a time-domain method, and is able to define material properties at all points within the domain, an ideal attribute for simulating complex geometries with mixed material. However it can be computationally expensive when a very large number of cells is needed. Indeed, for simulations at microwave frequencies in a space as large as a 100 m 2 lab, which has a volume of approximately 200 m 3 and the equivalent of about 20 million FDTD cells, it would take tens of hours to complete one simulation even with the aid of additional hardware accelerators. Thus to approximate the first-order effects of a real multipath channel without resorting to lengthy resource-challenge simulations, the simulations are simplified to simulate 2D channels. To accomplish this, a 3D structure is constructed with minimal height and number of cells in the vertical direction. The simulation is configured to allow a T E z mode to be excited by setting the +z and -z boundary conditions to PECs and using cylindrical sources. To simulate a 2 2 MIMO system, two sources are placed in the simulation. The

39 Chapter 3. Simulations and System Performance Estimation 23 first source is placed at (0.5 m, 0.5 m) and the second one is placed at (0.57 m, 0.5 m), so that the two sources are half wavelength apart. Both are cylindrical sources that radiate omni-directionally. To obtain the separate impulse responses between each source and a field point in the simulation, the simulation is performed twice with one source enabled while disabling the other one. In order to examine the performance of the system thoroughly in the simulation, the receiver has to be relocated to obtain a large number of sampling locations. To accomplish this, the simulation sets up a field sensor region that covers the entire area where data is collected on the electric field values throughout the space. This method does not simulate the effects of physical receiving antennas including mutual coupling and modification to the field introduced by the antenna structures, and hence is idealized. The sampling area is depicted in Figure 3.1. Each grid point in the field sensor can be treated as an omni-directional receiving antenna element as it performs superposition of waves arriving from all directions without any amplitude or phase modification. The sensor provides over sampling points and adds flexibility in choosing the sampling regions during post-processing for statistical analysis, i.e. an arbitrarily positioned 2-element receiving antenna can be simulated by selecting two appropriately-spaced points from the sampling region. 3.2 Evaluation Method and Figure of Merit The figure of merit in judging the performance of the system is the channel capacity. Recall that in an AWGN channel, the capacity of a SISO system can be obtained using Equation Using Equation 2.16 to calculate the MIMO capacity of two systems in the same setup can provide comparable results between different configurations, where the H in the equation is the channel matrix consisting of 2 2 transfer functions captured by the field sensors. However it should be noted that each element in the H matrix

40 Chapter 3. Simulations and System Performance Estimation 24 absorbers (12.5, 8.3) sampling area (4.57, 2.90) (8.23, 2.90) y sources (0, 0) x Figure 3.1: The sampling region in simulation, with coordinates in metres

41 Chapter 3. Simulations and System Performance Estimation 25 in Equation 2.16 is the transfer function of the link between transmitter and receiver that involves the channel as well as the antenna patterns and gains at both ends of the radio link. To isolate factors contributing to the performance advantage of one receiver configuration over another, changes to the equation must be made. To determine the capacity advantage of one receiver configuration over another based solely on the diversity gain from its antenna patterns, any difference in power gain due to antenna gain or path loss must be nullified. This can be accomplished by normalizing the H matrices against the total power received ( {P H 2 }), and Equation 2.16 can be rewritten as ( C = log 2 [det I + P )] (HRHH ), (3.1) {P H 2 } N 0 where ( ) 2 is the element-wise array power operation. This normalization process eliminates any power disparity between different channel matrices and the resulting capacity from the equation is solely based on the determinant of the term (and thus the eigenvalues of the matrix by computing HRH H ). As a convention in contrast to the transmitting SNR defined in 2.16, we define the corresponding SNR value presented by the expression P (HRH H ) {P H 2 }N 0 in Equation 3.1 as the receiving SNR. In order to present the channel capacity results effectively, we employ the complementary cumulative distribution function (CCDF) of channel capacity. A capacity CCDF describes the probability of achieving a certain capacity level or higher among the samples taken. Construct a capacity CCDF starts with calculating the capacity at each spatial location and the associated probability when other locations produce a higher capacity than it, which is called the outage probability. The capacity values at all locations are arranged according to their outage probability in descending order. The capacity CCDF is then plotted with the capacity as the abscissa, and outage probablity as the ordinate. With a large number of sampling points available, this allows the capacity CCDF of simulation results to more accurately present the probability of hitting a certain capacity inside the sampling region.

42 Chapter 3. Simulations and System Performance Estimation Simulations with Omni-directional Receiving Antennas Simulation Setup Knowing the omni-directional nature of the field sensors, the output from the sampled points can be directly taken and treated as results from theoretical omni-directional antennas. The voltage amplitude is simply the product of the field magnitude and the height between the two bounding planes. Using Equation 2.15, we can obtain the SISO capacities from all sensor locations and construct the CCDF curve. For calculating MIMO capacity, a 2 2 channel matrix need to be constructed. This means that it requires the field responses from both sources in the simulation. Each source generates the same amount of voltage, therefore comparing to the SISO system, the MIMO system receives double the amount of transmit power Simulation Results Figure 3.2 shows the channel capacity CCDF curves of both SISO and 2 2 MIMO. The CCDF curves are compared at a fixed transmitting SNR level of 10 db for both cases. It is evident that the 2 2 MIMO capacity is higher than SISO capacity at any outage probability. There is an advantage of 148% at 90% outage probability, and this advantage decreases to 40% at 50% outage probability, and maintains as the outage probability decreases.

43 Chapter 3. Simulations and System Performance Estimation x2 MIMO (omni) SISO (omni) 0.8 Probability (C > abscissa) Capacity (bit/s/hz) Figure 3.2: Channel Capacity CCDF, SISO vs 2 2 MIMO 3.4 Simulations with Synthetic Pattern Reconfigurable Receiving Antennas Simulation Setup After establishing a performance guideline with MIMO employing omni-directional antennas, we analyze the impact of antenna patterns to the performance of a MIMO system. Unlike an omni-directional antenna, a directive antenna has a distinct main lobe and its pattern is associated with the terms beamwidth and beam angle. The beamwidth is in general defined as the width, in degrees, between the half power (-3 db) points of the main lobe. The beam angle is the angle, in a predefined spherical coordinate system associated with the antenna s current orientation, where the gain reaches the maximum. To simulate the effects of antenna patterns, several options become available. One

44 Chapter 3. Simulations and System Performance Estimation 28 d x 2 dy N y 3 ϕ M x Figure 3.3: A 2D planar array in the azimuth plane option is to implement the interaction of antenna pattern with the general environment using FDTD-based antenna macromodeling technique [16], since the simulation software SEMCAD X is FDTD-based and could be incorporated with such a technique. However direct implementation requires one simulation per configuration per sampling location, and given the large dimensions of the computational space, this inevitably places a lot of stress on time and computing resources needed to complete the simulations. To make it less time consuming and more flexible, the antenna patterns can be synthesized during post-processing stage to avoid re-capturing the raw data from SEMCAD X. By treating the entire field sensor as an array of individual omni-directional antenna elements, array theory which has been well documented and studied [17] can be applied to a small group of closely spaced sensors for forming the desirable radiation patterns. Figure 3.3 shows a 2D planar array in the azimuth plane with M by N elements. Using array theory, a 2D array factor for the planar array can be written as: N M AF (ϕ) = I 1n e j(n 1)(kd y sin(ϕ)) [ I m1 e j(m 1)(kd x cos(ϕ)) ]. (3.2) n=1 m=1 where k is the wavenumber in free space, d x and d y are the spatial distances between each element in the x- and y-directions respectively, and I m1 and I 1n are the excitation coefficients of each element in the x- and y-directions respectively. In order to simulate an antenna pattern that is relatively directive, the pattern is

45 Chapter 3. Simulations and System Performance Estimation 29 chosen to have a cardioid shape with very small side and back lobes that represents the representative pattern from patches and small arrays. To synthesize such a pattern, a 3 3 planar array with 0.2λ inter-element spacing (d x and d y ) is sufficient. Each synthesized sensor array can be treated as one single receiving antenna in free-space after the weighted sum of each individual element is calculated. Such a weighting matrix for generating ϕ = 210 pattern can be written as j j I mn = j j j (3.3) j j j where the each weight or coefficient in the matrix is applied to the field value at the corresponding element in the 3 3 array, and the complex sum is the final output. The synthesized pattern is set to have a directivity 3 times that of the omni-directional pattern, and rotates 60-degree uniformly in the positive ϕ direction to mimic the pattern rotation of a reconfigurable antenna. Since the back wall of the inner room, where the sampling region is in Figure 3.1, is covered with absorbers, we only consider the patterns that have their main beams pointing away from the back wall. Figure 3.4 shows the plane of coverage by the receiver patterns, and the three principal patterns used, with main beams in the ϕ = 210, 270, and 330 directions respectively, where ϕ is defined nominally from the x-axis within the coordinate system shown in the figure. Since the receiving antennas are intended to have independent radiation patterns, this results in 9 combinations for the simulation. Figures 3.5(a) to 3.5(i) illustrate the synthetic radiation patterns that are generated with respect to ϕ. Each synthesized array combines the response at the 3 3 sub-elements to produce one channel response, therefore the channel matrix for a 2 2 MIMO system is still in the form of a 2 2 matrix, and the channel capacity can be computed using the same equations. To find the effects of antenna pattern on diversity gain, the capacity with a normalized H matrix is also calculated.

