Broadband High Power Amplifiers Using Spatial Power Combing ...
UNIVERSITY OF CALIFORNIA Santa Barbara Broadband High Power Amplifiers Using Spatial Power Combing Technique
A dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy in Electrical and Computer Engineering by Pengcheng Jia
Committee in charge: Professor Robert A. York, Chair Professor Umesh K. Mishra Professor Steve Long Dr. Yifeng Wu
December 2002
Broadband High Power Amplifiers Using Spatial Power Combing Technique
Copyright © 2002 by Pengcheng Jia
iii
Dedicated to my parents, and to my love, Xiangming
iv
ACKNOWLEDGEMENTS
Pursuing Ph.D in UCSB is one of the most valuable and exciting experiences in my education. The knowledge I learned and the confidence I gained in the 4 years’ study will be beneficial to my whole life. I am greatly indebted to my advisor, Prof. Bob York, who takes so much effort and patience in mentoring me to become a qualified researcher. From leading me into the wonderful spatial power combining field, to revise my poorly written papers and badly edited presentations, Prof. York gave me direction in every step of my thesis work, while never forgetting trivial details. It is his insight and wide knowledge that guided me to the completion of this thesis work; and it is his broad research interests that gave me the opportunities to explore many interesting projects in microwave field besides my thesis work. I also want to thank other members in my committee: Prof. Umesh Mishra for his interest in spatial power combining and support with Navy funding; Prof. Steve Long who was always there to help me with his profound knowledge and generous kindness; and Dr. Wu who played the role both as the advisor and as a friend, and gave so many good suggestions with his wisdom in both thermal and electrical engineering. I always feel lucky to be with so many excellent researchers in the York group. I sincerely thank Yu Liu for being a good partner in almost all the classes and an earnest colleague with whom I had so many intensive discussions on countless topics. Many thanks go to Paolo Maccarini, whose enormous contributions to the measurement lab greatly facilitated the progress of my projects. I’d like to thank Nick
v
Cheng for inspiring my interest in spatial power combining research and Jane Xu for teaching me all the skills in cleanroom. I am grateful also to other members in the York group for their support: Amit, Angelos, Andrea, Baki, Chris, Eric, Hongtao, Justin, Joe, Jim, Nadia, Pete, Troy and Vicki. I am thankful for all the members in Mishra group for sharing the cozy office, with whom I enjoyed the opportunity to be exposed to a variety of research projects on semiconductor material, device and circuitry: Ale, Ariel, Can, Dan, Dario, DJ, Gia, Haijiang, Huili, Ilan, Jae, Jason, Jeff, Lee, Likun, Mary, Naiqian, Peter, Prashant, Primit, Rama, Rob C., Rob U., Sten, Tim and Yingda. I’d like to especially thank Rob Coffie, who brought me high quality MMIC amplifiers that were the key components in the medium power amplifier and shared with me his thorough understanding in solid-state devices. I would like to thank many other students in the semiconductor group, from whom I benefited a lot through stimulating discussions: Yun Wei, Shouxuan, Paidi, P.K., Miguel and Jingshi. Lastly, I would like to thank my family and friends. I am very grateful to my Dad and Mom who encourages and supports me to realize all my dreams even though they have to sustain the long time separation from their beloved son. I owe all my achievement to my love, Xiangming, who shares all my joy and bitterness every day and night. Again, thanks to all my friends, for the happiness you brought into my life: Huili, Yang, Ping, Hongyuan, Kelly, Likun, Xiaojie, Songming, Rui, and all my basketball pals.
vi
VITA June 14, 1974
Born in Tianjin, China
June 1995
Bachelor of Science, Electronics Science and Information System, Nankai University, Tianjin, China
September 1995
Research Assistant, Dept. of Electrical Engineering, Tsinghua University, Beijing, China
June 1998
Master of Science, Electrical Engineering, Tsinghua University, Beijing, China
September 1998
Graduate Student Researcher, Dept. of Electrical and Computer Engineering, University of California, Santa Barbara
December 2002
Doctor of Philosophy, Electrical and Computer Engineering, University of California, Santa Barbara PUBLICATIONS
1. P. Jia, R.A. York, “Multi-Octave Spatial Power Combining in Oversized Coaxial Waveguide”, IEEE Trans. Microwave Theory and Tech, vol.50, (no.5), IEEE, May 2002. p.1355-60. 2. P. Jia, R.A. York, “A Compact Coaxial Waveguide Combiner Design For Broadband Power Amplifiers”, IEEE MTT-S International Microwave Symposium Digest, Pheonix, USA, May 2001. p.43-6 vol.1. 3. P. Jia, L.-Y. Chen, N.-S. Cheng, and R.A. York, “Design of Waveguide Finline Arrays for Spatial Power Combining”, IEEE Trans. Microwave Theory and Tech., vol.49, (no.4, pt.1), IEEE, April 2001. p.609-14. 4. P. Jia, Y. Liu, R.A. York, “Analysis of A Passive Spatial Combiner Using Tapered Slotline Array in Oversized Coaxial Waveguide”, 2000 IEEE MTT-S International Microwave Symposium Digest, Boston, MA, USA, June 2000. p.1933-6, Vol.3.
vii
5. N. –S. Cheng, P. Jia, D. B. Rensch and R. A. York, “A 120-Watt X-Band Spatially Combined Solid-State Amplifier”, IEEE Trans. Microwave Theory and Tech., vol. 47, (no. 12), IEEE, Dec. 1999. p.2557-61.
viii
ABSTRACT Broadband High Power Amplifiers Using Spatial Power Combing Technique by Pengcheng Jia
High power, broad bandwidth, high linearity and low noise are among the most important features in amplifier design. Realizing all these features in one amplifier remained as a big challenge for RF engineers. Broadband spatial power combining technique addresses all these issues by combining the output power of a large quantity of Microwave Monolithic Integrated Circuit (MMIC) amplifiers in a broadband environment, while maintaining good linearity and improving phase noise of the MMIC amplifiers. The intent of this research is to extend the waveguide based combiner design to broadband applications with emphasis on linearity improvement and phase noise reduction. Coaxial waveguide was used as the host of the combining circuits for broader bandwidth and better uniformity by equally distributing the input power to each element. The goal also includes the standardization of the modeling technique. Meanwhile, thermal management, efficiency, noise figure, phase noise and linearity analyses are all covered in this work. A broadband low noise medium power amplifier is first presented. The coaxial waveguide combiner is utilized to combine the output power of 32 low noise MMIC
ix
amplifiers. A bandwidth from 3.5 to 14 GHz is achieved with a maximum power of 1 watt. The residue noise of the amplifier is lower than –150 dBc at a 10 KHz offset from the carrier with 15 dB reduction compared to the residual phase noise of a MMIC amplifier. A new compact coaxial combiner with much smaller size is further investigated. Broadband slotline to microstrip line transition is integrated for better compatibility with commercial MMIC amplifiers. Thermal simulations are performed and a new thermal management scheme is employed to improve the heat sinking in high power application. A high power amplifier using the compact combiner design is built and demonstrated to have a bandwidth from 6 to17 GHz with 45-watt maximum output power. Linearity measurement has shown a high IP3 of 54 dBm. Residual phase noise is –140 dBc at a 10 KHz offset from carrier.
x
TABLE OF CONTENTS Preface........................................................................................................................... 1 1. Overview of Spatial Power Combining Technology ............................................ 4 1.1 Introduction ................................................................................................... 4 1.2 TWT Amplifiers............................................................................................ 6 1.3 Solid-State Amplifiers and Power Combining Technology.......................... 9 1.4 Spatial Power Combining............................................................................ 11 References ............................................................................................................... 21 2. Electromagnetic Modeling .................................................................................. 23 2.1 Modeling of the Rectangular Waveguide Combiner................................... 24 2.2 Modeling of the Coaxial Waveguide Combiner ......................................... 38 References ............................................................................................................... 51 3. Broadband Medium Power Amplifier Using Coaxial Waveguide Combiner..... 52 3.1 Passive Combiner Measurements................................................................ 53 3.2 Performance of the Active Combiner.......................................................... 60 References ............................................................................................................... 66 4. Design of High Power Amplifier Using the Coaxial Waveguide Combiner ...... 67 4.1 Motivation ................................................................................................... 68 4.2 Coaxial Waveguide Design......................................................................... 69 4.3 Synthesis of Waveguide Finline Array ....................................................... 70 4.4 Slotline to Microstrip Line Transition......................................................... 71 4.5 Compact Passive Structure of Coaxial Waveguide Combiner.................... 75 4.6 Leakage from Output to Input..................................................................... 76 4.7 Uniformity................................................................................................... 78 4.8 Fabrication Procedure ................................................................................. 85 4.9 Circuit Tray & Bias Line............................................................................. 87 4.10 Efficiency, Reliability and Thermal Analysis ............................................. 88 References ............................................................................................................... 98 5. Performance of High Power Amplifier Using Compact Coaxial Waveguide Combiner..................................................................................................................... 99 5.1 Measurement System ................................................................................ 100 5.2 MMIC Amplifier Characterization............................................................ 105 5.3 Output Power............................................................................................. 107 5.4 Linearity .................................................................................................... 113 5.5 Noise Figure .............................................................................................. 125 5.6 Spurious-Free Dynamic Range ................................................................. 128 5.7 Phase Noise of Combiner.......................................................................... 129 5.8 Summary ................................................................................................... 136 References ............................................................................................................. 137 6. Conclusion and Future Works........................................................................... 138
xi
Preface
PREFACE
High power, broadband, high linearity and low noise are among the most important features in amplifier design. When I started this project, realizing all these features in one amplifier remained as a big challenge for RF engineers. Research in spatial power combining led me into the wonder world of new technologies to solve these enigmas. To achieve all the features in one innovative design becomes the ultimate goal of my thesis work. Although most of the spatial power combining research groups are originated from Caltech’s Rutledge group, UCSB’s microwave group differentiates itself from other research groups by investigating in the waveguide based spatial power combining design with the guidance of Prof. Bob York. Dr. Angelos Alexanian and Dr. Nick Cheng have presented several exciting designs in the past years. The intent of this research is to extend their results to broadband applications with high output power while maintaining good linearity and low noise. The goal also includes the
1
standardization of the modeling technique. Meanwhile, efficiency, noise figure, phase noise and linearity analyses are added to supplement previous works. The thesis is organized in the following way. Chapter 1 compares waveguide based spatial power combining technique with the traveling tube amplifiers, corporate power combining and other spatial power combining techniques. The benefits of the coaxial waveguide design over the rectangular waveguide design are also explained. Chapter 2 follows with the modeling of the waveguide based spatial power combiners. Modeling for the rectangular waveguide combiners is covered first since it is the basis for all waveguide based combiner design. Waveguide model is then revised for coaxial waveguide application. Comparison with results from commercial software HFSS verifies the effectiveness of the modeling. Using the models and design parameters derived from the previous chapter, chapter 3 develops a practical design of coaxial waveguide combiner. 32 low power MMIC amplifiers are integrated into the combiner to build an active amplifier. 3.5 to 14 GHz bandwidth is achieved with 1watt output power. Chapter 4 addresses the size issue and leads to a more compact version. New slotline to microstrip line transition is characterized and combined into the system for better connections with commercial MMIC amplifiers. Fabrication process and thermal analysis is also covered in this chapter. Chapter 5 introduces the measurement system. Measurement result shows that the amplifier has 44-watt output power capacity and 6 to 17 GHz bandwidth. A variety of
2
practical issues that arise in the design of PA circuits and system, including linearity, noise figure and phase noise are also treated in this chapter. In Chapter 6, conclusions are made and potential improvements are presented as future works.
3
1.
Overview of Spatial Power Combining Technology
CHAPTER 1 Overview of Spatial Power Combining Technology 1.1
Introduction The history of microwave technology is a history of progressive advances in the
techniques used to generate, amplify, and process signals at microwave frequencies. Invented in the 1940s, the Traveling Wave Tube (TWT) has become a key element in microwave systems for radar, satellite communication and wireless communication. Currently, the Traveling Wave Tube Amplifier (TWTA) is the dominant choice for signal amplification subsystems in communication with operational frequencies higher than 4 GHz and power level greater than 10 Watt. An alternative to the TWTA is a power combiner. Power combining technique has been exploited extensively to improve the output power level from solid-state devices[1]. Two types of power combining techniques are the corporate combiner and spatial power combiner. The corporate combining technique will lead to very high 4
combining loss when integrate large amount of amplifiers, while spatial power combining technique was proposed with the goal to combine the a large quantity of solid-state amplifiers efficiently and improve the output power level to be competitive with TWTA. Many research groups including the Caltech’s Prof. Rutledge’s group, has done extensive research on the planar “tile” combing approach[2]. While UCSB’s microwave group attempted a “tray” scheme inside waveguide to achieve broader bandwidth, better thermal management and more efficient power collection[3-5]. Employing the “tray” combining scheme, we demonstrated an X band power amplifier with 150 Watt output power using oversized WR-94 rectangular waveguide. The rectangular waveguide combiner is easier to be fabricated and also better for thermal management. But the dominant TE10 mode inside rectangular waveguide will lead to non-uniform illumination of the loaded antenna trays inside the waveguide. Hence, the output power will experience a soft saturation that will deteriorate the linearity or lead to large back off of output power to satisfy the requirement of linearity. To meet the requirement of high linearity in many broadband communication systems, we extend the “tray” approach from rectangular waveguide to coaxial waveguide. A multi-octave bandwidth amplifier achieved bandwidth from 3.5 to 14 GHz with good linearity using oversized coaxial waveguide combiner[6]. A modified amplifier using more compact coaxial waveguide combiner design has shown 6 to 17 GHz bandwidth with 45 Watt maximum output power while with good linearity and high dynamic range. That enables it a good rival for current dominant TWT amplifiers. 5
1.2
TWT Amplifiers Vacuum electronic amplifiers are used in a wide variety of military and
commercial applications requiring high power at microwave and millimeter wave bands. From the richness of device concepts investigated through the 1960s, the helix and coupled cavity traveling wave tubes (TWTs), klystron, magnetron, and crossed field amplifier (CFA) emerged as the primary products of today’s industry[7]. The traveling wave tube amplifier (TWTA) is the most widely used vacuum electronic amplifier in communication systems that require wide bandwidth. It was invented in 1940s by Rudolph Kompfner and has had a long history in playing a key technology role in a variety of applications ranging from military and radar systems to commercial communication systems. Advances in TWT amplifiers (TWTA) have made satellite communications a huge success. Among them, helix TWT is widely used for communication applications, while coupled cavity TWT is better for narrowband systems requiring higher power. Along with the advances in tube and solid-state devices, the amplifier manufacturers have made great strides in improving the complete TWTA package and subsystems. TWTA sizes have shrunk considerably, power handling is better, costs have come down, microprocessor control is now standard, other features have increased and reliability is better than ever. A basic helix TWTA block diagram is shown in Figure 1.1.
6
Figure 1.1 Diagram of a traveling wave tube amplifier.
All TWTs possess four major subassemblies: An electron gun that produces a high density electron beam; A microwave slow-wave circuit that supports a traveling wave of electromagnetic energy with which the electron beam can interact; The collector that collects the spent electron beam emerging from the slow-wave circuit; The TWT package, which provides points for attachment to the using system, provides cooling for power dissipated within the TWT, and, in some cases, includes all or part of the beam focusing structure. The TWTA is a very complicated system and needs a lot of touch labor. It can only be manufactured in hundreds each month. But currently TWTA is the only high power amplifier that can work over a broad bandwidth. It can cover C, X, Ku band or a big fraction in Ka bands with power level from 10 Watts to 3000 Watts. The large
7
bandwidth and high power levels have made them dominant power amplifier for satellite communication. The advance in tube technology has improved the efficiency of the TWT amplifier up to 70% for narrow band and 50% for broadband, which is the present best solution for space satellite transponders. However, the drawbacks of the TWT amplifiers are also obvious, such as considerable size and weight. Tube amplifiers also need the high voltage drive - Electronic Power Conditioner (EPC) that requires additional complex accessory circuit and involves high voltage risk. Moreover, the tube amplifier is always rated with saturation power, which leads to bad linearity and is not good for broadband communication. To work linearly, the TWT amplifier is normally backed off from its saturated output power or additional linearization circuits are added. Linearization circuitry results in dramatically increase of system complexity and cost. Moreover each small increase in efficiency is very expensive. A high efficiency high power TWT amplifier in satellite may cost up to half a million dollars. Recently vacuum tube engineers have taken advantage of the MMICs to develop the MPM (Microwave Power Modules)[8, 9]. In brief, an MPM is the combination of a solid-state exciter and TWT amplifier that has similar gain-bandwidth characteristics as a conventional TWT but is much smaller and has superior noise, efficiency and linearity characteristics. As such, it is much smaller than a conventional TWT and can operate with much lower supply voltage. However, MPMs still require a great deal of "touch labor" in their assembly and testing and, hence, are prone to higher costs and lower production rates.
8
1.3
Solid-State Amplifiers and Power Combining Technology Improvements in solid-state material and amplifier have pushed the output power
level of a single MMIC (Microwave Monolithic Integrated Circuit) to the watt level, but there is still no commercially available MMIC amplifier that can output more than 5 Watts over X band. Even with the advent of promising high-power solid-state devices based on wide bandgap semiconductor materials such as gallium nitride (GaN) and silicon carbide (SiC)[10, 11], it is still difficult and costly at the present time to realize significant RF output power at a single device level. There is no surprise that vacuum electronics are still the dominant technology for high power applications. However, solid-state electronics are generally more desirable in terms of size, weight, reliability, and manufacturability. Economic considerations can also favor solid-state systems that can be mass-produced using modern IC technology.
Figure 1.2 A corporate structure for power combining. (From reference [1], IEEE).
9
Power combining techniques has been investigated extensively. The corporate combiner is one of the classic and most popular structures[1]. To satisfy system requirements, power from many individual devices must be added coherently. As shown in Figure 1.2, the outputs from a number of circuits are successively combined using two-way adders such as Wilkinson combiners. The number of individual devices is 2N, where N is the number of stages. The combining efficiency is therefore η = LN, where L is the insertion loss of each stage. Note that the physical layout of the corporate combiners with many elements causes the transmission lines in the last stages of combining to become very long. As the number of devices increases, the losses in these lines become insurmountable. As shown in Figure 1.3, loss of the combining circuit will increase dramatically and the output power of the combiner will even decrease when the number of devices is very large. 50
Power Added Efficiency, %
45 40 35 30 L = 0.1dB (Corporate) L = 0.2dB (Corporate) L = 0.3dB (Corporate) L = 0.5dB (Spatial) L = 1.0dB (Spatial) L = 1.5dB (Spatial)
25 20 15 1
10
100
1000
Number of Amplifiers
Figure 1.3 Output power available from combiners as a function of the number of elements.
10
High losses associated with circuit combining schemes can be avoided with spatial power combining technique since the energy is combined in the lossless space or low loss electric medium. In Figure 1.3’s example, the spatial power combining is superior when there is a large quantity of elements.
1.4
Spatial Power Combining
Overview of the spatial power combining technology
Spatial power combining was reported as early as 1968 with the construction of a 100-element spatially fed/spatially combined array for operation between a pair of electrically short monopole antennas[13]. This active array approach has dominated the research on spatial power combining recently while the antenna array has improved to be smaller and more broadband[2, 14]. Their active arrays interact with propagating beams in free space[15]. The incident, reflected and transmitted beams are guided and manipulated via conventionally optical components such as mirrors, polarizers and lenses. Therefore optical techniques are applied to systems operating far below the optical spectrum, hence the name quasi-optics. The natural configuration for such a system is therefore a Gaussian-beam waveguide, with the array placed at a beam waist where the phase front is planar. Many of the array concepts are designed for use in closed metallic waveguide to provide good packaging[16, 17]. It is attractive at frequencies below 100 GHz for several reasons: diffraction losses and focusing errors are minimized or eliminated, since all of the energy is confined by the waveguide walls; the metal walls provide a
11
convenient heat sink for the arrays; waveguide components are readily available in this frequency range and maybe more economical than quasi-optical components; and using quasi-optical arrays in this way allows them to be retrofitted into existing waveguide systems. A big disadvantage of this design is the nonuniform field profile in the waveguide leading to the edge elements on the array not coupling efficiently to the waveguide mode resulting in lower efficiency and linearity. This can be addressed using dielectric loading or sidewall corrugations[18, 19]. Furthermore, standard waveguide may not accommodate a sufficient number of devices unless the waveguide cross-section is enlarged. A higher-order mode suppression technique is also necessary in oversized waveguides. Although without optical beam guiding components, it is still be distinguished as quasi-optical combiner since it uses spatially fed/spatially combined array. One of the most popular spatial combining architecture is a tile approach. This approach denotes configurations that use relatively thin modules where the RF circuitry and active devices are placed on circuits parallel to the face of the array. The tile approach includes 2 popular designs, active array amplifier and grid amplifier. These designs, illustrated in Figure 1.4, are quite different, and each has its merits. The grid amplifier is an array of closely spaced differential transistor pairs. The input and outputs are cross polarized, and off-chip polarizers are used for tuning. The drawback of grids is that the small cell sizes limit the gain and power per cell to that available from a single differential pair. Because the active devices are very dense, the grid amplifier can be monolithically fabricated; this makes grids a very attractive 12
technology for moderate gain and power applications that demand a single chip massproducible solution. Active arrays, on the other hand, use larger unit cells with more conventional antennas like patches or slots. This larger unit cell allows integration of multi-stage MMICs with higher gain and output power. The passive radiating and tuning elements do tend to occupy a significant fraction of the active array and tray amplifier’s area; the most economical solution is to attach active MMICs to passive antennas. Active arrays may find use in very-high power or gain applications.
Figure 1.4 Two tile architectures: (a) grid amplifiers, (b) active array amplifiers. (From reference [12],IEEE.)
Attempts have already been made to apply the grid amplifier design to industrial products since it can easily fit all of the required components on a GaAs substrate. This approach has been well developed by research group in CalTech and achieved 5 Watt at Ka band from a GaAs chip. But heat removal or thermal management
13
becomes the biggest design challenge. The design can be modified to contain a heatconducting layer to channel the heart from the center array to the edges, but this heatconducting layer will be relatively thick to obtain adequate thermal conductance, then complicates the transfer of the signal from the receive side of the array to the transmit side. The thermal glue used to bond the solid-state chip to the heat-conducting layer will also add a high junction thermal resistance. The thermal management for output power of more than 20 Watt will be very difficult. A tray approach was then developed aiming to provide better thermal management and increase the bandwidth while achieving higher power[3]. This architecture provides more space for the RF circuitry and active devices. It can use amplifiers with higher gain and output power by integrating amplifiers in the longitude direction. Another advantage of this configuration is that the metal carrier of each circuit trays permits good heat conduction. The biggest disadvantage of the tray approach is the length of the system. However there are a lot applications where this is not an issue. Details of progress with the tray approach in UCSB will be discussed in the next section. Active spatial power combining has become a dynamic research field recently[2025]. Substantial power in the 100-Watt range at X band has been achieved by UCSB[5]. Sanders report a combiner with 272 MMICs generating 35 Watt at 61 GHz[26]. Researchers at Lockheed Martin and North Carolina State University have recently demonstrated a 25 Watt “tiled” combiner system at Ka Band (34GHz)[27]. Researchers at the California Institute of Technology have developed a 5 Watt single14
chip monolithic grid amplifier using Rockwell pseudomorphic high electron-mobility transistor (pHEMT) technology[28]. Moreover most of the basic components in microwave receiver and transmitters have been demonstrated as spatial combining circuits. These include amplifiers, oscillators[29], mixers[30], multipliers and switches[31]. Quasi-optical oscillators and amplifiers have been tested in wireless communications circuits, transponders, and beam switching systems. Aggressive developers now have the opportunity to build quasi-optical systems for a new generation of communication and radar equipments.
Rectangular Waveguide Spatial Power Combiner
Figure 1.5 (a) Schematic plot of the rectangular waveguide combiner (b) Layout of a single tray[4].
In the tray approach of spatial power combining, the RF circuitry and active devices are placed on circuits perpendicular to the face of the array and they usually
15
contain end fire antenna elements (such as a linearly tapered slot or Vivaldi antenna). Several tray approaches have been introduced, but the most successful one is the rectangular waveguide combiner developed by Angelos Alexian and Nick Cheng of UCSB[32]. The concept is illustrated schematically in Figure 1.5; we exploit the inherent spatial distribution of the field energy in the dominant waveguide mode to distribute and collect power to and from a dense array of amplifiers. Transitions between the amplifier and waveguide mode are made via electrically close taperedslot antennas (or finline structures). The combiner design is compact, but large enough to accommodate the desired number of amplifiers. The combiner is also well designed for thermal management for output power levels in hundreds of watts. The tray is made of copper with power MMIC amplifier sitting in the middle. The wasted heat is efficiently transferred from the center of tray to the outside waveguide surface, and then dissipated through forced air convection. The enclosed waveguide provides an excellent heat-sinking environment for the power devices and is a natural choice for most high-power applications. Another extraordinary feature of this design is the successful integration with broadband taper finline antenna. The frequency response of the passive structure is only limited by the cut-off frequency of the rectangular waveguide. Recent research activities include Nick Cheng’s 150 Watt X band amplifier, Vicki Chen’s K-band amplifier[33] and Jinho Jeong’s 3.3 Watt 24GHz amplifier using antipodal finline to microstrip line transition[34]. 16
With all these merits, the rectangular waveguide combiner is facing a big challenge of non-uniform illumination. The sinusoidal field distribution of the dominant waveguide mode TE10 will lead to different drive power at the input the MMIC array. When the input power increases and the MMIC amplifiers at outside trays begin to reach P1dB output power, the MMIC amplifiers at inner trays are already overdriven into deep saturation. The non-uniform drive will lead to a soft saturation of the amplifier, which is shown in Figure 1.6. If maximum output power is reached, intermodulation components will be very high due to the highly nonlinear operation of amplifier in inner trays. It will not be qualified for modern broadband communication systems that have high criteria for linearity.