46 Chapter 3. Simulations and System Performance Estimation 30 absorbers (12.5, 8.3) region of coverage (4.57, 2.90) (8.23, 2.90) y entrance (0, 0) x sources (0.5,0.5) metal file cabinets Figure 3.4: The floor plan of BA8177 showing the plane of coverage by the receiver patterns and the three principal patterns (in red, blue and green) tested

47 Chapter 3. Simulations and System Performance Estimation 31 (a) 210, 210 (b) 210, 270 (c) 210, 330 (d) 270, 210 (e) 270, 270 (f) 270, 330 (g) 330, 210 (h) 330, 270 (i) 330, 330 Figure 3.5: Synthetic patterns of the 3 3 array for simulating directive antennas with the beam angles indicated in degrees

48 Chapter 3. Simulations and System Performance Estimation Simulation Results Figure 3.6 shows the channel capacity CCDF curves of all SISO configurations, and Figure 3.7 shows the results of all 2 2 MIMO configurations with synthesized directive radiation patterns as well as the omni-directional reference pattern. Figure 3.8 summarizes the best and worst case performance for both SISO and MIMO configurations and compares to the reference. Not all combinations are presented here for brevity, and a complete plot can be found in Appendix A. From the results in Figure 3.7 and Figure 3.8, we can observe that configurations with beam angle pointing toward ϕ = 330 perform better than the others, and configuration (330, 330) produces the highest 90% outage capacity among all 9 configurations. Analyzing the electric field distribution inside the sampling region reveals that the angles of arrival are not uniformly distributed at the receiver and there are strong incoming waves arriving between ϕ = 270 and ϕ = 360. Capacity calculations from the H matrices reveal that the high directivity of a directive antenna can be very advantageous when the beam angle is pointing in the direction of arrival of the incoming signal. This suggests that favorably directed antennas could significantly outperform ones directed randomly. The additional directivity helps the antenna to receive more power and therefore achieve a better receiving SNR than an omni-directional antenna under the same situation. From Figure 3.8, it is also interesting to observe that, for a SISO configuration that is pointing towards ϕ = 330, the capacity achieved is higher than the worst performing MIMO configuration and far exceeds the channel capacity of omni-directional antennas in MIMO, suggesting that configuring the radiation pattern for optimal power reception is a crucial aspect in increasing channel capacity and can bring more benefit to the system than relying on diversity alone. Besides relating performance and beam angles, there is also symmetry present in the results. For instance, both configurations (210, 330) and (330, 210) produce near identical capacity CCDF curves and outage capacity since the spatial correlation between

49 Chapter 3. Simulations and System Performance Estimation SISO (270) SISO (210) SISO (330) SISO (omni) Probability (C > abscissa) Capacity (bit/s/hz) Figure 3.6: Channel Capacity CCDF, SISO, at fixed transmitting SNR. Synthesized radiation patterns are introduced into the simulations, with the corresponding beam angle in degrees shown in parentheses in the legend. the subarray elements ; similar observations can be made about (270, 330) and (330, 270) and the other pairs. Comparing to the omni-directional receiver, these first-order simulation results show that directive antennas in a MIMO system deliver a substantial gain in capacity, with most pattern configurations perform much better, and picture a promising performance for the subsequent experiments. In the best case scenario, the capacity improvement in MIMO at 90% outage probability can be as high as 71%. To find the diversity gain from the MIMO configurations, we analyze the capacity with respect to the receiving SNR where the effect of power gain has been taken out of the capacity calculation. For brevity, Figure 3.9 shows selected channel capacity CCDF results at a fixed receiving SNR for the SISO and MIMO configurations respectively. A

50 Chapter 3. Simulations and System Performance Estimation 34 Probability (C > abscissa) x2 MIMO (210, 210) 2x2 MIMO (210, 270) 2x2 MIMO (210, 330) 2x2 MIMO (270, 210) 2x2 MIMO (270, 270) 2x2 MIMO (270, 330) 2x2 MIMO (330, 210) 2x2 MIMO (330, 270) 2x2 MIMO (330, 330) 2x2 MIMO (omni) Capacity (bit/s/hz) Figure 3.7: Channel Capacity CCDF, 2 2 MIMO, at fixed transmitting SNR. Synthesized radiation patterns are introduced into the simulations, with the corresponding MIMO element beam angles in degrees shown in parentheses in the legend. complete plot can be found in Appendix A. Note that from the figure, all the SISO configurations perform equally and produce the same capacity value throughout all spatial locations at the same receiving SNR level, which is expected as the transfer functions are being normalized to the same value. The MIMO configurations on the other hand, have a wide spread range of capacity performance, starting off at the same capacity as the SISO system, to almost double the SISO capacity. This indicates that under the right situations, the MIMO system is able to exploit the additional antenna, receiving two independent versions of the signal. The omni-directional antennas in MIMO offer decent performance compare to the SISO case, but they are no match to most of the directive antenna configurations. At 90 % outage probability, 6 out of 9 directive antenna configurations are able to perform better due to higher diversity gain. It is interesting to

51 Chapter 3. Simulations and System Performance Estimation 35 Probability (C > abscissa) SISO (270) worst SISO (330) best SISO (omni) 2x2 MIMO (330, 330) best 2x2 MIMO (270, 270) worst 2x2 MIMO (omni) Capacity (bit/s/hz) Figure 3.8: Channel Capacity CCDF, SISO versus 2 2 MIMO, with only the best, worst and reference cases at fixed transmitting SNR are shown. Synthesized radiation patterns are introduced into the simulations, with the corresponding MIMO element beam angles in degrees shown in parentheses in the legend. observe the (330, 330) case which has the best performance versus transmitting SNR but the worst versus receiving SNR, indicating that pointing both the beams in the direction of arrival can obtain higher SNR due to antenna gain, however with the trade off in losing diversity gain. This finding can be extreme useful when operating in an environment when there is no dominant AOA. In such case, it may be wise to optimize the system in terms of eigenvalue spread as the diversity gain may become the dominant factor in affecting the performance of the system. Next, the associated DOF for each of the sampling locations and configurations are calculated. Figure 3.10 shows the DOF CCDF; a zoomed-in version of the plot is shown in Figure The performance ranking of the antenna configurations is essentially the

52 Chapter 3. Simulations and System Performance Estimation 36 Probability (C > abscissa) x2 MIMO (270, 210) best 2x2 MIMO (330, 330) worst SISO (210) SISO (270) SISO (330) 2x2 MIMO (omni) Capacity (bit/s/hz) Figure 3.9: Channel Capacity CCDF, SISO versus selected 2 2 MIMO, at fixed receiving SNR. The H matrices are normalized to eliminate power disparity between configurations. same as their capacity rankings, which is understandable as both figures portrait the system performance versus diversity order. Almost all of the configurations are able to achieve nearly 2 degrees of freedom at their maximum. However on average they are much lower than 2, with most numbers of DOF range between 1 and 1.1, while in the same probability range, the capacity change is significant. This indicates that a small change in the eigenvalue spread can lead to dramatic increase in performance, and there is a diminishing effect as the DOF approaches the theoretical maximum.

53 Chapter 3. Simulations and System Performance Estimation 37 Probability (DOF > abscissa) x2 MIMO (210, 210) 2x2 MIMO (210, 270) 2x2 MIMO (210, 330) 2x2 MIMO (270, 210) 2x2 MIMO (270, 270) 2x2 MIMO (270, 330) 2x2 MIMO (330, 210) 2x2 MIMO (330, 270) 2x2 MIMO (330, 330) 2x2 MIMO (omni) Degrees of Freedom Figure 3.10: Degrees of freedom CCDF, 2 2 MIMO. This metric indicates the number of channels the system effectively utilizes.

54 Chapter 3. Simulations and System Performance Estimation 38 Probability (DOF > abscissa) x2 MIMO (210, 210) 2x2 MIMO (210, 270) 2x2 MIMO (210, 330) 2x2 MIMO (270, 210) 2x2 MIMO (270, 270) 2x2 MIMO (270, 330) 2x2 MIMO (330, 210) 2x2 MIMO (330, 270) 2x2 MIMO (330, 330) 2x2 MIMO (omni) Degrees of Freedom Figure 3.11: Degrees of freedom CCDF, 2 2 MIMO. This metric indicates the number of channels the system effectively utilizes. The plot is zoomed in between 1 and 1.1 on the abscissa.

55 Chapter 4 Reconfigurable Antennas for MIMO 4.1 Motivation The indoor NLOS MIMO simulation results from Chapter 3 have shown promising MIMO capacity improvement with directive antennas over omni-directional ones. The antennas employing directive radiation patterns outperform the omni-directional antennas by as much as 75% at 90% outage probability in the selected sampling region. This gives us the motivation to pursue the study with a similar experimental demonstration utilizing directive antennas. The channel of a wireless system rarely stays stationary, and this requires the antennas to adapt their patterns to the environment in real-time to improve the instantaneous SNR and data rate. This can be achieved with pattern reconfigurable antennas. Much like a traditional antenna array where beam steering is achieved with reconfigurable phase shifters, this type of antenna employs tunable electronic components which can alter the antenna s radiation characteristics in the far-field. This allows the antenna pattern to be reconfigured remotely and quickly. Even though many reconfigurable antennas have been proposed for MIMO- and diversity-related applications [18] - [26], which are discussed shortly, few have been sim- 39

56 Chapter 4. Reconfigurable Antennas for MIMO 40 ulated in a channel, and even less have demonstrated the improvement experimentally in a lab setup [3] or in the field [19]. Therefore, the goal of our study is to physically demonstrate the throughput improvement by employing pattern reconfigurable antennas in a hardware testbed, and move the antennas across multiple locations to verify the improvements that are seen in previous channel simulations. This allows us to obtain a more generalized idea of their real-world performance. Therefore choosing the right antenna design is crucial in facilitating this demonstration. Before proceeding with the selection of an antenna design, a study of existing reconfigurable antennas proposed for MIMO applications is conducted. 4.2 Common Reconfigurable Antennas for MIMO There are several types of pattern reconfigurable antennas that have recently emerged from literature related to MIMO and diversity-related applications. While most of these antennas are designed to be compact in size and planar in structure for mobile applications, some are for conceptual demonstrations showing their potential capabilities. One common type of reconfigurable antenna utilizes PIN diodes to control the current flow within the structures, creating multiple modes when the PIN diodes are configured to different states [18] - [22]. This type of antenna usually features planar designs such as dipoles or slotted structures, which are divided into multiple sections connected with PIN diodes. By switching the PIN diodes, the current distribution on the antenna changes producing different antenna patterns. Some approaches intentionally involve the effect of mutual coupling. An example of such design is an antenna array consisting of closely spaced dipoles is introduced for adaptive MIMO applications [19]. The use of PIN diodes embedded in the array elements changes their electrical length and alters the mutual coupling between adjacent elements, and therefore the radiation patterns. More recently, microelectromechanical systems (MEMS) have seen increased adoption in this type of

57 Chapter 4. Reconfigurable Antennas for MIMO 41 antenna since its introduction as a replacement for PIN diodes [27]. They are physically small, have lower loss, lower power consumption, higher linearity and often provide better integration with substrate than conventional lumped elements, but at the expense of involving more complicated fabrication process. Another type of reconfigurable antenna also uses PIN diodes but consists of multiple antenna elements on the same substrate [23, 24]. However instead of changing their current distribution, the PIN diodes act as simple switches at the level of feeding network. Each individual antenna element has its own distinct pattern or polarization characteristics. By switching the PIN diodes on and off, a different set of elements can be selected and toggles between different radiation patterns and even different polarizations. The third type of reconfigurable antenna utilizes an input switching technique [25, 26], where the signal is fed into different input ports for selecting or creating diverse antenna patterns. Although the end result is similar to the previous type as both can be called beam switching antennas, the working principle is completely different. For example, an antenna array of four reconfigurable elements is built with PIN diodes and separate feeding ports for each element [25]. Depending on the ports that are been used and the state of PIN diodes on the corresponding elements, the antenna produces a combined radiation pattern from all active elements that varies between configurations. In another example, a novel loop antenna is proposed that is capable of producing directive beam patterns in the azimuth plane when the feed point changes along the contour of the loop [26]. The design is not very compact as the diameter of the loop and thickness of the substrate must satisfy certain design criteria, the concept is plausible for its uniqueness in the implementation of beam steering capability. The fourth type of reconfigurable antenna is built based on parasitic array and uses switches such as PIN diodes or MEMS to change the characteristics of parasitic elements [28] - [30]. The switches fundamentally work the same way as the ones in the first category, except that instead of altering the current distribution on the driven element,

58 Chapter 4. Reconfigurable Antennas for MIMO 42 they change the characteristics of the parasitic elements, and alter the far-field radiated pattern. For example, a pattern reconfigurable microstrip parasitic array uses switches to change the electrical lengths of its two parasitic elements [29], and much like the classic Yagi-Uda antenna, the parasitic elements behave either as a director or as a reflector depending on their configuration. There are other types of reconfigurable antennas. For example, a pattern reconfigurable antenna based on cylindrical electromagnetic bandgap (CEBG) structures is proposed that is capable of producing highly directive antenna patterns using multiple layers of of parasitic elements that can be configured either as a reflector to the radiated waves or as transparent elements to let the wave pass through [31]. The antenna has a 3D structure and isn t of a compact design, however it demonstrates its excellent beam forming ability and beam steering capability. 4.3 Selection Criteria for Performance Studies In order to demonstrate the usefulness of pattern reconfigurable antennas as receiver in a diversity system (SIMO/MIMO), we must select a suitable antenna to be designed and fabricated. Therefore to effectively and convincingly demonstrate the benefit of using reconfigurable antennas, the antenna design must meet the following constraints and requirements. For the purpose of confirming the simulation results, the antenna should be able to steer azimuthally. Ideally the angle separation between each pattern should be equal, so that a uniform sweep can be performed. Thirdly, the patterns produced should be similar to each other and only differ by the beam angle, eliminating as much uncertainty as possible when comparing the experimental results between various configurations. Lastly, the antenna should be easy to fabricate. The use of high complexity components such as MEMS to achieve pattern reconfigurability brings many advantages such as better

59 Chapter 4. Reconfigurable Antennas for MIMO 43 linearity and lower loss, however they also require more complex fabrication techniques. Many existing reconfigurable antennas use common lumped components such as varactor diodes and PIN diodes, and this greatly simplifies the fabrication process. 4.4 ESPAR Antenna Physical Design and Characteristics After careful considerations, the electronically steerable passive array radiator (ESPAR) antenna is chosen to be used as the receiving antennas in the physical experiments. It satisfies all the criteria listed in the previous section and is an ideal candidate for experimental demonstrations. The antenna has been well studied and documented in the literature over the past 11 years [32, 28, 33]. It consists of a single active radiator surrounded by a ring of equispaced passive elements. In this case, the active and passive elements are quarter wavelength monopole antennas, which is a common choice among ESPAR designs. The passive antennas are terminated with a passive load. The loads are primarily reactive, and their values determine the radiation pattern of the ESPAR antenna. In general, an inductively loaded element acts as reflector, and a capacitively loaded element acts as director. The number of passive elements and their associated radiation characteristics have been investigated [32]. It has been stated in literature that at least six passive elements should be used to generate a reasonable beam steering flexibility [32]. Therefore for the purpose of our study, the two ESPAR antennas used in the multi-antenna receiver are each composed of six parasitic elements. Each ESPAR antenna is able to generate six patterns with the beam angles 60 degrees apart azimuthally, covering the entire horizontal plane. The ESPAR antenna represents a design problem with many degrees of freedom. The parameters in an ESPAR design can be categorized into two kinds: structural parameters, which define its physical characteristics, and load parameters, which define its radiation

60 Chapter 4. Reconfigurable Antennas for MIMO 44 Table 4.1: ESPAR design parameters Category Parameter Description R sub R 0 Radius of substrate Radius of active monopole Structural R 1 - R 6 Radius of passive monopoles 1 to 6 H 0 height of active monopole extruding from the top of the substrate H 1 - H 6 height of passive monopoles extruding from the top of the substrate D horizontal distance from active monopole to surrounding passive monopoles Load - Z 6 Loading of parasitic monopoles #1 to #6 respectively characteristics. The parameters are listed in Table 4.1, and Figure 4.1 illustrates the orthogonal view of a six-element ESPAR antenna labeled with the design parameters. There are many design parameters. These parameters can vary in value independently resulting in an infinite number of combinations. To simplify the design procedure and to achieve omni-directional beam scanning, identical structural parameters are used for all passive elements. This reduces the total structural parameters from the original 16 to 6: R sub, R 0, R 1, H 0, H 1, and D. A second design consideration that the driven element and passive elements are identical further reduces the number of parameters down to 4: R sub, R 0, H 0, and D. To optimize the parameters, simulations are required to examine the relationship of parametric changes to radiation characteristics.

61 Chapter 4. Reconfigurable Antennas for MIMO 45 R 0 R H 0 1 H D R sub Figure 4.1: Diagram of an ESPAR antenna showing the corresponding design parameters. The driven element is highlighted in gray Monopole Simulation The basic building elements for an ESPAR antenna are the monopoles for the driven and parasitic elements. Therefore to begin the first step of the antenna design process, the parameters for the monopoles must be determined. Using Ansoft HFSS software employing finite element method, a monopole is simulated on top of a FR-4 substrate with ϵ r = 4.4, h = 1.6 mm, and a mm copper cladding. The radius of monopole, R 0, is assumed to be a fixed value of 0.5 mm representing a 12 AWG (American Wire Gauge). It is placed mm above the copper layer, with a 50Ω lumped port as excitation placed between the gap. Substrate sizes of R sub = 0.5λ, 0.75λ, 1λ are tested to determine the corresponding monopole heights and resonant input impedance at at test frequency 2.28 GHz. These parameters are used in the subsequent ESPAR simulations to determine the optimal substrate size for achieving high directivity. Table 4.2 lists the corresponding monopole heights and input impedances found at resonance corresponding to the three R sub values.