Figure 1.6 Effect of non-uniform illumination of incident power on output power. The inset plot outlines the actual positions of the trays. Sinusoidal field distribution is assumed for the non-uniform case[4].
Other modifications of the rectangular waveguide are also investigated to improve the uniformity. The performance can be improved using longitudinal corrugations. The unique characteristic of the UC-PBG (uniplanar compact photonic bandgap) 17
structure allows for the possibility of building a TEM waveguide when used as a planar reflector. The UC-PBG reflector behaves like a magnetic surface at its stopband frequency where the periodic loading changes the surface impedance to an open-circuit condition. When the two sidewalls of a rectangular waveguide are replace by the UC-PBG structure, a parallel-plate model will be established by magnetic boundary conditions leading to better uniformity inside the waveguide. A relatively uniform field distribution was demonstrate from 9.4 to 10.4 GHz by UCLA[35]. To extend the bandwidth, an active UC-EBG (uniplanar compact electromagnetic bandgap) structure is proposed with varactors to electrically tune the stopband frequency[36]. However, the present UC-PBG/EBG structure still can’t produce a satisfactory parallel plate mode with uniform illumination to all the elements. Moreover the bandwidth is limited and the active tuned structure increases the complexity of system. So the best solution to provide uniform illumination for all elements will be a coaxial waveguide combiner design.
Coaxial Waveguide Spatial Power Combiner
Coaxial waveguide combiner was exploited as the cylindrical resonant cavity combiner in the 70s[1]. Angelos Alexanian first applied the coaxial structure to spatial power combiner field with a preliminary demonstration of the idea using passive elements.
18
Here we extend the tray approach to the coaxial waveguide by radially dividing a coaxial waveguide section into 16 identical wedge trays and integrating with similar broadband tapered finline antennas as the rectangular waveguide combiner. In this combiner, MMIC amplifiers are placed on a bridge in the middle of each tray, and then connected with input and output antenna that couples energy from and to the waveguide. When the 16 tray sections are stacked together, the coaxial waveguide opening is formed. Then input and output coaxial waveguide section will be used to transit from the center stacked section to type N connectors. The input/output transition is optimized with a taper design to minimize the reflection with maximum bandwidth. Details of the design will be covered in the following chapters. There is very small dispersion since the dominant mode that propagates along the coaxial line is TEM mode. And the dense antenna array helps to suppress the higher modes in the oversized waveguide. We maintain all the benefits of the tray approach; while we future broaden the bandwidth of the combiner by fully exploiting the bandwidth of the tapered finline array because the coaxial waveguide doesn’t have any cut-off frequency. Moreover, the uniform illumination of all the antennas helps to maintain the amplifier’s linearity same as each MMIC amplifiers. The coaxial waveguide combiner has the potential of achieving multi-octave bandwidth. And the minimum spacing between the trays will be the MMIC amplifier’s size. So the coaxial waveguide combiner has the benefit of large-scale integration and can provide very high output power. High thermal dissipation
19
capacity of the tray approach enables the large-scale power combining feasible by providing reliable heat removal solutions. When the amplified signals from different MMIC amplifiers are combined, the power are added since all the signals are in phase. But the residue phase noises from each MMIC amplifier are irrelevant random variables. The sum of the residue phase noises will have the same power as that of a single MMIC. For a N-elements combiner, we will observe N times residue phase noise reduction in the output comparing with using only one MMIC amplifier. Table 1.1 Comparison between TWTA and QO amplifiers
TWTA
Quasi Optical Amplifiers Grid
Rect. WG
Coax. WG
5 Watt
150 Watt
50 Watt
Power
10-3000 Watt
Freq Band
C, X, Ku, Ka, Ka, V V
C, X, Ku, V
C-Ku
Efficiency
40-50%
15%
20%
17%
Linearity
Back off 7 dB N/A from rated power
N/A
Back off 3 dB from P1dB
Low
Low
IMD 2 A . Assuming the propagation constant is a monotonically increasing function of frequency, the lowest operating frequency is therefore defined by
θt ( f0 ) = 2 A
(4.4)
which is an implicit relationship between the taper length L , the lower cutoff frequency f 0 , and the maximum reflection coefficient Γ m . The main difficulties in applying the above results are the frequency dependence of the wave impedance and propagation constant, the difficulty in translating the impedance as a function of θ into a function of z , and the subsequent determination of the physical parameters required to synthesize the impedance taper. The frequency dependence of the wave impedance and propagation constant means that in general, the result (4.3) will require a different physical taper at each frequency, which is obviously not a possible implementation.
However, the normalized impedance
Z (θ ) / Z 0 of the finline transition is found to be a relatively weak function of frequency. We can therefore design the taper at a fixed frequency, chosen to be f 0 . In addition, it has been found that waves propagating along the finline structures are
28
approximately TE in character, allowing us to relate the wave impedance and propagation constant as
Z=
ωµ β
(4.5)
Using wave impedances instead of the characteristic impedance, (4.3) can now be rewritten in terms of β
β ( f0 , z) =
2θ ( f0 , z )
β L β 0 exp −Γ m A2 F
θt
− 1, A
(4.6)
where β , β L and β 0 correspond to Z , Z L and Z 0 , respectively, using (4.5). To compute the required propagation constant as a function of the position along the taper, β ( z ) , the taper structure is divided into N sections of length ∆z = L / N . θ can be approximated by i −1
θ ( zi ) ≈ ∑ 2 β ( zk )∆z = θ ( zi −1 ) + 2 β ( zi −1 )∆z
(4.7)
k =0
where zi = i∆z . Noting that θ (0) = 0 , we first evaluate β (0) from (4.6), and then use the approximation (4.7) in (4.6) to evaluate all subsequent values of β ( zi ) along the taper. We then repeat the iterative process until the solution set of β converges. The resulting procedure is similar to that used in [3] for single finline structures. Note that the propagation constants at the ends of the taper do not match the target values, β (0) ≠ β 0 and β ( L) ≠ β L . Using (4.6) it can be shown that
29
β ( L) = β L eΓ
(4.8)
m
and this either fixes the maximum reflection coefficient for a given β ( L ) , or determines β ( L) for a given Γ m . For a dense finline array, there is a potentially large impedance discontinuity in the transition from an unloaded waveguide to a loaded waveguide, so it is necessary to manipulate the substrate material, thickness, tray locations, and local waveguide width in order to satisfy (4.8) for a desired Γ m . Another possibility is to include a quarter-wave “notch” transformer as part of the finline transition [4], but this proved impossible in the present work due to the use of ceramic substrates which were difficult to machine.
Propagation Constant of a Finline Array
y PMC
b/2 t1 t2 1
g
2
X=d
t3 3
x
0 -a/2
c
a/2 PEC
-b/2
Figure 2.4 Cross section of a 2×2 finline array in a standard waveguide environment.
In this work the Spectral Domain Method (SDM) [5, 6] was used to find the relationship between the propagation constant and the geometrical parameters of the fin-line, most importantly the slot width. For simplicity, a 2×2 fin-line array was
30
analyzed, as shown in Figure 2.4. We assume perfect contact between the finline and the waveguide walls. Symmetries along the major axes were used to reduce the computation domain to the upper right quadrant of Figure 2.4. In the SDM, the electric fields and currents in each region are expanded as a Fourier series in y . Denoting the electric field in the i th region as Ei , Ei =
∞
∑ Ee α
n =−∞
j
i
ny
where α n =
2nπ b
2 b/2 Ei = ∫ Ei e jα n y dy b 0
(4.9)
Applying the boundary conditions at the interface x = d gives two algebraic equations Yyy E y + Yyz Ez = jωµ0 J y , Yzy E y + Yzz Ez = jωµ0 J z .
(4.10)
where the J are the unknown currents on the fins. Using the equivalent transmissionline “immitance” concept [6], we find Yyy = jωµ0 (YTE cos 2 θ + YTM sin 2 θ ) Yyz = −Yzy = −ωµ0 sin θ cos θ (YTM − YTE ) Yzz = jωµ0 [YTE sin 2 θ + YTM cos 2 θ ] where
31
(4.11)
β
sin θ = Γ ni =
β 2 + αn
cosθ =
2
αn β 2 + αn
2
,
β 2 + α n − ω 2 µ 0ε i 2
YTMi = jωε i / Γ ni ,
YTEi = Γ ni / jωµ 0 ,
YTML = YTM 1 tanh(Γ n1t1 ), YTEL = YTE1 tanh(Γ n1t1 ), YTM = YTM 2 YTE = YTE 2
YTML + YTM 2 tanh(Γ n 2t2 ) YTM 2 + YTML tanh(Γ n 2t2 ) YTEL + YTE 2 tanh(Γ n 2 t2 ) YTE 2 + YTEL tanh(Γ n 2 t2 )
+ YTM 3 coth(Γ n 3t3 ),
+ YTE 3 coth(Γ n 3t3 ).
The unknown aperture fields E y and Ez are expanded in terms of a basis set of rectangular pulses ξ i and ηi , which is more convenient for the wide slot portion in the finline, and then Fourier transformed to Ny
E y (α n ) = ∑ ciξ i (α n ), i =1
Nz
(4.12)
Ez (α n ) = ∑ diηi (α n ). i =1
Substituting (4.12) into (4.10) and integrating gives a homogeneous matrix equation Ny
Nz
i =1
j =1
Ny
Nz
∑ K piyy ( β )ci + ∑ K pjyz ( β )d j = 0, p = 1… N y ∑K i =1
zy qi
(4.13)
( β )ci + ∑ K qjzz ( β )d j = 0. q = 1… N z j =1
The propagation constants over the normalized gap 2 g / b are then found from the characteristic equation obtained by setting the determinant of the coefficient matrix in
32
(4.13) to zero. Figure 2.5 shows the results of this calculation for a representative physical situation of interest, corresponding to two 10-mil thick aluminum nitride (AlN) substrates, with a separation of c = 5 mm, placed in an X-band (WR-90) waveguide with dimension a = 0.9”, b = 0.4”. A single “pulse” basis function was used in this calculation.
e ffe c ti v e pe rmi ttiv ity
3.5 3 2.5 2 1.5 1 0.5 0.2
0.4
0.6
0.8
1
normalized gap width
Figure 2.5 Effective permittivity versus normalized gap width for a 2x2 finline arary in WR90 waveguide.
Using the results in Figure 2.5, an “optimized” taper was computed for an input reflection coefficient of –20dB. The shape of the optimized tapered finline is shown in Figure 2.6. no rmalized gap width
1 0.8
optimized taper
0.6 0.4 0.2 0 0
0.25
0.5
0.75
1
1.25
1.5
1.75
Position along taper, cm
Figure 2.6 Normalized gap width vs. location along the optimal tapered finline.
33
Scaling, Losses, and Combining Efficiency
The optimized taper design in Figure 2.6 was used as the basis of several combiners.
The 2x2 array taper design was scaled up to larger arrays. In the
combiner reported in [7], two additional trays were added to form a 4-tray 2x4 array. Not surprisingly, some degradation in return loss is observed as compared with the 2tray 2x2 array results.
50 Ω microstrip line
Finline taper 70-to-50 Ω microstrip taper
0
0
-5
-5
-10
-10
-15
-15
-20
-20
-25
S11: 2 Tray, dB S11: 6 Tray, dB
S21: 2 Tray, dB S21: 6 Tray, dB
S21, dB
S11, dB
Figure 2.7 Tray layout for a through measurement.
-25 -30
-30 8.5
9.0
9.5
10.0
10.5
11.0
11.5
12.0
Frequency, GHz
Figure 2.8 2-port measurements for the slotline arrays with 2 trays and 6 trays.
34
The 2-tray 2x2 array was then scaled up to a 6-tray 4x6 array design. In this case the number of finline transitions on each tray was doubled to four by linearly scaling the dimensions of the 2x2 design. The number of trays was increased to six, which required the use of a thinner tray and hence closer tray-to-tray spacing. In this case, 2port measurements were performed in order to examine the return loss and insertion loss characteristics of the passive structure. The layout of the test circuit (Figure 2.7) consists of back-to-back finline transitions with 50Ω microstrip lines used in place of active elements. As a consequence of the reduced tray spacing, the terminating resistance yielding the best impedance match was lowered to 70Ω, so a 70-to-50Ω taper was included in the microstrip through line. Measured reflection and transmission for the 2-tray (4x2) and 6-tray (4x6) systems are shown in Figure 2.8.
The 6-tray system with 24 tapers suffers
significantly increased reflection losses. Since the tray separation in the 4x6 system is the same as in the 4x2 system, the deterioration in performance results from the nonoptimized taper rather than the inter-tray coupling. Nevertheless, an important observation can be made with respect to insertion loss: ignoring the effects of reflection losses (which can potentially be recovered by properly optimizing the finline taper for the 4x6 array), the insertion losses appear approximately constant in the 2-tray and 6-tray system. This is a natural result of the parallel nature of the transitions, but has important consequences in combining efficiency when scaling the combiners to large numbers of devices. Whereas traditional combiners suffer a
35
reduction in combining efficiency as the number of devices is increased, the spatial combiners exhibit a loss that is roughly independent of the number of devices.
Insertion Loss, dBdB Dissipative Loss,
0
-2
-4 2 Tray, dB 6 Tray, dB 8 Tray, dB
-6
-8
-10 8.5
9.0
9.5
10.0
10.5
11.0
11.5
12.0
Freqency, GHz
Figure 2.9 Measured dissipative loss for 2,6, and 8-tray finline array with though lines.
To empirically quantify this assertion using the measured results of Figure 2.8, we compute a “dissipative loss” as Loss =
Pload Pload | S21 |2 = = Pforward Pin − Preflection 1− | S11 |2
(4.14)
The losses computed from (4.14) for the 2,6, and 8-tray configurations, plotted in Figure 2.9, show that the dissipative loss in each case is approximately constant as a function of frequency, and independent of the number of trays. It can then be shown that the maximum combining efficiency ηc of a spatial combiner structure is
ηc ≈ Lo
36
(4.15)
where Lo is the loss associated with the combiner circuit (post-amplification losses). In the present case, any loss can be attributed equally to the input and output antennas, which gives us an estimate of the maximum potential combining efficiency as
ηc ≈
| S21 |2 1− | S11 |2
(4.16)
Loss and efficiency based on (4.14) and (4.16) are shown in Figure 2.10, using the average loss over the band from the measured 2-port results of Figure 2.8 and similar measurements for an 8-tray system. The structure can be scaled up to accommodate more devices. Figure 2.10 shows a very small reduction in combining efficiency when the number of elements increases from 8 to 32. The separation between trays and size of the waveguide
100
0
80
-1
60
-2
40
-3
20
-4
0
Total Loss, dB
Efficiency, %
determine the maximum number of elements that the structure can accommodate.
-5
5.0
10.0
15.0
20.0
25.0
30.0
35.0
40.0
Number of Amplifiers
Figure 2.10 Efficiency & total insertion loss of power combiner vs. number of elements in finline array.
37
The impact of failed elements on the total loss of the spatial power combiner system is analyzed in [8]. The array will degrade gracefully, in agreement with the theoretical analysis. The design procedure for dense finline arrays has been applied to the design of active combiner systems forming the basis of successful combiner implementations reported in [9] and [10]. We have shown that the dissipative loss in such slotline arrays is approximately independent of the number of transitions used providing a compelling argument for the use of spatial combiner systems to combine the power from large numbers of active devices. We also explored the limits of scaling tapered finline structures to increasingly dense configurations, showing that the degradation in return loss is graceful but can become significant when designs are scaled by large multipliers. We will need to optimize the design accordingly or use more sophisticated optimization procedures [11] to improve the design of the arrays for dense configurations.
2.2
Modeling of the Coaxial Waveguide Combiner
The concept of using the coaxial waveguide for spatial power combining was first introduced by Dr. Alexanian and Dr. York in [7], with a preliminary demonstration of the idea using passive elements. The center coaxial waveguide is oversized and populated with a finline array and active amplifiers. Optimized coaxial waveguide transitions provide the connection from a type N connector to the oversize waveguide. In this section, we will discuss synthesizing the optimum waveguide 38
transition using small reflection theory.
We will then describe the synthesis of
optimized finline structures for realizing multi-octave operation of coaxial combiner structures, using an adaptation of the procedures developed for rectangular waveguide combiners. The design is applied to a 32-tray system and verified with HFSS simulations.
Design Concept
High-power amplifiers with multi-octave bandwidths are difficult to realize in MMIC technology, particularly when the power requirements call for combining the outputs of multiple MMIC amplifiers simultaneously. In the previous section, we have described an efficient combining technique using finline arrays in a rectangular waveguide, capable of operation over the full waveguide band. In principle it is possible to extend the operating frequency and amplifier capacity of such combiners by operating in a multi-mode environment; in fact, we have found that the use of very dense finline arrays can act to suppress high-order modes in such structures. But there are other difficulties; since the array is excited by a TE10 mode, the amplifier elements are driven non-uniformly, which can reduce the efficiency and distort the saturation characteristics of the system. In addition, the rectangular waveguide environment is dispersive, which complicates broadband impedance matching over an extended frequency range. These difficulties can be addressed by adapting the design to a TEM waveguide environment, such as in a coaxial waveguide. Figure 2.11 illustrates how this might
39
be done, using radial tapered-finline structures distributed uniformly in the annular aperture of an oversized coaxial line. The combiner is fed by gradually flared coaxial lines, which taper down to standard coaxial connectors at either end. Not only can this structure accommodate a large number of amplifiers and provide uniform illumination of the array, but it can also be designed for ultra-wideband operation. Slotline Array
Inner Conductor
MMIC Amplifiers Type N Connector
Outer Conductor
Center Section
Waveguide Taper
Figure 2.11 Schematic of an oversized coaxial waveguide combiner housing a dense finline array, with tapered transitions from type-N connector.
Optimum Waveguide Transition
As shown in Figure 2.11, an optimized coaxial waveguide taper is applied at both ends of the center section to transform from a standard 50-Ohm type N connector to the flared coaxial line. When finline array is loaded in the waveguide, the input impedance of each finline taper is the number of channels times the waveguide impedance. Lower waveguide impedance leads to a smaller waveguide aperture, which is helpful in suppressing higher modes, and keeps the finline taper shorter, resulting in lower conductive loss. Thus, the impedance of the center section was chosen to be 30 Ohms.
40
The reflection from the type N connector to flared waveguide line is minimized by the optimized coaxial waveguide transition. The gradual waveguide taper is synthesized using the small reflection theory of TEM lines, and has an input reflection coefficient Γin ( f ) =
1 θt − jθ d Z (θ ) ln e dθ 2 ∫0 dθ Z 0
(4.17)
where β is the propagation constant, θ t is the round-trip phase delay to a point z along the taper, L is the taper length, and θ t = 2 β L . In order to maintain a target input reflection Γin ( f ) over the desired bandwidth, it has been shown in [5,6] that Z (θ ) must take the form
ln
where A = Cosh −1
2θ Z (θ ) 1 Z L = ln + Γ m A2 F − 1, A . Z0 2 Z0 2θ t
(4.18)
Γ0 Z − Z0 1 Z L , Γ0 = L ≅ ln , and Γ m is the target reflection. Γm Z L + Z0 2 Z0
The characteristic impedance is defined as Z=
1 2π
µ0 D ln( o ) Di ε0
(4.19)
where Do and Di are the outer and inner diameters of the coaxial waveguide opening respectively.
41
Special attention must be paid to the definition of Γ 0 . To get an accurate impedance transition, the approximation of
1 ZL ln must be used because the 2 Z0
equation (4.18) comes from empirical results. An iteration process similar to that described in Section 2.1 is applied to optimize the taper with minimum reflection. The taper dimensions are shown in Figure 2.12 and the reflection coefficient is shown in Figure 2.13. 30
30
25
25 Di (mm)
Do (mm)
20
20
15
15
10
10
5
5
0
0 10
20
30
40
50
Z (mm)
Figure 2.12 Inner and Outer diameter of the optimized waveguide transition.
No solder or epoxy is used in assembling the system; the center conductor of the combiner is designed to mate directly to the center conductor of a type N connector for easy assembly. Unfortunately, this design results in additional loss and reflection. To reduce this loss, solder will be used to permanently connect them when it is fabricated as a product.
42
0 -10 -20 -30 -40 -50 -60 -70 -80 0
5
10
15
20
Frequency [GHz]
Figure 2.13 Reflection coefficient of the optimized transition.
Spectral Domain Modeling of the Finline Array
The finline array can be easily analyzed with a modern Electromagnetic (EM) simulator.
The procedure for synthesizing a broadband impedance-matching
transformer, described in the following section, requires an efficient code that can rapidly and iteratively evaluates the propagation constant of the structure over a range of frequencies and physical dimensions. The Spectral Domain Method (SDM) is well suited for this purpose[12]. A cross section of the loaded coaxial line is shown in Figure 2.14. We assume that each substrate carries two separate finline tapers, and that there is intimate electrical contact with the inner and outer coaxial conductors. Due to the symmetric loading and a dominant mode excitation, the computation domain can be reduced to a single waveguide cell, with PEC (Perfect Electrical Conductor) and PMC (Perfect Magnetic Conductor) boundary conditions as shown in Figure 2.14. The PEC boundary 43
condition is applied again to divide the waveguide cell radially into two unit cells. Each unit cell is left with one tapered finline and a constant ratio of outer radius to inner radius, thereby maintaining identical characteristic impedances. This unit cell could be modeled in cylindrical coordinates, but we choose to approximate each unit cell as a parallel plate waveguide that our previously developed methods for rectangular structures are applicable. y L Slotline taper
z
PEC PMC
PMC
y b PEC g
PMC PEC Loaded Waveguide
Waveguide Cell
PMC a PEC x Unit Cell
Figure 2.14 Schematic cross section of a coaxial waveguide with a uniform loading of radial finline structures.
At this point, the SDM approach is virtually identical with our approach for rectangular waveguide finline arrays in the previous section. The difference is that the sidewall boundary conditions are changed, which is a simple step in the SDM method.
44
3.5
y b
3
εr
g
2.5
12GHz
2
0 c
1.5
a
x
1
7GHz
0.5 0.2
0.4 0.6 normalizedslot
(a)
0.8
1
3 .5 3
εr
SDM 18GHz
2 .5 2 1 .5 1
4 GHz HFSS@8 GHz
0 .5 0 .2
0 .4
0 .6
0 .8
Normalized slot width g/b
1
(b) Figure 2.15 Effective permittivity from SDM and HFSS for varying slot width and frequency (a) inside rectangular waveguide with 2 trays (b) inside coaxial waveguide with 32trays.
The effective permittivity ( ε r = ( β / k0 ) ) versus normalized slot width for a range 2
of frequencies from 4-18GHz is shown in Figure 2.15(b), assuming a 32-tray system with 10-mil Aluminum Nitride substrates. The SDM simulation for a rectangular waveguide is shown in Figure 2.15(a) for comparison. Clearly there is little variation of the propagation constant with frequency for the coaxial waveguide finline array, indicating the desired quasi-TEM behavior. The magnetic field distribution of the slotline also verifies the TEM characteristic. As a numerical check on the SDM code, we analyzed the structure at one particular frequency using Agilent’s High Frequency Structure Simulator (HFSS).
45
The results shown in Figure 2.15 indicate good agreement with the SDM simulation. Furthermore, we observe the Electrical Magnetic field distribution inside the waveguide cell with HFSS simulator. The field shown in Figure 2.16 verifies that dominant mode is TEM mode.
(a)
(b)
Figure 2.16 (a) E field (b) H field of a cross section at 2 narrow slot end inside the waveguide cell.
Test of an Exp-Sin Shape Finline Array y Lt
50 Ohm termination
Slotline taper
z
Figure 2.17 Test circuit of an Exp-Sin shape slotline array.
To verify the simulation results, a 32 tapered finline array was fabricated. The end of the finline was terminated with 50-Ohm thin film resistors. The taper, shown in
46
Figure 2.17, uses the shape of Exp[ Sin(
πz 2 Lt
)] , where Lt is the length of the taper. The
bandwidth is determined by the length of the taper line: the longer Lt, the lower the cutoff limit. For this reason we chose Lt to be 1.2 inches. The finline openings have the same outer to inner radius ratio, resulting in the even distribution of energy between the upper taper and lower taper. The ends of the finlines are terminated with 50-Ohm thin film resistors, which have low parasitic inductances. S parameter measurements were performed. The return loss was plotted in Figure 2.18. The data shows that the return loss was lower than –10dB from 4 to 16 GHz. Good agreement between the measured results and an HFSS simulation prove that this structure can be used as a broadband power combiner.