62 Chapter 4. Reconfigurable Antennas for MIMO 46 Table 4.2: Simulated monopole heights and input impedances (Ω) R sub H 0 Input Impedance 0.5λ 0.227λ j λ 0.236λ j λ 0.228λ j Single ESPAR Simulations The first-order simulations of the ESPAR are initially carried out using the method of moments (MoM), a numerical computational method that solve integral field equations and solves for currents on the conducting bodies of the antenna. This method has low computational intensity and offers the ability to easily sweep multiple parameters and is ideal for first-order simulations. The three substrate sizes R sub = 0.5λ, 0.75λ, 1λ are investigated, and the ESPAR antenna models are built based on the corresponding established monopole parameters in Table 4.2. Figure 4.2 illustrates the top view of the meshed ESPAR structure created in Gmsh, a 3D finite element mesh generation software, with R sub = 1λ. The cylindrical monopoles are modeled as equivalent strips with a width of 4R 0 and height of H 0. To aid the design process, some conventions are established. First, we number the six passive elements from 1 to 6. They can be independently loaded with a reactance. The pattern of the ESPAR antenna is largely determined by the position and value of these reactive loads, i.e., the arrangement of reflectors and directors and their corresponding inductance/capacitance. For convenience, we define 0 as an inductively loaded passive monopole or a reflector, and 1 as a capacitively loaded passive monopole or a director. The loading scheme of the ESPAR antenna can be easily represented by a 6-bit binary string. For example, represents a loading scheme of monopole number 1 to 3 being loaded with capacitive loads, and number 4 to 6 with inductive loads. To find parameters that generate desirable radiation patterns, MoM simulations are

63 Z Y X Chapter 4. Reconfigurable Antennas for MIMO 47 Figure 4.2: Top view of ESPAR structure generated and meshed using Gmsh

64 Chapter 4. Reconfigurable Antennas for MIMO 48 conducted to determine the effects of ground plane size, the proportion of reflectors and directors (reflector-to-director ratio) and their corresponding loads at the base of the parasitic monopoles on the antenna pattern. The detailed procedures are documented in Appendix B, and from the results it is concluded that the best pattern is obtained with 1:1 reflector-to-director ratio, a substrate radius of R sub = 0.75λ, and loading values Z r = j30ω at the base of the reflectors and Z d = 300jΩ at the base of the directors Dual ESPAR Simulations While the MoM is ideal for simulations that require variable sweeping, it is not very suitable for simulations where realistic physical properties are involved. With the parameters established, a more realistic model of the ESPAR antenna can be built and simulated using the Ansoft HFSS software. HFSS provides a graphical user interface and the inclusion of material properties and variable based model structures makes it is a very powerful EM simulation package. The first step is to verify the parameters obtained with MoM. The antenna model is built identically to the one from MoM with six equidistant passive monopoles surrounding the driven element at a distance of 0.25λ at 2.28 GHz. Figure 4.3 illustrate the top and side views of the model. Following previous conventions, each monopole is numbered from 1 to 6, as shown in Figure 4.3(a). The loading scheme is used, where monopoles numbered 1, 2, and 6 are loaded with Z r, and monopoles numbered 3 to 5 are loaded with Z d. To add reactance to the input impedance of the passive monopoles in the simulation, six lumped RLC surfaces are placed vertically between the bottom of the monopoles and the top of ground plane surface. The lumped RLC condition in HFSS simulates a simple parallel RLC circuit. With Z r and Z d being purely reactive, the lumped RLC is programmed to represent Z r with an equivalent inductance, and Z d with an equivalent capacitance at 2.28 GHz. The radiation pattern, directivity, input impedance, and S 11 are examined from simulation results. The HFSS models produce similar but not identical radiation patterns

65 Chapter 4. Reconfigurable Antennas for MIMO 49 y z x x (a) Top view (b) Side view Figure 4.3: Simulation model of the ESPAR in HFSS to previous results obtained using the MoM. The normalized E θ patterns in the θ = 90 plane are compared in Figure 4.4. Note that the patterns are the most different at R sub = 1λ; decreasing Z r by j10ω brings the results a lot closer to that obtained using the MoM. Using R sub = 0.75λ consistently produces narrower main lobe beamwidth with comparable back lobe level, giving it a slight advantage over R sub = 1λ. Figure 4.5 shows its gain pattern for the θ polarization in constant ϕ and θ cuts. Due to the finite ground size, the pattern has its maximum gain of 8.8 db tilted slightly above the horizontal plane at θ = 55 and has a significant amount of radiation below ground plane. This fan shaped beam can be a desirable characteristic when using the antenna as a receiving antenna in the MIMO system since it has nearly equal gain in θ direction where reflections off the ceiling and floor may contribute signicantly to the receiver diversity. The back lobe level is well-controlled, and the front-to-back ratio is generally greater than 15dB. The input impedance at 2.28 GHz is j29.811ω, resulting in a S 11 of db; a matching

66 Chapter 4. Reconfigurable Antennas for MIMO 50 0 Normalized E θ θ=90 o plane θ MoM, R sub =0.75λ, =30j, Z 2 MoM, R sub =1λ, =50j, Z 2 HFSS, R sub =0.75λ, =30j, Z 2 HFSS, R sub =1λ, =50j, Z 2 HFSS, R sub =1λ, =40j, Z φ Figure 4.4: Simulated radiation pattern of ESPAR in HFSS, R sub = 0.75λ network is required to achieve an acceptable matching level. With the intention of using a pair of ESPAR antennas as receivers, they will be placed as close as 0.5 λ 0 between the driven elements. This introduces mutual coupling between the antennas. In order to study the effect of coupling on the antenna pattern, a model of dual ESPAR antennas sharing the same substrate and with 0.5 λ 0 spacing between the driving elements is built in HFSS. In order to generate identical set of patterns from both ESPAR antennas, there are mainly two ways to place them. Based on the geometry shown in Figure 4.3(a), one way is to place them side by side along the x-axis at 0.5 λ 0 apart. A problem arises when one passive monopole must be shared between the ESPAR antennas, making some configuration combinations impossible to realize. The other way is to place them along the y-axis, and this way there is enough space to retain all the parasitic elements of the two antennas. The dual-espar model in the simulation is built using the second placement method, and is illustrated in Figure 4.6. The passive

67 Chapter 4. Reconfigurable Antennas for MIMO 51 Radiation Pattern Curve Info db(realizedgaintheta) Freq='2.28GHz' Theta='70deg' db(realizedgaintheta) Freq='2.28GHz' Theta='80deg' db(realizedgaintheta) Freq='2.28GHz' Theta='90deg' db(realizedgaintheta) Freq='2.28GHz' Theta='100deg' db(realizedgaintheta) Freq='2.28GHz' Theta='110deg' driven element reflector director (a) Constant θ cuts Radiation Pattern 0 30 Curve Info db(realizedgaintheta) Freq='2.28GHz' Phi='90deg' db(realizedgaintheta) Freq='2.28GHz' Phi='120deg' db(realizedgaintheta) Freq='2.28GHz' Phi='150deg' db(realizedgaintheta) Freq='2.28GHz' Phi='180deg' driven element reflector director -180 (b) Constant ϕ cuts Figure 4.5: Gain patterns of HFSS simulated ESPAR antenna with 3 directors and 3 reflectors

68 Chapter 4. Reconfigurable Antennas for MIMO 52 Table 4.3: Dual ESPAR setup test configurations Test case ESPAR no. 1 configuration ESPAR no. 2 configuration monopoles are numbered from 1 to 12 as illustrated. In the simulation setup, ESPAR no. 1 is excited by a 50 Ω lumped port, while no. 2 is terminated with a 50 Ω lumped RLC. Each antenna has its own six possible loading scheme based on circular shifting scheme described earlier, and two antennas amount to a total of 36 possible combinations. Not all patterns need to be tested. Firstly due to the lab that the antennas will be placed in, the back wall is covered with absorbers preventing any reflections from that direction reaching the antennas, therefore only 180 degrees of beam coverage needs to be tested. Secondly, due to the symmetrical patterns the two identically-modeled ESPAR antennas generate, only one antenna needs to be analyzed, and this helps to further reduce the number of test cases down to six. They are listed in Table 4.3. Figure 4.7 shows the radiation patterns of the six test cases. What can be observed from test cases 3, 5, and 6 is the effect of mutual coupling between the antennas. Due to their close physical placements, in certain situations the parasitic elements of the neighbouring antenna may cause undesirable interference in the radiation pattern. For example, in test case 3, the coupled pattern in Figure 4.7(b) shows a shift of 30 in main beam direction. This can be explained by the antenna configurations, where monopole no. 7, 11, 12 acts as reflectors in ESPAR no. 2. Monopole no. 11 is so closely placed

69 Chapter 4. Reconfigurable Antennas for MIMO 53 ESPAR No ESPAR No. 1 y x Figure 4.6: Simulation model of the dual ESPAR in HFSS

70 Chapter 4. Reconfigurable Antennas for MIMO 54 to one of the directors of ESPAR no. 1, it reflects the field directed by monopole no. 1 and therefore creates a pattern equivalent to that of ESPAR no. 1 being configured as In test case 5, the shift of main beam is not as pronounced as in test case 3, however the back-lobe level increases due to a similar reason. Even though the close spacing takes a toll in the front-to-back ratio in this case, the simulated antenna pattern remains usable. Test case 6 represents the worst-case scenario, where the radiated waves from ESPAR no. 1 directly hit and bounce off the reflectors from ESPAR no. 2, resulting in reduced gain and a broadened and shifted main beam. At below 20 db, the S 21 is very unlikely to have any contribution to the pattern distortions. Similar scenarios to test case 6 can be avoided so long the antennas are placed further apart. However in order to keep the the driven inter-antenna spacing at 0.5λ 0, other solutions need to be sought out. The proposed solution is to change the ESPAR no. 2 configuration to Such a configuration uses monopole no. 11 as a director and therefore avoids the issue previously mentioned in test case 6. There will be a slight shift in the main beam angles of both ESPAR antennas, however with this small compromise both antennas can now produce very usable patterns. Figure 4.8 compares the radiation pattern of both ESPAR antennas against the ideal configuration (011100). With the problem solved and all radiation patterns simulated and verified, the dual ESPAR antennas is ready for fabrication Fabrication The dual-espar configuration is fabricated on a 1.6 mm double-sided copper cladded FR-4 substrate. The top cladding acts as the ground and also hosts the pads for the tunable loads; the bottom cladding is used for forming DC biasing traces. On the top layer, a small circular area surrounding each passive monopole is milled out to accommodate the load. The tunable load for each passive monopole is realized by using a PIN diode. PIN