Return Loss
Measurement HFSS Simulation
0 -5 -10 -15 -20 -25 -30
0
2
4
6 8 10 12 14 16 Frequency [GHz]
Figure 2.18 Measurement and HFSS simulation of return loss for 32 cards passive slotline array with 50 Ohm load.
Although the simulation results and test circuit measurements were in good agreement, the reflection is still too high for high performance amplifier systems
47
when the possibility of further deterioration from bonding wires and other parasitic effects is taken into account. The next section describes an optimization process to design a system with the best overall performance.
Synthesis of Optimized Tapers
In a 32-tray combiner the input impedance of each circuit tray is 32 times the characteristic impedance of the flared coaxial waveguide, which was chosen to be 30 Ohms. Accordingly, at the waveguide opening end of the finline taper, the terminal impedance is 480 Ohms. At the other end of the finline taper, we must connect to a 50 Ohm MMIC amplifier, which sets the target gap size. The design challenge is therefore to realize a broadband 9.6:1 impedance transformation to couple energy from the coax into a set of 50 Ohm MMIC amplifiers.
Ha l f ga p w i dth, mm
2.5 2 1.5 1 0.5 0
5
10
15
20
Position along taper, mm
Figure 2.19 Normalized gap width vs. location along the optimal tapered slotline for a 4-18GHz, 32-tray system of finlines on 10-mil AlN.
The design problem is analogous to the synthesis of tapered transmission-line impedance transformers. We have previously reported the iterative procedure in Section 2.1 for computing the taper shape of a non-TEM transmission line. This 48
method yields the shortest transformer for a specified cutoff frequency and return loss. After replacing the non-TEM parameters with frequency independent ones, we use the same procedure and the SDM results of Figure 2.15 to synthesize an optimized taper for a 4-18GHz, 32-tray system with a specified return loss of -15dB. The synthesis result is shown in Figure 2.19. 0 -5 S11_HFSS S11_SDM
-10 -15 -20 -25 -30 -35 -40
0
5
10 Freqency [GHz]
15
Figure 2.20 Reflection coefficient comparison between SDM and Agilent HFSS for the optimal taper design of Figure 2.19.
The analytical simulation results using the theory of small reflection are shown in Figure 2.20, confirming with our design criteria. Once the taper shape is known, the frequency response can be computed using an EM simulator such as HFSS. The HFSS result is also shown in Figure 2.20 with good agreement with analytical results. It should be stressed again that the sophisticated EM simulator provides an important check on the validity of the synthesis procedure, and can help fine-tune a design once
49
a near-optimal solution has been obtained by more computationally efficient means as described in [13]. Higher modes are investigated with HFSS. The results shown in Figure 2.21 reveal that they are suppressed effectively by the dense finline array. We observed from HFSS simulation that the next higher mode excited in the unit cell is more than 10 dB lower in magnitude than the dominant mode, and the third higher mode is 30 dB lower. The size of the waveguide opening and spacing between the trays play a key role in keeping the higher modes low. Those dimensions should be treated carefully when the design is modified. 0 Second and Third Higher Modes
-15
S11_2nd [dB] S11_3rd [dB]
-30
-45
-60
0
6 12 Freqency [GHz]
18
Figure 2.21 Higher modes inside the waveguide cell.
The design procedures elaborated in this chapter form the basis of the finline array design we used for both medium power and high power amplifiers using coaxial waveguide combiners. Experimental coaxial waveguide power combiner results will
50
be shown in following chapters. The theoretical analysis in this chapter enables the design of high performance waveguide combiners, while giving us the flexibility to optimize the system for different dimensions and also for different tray configurations.
References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13.
Pozar, D.M., Microwave Engineering. 2nd Ed. 1998, New York, NY: John Wiley & Sons. Klopfenstein, R.W., A Transmission-Line Taper of Improved Design. Proc. IRE, 1956. 442: p. 31-35. Schieblich, C., J.K. Piotrowski, and J.H. Hinken. Synthesis of optimum finline tapers using dispersion formulas for arbitrary slot widths and locations. 1984. Verver, C.J. and W.J.R. Hoefer. Quarter-wave matching of waveguide-to-finline transitions. 1984. Schmidt, L.P. and T. Itoh, Spectral domain analysis of dominant and higher order modes in fin-lines. IEEE Transactions on Microwave Theory and Techniques, 1980. MIT-28(9): p. 981-5. Itoh, T., Spectral domain immitance approach for dispersion characteristics of generalized printed transmission lines. IEEE Transactions on Microwave Theory and Techniques, 1980. MTT-28(7): p. 733-6. Alexanian, A. and R.A. York. Broadband waveguide-based spatial combiners. 1997. Rutledge, D.B., et al., Failures in power-combining arrays. IEEE Transactions on Microwave Theory and Techniques, 1999. 47(7, pt.1): p. 1077-82. Nai-Shuo, C., et al., 40-W CW broad-band spatial power combiner using dense finline arrays. IEEE Transactions on Microwave Theory and Techniques, 1999. 47(7, pt.1): p. 1070-6. Nai-Shuo, C., et al. A 120-W X-band spatially combined solid-state amplifier. 1999. Vale, C.A.W. and P. Meyer. Designing high-performance finline tapers with vector-based optimization. 1999. Pengcheng, J., et al. Analysis of a passive spatial combiner using tapered slotline array in oversized coaxial waveguide. 2000. Pengcheng, J., et al., Design of waveguide finline arrays for spatial power combining. IEEE Transactions on Microwave Theory and Techniques, 2001. 49(4, pt.1): p. 609-14.
51
3.
Broadband Medium Power Amplifier Using the Coaxial Waveguide Combiner
CHAPTER 3 Broadband Medium Power Amplifier Using the Coaxial Waveguide Combiner The finline array is the most important element for amplifiers using coaxial waveguide combiner. In previous chapter, a finline array for a 2-inch opening coaxial waveguide is synthesized following the design of coaxial waveguide transitions. All the parameters are optimized with small reflection theory and verified with HFSS. In this chapter, a passive waveguide housing and a finline array are fabricated and tested. The unloaded waveguide system is measured first. Then the optimized finline array is tested with 50-Ohm termination and through line. Inexpensive, low-noise traveling-wave amplifier (TWA) MMIC amplifiers were chosen as a demonstration vehicle. The broadband amplifier using the coaxial waveguide combiner reproduced the individual MMIC amplifier’s frequency response from 3.5-14 GHz, with a 75% combining efficiency[1].
52
3.1
Passive Combiner Measurements 6”
1.77”
2” Type N Connector
Waveguide Taper
6”
1.2”
Center Section
Waveguide Taper
Figure 3.1 Cross section of the unloaded waveguide structure.
The flared coaxial waveguide section was chosen to be 2 inch in the outer diameter of the opening and 1.2 inch in the inner diameter, corresponding to a 30 Ohm line[2]. A bulk center coaxial section was first used for waveguide characterization. The center coax is 1.77-inch long with a pair of 6-inch long coaxial waveguide taper connected on both sides. The waveguide structure was first assembled without finline array loaded inside. Metal and connector loss is shown as the S21 curve in Figure 3.2. Strong mismatch in S parameter can be found at 15 GHz that is caused by higher modes. Besides the TEM mode, waveguide modes can be excited in the oversized center section when the imperfect connections lead to discontinuities. These higher modes will be suppressed when the finline array is loaded. The finline circuitry will divide the waveguide into much smaller waveguide cells that can cut off waveguide modes.
53
0
0
S21 MAG [dB]
S11 MAG [dB]
-10
-10
-20
-20
-30
-30
-40
-40 0
2
4
6
8
10
12
14
16
Frequency [GHz]
Figure 3.2 S Parameter of the unloaded waveguide. 2.0“
0.6" 1.0"
1.77"
`
`
11.25 Degree
Figure 3.3 Mechanical drawing of circuit tray.
After testing the waveguide housing, we begin the process to fabricate circuit trays. We follow the 32-tray scheme developed in last chapter. The Metal tray, which is 1/32 of the center oversized waveguide section, is machined as shown in Figure
54
3.3. The tray is the carrier of the finline antenna and the MMIC amplifier. The finline antenna is realized on a ceramic substrate, and rests over a notched opening in the wedge-shaped metal tray, providing broadband impedance match from the coaxial waveguide to the MMIC amplifiers. A single tray with antenna and MMICs is illustrated in Figure 3.4. When the trays with antenna and amplifiers are stacked radially, the metal trays form a coaxial waveguide aperture populated with the finline tapers. The metal trays are clamped together, then connected with the coaxial waveguide tapers. The radius of the tray is 2 inches, although the radius of the effective opening is only 1 inch. MMIC Amplifier Tapered Slotline Antenna
Figure 3.4 Tray design for the modular coaxial combiner system.
As shown in Figure 3.3, notches are machined on the tray in the direction of wave propagation for accommodation of the circuit substrate. When stacked together, the notches will form slots on the waveguide walls. While the current for dominant mode is along the propagation direction of the wave, currents for higher modes are along the transverse direction. Those slots have no effect on the dominant mode, but can help to suppress the higher modes.
55
50 Ohm Single Wrap Resistor
Figure 3.5 Finline circuit card with 50 Ohm termination.
The optimized finline antennas are first tested using an finline array terminated with 50 Ohm resistors. Figure 3.5 shows a single finline circuit card that has 2 finline tapers onside. These finline tapers were fabricated on 10mil AlN substrates with 3µm gold metalization. Single wrap chip resistors were wire-bonded to the end of each finline taper. Those resistors have the smallest size, 30 mil by 20 mil, and have a ground plane that connects with one side of the resistor. The ground of the resistors is epoxyed to the ground plane of circuit. By close placement of the resistors with the pick & place tool, the length of the bonding wires can be minimized. Figure 3.6 demonstrates the reflection coefficient measurement results for 32-tray and 16-tray systems, which have 32 finline circuit cards and 16 finline circuit cards respectively. The measurement results show good qualitative agreement with the theoretical calculations in Chapter 2, although the maximum reflection coefficient is higher to some extend in the passband (~ -10dB). There is evidence of mismatch at the interface to the type-N connector, leading to the rapid undulations in the frequency response. This mismatch can be attributed to poor electrical connection between the type-N connector and the center-conductor of the coaxial waveguide taper. Meanwhile, the bonding wires connecting with resistors also contribute to the mismatch. Figure 3.7 shows the field distribution from HFSS for a finline circuit with 50-Ohm terminations. Bonding wires for resistors are included in the simulation. The
56
light spots represent peaks of the standing wave caused by mismatch at the termination. Obviously, the mismatch was mainly caused by the parasitic effect of the bonding wires. 0
S11 _16 Tray [dB] S11_32 Tray [dB]
-4
-8
-12
-16
-20
0
6 12 Frequency [GHz]
18
Figure 3.6 Return loss measurement for 16-tray and 32-tray combiner with 50 Ohm terminations on the finline circuits.
50Ohm Resisstor Bonding wires
Figure 3.7 Field distribution of waveguide cell with 50 Ohm resistor and bonding wire integrated.
Low combining losses are required to maintain good combining efficiency. A set of back-to-back finline cards are fabricated, with a 50-Ohm microstrip through line bonded in place of the MMIC amplifier to connect the input and output antennas as
57
shown in Figure 3.8. The commercially available MMICs all use microstrip line, so we use a microstrip though line with same dimension as the dummy amplifier. The measurement of finline array with through lines will be effective to estimate the loss of a combiner with MMIC integrated. The microstrip through lines are soldered to the surface of circuit card by eutectic alloy, then bonded to the end of the finline taper by 1 mil gold wires, which are around 300 µm in length.
Figure 3.8 Finline circuit card with through line connected in the middle. 0
-2 Loss_32 Tray [dB] Loss_16 Tray [dB]
-4
-6
-8
0
6 12 Freqency [GHz]
18
Figure 3.9 Dissipative loss for 16-tray and 32-tray combiners with 50 Ohm microstrip through-line in place of the active device. Inset shows circuit configuration.
Figure 3.9 shows the measured dissipative loss for 16 and 32-tray combiners using the 50-Ohm microstrip through line. The loss increases approximately with
58
f as expected. Again, we observed mismatch from the type-N connectors that leads
to small periodic dips in the loss. Higher-order modes are also observed at the higher frequencies and lead to dips in the S21 curve. However, those dips are still within the acceptable range and less of a concern comparing with the broad bandwidth of the system. Although the S21 measurement result is not the simple sum of the loss of input and output array, it allows us to roughly quantify the output array’s combining loss as half of the measured total loss.
Thus, the combining loss varies from
approximately 0.3dB at 4GHz to 1dB at 18GHz.
This translates to a potential
combining efficiency in excess of 75% over the entire band. 5
S11_16 Tray[dB] S11_32 Tray [dB]
0
-5
-10
-15
-20
0
6
12
18
Frequency [GHz]
Figure 3.10 Return loss for 16-tray and 32-tray combiners with 50 Ohm microstrip through-line in place of the active device.
Return loss measurement is shown in Figure 3.10. We observe stronger reflection than the measurement with 50-Ohm terminators. When we use through lines to connect input and output finline tapers, reflection from both the input and output
59
finline transition will add together at the input port. That will lead to about 3 dB increases in the reflection coefficient. It is important to note that the dissipative loss stays constant when we double the number of elements in the waveguide finline array, consistent with the observation in last chapter and [3] that dissipative loss is approximately independent of the number of finline transitions used. This characteristic makes the coaxial combiner promising for combing the power from large number of active devices. From the comparison of 16-tray and 32-tray configuration, we can see the 2 schemes have very similar results. The finline array is designed for 32-tray configuration; but when we only load 16-tray inside, the impedance transformation ratio for each finline taper is reduced from 9.6:1 to 4.8:1. Simulation with HFSS shows that 16-tray scheme has a little better impedance matching over 32-tray scheme. However, the parasitic effect of the bonding wires and other discontinuity overrules the small differences. Since the difference is smaller, we choose 16-tray configuration in the demonstration of the active amplifier.
3.2
Performance of the Active Combiner
For final testing of the combiner system, we use a set of 32 broadband MMIC amplifiers to build an active broadband amplifier. In this section, we use the term combiner referring to the broadband amplifier using the coaxial waveguide combiner. The Triquint TGA8349 TWA MMIC amplifiers are selected as the demonstration vehicles. Typical input SWR of TGA8349 is 1.2:1, and output SWR is 1.3:1. This 60
MMIC can generate 16dBm output power at 1 dB compression point. The small signal bandwidth of this MMIC amplifier is from DC to 14 GHz.
Bias Pad
Bias Capacitor
Input Taper
MMIC
Output Taper
Figure 3.11 The circuit tray with MMIC amplifiers.
A circuit tray with antennas and MMIC amplifiers is shown in Figure 3.11. Each tray carries a finline circuit card with 2 MMIC amplifiers that are soldered to the surface of the AlN circuit card. Since AlN has very high thermal conductivity, the heat generated by the low power MMIC amplifier can be effectively dissipated to the waveguide through the AlN board. A total of 32 MMIC amplifiers are integrated into the 16-tray system. The biasing pads are epoxyed at the side of the circuit tray. The dual-gate GaAs FET MMIC amplifier needs 4 bias voltages for gate, drain, second gate control and ground separately. Single layer capacitors are placed for filtering AC signals on DC bias line. Bonding wires are used for the DC and RF connection. Multiple wires are placed for RF connections to reduce parasitic inductance. The 10 mil thick AlN substrate sits on the notch of the aluminum circuit tray, carrying all of the finline tapers, MMIC amplifiers and bias capacitors. 61
Figure 3.12 Side view of the loaded section.
Input Waveguide Transition
Loaded Section Output Waveguide Transition
Bias Lines Circuit Tray
Figure 3.13 The overview of the active combiner.
An open view of the loaded center section is shown in Figure 3.12. The coaxial waveguide opening is formed when we stack all the circuit trays together.
Figure
3.13 shows the completely assembled combiner system including broadband coaxial tapers for feeding the loaded section. Bias lines connect the pins on the circuit tray to a bias board. 8-channel KEPCO power supply provides drain voltage and current. Each channel drives 2 circuit trays that have 4 MMIC amplifiers in total. The
62
separation of DC power supply maintains isolation between the bias lines that will help to protect other circuits when some MMIC amplifiers fail. 20
20
S21_Combiner [dB]
S21_MMIC [dB]
15
15
10
10
5
5
0
0 2
4
6
8 10 12 Frequency [GHz]
14
16
Figure 3.14 Measured small-signal gain of the active combiner and individual MMIC amplifier.
The system is very stable and no additional circuits or biasing capacitors are needed to suppress oscillations. Figure 3.14 and Figure 3.15 show the small-signal gain and reflection coefficient of the completed active combiner system. The broadband property is verified by both of the figures. The combiner has 10 to 11.5 dB gain over a broadband from 3.5 GHz to 14GHz. The passive combiner itself has a lower cut-off frequency. It has potential to operate up to 18 GHz. The upper end of the bandwidth of the combiner is limited by the frequency response of the MMIC. The loss of the system is consistent with the loss of the passive combiner. The total loss of the combiner including ohmic and mismatch losses is nearly a constant of 2 dB over the entire band.
This corresponds to 1dB output loss and hence 75%
combining efficiency. 63
0
0 S11 [dB]
S22 [dB]
-5
-5
-10
-10
-15
-15
-20
-20 5
10 Frequency [GHz]
15
Figure 3.15 Reflection coefficient of both input and output ports for the active combiner. 35
20 Gain [dB]
Output Power [dBm]
30
15
25
10
20
5
15
0 5
10
15 Input Power [dBm]
20
25
Figure 3.16 Output power and gain vs. input power.
Large-signal measurements were also recorded, using a TWT amplifier to drive the array into compression. Two directional couplers were connected at both the input and output port. Power sensors are connected to the couplers to measure the input and output power. The loss of this configuration is calibrated first. Then power
64
measurements are carried out at 10 GHz. Measurement setup will be introduced in more details in chapter 5. As shown in Figure 3.16, the 16-tray combiner system generated 1-Watt CW power at the 1 dB compression point. Using a nominal output power of 16dBm for each MMIC amplifier, this translates to a measured combing efficiency of 80% at this frequency, in good agreement with previous estimates based on system loss. The power-frequency response was characterized from 4 to 15 GHz using a fixed input power of 20 dBm. The result is shown in Figure 3.17. 29 to 30 dBm output power is reached in most of the band. It also follows the expected output power curve of the MMIC amplifier, indicating that the higher stopband of the combiner is determined by the MMIC amplifier. 35
15
30
10
25
5
20
0
-5
15 2
4
6
8 10 12 Frequency [GHz]
14
16
Figure 3.17 Output power and gain vs. frequency.
The 32-MMIC combiner, which has 16 circuit cards, has been demonstrated and 1-Watt output power is achieved. It verifies that the optimized broadband transition
65
designs based on SDM analysis are effective for large-scale power combining inside coaxial waveguide. The combiner system has a capacity of integrating as many as 64 MMIC power amplifiers. A good impedance match is achieved from 3.5 GHz up to 18 GHz. More characteristics such as residue phase noise will be covered in chapter 5. We will continue our work to optimize the system and demonstrate a higher power module in the next chapter.
References 1. 2. 3.
Pengcheng, J., et al., Multioctave spatial power combining in oversized coaxial waveguide. IEEE Transactions on Microwave Theory and Techniques, 2002. 50(5): p. 1355-60. Alexanian, A., Planar and Distributed Spatial Powe Combiners. ECE Technical Report, 1997. #97-20. Pengcheng, J., et al., Design of waveguide finline arrays for spatial power combining. IEEE Transactions on Microwave Theory and Techniques, 2001. 49(4, pt.1): p. 609-14.
66
4.
Design of High Power Amplifier Using the Coaxial Waveguide Combiner
CHAPTER 4 Design of High Power Amplifier Using the Compact Coaxial Waveguide Combiner In this chapter we report a high power amplifier design using enhanced broadband passive combiner structure. A significant reduction in size has been achieved while maintaining a 6-18GHz bandwidth and capacity for 32 MMIC amplifiers. A broadband slotline to microstrip line transition was developed and monolithically integrated with the finline antennas, to eliminate a troublesome bond-wire transition in earlier design and provide better compatibility with commercial MMIC amplifiers. The Spectral Domain Method (SDM) is applied to compute the field in the structure, and small reflection theory is applied again to synthesize the waveguide taper and optimize finline taper array.
67
4.1
Motivation
As shown in Figure 4.1, our previous design is undesirably large in both the diameter of the center section and the length of waveguide taper. A new compact combiner structure, which provides the same combining performance as the larger one, was developed for the high power module. N Connector
Center Section Previous Design
Waveguide Taper
New Design
Figure 4.1 Comparison between previous design and the new compact design.
For low power demonstrations, the MMIC amplifiers were die-attached to the AlN substrate on which the finline antennas were processed. However, junction thermal resistance between AlN substrate and metal carrier will lead to a high temperature step when the MMIC amplifiers generate a large amount of heat. In the high power module, MMIC amplifiers will need to be attached to the metal carrier for better thermal management. In Chapter 3, bonding wires were used to connect from MMIC amplifier’s input/output pads to the end of the finline. Since the MMIC amplifiers sit on the ground of the finline circuit, the bonding pads of MMIC amplifier are 100 um (MMIC thickness) higher than those on the finline circuit. The bonding wires have to go down a slope for connection. It increases the difficulty for
68
assembly, also leads to comparably longer boding wires. The parasitic inductance of the wire bonding deteriorates the impedance matching. Therefore it is necessary to keep the connection pads of the antennas and the commercial MMIC amplifiers in the same plane and minimize the length of bonding wires. A monolithic slotline to microstrip line transition was developed to serve this purpose. The new transition has a broad bandwidth and doesn’t limit the overall frequency response of the system. If needed, the transition design can be easily extended to higher frequencies. The new combiner’s passband was changed to 6-18 GHz since the broadband MMIC amplifiers we are going to use work exactly at this band. The waveguide housing and the finline antenna array were synthesized again to meet the changes on the bandwidth and dimensions. Same performances as the original design were maintained through the optimization process.
4.2
Coaxial Waveguide Design
To re-engineer the combiner and minimize the physical size of coaxial waveguide, we applied the small reflection theory to the coaxial waveguide taper again. We succeeded in reducing the diameter of the center section from 4 inches to 2.2 inches, and the length of each waveguide taper from 6.2 inches to 2.2 inches. The new compact waveguide housing were machined and tested. The simulation and measurement results of the unloaded coaxial waveguide are shown in Figure 4.2. Similar impedance matching and loss as the original waveguide housing are achieved for the new design. One improvement is that we don’t observe the same spike at 15 69
GHz as the previous result in chapter 3. The reduction in waveguide size helps to move the higher modes out of the 6-18 GHz band. Two N-to-SMA connectors were used in measurement, which is not considered in the simulation. The loss and reflection of the N-to-SMA connector are also shown in Figure 4.2 and contribute partly to the discrepancy between simulation and measurement. 0
0
-10
-2
-20
-4
-30
-6
-40
System S11 [dB] Simulation S11 [dB] Connector S11 [dB]
System S21 [dB] Simulation S21 [dB] Connector S21 [dB]
-8
-10
-50 0
5
10
15
Freqency [GHz]
Figure 4.2 S parameter of unloaded coaxial waveguide.
4.3
Synthesis of Waveguide Finline Array
Figure 4.3 Layout of the finline taper and the slotline to microstrip line transition.
70
Sixteen finline circuit cards were radially loaded inside the center waveguide opening. The symmetry and the thinness of the circuit substrate allow us to focus analytical attention on a unit cell. The unit cell was further approximated by a parallel-plate waveguide as described in chapter 3 and [1]. The Spectral Domain Method and the small reflection theory were used again to optimize the finline taper. The layout of the new finline taper is shown in Figure 4.3. Each circuit tray carries 2 finline tapers. To improve the linearity of the combiner, power should be distributed evenly to each taper. However, the field inside the waveguide is not radially uniform. So each of the finline taper on a single tray is designed with a different slot opening to equalize the power. When we put the finline array inside the waveguide, they will have the same outer radius to inner radius ratio.
4.4
Slotline to Microstrip Line Transition
In Nick Cheng’s design, a separate microstrip line transition was integrated to provide connection from the end of finline to MMICs. This approach needs a separate microstrip line circuitry and carefully bonding of the circuitry to finline substrate. In our previous design discussed in chapter 3, we attached MMIC to the finline circuit and directly connected the MMIC to the end of finline. This approach simplifies the circuitry; but since the connection pads are not in the same plane, it increases the difficulty for wire bonding. Moreover, this design is not effective for high power MMICs that will generate a lot of heat. To simplify the assembly and reduce parasitic
71
inductance, a monolithic slotline to microstrip transition was employed in the new design [2]. As shown in Figure 4.3, the finline taper was processed on the back of the AlN substrate, with a 90-degree slotline short stub at the end. A 90-degree microstrip line open stub is aligned to the slotline stub on the top of the substrate. The center of the 2 stubs are on the same line perpendicular to the surface, and their edges are parallel to each other. When put onto a metal carrier, the slotline becomes the ground of the microstrip line, which is in the same plane as the ground of the MMIC amplifiers. Due to the space limitation inside the compact structure, the stubs have to be bent 15degree inwards, and the microstrip line detours around the slotline stub in a small loop.
Zs
Zs
Zm
1:n jXom
Zm
jXss Ys
Ys Γin
jBs
jBm
Gm
Figure 4.4 Circuit model of slotline to microstrip line transition.