71 Chapter 4. Reconfigurable Antennas for MIMO 55 Gain θ (db) Gain Pattern For Configuration (000111, ) 40 ideal 45 coupled φ Gain θ (db) Gain Pattern For Configuration (000111, ) 40 ideal 45 coupled φ (a) Test case 1 (b) Test case 2 Gain θ (db) Gain Pattern For Configuration (001110, ) 40 ideal 45 coupled φ Gain θ (db) Gain Pattern For Configuration (001110, ) 40 ideal 45 coupled φ (c) Test case 3 (d) Test case 4 Gain θ (db) Gain Pattern For Configuration (011100, ) 40 ideal 45 coupled φ Gain θ (db) Gain Pattern For Configuration (011100, ) 40 ideal 45 coupled φ (e) Test case 5 (f) Test case 6 Figure 4.7: Radiation pattern of coupled dual ESPAR antennas. Red dotted lines represent the ideal (uncoupled) patterns, blue solid lines represent the coupled patterns.

72 Chapter 4. Reconfigurable Antennas for MIMO Gain Pattern For Configuration (011100, ) Gain θ (db) ideal 45 coupled, ESPAR1 coupled, ESPAR φ Figure 4.8: Radiation pattern of coupled dual ESPAR antennas in modified test case 6. The red dotted line represents the ideal (uncoupled) pattern, the blue solid line represents the coupled pattern of ESPAR no. 1, black solid line represents the coupled pattern of ESPAR no. 2

73 Chapter 4. Reconfigurable Antennas for MIMO 57 RP LS RS LS CR (a) Forward biased (b) Reverse biased Figure 4.9: Equivalent circuit of a PIN diode under forward and reverse bias conditions. When forward biased it has a positive reactance; when reverse biased, the reactance becomes negative. diodes are the ideal candidates as they possess the ability of switching sign of their reactive impedance based on the DC biasing voltage. The simple equivalent circuits of a PIN diode under forward and reverse bias is shown in Figure 4.9: when forward biased, the PIN diode can be represented by a small resistor in series with an inductor, thus providing a positive reactance; when reverse biased, the PIN diode can be represented by a small capacitor in parallel with a large resistor, and then in series with an inductor. The capacitance is small, resulting in a large negative reactance that overwhelms the inductance and make the passive element a director. The PIN diode of choice is a SMP LF by Skyworks. When forward biased, it has a series resistance of R S = 3.5Ω at I f = 1 ma. When reverse biased, it has a capacitance of 0.14 pf at V R = 1 V. The diode has a series inductance of L S = 0.45 nh, which translates to a reactance of j6.44ω. In order to reach the desired value of Z r = j10ω, an additional series inductor of 0.3 nh is added. This additional inductor has very little effect on Z d when its negative reactance is dominated by the capacitance of the reverse biased PIN diode. The DC biasing traces are etched on the bottom layer of the ESPAR. The DC path includes an RF choke that isolates DC bias point from RF signal, and a series resistor of 1000 Ω to limit the DC current flowing through the PIN diodes. Figure 4.10 shows the loading and biasing scheme for each passive monopole.

74 Chapter 4. Reconfigurable Antennas for MIMO 58 monopole pin diode inductor GND plane FR-4 dielectric 4 mm Top layer DC input point resistor RF choke GND plane Bottom layer FR-4 dielectric 10 mm antenna RF choke 0.3 nh 1000 Ω control Simplified equivalent circuit Figure 4.10: Loading and biasing scheme for the passive monopoles. The loads are on the top layer, while the DC biasing traces are on the bottom layer.

75 Chapter 4. Reconfigurable Antennas for MIMO 59 Figure 4.11: A photo of the fabricated duo-espar antennas on a single FR-4 substrate. The two RJ45 ports used to control the diode voltages are visible in the photo at the far edge of the substrate and pointing away from the camera. The monopoles are constructed from 18 gauge (roughly 1 mm) solid copper wires. Each is cut to the length of 0.236λ above the ground plane. A photograph of the fabricated antenna is shown in Figure To bias the twelve diodes properly, long conductors need to be routed from the voltage sources to the bias points on the antenna. Two RJ45 ethernet ports are integrated into the antenna substrate, each provides connectivity of up to eight distinct channels. Standard RJ45 cables can carry sufficient current to the PIN diodes, and are fully replaceable in case a longer cable is required to travel to the antennas. Two distinct biasing voltages are supplied to the diodes, one positive and one negative, and a manual biasing scheme is used.

76 Chapter 4. Reconfigurable Antennas for MIMO Evaluation The antennas are tested to verify the design and evaluate their performance. Similar to the simulations, the antenna testing is deliberately limited to the three configurations that cover the half plane away from the back wall for each of the ESPAR antennas, resulting in 9 combinations listed in Table 4.4. As a convention, these 9 combinations are denoted C1 to C9 in the remaining sections of this chapter. Table 4.4: List of ESPAR Test Configurations and S-parameter Measurements; Configurations are denoted from C1 to C9 ESPAR1 ESPAR2 Combination# Config. Beam Angle S 11 (db) Config. Beam Angle S 22 (db) S 21 (db) C ϕ = ϕ = C ϕ = ϕ = C ϕ = ϕ = C ϕ = ϕ = C ϕ = ϕ = C ϕ = ϕ = C ϕ = ϕ = C ϕ = ϕ = C ϕ = ϕ = Note that some configurations are slightly adjusted from their original intended value to compensate interference from passive elements of the neighbouring ESPAR. This is due to the considerations mentioned in Section 4.4.4, and in practice provides the closest pattern and main-beam direction to the originally intended configuration without any severe pattern distortions observed in previous simulations. The far-field patterns of the fabricated ESPAR antennas are measured in an anechoic chamber. Both ESPAR antennas are tested for the nine configuration combinations listed. Overall the measured patterns are very close to the simulated results, and in some

77 Chapter 4. Reconfigurable Antennas for MIMO Gain (db) ESPAR No.1 ESPAR No.2 25 ESPAR No.1 Sim ESPAR No.2 Sim φ (degree) Figure 4.12: Far-field radiation patterns of the dual-espar antennas Gain (db) ESPAR No.1 35 ESPAR No.2 40 ESPAR No.1 Sim ESPAR No.2 Sim φ (degree) Figure 4.13: Far-field radiation patterns of the dual-espar antennas Gain (db) ESPAR No.1 ESPAR No.2 25 ESPAR No.1 Sim ESPAR No.2 Sim φ (degree) Figure 4.14: Far-field radiation patterns of the dual-espar antennas

78 Chapter 4. Reconfigurable Antennas for MIMO Gain (db) ESPAR No.1 35 ESPAR No.2 ESPAR No.1 Sim 40 ESPAR No.2 Sim φ (degree) Figure 4.15: Far-field radiation patterns of the dual-espar antennas Gain (db) ESPAR No.1 ESPAR No.2 ESPAR No.1 Sim ESPAR No.2 Sim φ (degree) Figure 4.16: Far-field radiation patterns of the dual-espar antennas Gain (db) ESPAR No.1 ESPAR No.2 ESPAR No.2 Sim ESPAR No.1 Sim φ (degree) Figure 4.17: Far-field radiation patterns of the dual-espar antennas

79 Chapter 4. Reconfigurable Antennas for MIMO Gain (db) ESPAR No.1 ESPAR No.2 25 ESPAR No.1 Sim ESPAR No.2 Sim φ (degree) Figure 4.18: Far-field radiation patterns of the dual-espar antennas 0 10 Gain (db) ESPAR No.1 ESPAR No.2 40 ESPAR No.1 Sim ESPAR No.2 Sim φ (degree) Figure 4.19: Far-field radiation patterns of the dual-espar antennas Gain (db) ESPAR No.1 ESPAR No.2 ESPAR No.1 Sim ESPAR No.2 Sim φ (degree) Figure 4.20: Far-field radiation patterns of the dual-espar antennas

80 Chapter 4. Reconfigurable Antennas for MIMO 64 configurations produce even narrower beamwidth and higher front-to-back lobe level than the simulations. Some configurations produce patterns with slight beam shifts, but they are part of configuration considerations that have been discussed extensively in Section and also shown in the simulated results. The gain from the measurements are in general comparable to those in the simulations, only lower by 1 db on average which is likely to be caused by the higher loss in the lumped elements. The S-parameters of the antennas are measured using a network analyzer. Table 4.4 summarizes the findings. Inspecting the results, it is clear that the S 21 between the two closely spaced ESPAR antennas are in general very low with all but one configuration above -15 db. On the other hand, the return loss of the antennas are not very good in comparison, a properly designed matching network can help mitigate this problem. However, they can be omitted since return loss has no direct affect on the radiation patterns but rather the power gain measured at the output of the antenna, this greater return loss can be fully compensated in post-processing stage during experimental performance evaluations.