The transition in Figure 4.3 is modeled in Figure 4.4. The short slotline stub and open microstrip stub can be treated as a series of straight sections with various widths that are cascaded together [3]. To improve the accuracy in modeling the structure
72
enclosed in waveguide, we used 3D simulator Agilent HFSS to compute the reactance of the slotline stub jXss and microstrip stub jXom. Then we applied the values into the circuit model and optimized other parameters in the transition. In the circuit model, Zm and Zs is the characteristic impedance of microstrip line and slotline respectively, and n is the transformer ratio, n=−
1 b2 b Ey ( h) dy , Vo ∫− 2
Ey (h) = −
Gm =
2π u 2π u Vo (cos h − cot qo sin h), b λo λo
n 2 Zm , Zm 2 + Xom 2
Bm = −
n 2 Xom , Zm 2 + Xom 2
Ys =
1 , Zs
Bs = −
(4.1)
1 . Xss
Here, Vo is the voltage across the slot and Ey(h) is the electric field of the slotline on the other surface of the substrate. The details of the calculation of n are given in [4]. The reflection coefficient can be expressed as: Γin =
Ys − Gm − j ( Bs + Bm) . Ys + Gm + j ( Bs + Bm)
(4.2)
Our goal is to achieve bandwidth from 6 to 18GHz. Simulation shows that the lower band is more sensitive to the dimensions. So we chose 10GHz as the center frequency, then optimized the transition at this frequency to satisfy Ys=Gm, and Bs=Bm.
73
The impedence of the microstrip line is fixed to be 50 Ohm, corresponding to a 278 um strip width on a 254 um thick AlN substrate. The characteristic impedance of the slotline times n2 should be close to 50 Ohm. We chose the width of the slotline to be 40 um to match Ys with Gm. Due to the limitation of space inside the waveguide structure; the radius of the slotline stub was selected to be 2000 um. To realize Bs= Bm and minimize the reflection coefficient, the microstrip open stub should have a radius of 1500 um. Further optimization with Agilent HFSS showed that a microstrip stub with a 1600 um radius has a wider bandwidth. Simulation results shown in Figure 4.5, indicates that the slotline to microstrip transition can achieve a bandwidth of more than 12 GHz, from 6 to 18 GHz. If scaled down properly, the transition can also work at higher bands. The parasitic effect is much smaller than the bonding wire connection used in earlier work.
Figure 4.5 S parameters of slotline to microstrip line transition from Agilent HFSS simulation.
74
4.5
Compact Passive Structure of Coaxial Waveguide Combiner
(a)
(b) Figure 4.6 (a) Open view of the passive coaxial waveguide combiner, (b) circuit card with back-to-back finline antenna and transition to microstip line.
Figure 4.6(a) shows the open view of the combiner loaded with 16 circuit cards for loss measurement. Each circuit card has 2 transitions placed back to back as shown in Figure 4.6(b). The microstrip line dummy circuit used in last chapter was replaced by straight microstrip line since the integration of the new slotline to microstrip line transition. Slots are machined along the walls of the center flared coaxial waveguide, and circuit cards are slid into it as shown in Figure 4.6(a). The performance of the overall passive structure, which includes connectors, waveguide tapers, a divider and a combiner, is shown in Figure 4.7. 6-18 GHz bandwidth is observed from both the simulation and the measurement. The majority of the
75
discrepancy comes from the connectors, which introduce more than 1 dB of loss at the higher band.
Figure 4.7 Comparisons between simulation and measurement for passive coaxial waveguide combiner.
4.6
Leakage from Output to Input
When the MMIC amplifiers are integrated, metal trays similar to the one introduced in last chapter will be machined except that the size will be much smaller. The trays will be the carriers for antennas and MMIC amplifiers. When the wedge shaped metal trays are stacked together, the waveguide is partially blocked by the bridges that connect the inner and outer part of the metal trays. The dominant mode cannot propagate along the waveguide except through the finline circuits. While the metal bridges leave some room for MMIC amplifiers when stacked together, they will also provide passages for some of the higher waveguide modes. Since the circuit card
76
shown in Figure 4.6(b) can pass the energy through it, it is not viable to valuate the leakage of the waveguide from its S parameter measurement. Here we use the previous measurement results of the bigger combiner, which is loaded with a finline array that is terminated with 50-Ohm resistors. The measured S12 of that combiner is equivalent to the leakage from the output to the input through waveguide modes. From Figure 4.8 we can see when the waveguide is less loaded with only 16 tray compared to 32 tray, the leakage through higher modes is more severe. At higher frequencies, the 16-tray scheme has more leakage. The reason is the waveguide cell is larger for less loaded combiner and its size is not small enough to cut-off all the higher modes when frequency is higher than 15 GHz. The compact version introduced in this chapter will help to increase the cut off frequency of a waveguide cell to 18 GHz. The in band (6-18GHz) isolation can be improvement for the compact design. 0 -10
S12_16 Tray [dB] S12_32 Tray [dB]
-20 -30 -40 -50 -60 0
5
10 15 Frequency [GHz]
Figure 4.8 Waveguide leakage from output to input.
77
20
Since we use very high gain MMIC amplifiers, high leakage from the output to the input will cause instability problems when the output is not well matched. To account for this, we put Emerson EM absorbent at the bottom of the metal tray’s bridge. When the trays are stacked together, the absorbent will be right on the top of the MMIC and can increase the attenuation through that passage. Figure 4.9 shows a flipped circuit tray and the place to put the absorbent.
Apply thin layer of EM absorbent
.05" 0.36"
0.44"
Figure 4.9 Thin layer of EM absorbent on the back of the tray.
4.7
Uniformity
A waveguide cell (1/16th of the coaxial waveguide) and its electrical field distribution along the finline are shown on the left side of Figure 4.10. Since the field inside the coaxial waveguide is radially distributed, the finline pair is designed with the same outer radius to inner radius ratio to keep the impedance the same.
78
Mode1 Out-of-Phase
Mode2 In-Phase
Figure 4.10 In-phase and out-of-phase modes at output port (the microstrip line end)
0 -5 -10
S21_Mode2 [dB] S21_Mode1 [dB]
-15 -20 -25 -30
4
6
8
10
12
14
16
18
20
Frequency [GHz] Figure 4.11 Strength of mode 1 and mode 2 at output port
The electrical filed distribution at the output port (microstrip end) is shown on the right side of Figure 4.10. In this design, each microstrip line couples energy from the complimentary side of each finline. The microstrip lines at output port will output
79
out-of-phase signals with the same strength that is mode 1 at the output port. But the signal strength at the two lines can’t be identical as ideally because of coupling between the two transitions. The effect can be as attributed to the combination of the out-of-phase mode (mode 1) and the in-phase mode (mode 2). Figure 4.11 shows the amplitude of the 2 modes. Result shows that mode 1 is the dominant mode, and mode 2 is not neglectable. The combination of 2 modes will lead to different amplitude and phase at the microstrip lines. Let vs1 and vs 2 represent the signals at each line, their amplitude and phase are vs1 = Am1 cos ω t + Am 2 cos(ω t + ∆θ m ) | vs1 |= ( Am1 + Am 2 cos ∆θ m ) 2 + ( Am 2 sin ∆θ m ) 2
φs1 = arcsin
(4.3)
Am 2 sin ∆θ m | vs1 |
vs 2 = Am1 cos ω t − Am 2 cos(ω t + ∆θ m ) | vs 2 |= ( Am1 − Am 2 cos ∆θ m ) 2 + ( Am 2 sin ∆θ m ) 2
φs 2 = arcsin
(4.4)
Am 2 sin ∆θ m | vs 2 |
where Am1 & Am2 are the amplitude of mode 1 & 2 respectively, and ∆θ m is the phase difference of mode 1 & 2. As shown in Figure 4.12, the undesired in-phase mode (mode 2) causes a maximum of 8% amplitude difference and maximum of 10-degree phase error
80
between the inner and outer finline. Hence, the 16 inner channels will have the same phase and amplitude difference as the 16 outer channels. 1.3
200 Slot1_Amp Slot2_Amp
1.2
Phase Difference [degree]
180
1.1
160
1
140
0.9
120
0.8
4
6
8
10
12
14
16
18
100 20
Frequency [GHz] Figure 4.12 Phase and amplitude difference at the 2 microstrip line.
Figure 4.13 Generic combiner system.
The variation in phase and amplitude causes imperfect summation and leads to reduction in combining efficiency. We need to calculate the output power in the form
81
of the sum of N channel signals to quantitatively analyze the change in efficiency due to non-uniformity. We assume the input signal is Ain = A cos(ω t )
(4.5).
Considering the phase and gain variation in each channel of the splitter, then the input and output signal of each amplifier will respectively be A(1 + δ Ai ) cos(ω t + δφi ) N A(1 + δ Ai )G (1 + δ Gi ) = cos(ω t + δφi + δϕ i ) N
ain ,i = bout ,i
(4.6)
where G is the nominal voltage gain, δ Ai and δφi are the amplitude and phase variation of each channel in the splitter, δ Gi and δϕ i are the gain and phase errors of each amplifier. If we assume the combiner has the same characteristic as the splitter, the sum of the N channel signals can be expressed in phasor form as 1 N AG N bout ,i = ri (1 + δ Ai ) 2 (1 + δ Gi ) cos(ω t + 2δφi + δϕ i ) ∑ ∑ N i =1 N i =1 N AG ri (1 + δ Ai ) 2 (1 + δ Gi )e j (2δφi +δϕi )e jω t = ∑ N i =1
Bout =
(4.7)
where we assume the combiner has the same phase and amplitude variation as the splitter. Here we also considered the statistical failure of the MMIC that is represented by ri .
82
The details of the analysis of phase and gain error introduced by amplifiers are covered in [5]. Here we add the effect of the phase and amplitude variation introduced by the splitter and combiner. 2
The output power is proportional to P = Bout . If we denote the “no-error” output power as Po = ( AG ) 2 , then the relative change in the presence of error is P 1 = 2 Po N
N
N
∑∑ r r (1 + δ A ) (1 + δ A ) (1 + δ G )(1 + δ G )e i =1 j =1
2
i j
i
2
j
i
j [2(δφi −δφ j ) + (δϕi −δϕ j )]
j
(4.8).
If we neglect the phase and amplitude error from the combiner and splitter, after taking the ensemble average and assuming the individual amplitude and phase error of each amplifier have the same variance, we have 2 2 2 P 1 2 − δϕ 2 − δϕ 2 − δϕ ] ≈ Pe e = Pe e + [ Pe (1 + δ G 2 ) − Pe e Po N
(4.9)
where Pe is the survival rate of devices and Pe = r . If we only consider the phase and amplitude from the combiner and splitter and neglect the error from amplifiers, the N channels will be divided into 2 groups, the inner channel group and the outer channel group. N/2 inner channels have the same amplitude and phase, and we assume the amplitude and phase error is δ A1 and δφ1 . Here we assume N is an odd number. The N/2 outer channels also have the same amplitude and phase, and we assume the amplitude and phase errors are δ A2 and
83
δφ2 . We must notice that the inner group and outer group are intrinsically out of phase, so δφ2 is the phase error relative to 180 degree phase. Then we have P 1 = 2 Po N
N
N
∑∑ (1 + δ A ) (1 + δ A ) e 2
2
i
i =1 j =1
j [2(δφi −δφ j )]
j
1 = [(1 + δ A1 )4 + (1 + δ A2 )4 + 2(1 + δ A1 ) 2 (1 + δ A2 ) 2 cos 2δφ ] 4
(4.10)
where δφ = δφ2 − δφ1 . Since δφ2 is the phase error relative to 180 degree phase, δφ is also the phase error relative to 180 degree phase. Using the amplitude and phase error values from Figure 4.12, we can calculate the combining network’s efficiency with equation (4.10). The efficiency is expressed as the output power over “no-error” output power Po. Figure 4.14 shows the efficiency over the frequency band from 5 to18 GHz. 1 0.9 0.8
P/Po
0.7 0.6 0.5
4
6
8
10
12
14
16
18
20
Frequency [GHz]
Figure 4.14 Efficiency of the combining network.
We can see that the non-uniformity in the inner and outer channel leads to reduction in combining efficiency. Asymmetric the input array and the output array 84
can compensate for the amplitude and phase error. It will improve the uniformity related combining efficiency to 100%.
4.8
Fabrication Procedure
The copper carriers were machined by the UCSB engineering machine shop. An Electric Discharge Machine (EDM) was used to achieve accuracy of 3 mils or less. Cost and fabrication time will drop dramatically if we use a die-cast process. The metal trays were then plated with high purity gold using an electrolyte-plating process. The gold layer will protect the copper from oxidation and reduce RF loss.
Lower Right
Left Upper
Left Triquint MMIC AuSn CuMo Subcarrier AuGe
Cu Carrier (Gold plated)
Right
Figure 4.15 Assembly procedure.
The MMIC amplifiers were next mounted onto the metal trays with Cu/Mo subcarriers by a eutectic solder. The coefficient of thermal expansion of GaAs is 85
much different from that of Cu, which will cause problems in both assembly and lifetime if they are directly mounted together. To increase the reliability, a Cu/Mo subcarrier is necessary since it has similar thermal expansion coefficient as the GaAs substrate. A small cavity with same size as the Cu/Mo subcarrier was machined on the metal tray for better alignment. Since the GaAs MMIC is destroyed when the temperature is higher than 320 oC, the Cu/Mo subcarrier was bonded to the metal tray first with a Au/Ge eutectic solder at 360 oC. Then the GaAs MMIC amplifier was bonded to the Cu/Mo subcarrier with a Au/Sn eutectic solder at 280 oC. The procedure is shown in Figure 4.15. Some GaAs MMIC assembly guidelines are:
• AuSn (80/20) solder with limited exposure to temperatures at or above 300 oC • Use an alloy station or conveyor furnace with a reducing atmosphere • No fluxes should be utilized • Coefficient of thermal expansion matching is critical for long-term reliability • Storage in dry nitrogen atmosphere The component placement and adhesive attachment assembly notes are:
• Vacuum pencils and/or vacuum collets are the preferred method of pick up • Air bridges during placement should be avoided.
86
• Force impact is critical during auto placement • Curing should be done in a convection oven; proper exhaust is a safety concern • Microwave or radiant curing should not be used because of differential heating • Coefficient of thermal expansion matching is critical
4.9
Circuit Tray & Bias Line
Figure 4.16 Circuit tray of combiner.
The assembled circuit tray is shown in Figure 4.16. The 2-channel MMIC amplifier sits on the bridge that connects inner and outer sections. Input and output antennas were epoxyed on both sides. Bonding wires connect the end of microstrip line to the input and output pads of the MMIC amplifier. Bias pins were epoxyed at the outer side of the tray. DC currents are input to MMIC amplifiers through biasing lines and bonding wires. The overall DC impedance from pins to the MMIC amplifier’s pads is from 0.2 Ohm to 0.3 Ohm.
87
4.10
Efficiency, Reliability and Thermal Analysis
Amplifier Efficiency
Solid state power amplifiers (SSPA) are superior to tube amplifiers in size and scalability. However, a great challenge for SSPAs is efficiency. Three definitions of efficiency are commonly used. Drain efficiency is defined as the ratio of RF-output power to dc-input power, i.e., η = Po / PDC . Power added efficiency (PAE) incorporates the RF-drive power by subtracting it from the output power, i.e., η = ( Po − Pi ) / PDC . PAE gives a reasonable indication of power amplifier (PA) performance when gain is high; however, it can become negative for low gains. An overall efficiency such as η = Po /( PDC + Pi ) is usable in all situations. Class A, B, AB and C amplifiers are widely used in PA designs, but their drain efficiency only ranges from 50% to around 85% theoretically. Innovative class D, E, and F amplifiers can improve the drain efficiency up to unit ideally. Recent achievements from Raytheon and TRW have shown class-E amplifiers with a power added efficiency of over 60% using a pHEMT and DHBT respectively at X band[6, 7]. Although the new class-E amplifier design has shown a wide bandwidth over 1 GHz, it is still mandatory to use a class A amplifier for broader bandwidth applications. In the lossless situation, class A amplifiers have a drain efficiency of 50%. However, considering the lossy mechanism inside the devices and the matching circuit, the power added efficiency is only around 30%.
88
If the combiner has an output combining efficiency of 75%, the amplifier’s overall PAE is only a little more than 20%. It means 4 times the output power is wasted in the form of heat. When there is no input signal to the amplifier, 5 times the rated power is converted into heat. A 50-watt output power rated amplifier must have the ability to dissipate more than 250 watt of heat effectively. The pressure will be much alleviated if narrowband high efficiency class E amplifiers are integrated into the combiner. Another modification is to use Class B push pull amplifiers that will increase the efficiency decently while maintaining broad bandwidth.
Amplifier Reliability
The combiner system integrates a large quantity of MMIC amplifiers. Although GaN and SiC amplifiers have been demonstrated with promising performance in research labs, there are still no mature commercially available products for frequencies higher than C band in high power applications. GaAs is still the dominant material for MMIC power amplifiers. The reliability of GaAs devices is the key parameter for a high power combiner system. GaAs device reliability involves probability statistics, time, and a definition of failure. Given the failure criteria, the most direct way to determine reliability is to submit a large number of samples to actual use conditions and monitor their performance against the failure criteria over time. Since most applications require device life times of many years, this approach is unfeasible. To acquire reliability
89
data in a shorter time, an acceleration factor must be used to quicken the failure process. In most cases this acceleration factor is high temperature. The rationale behind high temperature life tests is that most physical and chemical processes are accelerated by temperature. The rate of acceleration for each failure mechanism is a constant called the activation energy. Most GaAs semiconductor failure mechanisms follow the Arrhenius equation that relates the rate of failure to temperature, time and activation energy. The Arrhenius Equation and Activation Energy is expressed as:
Tf 2 = Tf1 e
Ea ( 1 - 1 ) k T2 T1
(4.11)
where Tf = time to failure, Ea = activation energy in electron-volts (eV), k = 8.6142 E-5 (eV/°K), T = absolute temperature in Kelvin (°C +273).
Figure 4.17 Typical TriQuint Texas MESFET, HFET, and PHEMT median life time data.
90
To predict lifetimes at normal operating temperatures, multiple high temperature lifetime tests were performed by TriQuint Semiconductor. The median life from each of the tests is plotted as a point on an Arrhenius graph. All points are fit on a line. The slope of the line is the activation energy. Median life at any temperature can then be determined. Typically three temperatures are used. This data is available for MESFET, HFET, and PHEMTs under typical bias conditions. The failure criteria commonly used is 1 dB RF output power degradation. The ¼ um PHEMT process of TriQuint has shown a median lifetime of 2e6 hours at channel temperature of 140 oC with failure criteria being 1 dB degradation of RF output power. The TriQuint MMIC amplifiers used in our combiner system were strictly tested. In the test process, one and a half mil thick 80%Au/20%Sn eutectic solder was used to mount the MMIC amplifier on a 20 mil thick Cu/Mo Carrier, then placed on a hot plate at a temperature of 70 oC. The worst case would be that no is RF applied and 100% of the DC power is dissipated. Under the condition of Vd = 8 V, Id = 2.4 A, Pdiss = 19.2 W and channel temperature =145 oC, the lifetime is tested to be 1.6 E+6 hours, which is equal to 180 yrs.
Thermal Analysis
As shown in last section, the lifetime of the GaAs MMIC amplifier is over 180 years if the hot plate temperature is kept below 70 oC. If we want the solid state amplifier to last 50 years which is much longer than TWTA’s 15 year lifetime, the plate temperature should be no higher than 85 oC as indicated in Equation (4.11). If
91
the temperature of the combiner’s metal tray is controlled to be lower than 85 oC, the amplifier will be able to serve most communication systems. Generally, the ambient temperature is 25 oC. Heat transfer occurs as a result of the temperature gradient between the amplifiers and the environment. The heat generated by the MMIC amplifiers is dissipated into the air by 2 modes: conduction and convection. Conduction is the transmission of heat through a substance without perceptible motion of the substance itself. Heat is conducted by the copper tray to its outside surface. Convection is the term applied to heat transfer due to the bulk movement of a fluid. When a fan blows air over the outside surface of the carriers, the air absorbs heat from the carriers by convection. The one dimensional heat conduction equation is
qx = − kA
where qx is the heat flow in the x direction,
dT dx
(4.12)
dT is the temperature gradient or slope dx
of the temperature curve, A is the area normal to the heat flow direction and k is the thermal conductivity of the material. The numerical value of thermal conductivity is an indication of how fast heat is conducted through a material and is a macroscopic representation of all the molecular effects that contribute to the conduction of heat through a material. Based on equation
92
(4.12), measurements can be made to determine the thermal conductivity for various substances. As shown in Figure 4.18, copper is only inferior to silver in thermal conductivity at room temperature and is 1.6 times better than aluminum. So we chose copper as the material to be used for the metal carriers in the high power combiner instead of aluminum that was used in the medium power combiner design. The temperature gradient is reduced at the price of higher weight and cost. Since the high power system is mostly used in base stations, the performance is more important than the weight and price.
Figure 4.18 Variation of thermal temperature for various metals[8].
93
conductivity
with
A Copper/Molybdenum subcarrier is used between MMIC amplifiers and copper tray. It can help avoid expansion problems with temperature changes. To minimize thermal resistance from MMIC to the outside surface, a eutectic solder is used for die bonding instead of epoxy.
Thermal Simulation
In a structure similar to that shown in Figure 4.19, heat is transferred from the plate to the air. The mechanism of heat transfer at the wall is conduction because the fluid velocity at the wall is zero. However, the rate of heat transfer depends on the slope of the T vs. y curve at the wall---dT/dy at y=0. A steeper slope indicates a greater temperature difference and is highly dependent on the flow velocity. The heat transferred by convection is found to be proportional to the temperature difference. It is qc = hc A(Tw − T∞ )
(4.13)
in which hc is called the average convection heat transfer coefficient or the film conductance. This coefficient accounts for the overall effects embodied in the process of convection heat transfer. The overbar notation indicates that the film conductance defined in Equation (4.13) is an average that is conventionally assumed to be constant over the length of the plate. Typical values of hc are shown in Table 4.1.
94
Figure 4.19 Uniform air flow past a heated plate[8]. Table 4.1 Typical values of the convective heat-transfer coefficient hc for various fluids[8].
Fluid and condition Air in natural convection
hc
W/(m2 K) 5-25
Air in forced convection
30-300
Oil in forced convection
60-1,800
Water in forced convection
300-6,000
The wedge tray has a limited surface area due to its small radius. To help dissipate heat, fins were machined into the outside surface. The entire irregular configuration makes the heat transfer of the structure hard to be calculated analytically. As a result, a mechanical software package, SDRC’s Ideas 8.0, was chosen to simulate the heat transfer. The TMG (Thermal Management) and ESC (Electrical System Cooling) functions are well designed, making them good choices for our purpose. In the thermal model, we only simulated 1/16th of the waveguide structure due to the
95
symmetry. Since all the MMIC amplifiers generate the same amount of heat, there is no heat transfer between metal trays. Thus it is safe to assume an insulation layer on the interface between trays. The heat is conducted to the fins on the outside surface and convectively dissipated into the air. On each metal tray, the MMIC amplifier has two channels, each of which consumes 1.2A current at 8V DC bias. The worst case is analyzed when all the energy is dissipated in the form of heat and the total power consumed is around 20 watts. In the thermal model, the Cu/Mo subcarrier and eutectic solder layers are all included. Those parts are the same as those used in TriQuint’s hot plate lifetime test process. A 20-watt heat source is applied to the surface of the subcarrier. The metal tray’s material is copper with thermal conductivity of 400 W/(m K). Air flows through the fins and has an average heat transfer coefficient of 200 W/(m2 K). Simulations will show the temperature of the metal tray. The wedge shaped tray shown in Figure 4.20(a) has an outside surface area of 640 mm2. The outside surface temperature is 211 oC, while the temperature of the metal tray under the subcarrier is 220 oC. In Figure 4.20(b), 3 fins are added to the outside surface and increase the surface area to about 4 times of that shown in Figure 4.20(a). Then, it can be seen that the temperature drops dramatically, ranging from 82 oC under the subcarrier to 71 oC on the outside surface. Using the relationship between the hot plate temperature and lifetime, we conclude that the MMIC amplifiers can work over 50 years since the hot plate temperature is only 82 oC. 96
No thermal conduction on inner faces due to symmetry
Temperature Range: From 211 oC to 226 oC
Forced Air Convection on outer face (200 W/m2C2)
(a)
No thermal conduction on inner faces due to symmetry
Forced Air Convection on outer face(200 W/m2C)
Temperature Range: From 71 oC to 84 oC
(b) Figure 4.20 Simulation results with Ideas 8.0.
97
Another advantage of the combiner system is its graceful degradation. The MMIC amplifier’s failure means 1-dB degradation in output power. The amplifier degrades favorably and only reaches to 1-dB degradation after several MMIC amplifiers have failed, which will help to increase the lifetime of the amplifier greatly. The design related issues are well discussed in this chapter including waveguide and antenna design, fabrication, uniformity, efficiency, reliability and thermal analysis. The performance of the high power amplifier system will be covered in the next chapter.
References 1. 2. 3. 4. 5. 6. 7. 8.
Pengcheng, J., et al. Analysis of a passive spatial combiner using tapered slotline array in oversized coaxial waveguide. 2000. Zinieris, M.M., R. Sloan, and L.E. Davis, A broadband microstrip-to-slot-line transition. Microwave and Optical Technology Letters, 1998. 18(5): p. 339-42. Shuppert, B., Microstrip/slotline transitions: modeling and experimental investigation. IEEE Transactions on Microwave Theory and Techniques, 1988. 36(8): p. 1272-82. Gupta, K.C., R. Garg, and I.J. Bahl, Microstrip Lines and Slotlines. 1979: Artech House, Norwood, MA. York, R.A., Some considerations for optimal efficiency and low noise in large power combiners. IEEE Transactions on Microwave Theory and Techniques, 2001. 49(8): p. 147782. Tayrani, R. A monolithic X-band class-E power amplifier. 2001. Quach, T., et al. Ultra-efficient X-band and linear-efficient Ka-band power amplifiers using indium phosphide double heterojunction bipolar transistors. 2001. Janna, W.S., Engineering Heat Transfer. Second ed. 2000: CRC Press.
98
5.
Performance of High Power Amplifier Using Compact Coaxial Waveguide Combiner
CHAPTER 5 Performance of High Power Amplifier Using the Compact Coaxial Waveguide Combiner In the last chapter, we explained the design and fabrication procedure for the compact coaxial waveguide combiner for high power applications. In this chapter the measurement of the power and bandwidth performance of the combiner system will be presented. Since the motivation for the spatial power combiner is its power handling capacity, the design emphasis on power capacity neglect several key parameters such as linearity, noise figure, phase noise in previous work. For commercial power amplifiers, those parameters are also important parameters in determining the applicability of the amplifiers to certain systems. These issues are thoroughly discussed and the performance of a high power amplifier using the compact coaxial waveguide combiner is measured in this chapter. For clarity’s sake,
99
we use the term combiner to refer to the high power amplifier using the compact coaxial waveguide combiner.
5.1
Measurement System
Measurement Setup
LabView Program
Signal Source
Power Meter HP438A
Power Supply
TWTA
Power Supply 8 Channel Supply
Coupler #2
Coupler #1
Power Sensor A
Power Sensor B
Gate Bias
Drain Bias
Spectrum Analyzer
Attenuator 20dB / 100
DUT
Figure 5.1 Power measurement setup.
Figure 5.1 shows the measurement setup, which is similar to the setup used by Nick Cheng[1]. A traveling wave tube amplifier (TWTA) follows the signal source to
100
generate the broadband CW signal with a power level between 20 dBm and 43 dBm. The high noise floor of TWTA means that the output power of the TWTA must be greater than 20 dBm for an adequate signal to noise ratio. The output power of the TWTA is between 20 dBm to 35 dBm in its linear region of operation. Power sensors record the input and output power levels. Instead of using 3.5 mm connectors as an Agilent8485A does, the output high power sensor, an Agilent 8481B, uses type N connectors. With an integrated 30 dB attenuator, its dynamic range spans from 0 dBm to 44dBm. The input power sensor, an Agilent 8485A, only has a dynamic range from –30 dBm to 20 dBm. It is very important to consider the sensitivity, settling time and protection when choosing an appropriate power level at the power sensors. A 20 dB power attenuator is used to reduce the output power to a value within the safe range of the power sensor. A spectrum analyzer monitors the output signal for oscillation through the output coupler. The combiner requires two different power supplies to drive the GaAs pHEMTs. The gate bias is provided by an Agilent 3631. Since the drain bias requires high current and good stability, we chose KEPCO RA55 8 channel power supply. Each channel of the RA55 is independent with a capacity of up to 10 Volts and 12 Amps. The average current for each MMIC amplifier is around 1.2 Amp at 8 Volts; each channel of the RA55 provides drain current for 4 MMIC amplifiers. Voltage drops on the bias line lower the drain voltage, resulting in reduced output power. With a remote-sensing capability, the KEPCO RA55 power supply can properly compensate the voltage drop between the power supply and the DC bias 101
board, but thick bias lines and gold plated DC pins are still needed for the connections between the DC bias board and combiner.
Automatic Control Program
Figure 5.2 Virtual instrument front panel of Labview Program.
The signal source, power supply, power sensor and spectrum analyzer are all controlled by a laptop computer through the GPIB bus. Using National Instrument’s Labview, a Microsoft Windows based program was developed for instrument configuration, calibration and data acquisition. National Instruments supplies virtual instrument (VI) libraries for a variety of popular test equipments. The Labview program displays virtual front panels of the instruments on the computer. Users can execute operations on the virtual front panel, 102
which sends commands to the appropriate instrument via GPIB connection. The Labview program is able to automatically initialize instruments and carry out very complex calibration and measurement procedures, which are normally tedious to be done manually. Figure 5.2 shows the virtual front panels of our measurement system. The panel at the bottom of the screen is the calibration section. Since the power sensors use different calibration factors at different frequencies, calibration factor tables for both power sensors are stored into the calibration panel. The calibration program automatically loads the data and outputs the calibration results in the calibration data table on this panel. Since the amplifier measurement system involves many instruments, computer automation improves measurement efficiency. Additionally, the Labview program can add overload protection in the event of a MMIC amplifier failure during a measuremnt, ttherebyand avoiding the failure of other MMIC amplifiers due to excess current flow.
Calibration Procedure
Calibration is a necessary procedure for all RF measurements. The goal is to deembed the parameters of the input and output network, and then move the reference plane directly to the input and output port of the DUT.
103
Power Sensor A SMA to N Adapter
Coupler #1
Power Sensor B
(a) Power Sensor A
Power Sensor B SMA to N Adapter
Coupler #1
Attenuator 20dB / 100 Watt
Coupler #2 Spectrum Analyzer
(b) Figure 5.3 Calibration procedures.
Unlike the new vector network analyzers, which have four directional couplers and vector receivers, our measurement setup has only two directional couplers and scalar power sensors. Loss of the input/output coupler, cable and connectors are calculated by the calibration procedure describe in Figure 1.3. In step (a), the difference between coupler #1’s coupling port and the type N port is measured with two power sensors. Using the data from step (a), the power difference between the coupler #2’s output port and the input of the attenuator are calculated in step (b). All of the calibration data is stored in the Labview program and used for post measurement data correction. Equal length SMA to type N adaptors with different genders are used for connection to the DUT. Even though different gender adaptors are used, the calibration data is still sufficiently accurate since we choose the adaptors from the same manufacturer.
104
5.2
MMIC Amplifier Characterization
The MMIC amplifier we chose is a TGA9092 EPU (Engineering Prototype Unit), manufactured by TriQuint Semiconductor. The TriQuint TGA9092 is a dual-channel, three-stage wide band HPA MMIC designed using TriQuint’s proven 0.25 µm power pHEMT process which supports a variety of high performance applications, including military EW programs, VSAT, and other applications requiring wideband high power performance. Each amplifier channel consists of one 1200 µm input device driving a 2400 µm intermediate stage, which in turn drives a 4800 µm output stage.
Figure 5.4 Layout of TriQuint TGA9092 EPU.
105
Figure 5.5 2 dB compression point output power of TGA9092 (from TriQuint data sheet). 30
30
25
25
20
20 MMIC1 MMIC2 MMIC3 MMIC4
15
MMIC5 MMIC6 MMIC7 MMIC8
15
10
10
5
5
0
0 6
8
10
12
14
16
18
Freq [GHz] Figure 5.6 Small signal S21 data of 8 TGA9092 from the same wafer (from TriQuint Semiconductor).
A 2 dB compression point output power measurement result is shown in Figure 5.5. The P2dB power measurement, measured at different input power levels, exhibits the power capacity of the MMIC. The small signal S parameters of eight TGA9092
106
MMIC amplifiers are shown in Figure 5.6. All of the MMIC amplifiers are biased with a 8V drain voltage and 0.31 V gate voltage. The average drain current is 1.2 A with +/- 5% variation. Since the MMIC amplifiers are from the same wafer, the maximum gain variation is less than 2.5 dB. To minimize the efficiency reduction due to gain variation, it is recommended to use MMIC amplifiers from the same wafer, or at least the same process batch, for the combiner system.
5.3
Output Power
Small Signal Modeling
As explained in the previous chapters, the performance of the waveguide structure and finline transition is simulated by HFSS, a 3D FEM simulator. Since HFSS can only simulate 3D passive structures, the overall active amplifier’s performance is not directly available from HFSS simulations. We exported the S parameter results from HFSS to S2p files, then imported them into Agilent Advance Design System (ADS). The ADS small signal circuit model of the combiner is shown in Figure 5.7. Waveguide and Slotline Transistion
Lossy Matching Network
MMIC
Waveguide and Slotline Transistion
Figure 5.7 Schematic for small signal modeling of the combiner.
107
Due to the complexity of the finline transition, the HFSS simulation requires tremendous memory resources. Even the most advanced computer in our research lab, with 1 GB of memory, is only useful for simulations of 1/16th of the whole finline array. Using symmetry, we equally divided the combiner into 16 parallel identical sections from the input end to output end. The method is viable because of the symmetry of the combiner. In the ADS model, we only simulated one of the 16 sections. The section included the input/output waveguide tapers, finline transitions, a lossy matching network and a MMIC amplifier. The one section simulation is accurate enough to represent the overall amplifier performance because the spatial power combining theory proved that the power is evenly distributed to and combined from each channel and the overall gain is the same as for a single channel. Slotline Transition
MMIC Input R R2 R R=60 Ohm R1 R=60 Ohm
WIRE Wire2 D=1.0 mil L=300 um Rho=1.0
WIRE Wire1 D=1.0 mil L=300 um Rho=1.0
Figure 5.8 Circuit schematic of lossy matching network.
The TGA9092 MMIC amplifier has a gain in excess of 25 dB which very easily cause oscillation problems in a packaged waveguide environment when the output to input isolation is only slightly higher than 20 dB at some frequency, as shown in chapter 4. The circuit becomes stable when the overall gain is reduced within the 20
108
dB limitation due to the insertion of the lossy matching network. Bonding wires were included in the lossy matching network circuit model to improve the accuracy of the simulation, the results of which are as shown in Figure 5.8. The good agreement between the measurement and the simulation, which is shown in Figure 5.9, verifies the effectiveness of the modeling. There is 8-dB difference in the gain between the MMIC amplifier and combiner, which arise from the lossy matching network. 30
Simulation & Measurment of Combiner
30
20
20
10
10
0
Measurement_S21 [dB] Simulation_S21 [dB]
MMIC_S21 [dB]
0
-10
-10
-20 5
10
15
20
-20
Frequency [GHz]
Figure 5.9 Comparison of simulation and measurement of the combiner and measurement of the MMIC amplifier.
The lossy matching network reduces the feedback loop gain and stabilizes the circuit. In the mean while, the lossy matching network improves input impedance match of the MMIC amplifier, which is initially very bad because the MMIC amplifier is optimized for wide bandwidth and high power. The comparison of S11 and S22 between the MMIC amplifier and the combiner is shown in Figure 5.10. Aside from the reduction in S11 for a combiner relative to a MMIC amplifier, we also 109
observe ripples in the S11 and S22 curve of the combiner, which are caused by the cancellation of the reflected signal from different end of the waveguide taper. 0
0
-10
-10
-20
-20 S11_MMIC [dB] S11_Combiner [dB]
S22 _MMIC [dB] S22 _Combiner [dB]
-30 5
10
15
20
-30
Frequency [GHz]
Figure 5.10 S11 & S22 of the MMIC amplifier and the combiner.
Power Measurement
Figure 5.11 Combiner with bias lines.
Figure 5.11 shows the assembly of the combiner system. The bias lines were connected from a biasing board to the 16 individual circuit trays. The KEPCO 8
110
channel power supply and Agilent power supply were connected to the biasing board. The calibration procedure was performed before the measurement and the calibration data were stored in the Labview measurement program. The measurement setup has been illustrated in section 5.1.
50
100
Pout=46.4 dBm 40
80
Pout [dBm]
30
Efficiency [%]
Gain [dB]
60
20
40
10
20
0
0 6
8
10
12
14
16
18
Freq [GHz] Figure 5.12 Frequency sweep at 30 dBm input power.
The input power level was chosen to be 30 dBm. A frequency sweep measurement result is shown in Figure 5.12. A maximum power of 44 Watts is obtained at 10 GHz. The gain curve followed a similar shape to the small signal gain curve in Figure 5.9, with the exception that gain compression may occur differently over the band. The 3 dB bandwidth is from 6 to 17 GHz. The output power curve varies slightly from the 2 dB compression output power curves in Figure 5.5 that were measured at different input power levels to reach 2 dB compression. We noted
111
that two MMIC amplifiers of the 32-MMIC combiner were nonfunctional during the measurement. The combiner’s output power was measured with 30 working MMIC amplifiers, which is 88% of the 32-MMIC combiner’s output power basing on the graceful degradation theory[2]. The KEPCO 8 channel power supply’s output currents were recorded through the Labview program and the data are shown in Table 5.1. Each power supply channel provided current for four MMIC amplifiers. The amplifier works in a Class AB state, close to Class A. When the input power increases, the current also increases to some extent due to waveform cutoff. The channels that are adjacent to the broken channel have a larger increase in current as a result of a more severe overdrive. This deteriorates intermodulation distortion. The best linearity can be achieved if all MMIC amplifiers are biased in class A with 1.2 amps biasing currents. Table 5.1 Current changes
Current (Amps) @Vds=8V,Vg=-0.4V Small signal before 2 MMICs are broken Small signal after 2 MMICs are broken Pin = 30 dBm after 2 MMICs are broken
I1
I2
I3
I4
I5
I6
I7
I8
Itotal
Iavg/ MMIC
3.95
4.11
3.77
3.93
3.73
4.52
3.48
3.52
31.00
0.97
4.02
4.23
3.78
1.84
3.78
4.58
3.53
3.58
29.34
0.98
4.21
4.61
3.88
2.42
4.71
5.05
3.62
3.45
31.95
1.07
PAE =
Pout − Pin Pout − Pin P −P = = out 8 in PDC VDS I D ,total VDS ∑ I D ,i i =1
112
(4.14)
The power added efficiency (PAE) is calculated from equation (4.14). The PAE is measured and shown both in Figure 5.12 and Figure 5.13. At 10 GHz, the small signal gain of the combiner is 18.2 dB, which is about 0.4 dB smaller than the small signal gain of the combiner when all MMIC amplifiers are functional. At 30 dBm input power, the gain is compressed by 1.8 dB. The power added efficiency was about 17% at an output power of 44-Watts. 50
50
40
40
Pout [dBm] Gain [dB]
30
Efficiency [%]
30
20
20
10
10
0
0 20
22
24
26
28
30
Input Power [dBm] Figure 5.13 Power sweep at 10 GHz.
5.4
Linearity
Linearity is important for broadband communication systems. A two-tone intermodulation distortion (IMD) measurement is used to evaluate the linearity of the amplifiers. The IMD is a ratio of the strength of the third order component produced by two adjacent fundamental signals to the strength of one of the fundamental signals. 113
The extrapolated cross point of the fundamental and the third order intermodulation component is known as the third order intercept point (IP3). Although the power level of the fundamental carrier can never be equal to that of the third order intermodulation component because of (you used too many “due to”) saturation, it is reflective of the amplifier’s linearity.
IP3
POutput
PSat
P1dB
3rd
5th
PInput Figure 5.14 Output power and harmonics.
Compared to TWTAs that work in the saturation mode, solid-state amplifiers offer better linearity by operating at P1dB point. To reach an IMD level of –25 dBc, a typical TWTA needs to back off more than 7 dB from the rated single carrier output power. A solid-state amplifier only needs to back off around 2 to 3 dB from P1dB to reach the same IMD level. The output signal’s voltage has following relationship with the input voltage: Vout =a 0 + a1v + a2 v 2 + a3v 3
114
(4.15)
For one tone signal, the amplitude at the 1 dB compression point vic is given by 3 3 a3vic 4 = 1 − 10−0.05 a1vic
(4.16)
Table 5.2 Two-tone distortion products
a1v
a3v 3
One tone: ω
1
3 4
2 tone: ω1 , ω 2
1
9 4
25 4
3 4
25 8
2 tone third order distortion product: 2ω1 ± ω 2 , 2ω 2 ± ω1
a5v 5
We now consider the same amplifier but with a two-tone signal applied to its input, both tones having equal amplitude. The amplitude of each carrier at the IP3, vip 3 , is by definition, a1vip 3 =
3 3 a3vip 3 . 4
(4.17)
The amplitudes of each IM3 product[3] is determined using Table 5.2. Note that the definition of IP3 used here relates the amplitude of a single IM carrier to the amplitude of a single input carrier. Combining equation (4.16) and (4.17) yields 2
vip 3 1 = −0.05 vic 1 − 10
115
(4.18)
which corresponds to a ratio of 9.2, or about 9.6 dB. We can conclude that if only the third order component is present, the IMD3 at the 1 dB compression point will be 18.2 dBc. The other phenomenon we observed is that the sum of the two tone’s signal output power at the 1 dB compression point in the two tone measurement is smaller than the P1dB power in one tone measurement. The relationship of P1dB for one-tone and two-tone signals can be proved using similar approaches. In the two-tone case, if the 1 dB compression amplitude of each carrier is vic 2 , the amplitude of either fundamental carrier will be 9 3 voc = a1vic 2 − a3vic 2 . 4
(4.19)
So that the two tone P1dB compression point is given by 2
vic =
4a1 (1 − 10−0.05 ) . 9a3
(4.20)
By comparison with (4.16), the ratio between the single-tone and two-tone signal amplitude at 1 dB compression point is 2
vic1 =3 vic 2
(4.21)
The 1.7 dB difference explains the discrepancy in the P1dB power in one-tone and two-tone measurements.
116
If only the third order component considered, the third-order distortion trace would be linear in input power. When the power is backed off from the IP3 point, the third-order products would fall at a rate of 3 dB for every decibel of output or input power reduction, or 2 dBc relative to carrier. So normally, an amplifier needs to back off about 3 dB from its one tone P1dB to achieve an IMD3 level of –25 dBc. However, the third order product is not simply linear for GaAs amplifiers. In practice, the nonlinearity is beneficial in that it allows for a larger IMD3 near P1dB than would be possible without other higher order contributions. IMD Measurement of MMIC 50 40
OIP3 k=1
30 20 10 0
k=3
-10 -20 -30 -20
-15
-10
-5
0
5
10
15
20
Pin [dBm]
Figure 5.15 Two-tone measurement of the MMIC amplifier.
The gm curve of GaAs MMIC amplifier under investigation is shown in Figure 5.16. It is measured from a small test device on the MMIC chip. A strong fifth order component is introduced from the nonlinearity of the gm curve which causes the concavity of the third order product in Figure 5.15. To analyze the linearity, we fit a
117
polynomial curve with the gm data. The fitted curve is the thinner trace in Figure 5.16. 40
30
20 gm [ms]
10
0
-10 -1.5
-1
-0.5
0
0.5
Vgs [V]
Figure 5.16 Gm Curve of the Triquint GaAs HEMT.
The expression for the fitted gm curve is: 2
3
4
5
g m (Vgs ) = a0 + a1Vgs + a2Vgs + a3Vgs + a4Vgs + a5Vgs + a6Vgs
6
(4.22)
Table 5.3 Polynomial coefficient for gm curve fitting
Polynomial Coefficients
a0
a1
a2
a3
a4
a5
a6
Value
36.8
-15
-6.3
35.5
-15
-20
-0.1
I ds (Vgs ) = ∫ g m (Vgs )dVgs
(4.23)
Since the gate is biased at –0.4V, we have Vgs = −0.4 + vin and 2
3
4
I ds (−0.4 + vin ) = c0 + c1vin + c2 vin + c3vin + c4 vin + c5vin
118
5
(4.24)
Table 5.4 Polynomial coefficient for Ids
Polynomial coefficients Value
c1
c3
c5
39.3
-16.8
5
The third order product can be expressed as 3 25 3 5 v2ω1 ±ω2 = ( c3vin + c5vin ) ⋅ Fscale ⋅ RL 4 8
(4.25)
where Fscale is the scaling factor between the real device in the amplifier and the testing device used for gm curve measurement, and RL is the load impedance. -15
-20
-25
-30
-20
-18
-16
-14
-12
-10
Figure 5.17 Third order component with contribution of fifth order component.
As shown in Figure 5.17, the third order product begins to decrease when the input power is strong and the contribution of fifth order component takes effect. The third order component increases again when the input signal’s envelope starts to saturate as shown in Figure 5.15.
119
10 8 6 4 2 -18
- 16
- 14
-12
- 10
Figure 5.18 Fundamental component with and without the contribution from fifth order products.
In Figure 5.18, the dashed black line is a linear curve; the red line and the blue dashed line represent the fundamental component with and without the contribution of fifth order products respectively. The fifth order products compensate the saturation of the fundamental component due to contribution to the third order products when the input power is close to P1dB, and improves the linearity as shown in the figure. For this reason, we expect our GaAs amplifier to have better linearity when the output power is close to the P1dB point. As shown in Figure 5.15, the GaAs amplifier only needs to back off less than 2 dB from the P1dB point to reach a IMD3 of –25 dBc. To evaluate the change of the IP3 point in power combining, we need to compare the third order intermodulation component (IM3) of a MMIC amplifier and the combiner. For a MMIC amplifier, we can express the fundamental and IM3 output power as Pout = Gm Pin IM 3 = A Pin
120
3
(4.26)
where Gm is the gain of a MMIC amplifier, and A is the coefficient for IM 3 . For MMIC:
Gm
Pin
Pout IM 3
For Combiner:
Li
Lm Lossy Matching Network
Pin
Li
Gm
Pin , e IM 3 ,e
Pout
N
IM 3 Lossy reflection Matching N Way Combiner
N Way Divider 2
Overall Gain G c = G m Li L m *We assume divider and combiner are identical Figure 5.19 Linearity analysis for the MMIC amplifier and the combiner.
At the IP3 point, the linearly extrapolated fundamental output power is equal to the IM3. The OIP3 is the output power at IP3 point where Pout =IM 3 . The OIP3 of a MMIC amplifier is 3
Gm 12 ) . OIP3m = ( A
(4.27)
For a combiner, we have 2
Gc = Gm Li Lm 2
Pout = Gc Pin = Gm Li Lm Pin .
121
(4.28)
For each MMIC amplifier in the combiner, we have Pin ,e = IM 3,e
Pin Li Lm N 3 = A Pin ,e
(4.29)
where N is the number of channels in the combiner and Lm is the loss of the lossy matching network. We assume the divider and combiner have the same loss Li . The IM 3,e from each MMIC amplifier are added in the same way as the fundamental signal. The sum of the IM 3,e at the output port is expressed in IM 3 as 3
IM 3 = N IM 3,e Li = N A Pin ,e Li
(4.30)
Then, we have Pout = IM 3 = N A ( 3
Pin Li Lm 3 ) Li n
G OIP3c = N Li ( m ) A
1 2
(4.31)
where OIP3c is the OIP3 of the combiner. Comparing equations (4.27) and (4.31), we conclude that OIP3c = N Li OIP3m.
(4.32)
For a 32-channel combiner with a Li of 1dB, the combiner will have a factor of 14 dB improvement in OIP3 over a MMIC amplifier. We note that the OIP3 has no relationship with the lossy matching network. We will observe the 14 dB improvement no matter whether we use the lossy matching network or not. The
122
relationship between the fundamental component and third order component remains the same for the combiner and a MMIC amplifier. The intermodulation distortion was measured by two tones at 10 GHz with a separation of 1 MHz in spectrum. The output signal was input to a spectrum analyzer. The spectrum analyzer attenuates its input signal with its internal attenuators to minimize the intermodulation by the mixers inside the spectrum analyzer. The measurement setup is shown in Figure 5.20. The system was calibrated with the procedure shown in Figure 1.3. After the DUT was added in the measurement setup, the fundamental and third order intermodulation components were read from the spectrum analyzer, and then corrected using the calibration data. Power Sensor A Source 1
SMA to N Adapter
Coupler #1
Source 2
DUT
Power Combiner Power Sensor B
Attenuator 20dB / 100 Watt
Coupler #2 Spectrum Analyzer
Figure 5.20 Intermodulation distortion measurement setup.
The IMD measurement result of the combiner with 30 working MMIC amplifiers is shown in Figure 5.21. There is no obvious concaved curve in this figure. The change of the IM3 curve is due to the unequally driving of the MMIC amplifiers because of the nonfunctioning of 2 MMIC amplifiers. The mismatch introduced by 123
the broken MMIC amplifiers causes the adjacent MMIC amplifiers to be driven by a larger than average power. Those adjacent MMIC amplifiers saturate in amplitude faster when the input signal increases. The amplitude saturation will offset the improvement due to the fifth order products shown in Figure 5.15. Since the IP3 point is determined by the linear part of the fundamental and IM3 traces, it will not be changed when some of the MMIC amplifiers are overdriven. 60 OIP3_Combiner
40
Combiner Single Tone k=1
OIP3_MMIC MMIC Single Tone
20
k=1
0 IMD3 IMD3 -20
-20
-10
0
10
20
30
40
Pin [dBm]
Figure 5.21 Comparison of IMD between the MMIC amplifier and the combiner.
At 10 GHz, the output IP3 (OIP3) is 52 dBm compared to 38 dBm of a single MMIC, which corresponds to a 14 dB improvement over a single MMIC amplifier. Figure 5.21 consolidates the conclusion from equation (4.32).