81 Chapter 5 Experimental Results 5.1 Test Methodology Existing Evaluation Methods in Literature With the reconfigurable antenna fabricated, experimental demonstration become possible. To measure the performance of a wireless system, many methods can be used. We examine the most commonly used methods, and based on the most suitable one, a hardware testbed is to be constructed for conducting indoor measurements. Capacity evaluation is a commonly used method of evaluating MIMO performance. There are many examples using this method[3, 34, 35, 36]. One common approach to this type of evaluation is to use channel simulations. Some use statistical models in the simulations, such as an ideal Rayleigh fading channel; while others use more deterministic models that mimic real-world environments [3]. This approach can be useful in analyzing the general performance of a system, and from Section 3 it has proven to be a useful tool in predicting possible performance gains from using reconfigurable antenna patterns. Another approach is to measure the channel response in a real channel, and calculate the Shannon capacity. Single turn-key solutions exist that can test, post-process and calculate the channel capacity [36]. Examples of cheaper solutions that do not involve 65

82 Chapter 5. Experimental Results 66 purchasing specialized MIMO capacity measurement equipment can also be found in literature [3, 34], where a network analyzer is used to measure the channel responses between TX and RX, and the data is imported directly into a computer to compute the channel capacity. This method can quickly evaluate the upper bound of a system in the given channel, including effects from antenna gain, mutual coupling, multipath propagation and path loss. BER, which was previously discussed in Section 2.2, has also been used as a method for measuring the performance of wireless systems including MIMO systems [37, 38], as well as for optical links and other types of communication systems. Measuring BER is a practical method in evaluating the performance and capacity of a communication system. Instead of statistically analyzing the wireless channel and system, it sends real bit streams and measures the number of bit errors at the receiving end. This involves many factors such as interference conditions, channel noise, distortion, path loss, wireless multipath fading, antenna gains, and receiver noise which are all reflected in the BER results. A better performing system will improve the overall signal quality by improving one or more of the above factors and produce fewer bit errors BER as a Method of Evaluation For our study on MIMO and reconfigurable antennas, we want a robust way of measuring the real-world performance of a wireless communication system. Channel capacity evaluations may provide a relatively easy and quick way of gauging the performance, however it represents the theoretical upper limit of the system throughput and does not have the same degree of realism BER measurements would provide as factors such as noise and interference that would be reflected in the BER measurements may not necessarily affect the capacity value. However, since BER measurements deal with sending and receiving real bit streams, they are much more time consuming and computationally expensive than simply measuring the channel matrix; careful implementation on the

83 Chapter 5. Experimental Results 67 testbed conducting the measurements is required. 5.2 BER Testbed In order to conduct BER measurements using various antenna configurations including the ESPAR antennas, a testbed is proposed and implemented as the testing platform. Details are given on the signal flow of the testbed, as well as its hardware implementation Testbed Signal Flow Due to the scope of this thesis, the testbed is intended to be used for testing a wireless system consisting of up to two transmitting antennas and two receive antennas. As such, it is designed from the ground up with these constraints in mind. At the transmitter, the signal flow starts with the generation of the signal that needs to be sent wirelessly over the air to the receiver. The signal consists of a sequence of uniformly distributed random numbers of 1 s and -1 s. In a SISO or SIMO system, a single transmitting antenna is used to send out all the bits sequentially. Quadrature phase-shift keying (QPSK) modulation is performed on the bit streams. The bits are separated into the odd bits and even bits. The odd bits form the in-phase (I) data waveform, and the even bits form the quadrature (Q) data waveform. Each waveform is pulse-shaped to occupy a finite and limited bandwidth in the channel, and subsequently modulated to a carrier frequency that is within the available bandwidth of the antennas under test (AUT) and less prone to interference inside the channel where testing is conducted. The pulse-shaped I and Q waveforms are combined to create the final waveform that is sent through the transmitting antenna. This allows two bits to be sent per symbol. Figure 5.1 and Figure 5.2 show the complete signal flow diagram of the proposed transmitter with a single transmitting antenna and with dual transmitting antennas respectively. In MISO and MIMO systems, a space-time block coding scheme is to be employed

84 Chapter 5. Experimental Results 68 bit stream even/ odd odd even pulseshaper I x 0 LO@2.28 GHz x Q I/Q modulator TX Figure 5.1: TX Pipeline Configuration with Single Transmitting Antenna bit stream coder I/Q mod I/Q mod I Q I Q pulse-shaper + TX1 TX2 + power amp Figure 5.2: TX Pipeline Configuration with Dual Transmitting Antennas

85 Chapter 5. Experimental Results 69 [c1 c2] TX pipeline 1 original bit stream even /odd c1 c2 -c2* c1* Alamouti Coder [-c2* c1*] TX pipeline 2 Figure 5.3: Implementation of Alamouti coding at the transmitter in MISO/MIMO testbed for the system to efficiently take the advantage of the multiple transmitting antennas; Alamouti coding is used for its simple and effective implementation. Details on the coding scheme has been described in Section Figure 5.3 shows how Alamouti coding can be implemented in a MISO/MIMO testbed at the transmitter. At the receiver, the signal flow starts with capturing the transmitted signal with the receive antenna. A bandpass filter that has a center frequency that matches the carrier frequency and bandwidth corresponding to that of the transmitted waveform is attached to the output of the antenna to provide strong attenuation of interference outside of the band that the test signal occupies. The filtered signal is then amplified and down-converted from the carrier frequency back to the form of separate I and Q data streams (I/Q demodulation). Phase detection is required to correct any rotation in the modulation constellation as a result of propagation in the channel and cables. The phase-corrected I and Q data streams are passed through filters matched to the transmit pulse shape to determine the bits received. It performs cross-correlation of the received data stream with the pulse that is used in pulse-shaping the original signal at the transmitting end. The outcome is a bit stream that contains either positive or negative numbers. To convert into binary stream of ±1, the result is compared to a threshold set at zero; any number that is greater than 0 becomes a 1, and any number less than 0 becomes a -1. The binary bit stream can be then directly compared to the original bit stream to determine the number of bit errors and therefore the error percentage or BER.

86 Chapter 5. Experimental Results 70 received signal + cos x LO@2.28 GHz x sin LPF downconverter I Q Timing Recovery & Channel Estimation I_corrected Q_corrected matched filter decode & compare BER Figure 5.4: RX Pipeline Configuration for Single Receiving Antenna downconverter received signal 1 received signal x 0 LO@2.28 GHz -90 x x 0 LO@2.28 GHz -90 x LPF LPF downconverter I Q I Q Timing Recovery & Channel Estimation Timing Recovery & Channel Estimation I_corrected Q_corrected matched filter I_corrected Q_corrected matched filter decode & compare BER Figure 5.5: RX Pipeline Configuration for Dual Receiving Antennas With two receive antennas, the maximal ratio combining (MRC) technique is used to increase the receive diversity. As introduced in Section 2.1.2, it requires knowledge of the relative amplitudes between receivers. This knowledge can be obtained during the step of phase-correction following the I/Q demodulation, by performing cross-correlation of received waveform with a training sequence that is known a priori. The complex response of the cross-correlation gives information on the phase and power of each branch signal so that they can be co-phased and weighted according to the MRC algorithm, respectively. Figure 5.4 and Figure 5.5 show the complete signal flow diagram of the proposed receiver with a single transmitting antenna and with dual transmitting antennas respectively. For MISO/MIMO system testing, a decoder is used at the receiver to correctly decode and combine the signal transmitted using Alamouti coding for BER analysis. Figure 5.6

87 Chapter 5. Experimental Results 71 Alamouti Decoder RX pipeline 1 converted bits even (C2') /odd (C1') R1= C1' + C2'* R2= C1' - C2'* even(r2) /odd (R1) MRC compute BER RX pipeline 2 converted bits even (C2') /odd (C1') R1= C1' + C2'* R2= C1' - C2'* even(r2) /odd (R1) decoded bits Alamouti Decoder Figure 5.6: Implementation of Alamouti coding at the receiver in MIMO testbed shows how Alamouti coding for a MIMO testbed can be implemented. The signal flow differs from SIMO s slightly by introducing Alamouti decoders after recovering the bits and before performing MRC Hardware Realization for SIMO The proposed signal flow in Section is implemented into a hardware testbed and is documented in this section. Due to limited hardware components available, the implemented transmitter can only transmit with one antenna at any given time. For MISO or MIMO testing where multiple antennas are required, the hardware testbed resorts to achieve a combined effect of multiple antennas in post-processing. The method is based on the concept of a virtual antenna array. By physically relocating the sole transmitting antenna between two designated locations, it emulates a two-antenna array with only one physical antenna. The effect of mutual coupling between the transmitting antennas can be approximated by placing a dummy element terminated in matched load close to the transmitting element, to mimic the presence of the second transmitting antenna. However this is not done as it is not part of the investigation in this thesis. Table 5.1 lists all the hardware components used to build the BER testbed. In this section, the detailed implementation of the BER testbed for SIMO processing

88 Chapter 5. Experimental Results 72 Table 5.1: List of BER Testbed Hardware Components Equipment Qty Details computer 1 installed with Matlab TX Agilent PSG 1 E8267D Vector signal generator Agilent AWG 1 N8241A Arbitrary Waveform Generator power amplifier 1 30 db gain, max output 1W bandpass filter (BPF) 2 passband: 2250MHz MHz low noise amplifier (LNA) 2 20 db gain, max output 20 dbm 2 way power splitter 2 Mini-Circuits ZX , down-converter RX component frequency mixer 4 Mini-Circuits ZX05-C42+, down-converter component lowpass filter (LPF) 4 Mini-Circuits VLF-180+, down-converter component 2 way-90 power splitter 2 Mini-Circuits ZX10Q-2-27+, down-converter component digital oscilloscope 1 Tektronix TDS5104