124
5.5
Noise Figure For MMIC:
N add N o = Gm N i + N add
Ni Gm
For Combiner: Li
Li
Lm N add Ni ,m
Lossy Matching Network
N o,m
Gm
Ni
No
N
Lossy Matching Network
N Way Combiner
N Way Divider 2
Gc = Gm Li Lm
Overall Gain
*We assume divider and combiner are identical Figure 5.22 Noise Figure of a MMIC amplifier and a combiner.
For a single MMIC, the noise figure is Fm =
Si / N i No G ⋅ N + N add = = m i So / N o Gm ⋅ N i Gm ⋅ N i
(4.33)
For a combiner, we assume the input and output matching networks are the same and the combiner’s gain is represented by Gc as shown in Figure 5.22.
125
The added noise from each channel is uncorrelated and can be treated as a random variable. When N uncorrelated random variables sum up at the output port, the total noise power will be the same as the noise power of a single channel. When the loss of the combiner Li is taken into consideration, the overall added noise for the combiner is N add ⋅ Li , where N add is the added noise of each MMIC amplifier. Since the input noise is correlated when distributed to each channel, the input noise from each channel is summed in phase at the output of the combiner. The overall noise figure Fc is then expressed as: Fc =
Gc ⋅ N i + N add ⋅ Li . Gc ⋅ N i
(4.34)
2
For a combiner system without the lossy matching network, Gc = Gm Li . In that case the combiner’s noise figure can be expressed as Fc = 1 + ( Fm − 1)
where Li =
1 Li
(4.35)
Gc −1 . If we assume Li ≈ 1 , Fc ≈ Fm Li . Gm
The measured noise figure Fc and calculated noise figure Fc * , are shown in Figure 5.23, along with the noise figure of a single MMIC amplifier for comparison. Symbol NF represents the noise figure F in dB. The measurement is based on a 2 x 2 array rectangular waveguide combiner. Good agreements verify the conclusion that when no lossy matching network is 126
integrated in this combiner system and the loss of the combiner is very low; the noise figure of the combiner and the MMIC amplifier are very close to each other. 25
25
20
20
15
NFm(dB) NFc(dB) NFc*(dB)
Gm(dB) Gc(dB)
15
10
10
5
5
0
0 8
8.5
9
9.5
10
10.5
11
Freq(GHz)
Figure 5.23 Measurement of noise figure of a single MMIC amplifier and a combiner and calculation from equation (4.35). 2
For a combiner system with an integrated lossy matching network, Gc = Gm Li Lm , and we have Fc = 1 + ( Fm − 1)
1 . Li Lm
(4.36)
If we assume Li ≈ 1 , Lm 1, then equation (4.36) can be approximated as −1
−1
Fc ≈ Fm Li Lm .
127
(4.37)
5.6
Spurious-Free Dynamic Range
Given the IP3 and noise figure, one can define another important property of the \amplifiers called the spurious-free dynamic range (SFDR). The SFDR represents the ability of a system to detect or boost signals in the presence of noise and other strong signals, and is important in several system applications such as transponding the multiple carriers that routinely pass through terrestrial base stations operating under code-division multiple access (CDMA) technology. Another example is the detection of a frequency-chirped radar return signal in the presence of strong clutter. The lower limit of the SFDR occurs when the input signal power equals the input band limited Rx noise power. To account for multiple input signals, the upper limit of the SFDR is often defined as the power when an IM3 tone (for two equal input tones) equals the output receiver noise power. This yields the expression 2 OIP3 SFDR = ( )3 F ·G·k B ·T0 · ∆f
(4.38)
where F is the receiver noise factor, G is the gain, ∆f is the instantaneous bandwidth, k B is Boltzman's constant and T0 is the ambient temperature[4].
From this expression, the scaling behavior of the SFDR in an amplifier is obvious. For a combiner system without an integrated lossy matching network, G and F don’t 2 3
scale, only OIP3 scales with N. The SFDR is proportional to N . This is shown in
128
Gc = Gm Li Fc ≈ Fm Li
2
−1
OIP3c = NLi OIP3m
(4.39)
2
SFDRc = N 3 SFDRm
where subscript c refers to combiner and m refers to MMIC amplifier. For a combiner with lossy matching network, we have 2
Gc = Gm Li Lm −1
Fc ≈ Fm Li Lm
−1
OIP3c = NLi OIP3m
(4.40)
2
SFDRc = N 3 SFDRm .
From equation (4.39) and (4.40), we conclude that the combiner will have a 10 dB improvement in SFDR over a single MMIC amplifier.
5.7
Phase Noise of Combiner
Residual Phase Noise
Phase noise is the fluctuation of the phase due to a resistor’s thermal noise, an active devices’ 1/f noise and shot noise. Residue phase noise is the added noise to a signal’s phase when the signal is processed by a 2-port device. It is commonly used to evaluate the phase noise characteristic of the 2-port device. Residual phase noise includes two basic noise mechanisms: additive noise and multiplicative noise. Additive noise is generated by the device and added linearly to
129
the signal. Multiplicative noise is the noise that modulates RF signal by the multiplication of baseband noise with the signal. The mixing is due to the nonlinearities in the 2-port network. The baseband noise may be generated by the active devices of the internal network, or it may come from low-frequency noise in the signal line or power supply.
Figure 5.24 Multiplicative residual noise model.
To measure the residual phase noise, we demodulate the RF signal and then analyze the noise at baseband. If we assume φpeak is ensemble average
(4.43)
δθ is Fourier transform Assuming that input noise and amplifier noise contributions are uncorrelated:
δθ in∗δϕi = 0 for all i, δϕi∗δϕ j = 0 for i ≠ j Excess noise from each amplifier is δϕ
δθ out
2
= δθ in
2
, making the output noise
2
+
1 2 δϕ N
(4.44)
(4.45)
The second part in equation (4.45) is the residual phase noise. It will be improved by a factor of N if all the added noise is uncorrelated. If the input phase noise is small compared to the residual phase noise, we will also see approximately a N times reduction in the output phase noise.
Residual Phase Noise Measurement
The residual phase noise is measured with a HP3048 phase noise system. As shown in Figure 5.26, the input signal is divided into two identical signals by a power divider. The phase of the two signals are adjusted into quadrature by a phase shifter. The HP11848 phase noise interface has a double balanced mixer as phase detector. As shown in Figure 5.26, HP 11848 demodulates the RF signal to base band noise signal and the base band signal’s spectrum is computed by HP3561A dynamic signal
132
analyzer. If the peak phase deviation is smaller than 0.2 radians, the RF signal phase noise spectrum is one half of the base band noise spectrum. 20 dB Attenuator 11848 Phase Noise Interface
DUT
+8dBm R Port
G=18dB
Source
V1Sin(2π f 0t + φ (t ))
π
L Port
Power Spliter
Kφ ⋅ φ (t ) = Vn (t )
V2 Sin(2π f 0t + φ (t ) + ) 2
+10dBm
Phase Shifter GPIB Control S (f) Sφ = n 2 Kφ L( f ) =
Sφ ( f ) 2
=
Sn ( f ) 2 2 Kφ
0.01Hz
100KHz
3561A Dynamic Signal Analyzer
Figure 5.26 HP3048 Phase noise measurement system. 0
0 L(f)_MMIC [dBc]
-20
L(f)_Combiner [dBc]
-20
-40
-40
-60
-60
-80
-80
-100
-100
-120
-120
-140
-140 -10 dB/decade
-160
-160 1
10
100 1000 Offset Freq [Hz]
10
4
Figure 5.27 Residual phase noise of a MMIC amplifier and the combiner for 1 Watt medium power combiner system.
133
The residual phase noise measurement results of a MMIC amplifier and the combiner in the 1-Watt medium power combiner system are shown in Figure 5.27. An average of 15 dB improvements in residual phase noise is observed. The residual phase noise is lower than –150 dBc at a 10KHz offset from the carrier frequency. In the MMIC amplifier’s measurement, some spikes are observed which are the spurs from the DC bias lines. The residual phase noise spectrum of the medium power combiner follows a 10 dB/decade curve from 1Hz to 1KHz offset from carrier frequency which is the characteristic of 1/f noise. In the medium power combiner, the MMIC amplifier is chosen to be a low noise amplifier. 1/f noise is the dominant source of phase noise. The 1/f noise is from the imperfect material of each GaAs device and is uncorrelated for each MMIC amplifier. From equation (4.45), 15 dB reductions in residue phase noise are expected which agrees well with the measurement results. The residual phase noise measurement results of a MMIC amplifier and the combiner in the high power combiner are shown in Figure 5.28. The residual phase noise level is around –140 dBc at a 10 KHz offset from the carrier frequency, which is about 10 dB higher than the measured value for the medium power combiner since the MMIC amplifiers were designed for high power applications. However, instead of 15 dB, only 5 to 6 dB residual phase noise reduction is observed from the phase noise spectrum. The reason of this lower reduction in phase noise is the partial correlation of the phase noise sources in each channel.
134
0
0
-10
-10 L(f)_Combiner [dBc]
L(f)_MMIC [dBc]
-20
-20
-30
-30
-40
-40
-50
-50
-60
-60
-70
-70
-80
-80
-90 -100
-90 -20 dB/decade
-100
-110
-110
-120
-120 -20 dB/decade
-130
-130
-140 -150 0.01
-140 -150
0.1
1 10 100 Offset Freq [Hz]
1000
10
4
Figure 5.28 Phase noise measurement of a single MMIC and the high power combiner.
KEPCO 8 Channel Power Supply
Figure 5.29 Power supply of the combiner system.
Since the 32 MMIC amplifiers are driven by KEPCO 8-channel DC power supply as shown in Figure 5.29, multiplicative noise from the power supply is partially
135
correlated. As shown in the figure, every four MMIC amplifiers have the same noise from power apply and there are other correlations from the sharing of the same ground line. And since the current is very high, the multiplicative DC noise is dominant in all the noise sources. We didn’t observe a 1/f curve from 1Hz to 1KHz offset from carrier frequency because the 1/f noise is inferior compared with the DC line noise. In stead, a low-pass characteristic curve is observed in the phase noise spectrum. That is because the capacitors used in the DC bias line forms low pass filters and the low pass filtered multiplicative noise spectrum is transferred to the carrier. If we integrate voltage regulator for each MMIC amplifier, asides from adding voltage protection feature to each MMIC amplifier, we will decorrelate the DC line noise. We will be able to achieve the same phase noise reduction as the medium power amplifier does.
5.8
Summary
The amplifier using the compact coaxial waveguide combiner shows the 3-dB bandwidth from 6 to 17 GHz with a maximum power of 44 Watt. We maintain the combiner’s linearity similar to that of a MMIC amplifier, while improving the OIP3 of the combiner to 52 dBm, which is 14 dB higher than that of a single MMIC amplifier used in the combiner. The SFDR range is also increase by 10 dB and residual phase noise is reduced by 5 to 7 dB. The residual phase noise floor is –140 dBc at a 10 KHz offset from carrier frequency. All these features enable this amplifier 136
a good candidate for high power amplifiers in wireless and satellite communication base stations.
References
1. 2. 3. 4.
Cheng, N.-S., Waveguide-Based Spatial Power Combiners, in Dept. of Electrical and Computer Engineering. June 1999, University of California: Santa Barbara, CA 93106. Rutledge, D.B., et al., Failures in power-combining arrays. IEEE Transactions on Microwave Theory and Techniques, 1999. 47(7, pt.1): p. 1077-82. Cripps, S.C., RF Power Amplifiers for Wireless Communications. 1999: Artech House. Brown, E.R. and J.F. Harvey, System characteristics of quasi-optical power amplifiers. IEEE Circuits and Systems Magazine, 2001. 1(4): p. 22-36.
137
6.
Conclusion and Future Works
CHAPTER 6 Conclusions and Future Works In this thesis, we discussed the modeling and fabrication technique of coaxial waveguide power combiner, and successfully demonstrated a low-noise medium power amplifier and a high power broadband amplifier with the combining technique, each integrating 32 MMIC amplifiers. This thesis proves that the power combining technique using coaxial waveguide structure is the most effective approach to integrate a large quantity of devices over a broadband width with high power combining efficiency. The high power broadband amplifier design will enable the power amplifier industry with a quick shift from traveling tube amplifiers to the solid-state amplifiers. The low-noise and high dynamic range properties in the medium power amplifier also show good applications in receiver design. Some challenges are still remained in the high power amplifier design. Preliminary thermal simulations are conducted in the design. Since the combiner
138
structure is very compact, the viscosity of the air can slow down the air flow speed inside metal fins. Longer fins with wider spacing are suggested to get better thermal dissipation. More extensive thermal analysis, which includes the cooling fans, is needed for a more reliable amplifier design. Oscillation is one of the biggest threats to high power amplifiers. We use lossy matching network to reduce the feedback loop gain to keep the amplifier stable in the high power combiner. But the medium power combiner, which uses much lower gain MMIC amplifiers, is oscillation-free. So choosing MMIC amplifier with proper gain is a key issue in the design phase. For application in communication systems, besides power, bandwidth, linearity and noise, efficiency is always combined together with those parameters in evaluating an amplifier system. Switch-type amplifiers are well investigated to achieve higher efficiency. But those types of amplifiers will have problems in linearity since the amplifiers are working at switching state. A class B push pull amplifier doesn’t have as high efficiency as Class D, E and F amplifiers, but its efficiency is about 28% higher than that of a class A amplifier. Furthermore, it is as well as suited for broadband application as a class A amplifier. The two outputs on each circuit tray in our combiner are designed with 180 degree phase difference which enable this architecture a perfect fit for class B push pull design. The challenge is to design the broadband balun which functions as a broadband transformer to combine the half period signal from each of the two amplifiers together. 139
The 2 amplifiers in the push pull amplifier work alternatively. It means during each half cycle one amplifier is loaded, while the other amplifier is open. The broadband balun needs to match the amplifiers with the load in either half cycle.
140
A dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy in Electrical and Computer Engineering by Pengcheng Jia
Committee in charge: Professor Robert A. York, Chair Professor Umesh K. Mishra Professor Steve Long Dr. Yifeng Wu
December 2002
Broadband High Power Amplifiers Using Spatial Power Combing Technique
Copyright © 2002 by Pengcheng Jia
iii
Dedicated to my parents, and to my love, Xiangming
iv
ACKNOWLEDGEMENTS
Pursuing Ph.D in UCSB is one of the most valuable and exciting experiences in my education. The knowledge I learned and the confidence I gained in the 4 years’ study will be beneficial to my whole life. I am greatly indebted to my advisor, Prof. Bob York, who takes so much effort and patience in mentoring me to become a qualified researcher. From leading me into the wonderful spatial power combining field, to revise my poorly written papers and badly edited presentations, Prof. York gave me direction in every step of my thesis work, while never forgetting trivial details. It is his insight and wide knowledge that guided me to the completion of this thesis work; and it is his broad research interests that gave me the opportunities to explore many interesting projects in microwave field besides my thesis work. I also want to thank other members in my committee: Prof. Umesh Mishra for his interest in spatial power combining and support with Navy funding; Prof. Steve Long who was always there to help me with his profound knowledge and generous kindness; and Dr. Wu who played the role both as the advisor and as a friend, and gave so many good suggestions with his wisdom in both thermal and electrical engineering. I always feel lucky to be with so many excellent researchers in the York group. I sincerely thank Yu Liu for being a good partner in almost all the classes and an earnest colleague with whom I had so many intensive discussions on countless topics. Many thanks go to Paolo Maccarini, whose enormous contributions to the measurement lab greatly facilitated the progress of my projects. I’d like to thank Nick
v
Cheng for inspiring my interest in spatial power combining research and Jane Xu for teaching me all the skills in cleanroom. I am grateful also to other members in the York group for their support: Amit, Angelos, Andrea, Baki, Chris, Eric, Hongtao, Justin, Joe, Jim, Nadia, Pete, Troy and Vicki. I am thankful for all the members in Mishra group for sharing the cozy office, with whom I enjoyed the opportunity to be exposed to a variety of research projects on semiconductor material, device and circuitry: Ale, Ariel, Can, Dan, Dario, DJ, Gia, Haijiang, Huili, Ilan, Jae, Jason, Jeff, Lee, Likun, Mary, Naiqian, Peter, Prashant, Primit, Rama, Rob C., Rob U., Sten, Tim and Yingda. I’d like to especially thank Rob Coffie, who brought me high quality MMIC amplifiers that were the key components in the medium power amplifier and shared with me his thorough understanding in solid-state devices. I would like to thank many other students in the semiconductor group, from whom I benefited a lot through stimulating discussions: Yun Wei, Shouxuan, Paidi, P.K., Miguel and Jingshi. Lastly, I would like to thank my family and friends. I am very grateful to my Dad and Mom who encourages and supports me to realize all my dreams even though they have to sustain the long time separation from their beloved son. I owe all my achievement to my love, Xiangming, who shares all my joy and bitterness every day and night. Again, thanks to all my friends, for the happiness you brought into my life: Huili, Yang, Ping, Hongyuan, Kelly, Likun, Xiaojie, Songming, Rui, and all my basketball pals.
vi
VITA June 14, 1974
Born in Tianjin, China
June 1995
Bachelor of Science, Electronics Science and Information System, Nankai University, Tianjin, China
September 1995
Research Assistant, Dept. of Electrical Engineering, Tsinghua University, Beijing, China
June 1998
Master of Science, Electrical Engineering, Tsinghua University, Beijing, China
September 1998
Graduate Student Researcher, Dept. of Electrical and Computer Engineering, University of California, Santa Barbara
December 2002
Doctor of Philosophy, Electrical and Computer Engineering, University of California, Santa Barbara PUBLICATIONS
1. P. Jia, R.A. York, “Multi-Octave Spatial Power Combining in Oversized Coaxial Waveguide”, IEEE Trans. Microwave Theory and Tech, vol.50, (no.5), IEEE, May 2002. p.1355-60. 2. P. Jia, R.A. York, “A Compact Coaxial Waveguide Combiner Design For Broadband Power Amplifiers”, IEEE MTT-S International Microwave Symposium Digest, Pheonix, USA, May 2001. p.43-6 vol.1. 3. P. Jia, L.-Y. Chen, N.-S. Cheng, and R.A. York, “Design of Waveguide Finline Arrays for Spatial Power Combining”, IEEE Trans. Microwave Theory and Tech., vol.49, (no.4, pt.1), IEEE, April 2001. p.609-14. 4. P. Jia, Y. Liu, R.A. York, “Analysis of A Passive Spatial Combiner Using Tapered Slotline Array in Oversized Coaxial Waveguide”, 2000 IEEE MTT-S International Microwave Symposium Digest, Boston, MA, USA, June 2000. p.1933-6, Vol.3.
vii
5. N. –S. Cheng, P. Jia, D. B. Rensch and R. A. York, “A 120-Watt X-Band Spatially Combined Solid-State Amplifier”, IEEE Trans. Microwave Theory and Tech., vol. 47, (no. 12), IEEE, Dec. 1999. p.2557-61.
viii
ABSTRACT Broadband High Power Amplifiers Using Spatial Power Combing Technique by Pengcheng Jia
High power, broad bandwidth, high linearity and low noise are among the most important features in amplifier design. Realizing all these features in one amplifier remained as a big challenge for RF engineers. Broadband spatial power combining technique addresses all these issues by combining the output power of a large quantity of Microwave Monolithic Integrated Circuit (MMIC) amplifiers in a broadband environment, while maintaining good linearity and improving phase noise of the MMIC amplifiers. The intent of this research is to extend the waveguide based combiner design to broadband applications with emphasis on linearity improvement and phase noise reduction. Coaxial waveguide was used as the host of the combining circuits for broader bandwidth and better uniformity by equally distributing the input power to each element. The goal also includes the standardization of the modeling technique. Meanwhile, thermal management, efficiency, noise figure, phase noise and linearity analyses are all covered in this work. A broadband low noise medium power amplifier is first presented. The coaxial waveguide combiner is utilized to combine the output power of 32 low noise MMIC
ix
amplifiers. A bandwidth from 3.5 to 14 GHz is achieved with a maximum power of 1 watt. The residue noise of the amplifier is lower than –150 dBc at a 10 KHz offset from the carrier with 15 dB reduction compared to the residual phase noise of a MMIC amplifier. A new compact coaxial combiner with much smaller size is further investigated. Broadband slotline to microstrip line transition is integrated for better compatibility with commercial MMIC amplifiers. Thermal simulations are performed and a new thermal management scheme is employed to improve the heat sinking in high power application. A high power amplifier using the compact combiner design is built and demonstrated to have a bandwidth from 6 to17 GHz with 45-watt maximum output power. Linearity measurement has shown a high IP3 of 54 dBm. Residual phase noise is –140 dBc at a 10 KHz offset from carrier.
x
TABLE OF CONTENTS Preface........................................................................................................................... 1 1. Overview of Spatial Power Combining Technology ............................................ 4 1.1 Introduction ................................................................................................... 4 1.2 TWT Amplifiers............................................................................................ 6 1.3 Solid-State Amplifiers and Power Combining Technology.......................... 9 1.4 Spatial Power Combining............................................................................ 11 References ............................................................................................................... 21 2. Electromagnetic Modeling .................................................................................. 23 2.1 Modeling of the Rectangular Waveguide Combiner................................... 24 2.2 Modeling of the Coaxial Waveguide Combiner ......................................... 38 References ............................................................................................................... 51 3. Broadband Medium Power Amplifier Using Coaxial Waveguide Combiner..... 52 3.1 Passive Combiner Measurements................................................................ 53 3.2 Performance of the Active Combiner.......................................................... 60 References ............................................................................................................... 66 4. Design of High Power Amplifier Using the Coaxial Waveguide Combiner ...... 67 4.1 Motivation ................................................................................................... 68 4.2 Coaxial Waveguide Design......................................................................... 69 4.3 Synthesis of Waveguide Finline Array ....................................................... 70 4.4 Slotline to Microstrip Line Transition......................................................... 71 4.5 Compact Passive Structure of Coaxial Waveguide Combiner.................... 75 4.6 Leakage from Output to Input..................................................................... 76 4.7 Uniformity................................................................................................... 78 4.8 Fabrication Procedure ................................................................................. 85 4.9 Circuit Tray & Bias Line............................................................................. 87 4.10 Efficiency, Reliability and Thermal Analysis ............................................. 88 References ............................................................................................................... 98 5. Performance of High Power Amplifier Using Compact Coaxial Waveguide Combiner..................................................................................................................... 99 5.1 Measurement System ................................................................................ 100 5.2 MMIC Amplifier Characterization............................................................ 105 5.3 Output Power............................................................................................. 107 5.4 Linearity .................................................................................................... 113 5.5 Noise Figure .............................................................................................. 125 5.6 Spurious-Free Dynamic Range ................................................................. 128 5.7 Phase Noise of Combiner.......................................................................... 129 5.8 Summary ................................................................................................... 136 References ............................................................................................................. 137 6. Conclusion and Future Works........................................................................... 138
xi
Preface
PREFACE
High power, broadband, high linearity and low noise are among the most important features in amplifier design. When I started this project, realizing all these features in one amplifier remained as a big challenge for RF engineers. Research in spatial power combining led me into the wonder world of new technologies to solve these enigmas. To achieve all the features in one innovative design becomes the ultimate goal of my thesis work. Although most of the spatial power combining research groups are originated from Caltech’s Rutledge group, UCSB’s microwave group differentiates itself from other research groups by investigating in the waveguide based spatial power combining design with the guidance of Prof. Bob York. Dr. Angelos Alexanian and Dr. Nick Cheng have presented several exciting designs in the past years. The intent of this research is to extend their results to broadband applications with high output power while maintaining good linearity and low noise. The goal also includes the
1
standardization of the modeling technique. Meanwhile, efficiency, noise figure, phase noise and linearity analyses are added to supplement previous works. The thesis is organized in the following way. Chapter 1 compares waveguide based spatial power combining technique with the traveling tube amplifiers, corporate power combining and other spatial power combining techniques. The benefits of the coaxial waveguide design over the rectangular waveguide design are also explained. Chapter 2 follows with the modeling of the waveguide based spatial power combiners. Modeling for the rectangular waveguide combiners is covered first since it is the basis for all waveguide based combiner design. Waveguide model is then revised for coaxial waveguide application. Comparison with results from commercial software HFSS verifies the effectiveness of the modeling. Using the models and design parameters derived from the previous chapter, chapter 3 develops a practical design of coaxial waveguide combiner. 32 low power MMIC amplifiers are integrated into the combiner to build an active amplifier. 3.5 to 14 GHz bandwidth is achieved with 1watt output power. Chapter 4 addresses the size issue and leads to a more compact version. New slotline to microstrip line transition is characterized and combined into the system for better connections with commercial MMIC amplifiers. Fabrication process and thermal analysis is also covered in this chapter. Chapter 5 introduces the measurement system. Measurement result shows that the amplifier has 44-watt output power capacity and 6 to 17 GHz bandwidth. A variety of
2
practical issues that arise in the design of PA circuits and system, including linearity, noise figure and phase noise are also treated in this chapter. In Chapter 6, conclusions are made and potential improvements are presented as future works.
3
1.
Overview of Spatial Power Combining Technology
CHAPTER 1 Overview of Spatial Power Combining Technology 1.1
Introduction The history of microwave technology is a history of progressive advances in the
techniques used to generate, amplify, and process signals at microwave frequencies. Invented in the 1940s, the Traveling Wave Tube (TWT) has become a key element in microwave systems for radar, satellite communication and wireless communication. Currently, the Traveling Wave Tube Amplifier (TWTA) is the dominant choice for signal amplification subsystems in communication with operational frequencies higher than 4 GHz and power level greater than 10 Watt. An alternative to the TWTA is a power combiner. Power combining technique has been exploited extensively to improve the output power level from solid-state devices[1]. Two types of power combining techniques are the corporate combiner and spatial power combiner. The corporate combining technique will lead to very high 4
combining loss when integrate large amount of amplifiers, while spatial power combining technique was proposed with the goal to combine the a large quantity of solid-state amplifiers efficiently and improve the output power level to be competitive with TWTA. Many research groups including the Caltech’s Prof. Rutledge’s group, has done extensive research on the planar “tile” combing approach[2]. While UCSB’s microwave group attempted a “tray” scheme inside waveguide to achieve broader bandwidth, better thermal management and more efficient power collection[3-5]. Employing the “tray” combining scheme, we demonstrated an X band power amplifier with 150 Watt output power using oversized WR-94 rectangular waveguide. The rectangular waveguide combiner is easier to be fabricated and also better for thermal management. But the dominant TE10 mode inside rectangular waveguide will lead to non-uniform illumination of the loaded antenna trays inside the waveguide. Hence, the output power will experience a soft saturation that will deteriorate the linearity or lead to large back off of output power to satisfy the requirement of linearity. To meet the requirement of high linearity in many broadband communication systems, we extend the “tray” approach from rectangular waveguide to coaxial waveguide. A multi-octave bandwidth amplifier achieved bandwidth from 3.5 to 14 GHz with good linearity using oversized coaxial waveguide combiner[6]. A modified amplifier using more compact coaxial waveguide combiner design has shown 6 to 17 GHz bandwidth with 45 Watt maximum output power while with good linearity and high dynamic range. That enables it a good rival for current dominant TWT amplifiers. 5
1.2
TWT Amplifiers Vacuum electronic amplifiers are used in a wide variety of military and
commercial applications requiring high power at microwave and millimeter wave bands. From the richness of device concepts investigated through the 1960s, the helix and coupled cavity traveling wave tubes (TWTs), klystron, magnetron, and crossed field amplifier (CFA) emerged as the primary products of today’s industry[7]. The traveling wave tube amplifier (TWTA) is the most widely used vacuum electronic amplifier in communication systems that require wide bandwidth. It was invented in 1940s by Rudolph Kompfner and has had a long history in playing a key technology role in a variety of applications ranging from military and radar systems to commercial communication systems. Advances in TWT amplifiers (TWTA) have made satellite communications a huge success. Among them, helix TWT is widely used for communication applications, while coupled cavity TWT is better for narrowband systems requiring higher power. Along with the advances in tube and solid-state devices, the amplifier manufacturers have made great strides in improving the complete TWTA package and subsystems. TWTA sizes have shrunk considerably, power handling is better, costs have come down, microprocessor control is now standard, other features have increased and reliability is better than ever. A basic helix TWTA block diagram is shown in Figure 1.1.
6
Figure 1.1 Diagram of a traveling wave tube amplifier.
All TWTs possess four major subassemblies: An electron gun that produces a high density electron beam; A microwave slow-wave circuit that supports a traveling wave of electromagnetic energy with which the electron beam can interact; The collector that collects the spent electron beam emerging from the slow-wave circuit; The TWT package, which provides points for attachment to the using system, provides cooling for power dissipated within the TWT, and, in some cases, includes all or part of the beam focusing structure. The TWTA is a very complicated system and needs a lot of touch labor. It can only be manufactured in hundreds each month. But currently TWTA is the only high power amplifier that can work over a broad bandwidth. It can cover C, X, Ku band or a big fraction in Ka bands with power level from 10 Watts to 3000 Watts. The large
7
bandwidth and high power levels have made them dominant power amplifier for satellite communication. The advance in tube technology has improved the efficiency of the TWT amplifier up to 70% for narrow band and 50% for broadband, which is the present best solution for space satellite transponders. However, the drawbacks of the TWT amplifiers are also obvious, such as considerable size and weight. Tube amplifiers also need the high voltage drive - Electronic Power Conditioner (EPC) that requires additional complex accessory circuit and involves high voltage risk. Moreover, the tube amplifier is always rated with saturation power, which leads to bad linearity and is not good for broadband communication. To work linearly, the TWT amplifier is normally backed off from its saturated output power or additional linearization circuits are added. Linearization circuitry results in dramatically increase of system complexity and cost. Moreover each small increase in efficiency is very expensive. A high efficiency high power TWT amplifier in satellite may cost up to half a million dollars. Recently vacuum tube engineers have taken advantage of the MMICs to develop the MPM (Microwave Power Modules)[8, 9]. In brief, an MPM is the combination of a solid-state exciter and TWT amplifier that has similar gain-bandwidth characteristics as a conventional TWT but is much smaller and has superior noise, efficiency and linearity characteristics. As such, it is much smaller than a conventional TWT and can operate with much lower supply voltage. However, MPMs still require a great deal of "touch labor" in their assembly and testing and, hence, are prone to higher costs and lower production rates.
8
1.3
Solid-State Amplifiers and Power Combining Technology Improvements in solid-state material and amplifier have pushed the output power
level of a single MMIC (Microwave Monolithic Integrated Circuit) to the watt level, but there is still no commercially available MMIC amplifier that can output more than 5 Watts over X band. Even with the advent of promising high-power solid-state devices based on wide bandgap semiconductor materials such as gallium nitride (GaN) and silicon carbide (SiC)[10, 11], it is still difficult and costly at the present time to realize significant RF output power at a single device level. There is no surprise that vacuum electronics are still the dominant technology for high power applications. However, solid-state electronics are generally more desirable in terms of size, weight, reliability, and manufacturability. Economic considerations can also favor solid-state systems that can be mass-produced using modern IC technology.
Figure 1.2 A corporate structure for power combining. (From reference [1], IEEE).
9
Power combining techniques has been investigated extensively. The corporate combiner is one of the classic and most popular structures[1]. To satisfy system requirements, power from many individual devices must be added coherently. As shown in Figure 1.2, the outputs from a number of circuits are successively combined using two-way adders such as Wilkinson combiners. The number of individual devices is 2N, where N is the number of stages. The combining efficiency is therefore η = LN, where L is the insertion loss of each stage. Note that the physical layout of the corporate combiners with many elements causes the transmission lines in the last stages of combining to become very long. As the number of devices increases, the losses in these lines become insurmountable. As shown in Figure 1.3, loss of the combining circuit will increase dramatically and the output power of the combiner will even decrease when the number of devices is very large. 50
Power Added Efficiency, %
45 40 35 30 L = 0.1dB (Corporate) L = 0.2dB (Corporate) L = 0.3dB (Corporate) L = 0.5dB (Spatial) L = 1.0dB (Spatial) L = 1.5dB (Spatial)
25 20 15 1
10
100
1000
Number of Amplifiers
Figure 1.3 Output power available from combiners as a function of the number of elements.
10
High losses associated with circuit combining schemes can be avoided with spatial power combining technique since the energy is combined in the lossless space or low loss electric medium. In Figure 1.3’s example, the spatial power combining is superior when there is a large quantity of elements.
1.4
Spatial Power Combining
Overview of the spatial power combining technology
Spatial power combining was reported as early as 1968 with the construction of a 100-element spatially fed/spatially combined array for operation between a pair of electrically short monopole antennas[13]. This active array approach has dominated the research on spatial power combining recently while the antenna array has improved to be smaller and more broadband[2, 14]. Their active arrays interact with propagating beams in free space[15]. The incident, reflected and transmitted beams are guided and manipulated via conventionally optical components such as mirrors, polarizers and lenses. Therefore optical techniques are applied to systems operating far below the optical spectrum, hence the name quasi-optics. The natural configuration for such a system is therefore a Gaussian-beam waveguide, with the array placed at a beam waist where the phase front is planar. Many of the array concepts are designed for use in closed metallic waveguide to provide good packaging[16, 17]. It is attractive at frequencies below 100 GHz for several reasons: diffraction losses and focusing errors are minimized or eliminated, since all of the energy is confined by the waveguide walls; the metal walls provide a
11
convenient heat sink for the arrays; waveguide components are readily available in this frequency range and maybe more economical than quasi-optical components; and using quasi-optical arrays in this way allows them to be retrofitted into existing waveguide systems. A big disadvantage of this design is the nonuniform field profile in the waveguide leading to the edge elements on the array not coupling efficiently to the waveguide mode resulting in lower efficiency and linearity. This can be addressed using dielectric loading or sidewall corrugations[18, 19]. Furthermore, standard waveguide may not accommodate a sufficient number of devices unless the waveguide cross-section is enlarged. A higher-order mode suppression technique is also necessary in oversized waveguides. Although without optical beam guiding components, it is still be distinguished as quasi-optical combiner since it uses spatially fed/spatially combined array. One of the most popular spatial combining architecture is a tile approach. This approach denotes configurations that use relatively thin modules where the RF circuitry and active devices are placed on circuits parallel to the face of the array. The tile approach includes 2 popular designs, active array amplifier and grid amplifier. These designs, illustrated in Figure 1.4, are quite different, and each has its merits. The grid amplifier is an array of closely spaced differential transistor pairs. The input and outputs are cross polarized, and off-chip polarizers are used for tuning. The drawback of grids is that the small cell sizes limit the gain and power per cell to that available from a single differential pair. Because the active devices are very dense, the grid amplifier can be monolithically fabricated; this makes grids a very attractive 12
technology for moderate gain and power applications that demand a single chip massproducible solution. Active arrays, on the other hand, use larger unit cells with more conventional antennas like patches or slots. This larger unit cell allows integration of multi-stage MMICs with higher gain and output power. The passive radiating and tuning elements do tend to occupy a significant fraction of the active array and tray amplifier’s area; the most economical solution is to attach active MMICs to passive antennas. Active arrays may find use in very-high power or gain applications.
Figure 1.4 Two tile architectures: (a) grid amplifiers, (b) active array amplifiers. (From reference [12],IEEE.)
Attempts have already been made to apply the grid amplifier design to industrial products since it can easily fit all of the required components on a GaAs substrate. This approach has been well developed by research group in CalTech and achieved 5 Watt at Ka band from a GaAs chip. But heat removal or thermal management
13
becomes the biggest design challenge. The design can be modified to contain a heatconducting layer to channel the heart from the center array to the edges, but this heatconducting layer will be relatively thick to obtain adequate thermal conductance, then complicates the transfer of the signal from the receive side of the array to the transmit side. The thermal glue used to bond the solid-state chip to the heat-conducting layer will also add a high junction thermal resistance. The thermal management for output power of more than 20 Watt will be very difficult. A tray approach was then developed aiming to provide better thermal management and increase the bandwidth while achieving higher power[3]. This architecture provides more space for the RF circuitry and active devices. It can use amplifiers with higher gain and output power by integrating amplifiers in the longitude direction. Another advantage of this configuration is that the metal carrier of each circuit trays permits good heat conduction. The biggest disadvantage of the tray approach is the length of the system. However there are a lot applications where this is not an issue. Details of progress with the tray approach in UCSB will be discussed in the next section. Active spatial power combining has become a dynamic research field recently[2025]. Substantial power in the 100-Watt range at X band has been achieved by UCSB[5]. Sanders report a combiner with 272 MMICs generating 35 Watt at 61 GHz[26]. Researchers at Lockheed Martin and North Carolina State University have recently demonstrated a 25 Watt “tiled” combiner system at Ka Band (34GHz)[27]. Researchers at the California Institute of Technology have developed a 5 Watt single14
chip monolithic grid amplifier using Rockwell pseudomorphic high electron-mobility transistor (pHEMT) technology[28]. Moreover most of the basic components in microwave receiver and transmitters have been demonstrated as spatial combining circuits. These include amplifiers, oscillators[29], mixers[30], multipliers and switches[31]. Quasi-optical oscillators and amplifiers have been tested in wireless communications circuits, transponders, and beam switching systems. Aggressive developers now have the opportunity to build quasi-optical systems for a new generation of communication and radar equipments.
Rectangular Waveguide Spatial Power Combiner
Figure 1.5 (a) Schematic plot of the rectangular waveguide combiner (b) Layout of a single tray[4].
In the tray approach of spatial power combining, the RF circuitry and active devices are placed on circuits perpendicular to the face of the array and they usually
15
contain end fire antenna elements (such as a linearly tapered slot or Vivaldi antenna). Several tray approaches have been introduced, but the most successful one is the rectangular waveguide combiner developed by Angelos Alexian and Nick Cheng of UCSB[32]. The concept is illustrated schematically in Figure 1.5; we exploit the inherent spatial distribution of the field energy in the dominant waveguide mode to distribute and collect power to and from a dense array of amplifiers. Transitions between the amplifier and waveguide mode are made via electrically close taperedslot antennas (or finline structures). The combiner design is compact, but large enough to accommodate the desired number of amplifiers. The combiner is also well designed for thermal management for output power levels in hundreds of watts. The tray is made of copper with power MMIC amplifier sitting in the middle. The wasted heat is efficiently transferred from the center of tray to the outside waveguide surface, and then dissipated through forced air convection. The enclosed waveguide provides an excellent heat-sinking environment for the power devices and is a natural choice for most high-power applications. Another extraordinary feature of this design is the successful integration with broadband taper finline antenna. The frequency response of the passive structure is only limited by the cut-off frequency of the rectangular waveguide. Recent research activities include Nick Cheng’s 150 Watt X band amplifier, Vicki Chen’s K-band amplifier[33] and Jinho Jeong’s 3.3 Watt 24GHz amplifier using antipodal finline to microstrip line transition[34]. 16
With all these merits, the rectangular waveguide combiner is facing a big challenge of non-uniform illumination. The sinusoidal field distribution of the dominant waveguide mode TE10 will lead to different drive power at the input the MMIC array. When the input power increases and the MMIC amplifiers at outside trays begin to reach P1dB output power, the MMIC amplifiers at inner trays are already overdriven into deep saturation. The non-uniform drive will lead to a soft saturation of the amplifier, which is shown in Figure 1.6. If maximum output power is reached, intermodulation components will be very high due to the highly nonlinear operation of amplifier in inner trays. It will not be qualified for modern broadband communication systems that have high criteria for linearity.
Figure 1.6 Effect of non-uniform illumination of incident power on output power. The inset plot outlines the actual positions of the trays. Sinusoidal field distribution is assumed for the non-uniform case[4].
Other modifications of the rectangular waveguide are also investigated to improve the uniformity. The performance can be improved using longitudinal corrugations. The unique characteristic of the UC-PBG (uniplanar compact photonic bandgap) 17
structure allows for the possibility of building a TEM waveguide when used as a planar reflector. The UC-PBG reflector behaves like a magnetic surface at its stopband frequency where the periodic loading changes the surface impedance to an open-circuit condition. When the two sidewalls of a rectangular waveguide are replace by the UC-PBG structure, a parallel-plate model will be established by magnetic boundary conditions leading to better uniformity inside the waveguide. A relatively uniform field distribution was demonstrate from 9.4 to 10.4 GHz by UCLA[35]. To extend the bandwidth, an active UC-EBG (uniplanar compact electromagnetic bandgap) structure is proposed with varactors to electrically tune the stopband frequency[36]. However, the present UC-PBG/EBG structure still can’t produce a satisfactory parallel plate mode with uniform illumination to all the elements. Moreover the bandwidth is limited and the active tuned structure increases the complexity of system. So the best solution to provide uniform illumination for all elements will be a coaxial waveguide combiner design.
Coaxial Waveguide Spatial Power Combiner
Coaxial waveguide combiner was exploited as the cylindrical resonant cavity combiner in the 70s[1]. Angelos Alexanian first applied the coaxial structure to spatial power combiner field with a preliminary demonstration of the idea using passive elements.
18
Here we extend the tray approach to the coaxial waveguide by radially dividing a coaxial waveguide section into 16 identical wedge trays and integrating with similar broadband tapered finline antennas as the rectangular waveguide combiner. In this combiner, MMIC amplifiers are placed on a bridge in the middle of each tray, and then connected with input and output antenna that couples energy from and to the waveguide. When the 16 tray sections are stacked together, the coaxial waveguide opening is formed. Then input and output coaxial waveguide section will be used to transit from the center stacked section to type N connectors. The input/output transition is optimized with a taper design to minimize the reflection with maximum bandwidth. Details of the design will be covered in the following chapters. There is very small dispersion since the dominant mode that propagates along the coaxial line is TEM mode. And the dense antenna array helps to suppress the higher modes in the oversized waveguide. We maintain all the benefits of the tray approach; while we future broaden the bandwidth of the combiner by fully exploiting the bandwidth of the tapered finline array because the coaxial waveguide doesn’t have any cut-off frequency. Moreover, the uniform illumination of all the antennas helps to maintain the amplifier’s linearity same as each MMIC amplifiers. The coaxial waveguide combiner has the potential of achieving multi-octave bandwidth. And the minimum spacing between the trays will be the MMIC amplifier’s size. So the coaxial waveguide combiner has the benefit of large-scale integration and can provide very high output power. High thermal dissipation
19
capacity of the tray approach enables the large-scale power combining feasible by providing reliable heat removal solutions. When the amplified signals from different MMIC amplifiers are combined, the power are added since all the signals are in phase. But the residue phase noises from each MMIC amplifier are irrelevant random variables. The sum of the residue phase noises will have the same power as that of a single MMIC. For a N-elements combiner, we will observe N times residue phase noise reduction in the output comparing with using only one MMIC amplifier. Table 1.1 Comparison between TWTA and QO amplifiers
TWTA
Quasi Optical Amplifiers Grid
Rect. WG
Coax. WG
5 Watt
150 Watt
50 Watt
Power
10-3000 Watt
Freq Band
C, X, Ku, Ka, Ka, V V
C, X, Ku, V
C-Ku
Efficiency
40-50%
15%
20%
17%
Linearity
Back off 7 dB N/A from rated power
N/A
Back off 3 dB from P1dB
Low
Low
IMD 2 A . Assuming the propagation constant is a monotonically increasing function of frequency, the lowest operating frequency is therefore defined by
θt ( f0 ) = 2 A
(4.4)
which is an implicit relationship between the taper length L , the lower cutoff frequency f 0 , and the maximum reflection coefficient Γ m . The main difficulties in applying the above results are the frequency dependence of the wave impedance and propagation constant, the difficulty in translating the impedance as a function of θ into a function of z , and the subsequent determination of the physical parameters required to synthesize the impedance taper. The frequency dependence of the wave impedance and propagation constant means that in general, the result (4.3) will require a different physical taper at each frequency, which is obviously not a possible implementation.
However, the normalized impedance
Z (θ ) / Z 0 of the finline transition is found to be a relatively weak function of frequency. We can therefore design the taper at a fixed frequency, chosen to be f 0 . In addition, it has been found that waves propagating along the finline structures are
28
approximately TE in character, allowing us to relate the wave impedance and propagation constant as
Z=
ωµ β
(4.5)
Using wave impedances instead of the characteristic impedance, (4.3) can now be rewritten in terms of β
β ( f0 , z) =
2θ ( f0 , z )
β L β 0 exp −Γ m A2 F
θt
− 1, A
(4.6)
where β , β L and β 0 correspond to Z , Z L and Z 0 , respectively, using (4.5). To compute the required propagation constant as a function of the position along the taper, β ( z ) , the taper structure is divided into N sections of length ∆z = L / N . θ can be approximated by i −1
θ ( zi ) ≈ ∑ 2 β ( zk )∆z = θ ( zi −1 ) + 2 β ( zi −1 )∆z
(4.7)
k =0
where zi = i∆z . Noting that θ (0) = 0 , we first evaluate β (0) from (4.6), and then use the approximation (4.7) in (4.6) to evaluate all subsequent values of β ( zi ) along the taper. We then repeat the iterative process until the solution set of β converges. The resulting procedure is similar to that used in [3] for single finline structures. Note that the propagation constants at the ends of the taper do not match the target values, β (0) ≠ β 0 and β ( L) ≠ β L . Using (4.6) it can be shown that
29
β ( L) = β L eΓ
(4.8)
m
and this either fixes the maximum reflection coefficient for a given β ( L ) , or determines β ( L) for a given Γ m . For a dense finline array, there is a potentially large impedance discontinuity in the transition from an unloaded waveguide to a loaded waveguide, so it is necessary to manipulate the substrate material, thickness, tray locations, and local waveguide width in order to satisfy (4.8) for a desired Γ m . Another possibility is to include a quarter-wave “notch” transformer as part of the finline transition [4], but this proved impossible in the present work due to the use of ceramic substrates which were difficult to machine.
Propagation Constant of a Finline Array
y PMC
b/2 t1 t2 1
g
2
X=d
t3 3
x
0 -a/2
c
a/2 PEC
-b/2
Figure 2.4 Cross section of a 2×2 finline array in a standard waveguide environment.
In this work the Spectral Domain Method (SDM) [5, 6] was used to find the relationship between the propagation constant and the geometrical parameters of the fin-line, most importantly the slot width. For simplicity, a 2×2 fin-line array was
30
analyzed, as shown in Figure 2.4. We assume perfect contact between the finline and the waveguide walls. Symmetries along the major axes were used to reduce the computation domain to the upper right quadrant of Figure 2.4. In the SDM, the electric fields and currents in each region are expanded as a Fourier series in y . Denoting the electric field in the i th region as Ei , Ei =
∞
∑ Ee α
n =−∞
j
i
ny
where α n =
2nπ b
2 b/2 Ei = ∫ Ei e jα n y dy b 0
(4.9)
Applying the boundary conditions at the interface x = d gives two algebraic equations Yyy E y + Yyz Ez = jωµ0 J y , Yzy E y + Yzz Ez = jωµ0 J z .
(4.10)
where the J are the unknown currents on the fins. Using the equivalent transmissionline “immitance” concept [6], we find Yyy = jωµ0 (YTE cos 2 θ + YTM sin 2 θ ) Yyz = −Yzy = −ωµ0 sin θ cos θ (YTM − YTE ) Yzz = jωµ0 [YTE sin 2 θ + YTM cos 2 θ ] where
31
(4.11)
β
sin θ = Γ ni =
β 2 + αn
cosθ =
2
αn β 2 + αn
2
,
β 2 + α n − ω 2 µ 0ε i 2
YTMi = jωε i / Γ ni ,
YTEi = Γ ni / jωµ 0 ,
YTML = YTM 1 tanh(Γ n1t1 ), YTEL = YTE1 tanh(Γ n1t1 ), YTM = YTM 2 YTE = YTE 2
YTML + YTM 2 tanh(Γ n 2t2 ) YTM 2 + YTML tanh(Γ n 2t2 ) YTEL + YTE 2 tanh(Γ n 2 t2 ) YTE 2 + YTEL tanh(Γ n 2 t2 )
+ YTM 3 coth(Γ n 3t3 ),
+ YTE 3 coth(Γ n 3t3 ).
The unknown aperture fields E y and Ez are expanded in terms of a basis set of rectangular pulses ξ i and ηi , which is more convenient for the wide slot portion in the finline, and then Fourier transformed to Ny
E y (α n ) = ∑ ciξ i (α n ), i =1
Nz
(4.12)
Ez (α n ) = ∑ diηi (α n ). i =1
Substituting (4.12) into (4.10) and integrating gives a homogeneous matrix equation Ny
Nz
i =1
j =1
Ny
Nz
∑ K piyy ( β )ci + ∑ K pjyz ( β )d j = 0, p = 1… N y ∑K i =1
zy qi
(4.13)
( β )ci + ∑ K qjzz ( β )d j = 0. q = 1… N z j =1
The propagation constants over the normalized gap 2 g / b are then found from the characteristic equation obtained by setting the determinant of the coefficient matrix in
32
(4.13) to zero. Figure 2.5 shows the results of this calculation for a representative physical situation of interest, corresponding to two 10-mil thick aluminum nitride (AlN) substrates, with a separation of c = 5 mm, placed in an X-band (WR-90) waveguide with dimension a = 0.9”, b = 0.4”. A single “pulse” basis function was used in this calculation.
e ffe c ti v e pe rmi ttiv ity
3.5 3 2.5 2 1.5 1 0.5 0.2
0.4
0.6
0.8
1
normalized gap width
Figure 2.5 Effective permittivity versus normalized gap width for a 2x2 finline arary in WR90 waveguide.
Using the results in Figure 2.5, an “optimized” taper was computed for an input reflection coefficient of –20dB. The shape of the optimized tapered finline is shown in Figure 2.6. no rmalized gap width
1 0.8
optimized taper
0.6 0.4 0.2 0 0
0.25
0.5
0.75
1
1.25
1.5
1.75
Position along taper, cm
Figure 2.6 Normalized gap width vs. location along the optimal tapered finline.
33
Scaling, Losses, and Combining Efficiency
The optimized taper design in Figure 2.6 was used as the basis of several combiners.
The 2x2 array taper design was scaled up to larger arrays. In the
combiner reported in [7], two additional trays were added to form a 4-tray 2x4 array. Not surprisingly, some degradation in return loss is observed as compared with the 2tray 2x2 array results.