89 Chapter 5. Experimental Results 73 bit stream (rand binary) coder even/odd computer/matlab odd even/odd even odd even pulse-shaper 1 2 Figure 5.7: SIMO TX Pipeline Configuration Part 1 consists functions implemented in Matlab script executed from a computer. The pipeline marked 1 and 2 continue to part 2 in Figure 5.8. is given. Utilizing available hardware instruments, the testbed can fully accommodate one transmitting antenna and two receiving antennas at the same time. The core of this testbed is a laptop running Matlab equipped with the Test and Measurement toolbox. With proper drivers installed, the toolbox can control all the hardware instruments via local area network connection or general purpose instrumentation bus (GPIB) interface. Figure 5.7 depicts the first half of the transmitter pipeline, where functions are implemented in Matlab script fully executable on a computer. Figure 5.8 shows the rest of the pipeline that utilizes hardware instruments fully controlled by Matlab script via the Test and Measurement toolbox and LAN interface. The transmitter pipeline starts with a sequence of uniformly distributed pseudorandom binary bits generated using Matlab. Every bit in the sequence is then pulseshaped in Matlab by multiplying it to a raised cosine pulse shape that has been truncated to a finite length and connect them sequentially without overlapping to form the waveform. A raised cosine pulse shape can be generated in time-domain with equation: h(t) = sinc( t T ) cos( πβt T ) 1 4β2 t 2 T 2 (5.1) where there are two variables: β, the roll-off factor, and T, the reciprocal of the symbol-

90 Chapter 5. Experimental Results 74 1 or 2 quantizer Agilent AWG I x GHz x Q Agilent PSG TX Figure 5.8: SIMO TX Pipeline Configuration Part 2 consists of hardware instruments which are used to realize and modulate the digital waveforms generated by Matlab script, fully controlled by Matlab script via imbedded Matlab toolbox and LAN interface rate. The raised cosine pulse shape used is generated with β = 1 and a T = 3 f br, where f br is the transmitted symbol rate, here 10 MS/s. This symbol rate guarantees that the intersymbol interference (ISI), which is the distortion of a signal created when one symbol interferes with the subsequent ones, is fully prevented. It reaches the requirement such that the symbol rate at the transmitter must be limited to no more than twice the channel bandwidth [39]. Given that the bandwidth of the signal is about 50 MHz, and the chosen signal rate in the testbed is 10 MS/s, this creates a flat-fading channel, a channel that is non-frequency selective, where there is essentially no ISI. Using QPSK modulation, it splits the bits into the I data stream containing the odd bits, and the Q data stream containing the even bits. The I and Q waveforms are transferred into the Agilent N8241A Arbitrary Waveform Generator (AWG) via a LAN connection, where they are physically synthesized by the AWG at a sampling rate of 1.25 GHz and output to the Agilent E8267D PSG Vector Signal Generator via two shortlength SMA cables at a symbol rate of 10 MSps with a resolution of 125 bits per symbol. The Agilent PSG modulates the I/Q waveform onto a 2.28 GH carrier, and sends out the modulated QPSK waveform to the next stage of the transmitter. The waveform is amplified first by a power amplifier attached before the transmitting antenna, and then

91 Chapter 5. Experimental Results 75 input waveform x GHz LO 0 x LPF I channel data stream Q channel data stream Figure 5.9: Illustration of Assembled I/Q Demodulator broadcast. At each of the two receiver chains, a receiving antenna captures the transmitted waveform, along with possible interference and sky noise. A carefully chosen bandpass filter is attached to the output to attenuate any interference outside of the desirable band. A low-noise amplifier (LNA) is used to boost the signal amplitude so there is sufficient signal headroom to work with during post-processing stage. The filtered and amplified signal then gets sent into a down-converter or I/Q demodulator. Each down-converter is a simple RF circuit composed of six individual components: one power splitter, one 90 degree power splitter, two mixers, and two low-pass filters. It is assembled according to Figure 5.9. The splitter splits the received signal equally into two streams that are fed into both I and Q branches of the down-converter and down-converts back to DC from the carrier frequency. A separate signal generator, a Rohde & Schwarz SMIQ 03B Signal Generator, supplies the required sinusoidal local oscillator (LO) input to the mixers, coupled by a 90 degree power splitter to give the I branch a phase advancement of 90 degrees over the Q branch as required to perform the down-conversion. A digital oscilloscope captures the down-converted waveforms, and sends them via a GPIB interface to the computer running Matlab for in-software post-processing. Figure 5.10 illustrates the hardware receiver pipeline described thus far. During the post-processing stage, the Matlab script first converts the ASCII streams from the oscilloscope into vector of numbers. It then performs a circular cross-correlation

92 + Chapter 5. Experimental Results 76 I/Q demodulator RX1 x 0 BPF LNA x -90 LPF signal generator 2.28 GHz GPIB interface 0 x RX2 BPF LNA x LPF I/Q demodulator amp sampler buffer digital oscilloscope Figure 5.10: RX Pipeline Configuration Part 1 consists of functions realized by hardware instruments and the pipeline continues in Figure 5.11 on the demodulated I/Q complex waveform against the known transmitted waveform, and the result is a complex waveform that contains information to the relative signal amplitude, symbol offset and phase offset. The offset values are used in the later stage to align the QPSK constellation. Equation 5.2 shows how circular cross-correlation is performed: B = F 1 [F(S received ) F(S ideal )] (5.2) where the asterisk denotes the complex conjugate, F is the Fourier transform, F 1 is the inverse Fourier transform, S received is the received demodulated waveform, S ideal is the ideal waveform, and B is the result of the cross-correlation in complex form. The functions fft and ifft in matlab are used to implement the Fourier transforms and inverse Fourier transforms in Equation 5.2. After the timing and phase recovery, matched filtering is performed in Matlab to decode the waveform into bit streams. The waveform is sampled at a the frequency of

93 Chapter 5. Experimental Results 77 I from RX1... Q I from RX2... Q GPIB interface Timing Recovery & Channel Estimation Timing Recovery & Channel Estimation I_corrected Q_corrected I_corrected Q_corrected matched filter decode & compare BER computer/matlab Figure 5.11: RX Pipeline Configuration Part 2 continuing from Part 1 in 5.10, consists of post-processing steps implemented in software the intended symbol rate of 10 MS/s with an initial delay equal to the amount of time found during the timing recovery phase. In a SISO setup, where the system has only a single transmitter and a single receiver, the extracted bits from the receiver are compared directly to the original bits generated to determine the percentage of bit errors. In a SIMO or MIMO system with two receiving antennas, the bit streams from each antenna are first combined using MRC technique before BER is determined. With MIMO, it takes one additional step prior to MRC to re-combine the waveform received from both antennas using Alamouti coding scheme. Figure 5.11 demonstrates the remaining receiver pipeline that mostly relies on post-processing using a computer and Matlab software. The assembled testbed is shown in Figure In the figure, from left to right, and top to bottom, are: 1. Agilent N8241A Arbitrary Waveform Generator (AWG) 2. Agilent E8267D PSG Vector Signal Generator 3. Rohde & Schwarz SMIQ 03B Signal Generator, as LO for IQ demodulation 4. Tektronix TDS5104 Digital Phosphor Oscilloscope All the instruments are synchronized to 10 MHz reference clock generated by the

94 Chapter 5. Experimental Results 78 Figure 5.12: The hardware implementation of BER Testbed with full-blown equipments at its current development stage. Note the two green cables are the outputs from receiving antennas, and there are four channels going into the digital oscilloscope representing the I and Q channels of both antennas. Agilent E8267D PSG through coaxial cables. Spatial averaging of the BER is performed as the last step in the testbed, in order to find and compare the diversity gain of different antenna configurations. Multiple BER curves are captured at different spatial locations and they are averaged to construct the final BER curve. In order to accomplish the spatial relocation of the antennas accurately and efficiently, they are to be mounted on the positioner shown in Figure The positioner is installed inside BA8177 and can move freely in a predefined vertical plane of 1.6 m in width and 1.6 m in height. Initial tests have shown little difference between the results captured at different heights, therefore for the experimental demonstration, all

95 Chapter 5. Experimental Results 79 results are measured at a constant height of 0.5 m. The positioner can be automatically controlled via a LAN interface by calling a script from Matlab Practical Implementation Issues and Proposed Solutions There are specific issues that may occur when the testbed is used for real-world testing in a realistic channel. Some are related to the nature of wireless propagation in a channel, while some are related to hardware limitations. Regardless of their nature, they must all be addressed before the testbed can be used reliably. The first issue is related to performing accurate spatial averaging of BER. Before going into further details, understanding of the limitation on the minimum BER the testbed is capable of measuring must be established. There are many hardware-imposed limitations. For instance, the digital oscilloscope can capture at most bits at a time, the transfer speed of the GPIB interface between the scope and the computer is very slow, and the computer running Matlab requires a finite amount of time to execute the post-processing functions in the script. The run times at each stage of the testbed add up to a significant amount, and the testbed will require slightly over 30 seconds of execution time to complete capture every bits and subsequent post-processing. For the BER versus SNR curve to be statistically meaningful, at least 100 bit errors have to be accumulated at each SNR point. As BER lowers, the total number of bits that require to be captured to accumulate at least 100 bit errors grows. Since this process of capturing and post-processing is very time consuming, the testbed would be designed such that it will stop after a point whenever the time spent at this transmit power level exceeds a certain threshold. A reasonable threshold is 3 minutes, or 5 captures at one power level, which is equivalent to capturing bits. This threshold sets the limit of the lowest BER the testbed can possibly capture to be on the order of This BER may be sufficiently low when looking at individual BER curves, however it becomes a problem when spatial averaging is performed. During spatial averaging, the BER at each

96 Chapter 5. Experimental Results 80 Figure 5.13: The positioner used as part of the testbed, manufactured by Nearfield Systems Inc. The positioner can cover a vertical plane of 1.6 m horizontally and 0.5 m vertically during testing.

97 Chapter 5. Experimental Results 81 transmit power level is averaged according to number of spatial locations, requiring every BER curve to span the same range of transmit power levels. Different spatial locations produce different BER curves, meaning that the minimum BER is achieved at different power levels for each curve. There are certain spatial locations where the minimum BER is reached at lower powers, meaning that when such curves are averaged with BER curves from other spatial locations, there will be insufficient data at the higher power levels to form an accurate average value. This leads to BER curves with poor accuracy at high power levels. One way of solving this problem is to extend the threshold of run time. It will capture more BER data, however the run time will increase exponentially with linear increase in power and quickly becomes unacceptable. Simple interpolation of missing data points can be performed by calculating and adding ideal BERs beyond the point where measurements ceased, but this type of interpolation does not bear any realism in the numbers added. A technique by using both interpolation and capturing additional BER data is proposed here and implemented in the testbed. This technique involves stitching sections of real BER measurements to extend the BER curve beyond the BER threshold without drastic increase in the run time. The idea exploits the difference in noise floor of the amplifiers inside the digital oscilloscope. When a larger vertical scale is selected on the oscilloscope, the oscilloscope switches to a higher gain but noisier amplifiers internally. This switching can be intentionally performed when the BER hits the threshold, it reduces the SNR at the receiver to provide more SNR headroom in BER measurements. This step is repeatedly performed whenever the BER hits the threshold again, allowing the testbed to continue producing BER data at higher transmit power levels that would have been otherwise absent. However this change in SNR creates an abrupt discontinuity in the BER curve, and adjustments are required to make the resulting BER curve continuous. Figures 5.14 and 5.15 show the flowcharts of this process during capturing and post-

98 Chapter 5. Experimental Results 82 increase transmit power capture BER Regular BER Testbed Flow terminate yes reached last power level? no no reached threshold? yes increase oscilloscope vertical scale setting Stitching Technique Flow Figure 5.14: A flowchart showing the implementation of the stitching technique during capturing stage for extending BER curves

99 Chapter 5. Experimental Results 83 check for BER discontinuity in original BER curve no discontinuity found? yes calculate equvalent ideal SNR right before and after the discontinuity find next discontinuity find the SNR difference and add to subsequent data points until reaching next discontinuity re-calculate BER data using new equivalent SNR values construct a new BER curve from original curve with the new BER data Figure 5.15: A flowchart showing the implementation of the stitching technique during post-processing stage for extending BER curves

100 Chapter 5. Experimental Results 84 processing stages respectively. To find the change in SNR caused by switching amplifiers, the two adjacent BER values using two vertical scale settings must be converted into comparable equivalent SNR values. Using QPSK modulation, the ideal BER curve follows Equation 5.3, where E b is the energy per bit, and N 0 represents the per-bit-snr. Q(x) is defined in Equation 5.4: 2Eb P b = Q( ) (5.3) N 0 is the noise power spectral density, and 2E b N 0 Q(x) = 1 2π x exp( u2 )du. (5.4) 2 It can also be expressed in terms of the error function in Equation 5.5: Q(x) = 1 2 erfc( x 2 ). (5.5) With Equation 5.5, the equivalent per-bit-snr can be found for a given SNR by transforming it into Equation 5.6: 2E b N 0 = [erfc 1 (P b 2)] 2, (5.6) and the measured BER values can be converted into equivalent theoretical SNR figures that are comparable between each other. Once the SNR difference between the two measurement points is found, this difference is added to all the equivalent SNR values for subsequent measurement points that use the same vertical scale setting, to bring the SNR in-line with the previous section of the BER curve. Adjusted BER values can be obtained at these points using Equation 5.3 and the new SNR value to replace the old values and remove the discontinuity in the BER curve. This procedure is repeated for all occurrences of vertical scale switching. Figure 5.16 illustrates an example of BER curves before and after performing the stitching technique. This method solves the problem we

101 Chapter 5. Experimental Results oscilloscope vertical scale settings: 1mV/div 5mV/div 20mV/div testbed BER capture limit BER discontinuities BER curve w/ stitching (raw) BER curve w/ stitching (post processed) BER curve w/o stitching Transmit Power Level (db) Figure 5.16: A detailed example illustrating BER curves before and after applying the stitching technique were facing earlier and provides enough accuracy for attaining acceptable BER spatial averaging results. However it is not a very good one as it does not generate realistic BER data captures below the minimum BER threshold, therefore can be considered only as a partial solution to this problem. Occasionally, the signal would get corrupted with random interference that causes one particular capture to exhibit significantly worse BER than the others made at the same transmit power level. This makes a measurable impact on the subsequent BER calculation and causes spikes in the BER curve that should otherwise be very smooth. The result of such spikes are often detected very late into the test when all the measurement locations have been tested and post-processed, and thus requires the entire BER curve at the troubled measurement location to be recaptured. This puts unnecessary delays in the test procedure. The implemented solution to this problem is to perform on-the-fly BER sanity checks. Knowing the BER at previous power level, and the increment in each step (thus the BER increase), the BER at the current power level can be estimated

102 Chapter 5. Experimental Results 86 using Equation 5.3. This BER value is used as a guidance to the actual BER value that is measured from the testbed. The actual BER may differ from the guidance value by a small amount, due to change in noise composition and distribution and fading in the channel. A threshold of tolerance can be set to the BER guidance value to determine if the measurement is within the acceptable range; and when the measurement falls out of the range, an immediate re-measurement at this particular setting can be conducted immediately without additional supervision and human intervention Testbed Validation The testbed is evaluated to ensure that both the hardware and software implementation function correctly to produce a BER curve that is close to the ideal one produced in Equation 5.3. A short cable is first used as a channel that is the most robust and less prone to being affected by variables and uncertainties from the environment. It directly connects the output of the transmitter to the input of the receiver. This eliminates any effects from multipath propagating, and possible interference from other sources. This measures the SISO capability of the testbed and its general accuracy using one of the two receiver pipelines. The next step involves testing with antennas in the EMC chamber at the University of Toronto, a controlled environment where interference from other sources is a non-issue. Three monopole antennas are used, one as the transmitting antenna and the other two as receiving antennas. It runs a check through the SIMO implementation of the testbed including the MRC algorithm and the second receiver pipeline. The ideal BER curve obtained using the Q-function in Equation 5.3 is compared against the results. Since we have not characterized the noise in the amplifiers and the digital oscilloscope, the receiving SNR is unknown to us. Therefore we attempt to align the ideal BER curve to the measured curve with a mere horizontal shift. Figures

103 Chapter 5. Experimental Results BER Testbed Sanity Check 10 1 testbed: SISO, cable Q function: QPSK 10 0 BER Testbed Sanity Check BER PSG power level (dbm) BER testbed: SISO, RX1 testbed: SISO, RX2 testbed: SIMO (MRC) Ideal QPSK BER curve PSG power level (dbm) (a) SISO cable results (b) SIMO antenna testing results Figure 5.17: Sanity check on the implementation of the BER testbed 5.17(a) and 5.17(b) show the results of both validation tests respectively. Both results match well to the shifted theoretical BER probability curve, indicating that reliable BER can be obtained with the BER testbed. The SIMO result also indicates that MRC implementation in the testbed works well and provides the expected advantage in SNR over a SISO system. 5.3 BER Results with ESPAR Antennas The experiments are conducted inside the BA8177 Antenna Lab in the Bahen Center of Information Technology at the University of Toronto. Due to time and equipment limitations, the test is conducted with a SIMO setup with one transmitting antenna. MIMO tests are feasible with manual displacement of transmitter to create a virtual transmit array, however this would increase the total experimental time by a factor of two. With a SIMO setup, it is still expected to show, as with the earlier CAD simulations that diversity and antenna gain play an important role in facilitating the higher capacity of reconfigurable antennas over their omni-directional counterparts. The transmitter uses a single monopole antenna operating at 2.28 GHz and radiating omni-directionally in the azimuth plane. The reference receiver setup has a pair of

104 Chapter 5. Experimental Results 88 Figure 5.18: ESPAR antennas mounted on the positioner. Two CAT5 ethernet cables provides power and correct biasing voltage to the diodes on the antenna for creating steerable patterns. monopole antennas identical to that used in the transmitter. The AUTs are the fabricated ESPAR antennas discussed in Section 4.4, mounted on the positioner shown in Figure Figure 5.19 shows the spatial locations of the transmitter and receivers with respect to the testing environment. Figures 5.20 to 5.28 illustrates how both ESPAR antenna patterns are oriented inside the lab, to give a better perspective of the antenna patterns with respect to the antennas physical locations and orientation. By design, each ESPAR antenna can reconfigure its pattern to cover all six 60-degree sectors. Certain patterns point the beam into the back wall of the room where absorbers are installed and the received amplitude will diminish dramatically. Therefore, the test deliberately limits the tunable range of each antenna to