50 Ω microstrip line
Finline taper 70-to-50 Ω microstrip taper
0
0
-5
-5
-10
-10
-15
-15
-20
-20
-25
S11: 2 Tray, dB S11: 6 Tray, dB
S21: 2 Tray, dB S21: 6 Tray, dB
S21, dB
S11, dB
Figure 2.7 Tray layout for a through measurement.
-25 -30
-30 8.5
9.0
9.5
10.0
10.5
11.0
11.5
12.0
Frequency, GHz
Figure 2.8 2-port measurements for the slotline arrays with 2 trays and 6 trays.
34
The 2-tray 2x2 array was then scaled up to a 6-tray 4x6 array design. In this case the number of finline transitions on each tray was doubled to four by linearly scaling the dimensions of the 2x2 design. The number of trays was increased to six, which required the use of a thinner tray and hence closer tray-to-tray spacing. In this case, 2port measurements were performed in order to examine the return loss and insertion loss characteristics of the passive structure. The layout of the test circuit (Figure 2.7) consists of back-to-back finline transitions with 50Ω microstrip lines used in place of active elements. As a consequence of the reduced tray spacing, the terminating resistance yielding the best impedance match was lowered to 70Ω, so a 70-to-50Ω taper was included in the microstrip through line. Measured reflection and transmission for the 2-tray (4x2) and 6-tray (4x6) systems are shown in Figure 2.8.
The 6-tray system with 24 tapers suffers
significantly increased reflection losses. Since the tray separation in the 4x6 system is the same as in the 4x2 system, the deterioration in performance results from the nonoptimized taper rather than the inter-tray coupling. Nevertheless, an important observation can be made with respect to insertion loss: ignoring the effects of reflection losses (which can potentially be recovered by properly optimizing the finline taper for the 4x6 array), the insertion losses appear approximately constant in the 2-tray and 6-tray system. This is a natural result of the parallel nature of the transitions, but has important consequences in combining efficiency when scaling the combiners to large numbers of devices. Whereas traditional combiners suffer a
35
reduction in combining efficiency as the number of devices is increased, the spatial combiners exhibit a loss that is roughly independent of the number of devices.
Insertion Loss, dBdB Dissipative Loss,
0
-2
-4 2 Tray, dB 6 Tray, dB 8 Tray, dB
-6
-8
-10 8.5
9.0
9.5
10.0
10.5
11.0
11.5
12.0
Freqency, GHz
Figure 2.9 Measured dissipative loss for 2,6, and 8-tray finline array with though lines.
To empirically quantify this assertion using the measured results of Figure 2.8, we compute a “dissipative loss” as Loss =
Pload Pload | S21 |2 = = Pforward Pin − Preflection 1− | S11 |2
(4.14)
The losses computed from (4.14) for the 2,6, and 8-tray configurations, plotted in Figure 2.9, show that the dissipative loss in each case is approximately constant as a function of frequency, and independent of the number of trays. It can then be shown that the maximum combining efficiency ηc of a spatial combiner structure is
ηc ≈ Lo
36
(4.15)
where Lo is the loss associated with the combiner circuit (post-amplification losses). In the present case, any loss can be attributed equally to the input and output antennas, which gives us an estimate of the maximum potential combining efficiency as
ηc ≈
| S21 |2 1− | S11 |2
(4.16)
Loss and efficiency based on (4.14) and (4.16) are shown in Figure 2.10, using the average loss over the band from the measured 2-port results of Figure 2.8 and similar measurements for an 8-tray system. The structure can be scaled up to accommodate more devices. Figure 2.10 shows a very small reduction in combining efficiency when the number of elements increases from 8 to 32. The separation between trays and size of the waveguide
100
0
80
-1
60
-2
40
-3
20
-4
0
Total Loss, dB
Efficiency, %
determine the maximum number of elements that the structure can accommodate.
-5
5.0
10.0
15.0
20.0
25.0
30.0
35.0
40.0
Number of Amplifiers
Figure 2.10 Efficiency & total insertion loss of power combiner vs. number of elements in finline array.
37
The impact of failed elements on the total loss of the spatial power combiner system is analyzed in [8]. The array will degrade gracefully, in agreement with the theoretical analysis. The design procedure for dense finline arrays has been applied to the design of active combiner systems forming the basis of successful combiner implementations reported in [9] and [10]. We have shown that the dissipative loss in such slotline arrays is approximately independent of the number of transitions used providing a compelling argument for the use of spatial combiner systems to combine the power from large numbers of active devices. We also explored the limits of scaling tapered finline structures to increasingly dense configurations, showing that the degradation in return loss is graceful but can become significant when designs are scaled by large multipliers. We will need to optimize the design accordingly or use more sophisticated optimization procedures [11] to improve the design of the arrays for dense configurations.
2.2
Modeling of the Coaxial Waveguide Combiner
The concept of using the coaxial waveguide for spatial power combining was first introduced by Dr. Alexanian and Dr. York in [7], with a preliminary demonstration of the idea using passive elements. The center coaxial waveguide is oversized and populated with a finline array and active amplifiers. Optimized coaxial waveguide transitions provide the connection from a type N connector to the oversize waveguide. In this section, we will discuss synthesizing the optimum waveguide 38
transition using small reflection theory.
We will then describe the synthesis of
optimized finline structures for realizing multi-octave operation of coaxial combiner structures, using an adaptation of the procedures developed for rectangular waveguide combiners. The design is applied to a 32-tray system and verified with HFSS simulations.
Design Concept
High-power amplifiers with multi-octave bandwidths are difficult to realize in MMIC technology, particularly when the power requirements call for combining the outputs of multiple MMIC amplifiers simultaneously. In the previous section, we have described an efficient combining technique using finline arrays in a rectangular waveguide, capable of operation over the full waveguide band. In principle it is possible to extend the operating frequency and amplifier capacity of such combiners by operating in a multi-mode environment; in fact, we have found that the use of very dense finline arrays can act to suppress high-order modes in such structures. But there are other difficulties; since the array is excited by a TE10 mode, the amplifier elements are driven non-uniformly, which can reduce the efficiency and distort the saturation characteristics of the system. In addition, the rectangular waveguide environment is dispersive, which complicates broadband impedance matching over an extended frequency range. These difficulties can be addressed by adapting the design to a TEM waveguide environment, such as in a coaxial waveguide. Figure 2.11 illustrates how this might
39
be done, using radial tapered-finline structures distributed uniformly in the annular aperture of an oversized coaxial line. The combiner is fed by gradually flared coaxial lines, which taper down to standard coaxial connectors at either end. Not only can this structure accommodate a large number of amplifiers and provide uniform illumination of the array, but it can also be designed for ultra-wideband operation. Slotline Array
Inner Conductor
MMIC Amplifiers Type N Connector
Outer Conductor
Center Section
Waveguide Taper
Figure 2.11 Schematic of an oversized coaxial waveguide combiner housing a dense finline array, with tapered transitions from type-N connector.
Optimum Waveguide Transition
As shown in Figure 2.11, an optimized coaxial waveguide taper is applied at both ends of the center section to transform from a standard 50-Ohm type N connector to the flared coaxial line. When finline array is loaded in the waveguide, the input impedance of each finline taper is the number of channels times the waveguide impedance. Lower waveguide impedance leads to a smaller waveguide aperture, which is helpful in suppressing higher modes, and keeps the finline taper shorter, resulting in lower conductive loss. Thus, the impedance of the center section was chosen to be 30 Ohms.
40
The reflection from the type N connector to flared waveguide line is minimized by the optimized coaxial waveguide transition. The gradual waveguide taper is synthesized using the small reflection theory of TEM lines, and has an input reflection coefficient Γin ( f ) =
1 θt − jθ d Z (θ ) ln e dθ 2 ∫0 dθ Z 0
(4.17)
where β is the propagation constant, θ t is the round-trip phase delay to a point z along the taper, L is the taper length, and θ t = 2 β L . In order to maintain a target input reflection Γin ( f ) over the desired bandwidth, it has been shown in [5,6] that Z (θ ) must take the form
ln
where A = Cosh −1
2θ Z (θ ) 1 Z L = ln + Γ m A2 F − 1, A . Z0 2 Z0 2θ t
(4.18)
Γ0 Z − Z0 1 Z L , Γ0 = L ≅ ln , and Γ m is the target reflection. Γm Z L + Z0 2 Z0
The characteristic impedance is defined as Z=
1 2π
µ0 D ln( o ) Di ε0
(4.19)
where Do and Di are the outer and inner diameters of the coaxial waveguide opening respectively.
41
Special attention must be paid to the definition of Γ 0 . To get an accurate impedance transition, the approximation of
1 ZL ln must be used because the 2 Z0
equation (4.18) comes from empirical results. An iteration process similar to that described in Section 2.1 is applied to optimize the taper with minimum reflection. The taper dimensions are shown in Figure 2.12 and the reflection coefficient is shown in Figure 2.13. 30
30
25
25 Di (mm)
Do (mm)
20
20
15
15
10
10
5
5
0
0 10
20
30
40
50
Z (mm)
Figure 2.12 Inner and Outer diameter of the optimized waveguide transition.
No solder or epoxy is used in assembling the system; the center conductor of the combiner is designed to mate directly to the center conductor of a type N connector for easy assembly. Unfortunately, this design results in additional loss and reflection. To reduce this loss, solder will be used to permanently connect them when it is fabricated as a product.
42
0 -10 -20 -30 -40 -50 -60 -70 -80 0
5
10
15
20
Frequency [GHz]
Figure 2.13 Reflection coefficient of the optimized transition.
Spectral Domain Modeling of the Finline Array
The finline array can be easily analyzed with a modern Electromagnetic (EM) simulator.
The procedure for synthesizing a broadband impedance-matching
transformer, described in the following section, requires an efficient code that can rapidly and iteratively evaluates the propagation constant of the structure over a range of frequencies and physical dimensions. The Spectral Domain Method (SDM) is well suited for this purpose[12]. A cross section of the loaded coaxial line is shown in Figure 2.14. We assume that each substrate carries two separate finline tapers, and that there is intimate electrical contact with the inner and outer coaxial conductors. Due to the symmetric loading and a dominant mode excitation, the computation domain can be reduced to a single waveguide cell, with PEC (Perfect Electrical Conductor) and PMC (Perfect Magnetic Conductor) boundary conditions as shown in Figure 2.14. The PEC boundary 43
condition is applied again to divide the waveguide cell radially into two unit cells. Each unit cell is left with one tapered finline and a constant ratio of outer radius to inner radius, thereby maintaining identical characteristic impedances. This unit cell could be modeled in cylindrical coordinates, but we choose to approximate each unit cell as a parallel plate waveguide that our previously developed methods for rectangular structures are applicable. y L Slotline taper
z
PEC PMC
PMC
y b PEC g
PMC PEC Loaded Waveguide
Waveguide Cell
PMC a PEC x Unit Cell
Figure 2.14 Schematic cross section of a coaxial waveguide with a uniform loading of radial finline structures.
At this point, the SDM approach is virtually identical with our approach for rectangular waveguide finline arrays in the previous section. The difference is that the sidewall boundary conditions are changed, which is a simple step in the SDM method.
44
3.5
y b
3
εr
g
2.5
12GHz
2
0 c
1.5
a
x
1
7GHz
0.5 0.2
0.4 0.6 normalizedslot
(a)
0.8
1
3 .5 3
εr
SDM 18GHz
2 .5 2 1 .5 1
4 GHz HFSS@8 GHz
0 .5 0 .2
0 .4
0 .6
0 .8
Normalized slot width g/b
1
(b) Figure 2.15 Effective permittivity from SDM and HFSS for varying slot width and frequency (a) inside rectangular waveguide with 2 trays (b) inside coaxial waveguide with 32trays.
The effective permittivity ( ε r = ( β / k0 ) ) versus normalized slot width for a range 2
of frequencies from 4-18GHz is shown in Figure 2.15(b), assuming a 32-tray system with 10-mil Aluminum Nitride substrates. The SDM simulation for a rectangular waveguide is shown in Figure 2.15(a) for comparison. Clearly there is little variation of the propagation constant with frequency for the coaxial waveguide finline array, indicating the desired quasi-TEM behavior. The magnetic field distribution of the slotline also verifies the TEM characteristic. As a numerical check on the SDM code, we analyzed the structure at one particular frequency using Agilent’s High Frequency Structure Simulator (HFSS).
45
The results shown in Figure 2.15 indicate good agreement with the SDM simulation. Furthermore, we observe the Electrical Magnetic field distribution inside the waveguide cell with HFSS simulator. The field shown in Figure 2.16 verifies that dominant mode is TEM mode.
(a)
(b)
Figure 2.16 (a) E field (b) H field of a cross section at 2 narrow slot end inside the waveguide cell.
Test of an Exp-Sin Shape Finline Array y Lt
50 Ohm termination
Slotline taper
z
Figure 2.17 Test circuit of an Exp-Sin shape slotline array.
To verify the simulation results, a 32 tapered finline array was fabricated. The end of the finline was terminated with 50-Ohm thin film resistors. The taper, shown in
46
Figure 2.17, uses the shape of Exp[ Sin(
πz 2 Lt
)] , where Lt is the length of the taper. The
bandwidth is determined by the length of the taper line: the longer Lt, the lower the cutoff limit. For this reason we chose Lt to be 1.2 inches. The finline openings have the same outer to inner radius ratio, resulting in the even distribution of energy between the upper taper and lower taper. The ends of the finlines are terminated with 50-Ohm thin film resistors, which have low parasitic inductances. S parameter measurements were performed. The return loss was plotted in Figure 2.18. The data shows that the return loss was lower than –10dB from 4 to 16 GHz. Good agreement between the measured results and an HFSS simulation prove that this structure can be used as a broadband power combiner.
Return Loss
Measurement HFSS Simulation
0 -5 -10 -15 -20 -25 -30
0
2
4
6 8 10 12 14 16 Frequency [GHz]
Figure 2.18 Measurement and HFSS simulation of return loss for 32 cards passive slotline array with 50 Ohm load.
Although the simulation results and test circuit measurements were in good agreement, the reflection is still too high for high performance amplifier systems
47
when the possibility of further deterioration from bonding wires and other parasitic effects is taken into account. The next section describes an optimization process to design a system with the best overall performance.
Synthesis of Optimized Tapers
In a 32-tray combiner the input impedance of each circuit tray is 32 times the characteristic impedance of the flared coaxial waveguide, which was chosen to be 30 Ohms. Accordingly, at the waveguide opening end of the finline taper, the terminal impedance is 480 Ohms. At the other end of the finline taper, we must connect to a 50 Ohm MMIC amplifier, which sets the target gap size. The design challenge is therefore to realize a broadband 9.6:1 impedance transformation to couple energy from the coax into a set of 50 Ohm MMIC amplifiers.
Ha l f ga p w i dth, mm
2.5 2 1.5 1 0.5 0
5
10
15
20
Position along taper, mm
Figure 2.19 Normalized gap width vs. location along the optimal tapered slotline for a 4-18GHz, 32-tray system of finlines on 10-mil AlN.
The design problem is analogous to the synthesis of tapered transmission-line impedance transformers. We have previously reported the iterative procedure in Section 2.1 for computing the taper shape of a non-TEM transmission line. This 48
method yields the shortest transformer for a specified cutoff frequency and return loss. After replacing the non-TEM parameters with frequency independent ones, we use the same procedure and the SDM results of Figure 2.15 to synthesize an optimized taper for a 4-18GHz, 32-tray system with a specified return loss of -15dB. The synthesis result is shown in Figure 2.19. 0 -5 S11_HFSS S11_SDM
-10 -15 -20 -25 -30 -35 -40
0
5
10 Freqency [GHz]
15
Figure 2.20 Reflection coefficient comparison between SDM and Agilent HFSS for the optimal taper design of Figure 2.19.
The analytical simulation results using the theory of small reflection are shown in Figure 2.20, confirming with our design criteria. Once the taper shape is known, the frequency response can be computed using an EM simulator such as HFSS. The HFSS result is also shown in Figure 2.20 with good agreement with analytical results. It should be stressed again that the sophisticated EM simulator provides an important check on the validity of the synthesis procedure, and can help fine-tune a design once
49
a near-optimal solution has been obtained by more computationally efficient means as described in [13]. Higher modes are investigated with HFSS. The results shown in Figure 2.21 reveal that they are suppressed effectively by the dense finline array. We observed from HFSS simulation that the next higher mode excited in the unit cell is more than 10 dB lower in magnitude than the dominant mode, and the third higher mode is 30 dB lower. The size of the waveguide opening and spacing between the trays play a key role in keeping the higher modes low. Those dimensions should be treated carefully when the design is modified. 0 Second and Third Higher Modes
-15
S11_2nd [dB] S11_3rd [dB]
-30
-45
-60
0
6 12 Freqency [GHz]
18
Figure 2.21 Higher modes inside the waveguide cell.
The design procedures elaborated in this chapter form the basis of the finline array design we used for both medium power and high power amplifiers using coaxial waveguide combiners. Experimental coaxial waveguide power combiner results will
50
be shown in following chapters. The theoretical analysis in this chapter enables the design of high performance waveguide combiners, while giving us the flexibility to optimize the system for different dimensions and also for different tray configurations.
References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13.
Pozar, D.M., Microwave Engineering. 2nd Ed. 1998, New York, NY: John Wiley & Sons. Klopfenstein, R.W., A Transmission-Line Taper of Improved Design. Proc. IRE, 1956. 442: p. 31-35. Schieblich, C., J.K. Piotrowski, and J.H. Hinken. Synthesis of optimum finline tapers using dispersion formulas for arbitrary slot widths and locations. 1984. Verver, C.J. and W.J.R. Hoefer. Quarter-wave matching of waveguide-to-finline transitions. 1984. Schmidt, L.P. and T. Itoh, Spectral domain analysis of dominant and higher order modes in fin-lines. IEEE Transactions on Microwave Theory and Techniques, 1980. MIT-28(9): p. 981-5. Itoh, T., Spectral domain immitance approach for dispersion characteristics of generalized printed transmission lines. IEEE Transactions on Microwave Theory and Techniques, 1980. MTT-28(7): p. 733-6. Alexanian, A. and R.A. York. Broadband waveguide-based spatial combiners. 1997. Rutledge, D.B., et al., Failures in power-combining arrays. IEEE Transactions on Microwave Theory and Techniques, 1999. 47(7, pt.1): p. 1077-82. Nai-Shuo, C., et al., 40-W CW broad-band spatial power combiner using dense finline arrays. IEEE Transactions on Microwave Theory and Techniques, 1999. 47(7, pt.1): p. 1070-6. Nai-Shuo, C., et al. A 120-W X-band spatially combined solid-state amplifier. 1999. Vale, C.A.W. and P. Meyer. Designing high-performance finline tapers with vector-based optimization. 1999. Pengcheng, J., et al. Analysis of a passive spatial combiner using tapered slotline array in oversized coaxial waveguide. 2000. Pengcheng, J., et al., Design of waveguide finline arrays for spatial power combining. IEEE Transactions on Microwave Theory and Techniques, 2001. 49(4, pt.1): p. 609-14.
51
3.
Broadband Medium Power Amplifier Using the Coaxial Waveguide Combiner
CHAPTER 3 Broadband Medium Power Amplifier Using the Coaxial Waveguide Combiner The finline array is the most important element for amplifiers using coaxial waveguide combiner. In previous chapter, a finline array for a 2-inch opening coaxial waveguide is synthesized following the design of coaxial waveguide transitions. All the parameters are optimized with small reflection theory and verified with HFSS. In this chapter, a passive waveguide housing and a finline array are fabricated and tested. The unloaded waveguide system is measured first. Then the optimized finline array is tested with 50-Ohm termination and through line. Inexpensive, low-noise traveling-wave amplifier (TWA) MMIC amplifiers were chosen as a demonstration vehicle. The broadband amplifier using the coaxial waveguide combiner reproduced the individual MMIC amplifier’s frequency response from 3.5-14 GHz, with a 75% combining efficiency[1].
52
3.1
Passive Combiner Measurements 6”
1.77”
2” Type N Connector
Waveguide Taper
6”
1.2”
Center Section
Waveguide Taper
Figure 3.1 Cross section of the unloaded waveguide structure.
The flared coaxial waveguide section was chosen to be 2 inch in the outer diameter of the opening and 1.2 inch in the inner diameter, corresponding to a 30 Ohm line[2]. A bulk center coaxial section was first used for waveguide characterization. The center coax is 1.77-inch long with a pair of 6-inch long coaxial waveguide taper connected on both sides. The waveguide structure was first assembled without finline array loaded inside. Metal and connector loss is shown as the S21 curve in Figure 3.2. Strong mismatch in S parameter can be found at 15 GHz that is caused by higher modes. Besides the TEM mode, waveguide modes can be excited in the oversized center section when the imperfect connections lead to discontinuities. These higher modes will be suppressed when the finline array is loaded. The finline circuitry will divide the waveguide into much smaller waveguide cells that can cut off waveguide modes.
53
0
0
S21 MAG [dB]
S11 MAG [dB]
-10
-10
-20
-20
-30
-30
-40
-40 0
2
4
6
8
10
12
14
16
Frequency [GHz]
Figure 3.2 S Parameter of the unloaded waveguide. 2.0“
0.6" 1.0"
1.77"
`
`
11.25 Degree
Figure 3.3 Mechanical drawing of circuit tray.
After testing the waveguide housing, we begin the process to fabricate circuit trays. We follow the 32-tray scheme developed in last chapter. The Metal tray, which is 1/32 of the center oversized waveguide section, is machined as shown in Figure
54
3.3. The tray is the carrier of the finline antenna and the MMIC amplifier. The finline antenna is realized on a ceramic substrate, and rests over a notched opening in the wedge-shaped metal tray, providing broadband impedance match from the coaxial waveguide to the MMIC amplifiers. A single tray with antenna and MMICs is illustrated in Figure 3.4. When the trays with antenna and amplifiers are stacked radially, the metal trays form a coaxial waveguide aperture populated with the finline tapers. The metal trays are clamped together, then connected with the coaxial waveguide tapers. The radius of the tray is 2 inches, although the radius of the effective opening is only 1 inch. MMIC Amplifier Tapered Slotline Antenna
Figure 3.4 Tray design for the modular coaxial combiner system.
As shown in Figure 3.3, notches are machined on the tray in the direction of wave propagation for accommodation of the circuit substrate. When stacked together, the notches will form slots on the waveguide walls. While the current for dominant mode is along the propagation direction of the wave, currents for higher modes are along the transverse direction. Those slots have no effect on the dominant mode, but can help to suppress the higher modes.
55
50 Ohm Single Wrap Resistor
Figure 3.5 Finline circuit card with 50 Ohm termination.
The optimized finline antennas are first tested using an finline array terminated with 50 Ohm resistors. Figure 3.5 shows a single finline circuit card that has 2 finline tapers onside. These finline tapers were fabricated on 10mil AlN substrates with 3µm gold metalization. Single wrap chip resistors were wire-bonded to the end of each finline taper. Those resistors have the smallest size, 30 mil by 20 mil, and have a ground plane that connects with one side of the resistor. The ground of the resistors is epoxyed to the ground plane of circuit. By close placement of the resistors with the pick & place tool, the length of the bonding wires can be minimized. Figure 3.6 demonstrates the reflection coefficient measurement results for 32-tray and 16-tray systems, which have 32 finline circuit cards and 16 finline circuit cards respectively. The measurement results show good qualitative agreement with the theoretical calculations in Chapter 2, although the maximum reflection coefficient is higher to some extend in the passband (~ -10dB). There is evidence of mismatch at the interface to the type-N connector, leading to the rapid undulations in the frequency response. This mismatch can be attributed to poor electrical connection between the type-N connector and the center-conductor of the coaxial waveguide taper. Meanwhile, the bonding wires connecting with resistors also contribute to the mismatch. Figure 3.7 shows the field distribution from HFSS for a finline circuit with 50-Ohm terminations. Bonding wires for resistors are included in the simulation. The
56
light spots represent peaks of the standing wave caused by mismatch at the termination. Obviously, the mismatch was mainly caused by the parasitic effect of the bonding wires. 0
S11 _16 Tray [dB] S11_32 Tray [dB]
-4
-8
-12
-16
-20
0
6 12 Frequency [GHz]
18
Figure 3.6 Return loss measurement for 16-tray and 32-tray combiner with 50 Ohm terminations on the finline circuits.
50Ohm Resisstor Bonding wires
Figure 3.7 Field distribution of waveguide cell with 50 Ohm resistor and bonding wire integrated.
Low combining losses are required to maintain good combining efficiency. A set of back-to-back finline cards are fabricated, with a 50-Ohm microstrip through line bonded in place of the MMIC amplifier to connect the input and output antennas as
57
shown in Figure 3.8. The commercially available MMICs all use microstrip line, so we use a microstrip though line with same dimension as the dummy amplifier. The measurement of finline array with through lines will be effective to estimate the loss of a combiner with MMIC integrated. The microstrip through lines are soldered to the surface of circuit card by eutectic alloy, then bonded to the end of the finline taper by 1 mil gold wires, which are around 300 µm in length.
Figure 3.8 Finline circuit card with through line connected in the middle. 0
-2 Loss_32 Tray [dB] Loss_16 Tray [dB]
-4
-6
-8
0
6 12 Freqency [GHz]
18
Figure 3.9 Dissipative loss for 16-tray and 32-tray combiners with 50 Ohm microstrip through-line in place of the active device. Inset shows circuit configuration.
Figure 3.9 shows the measured dissipative loss for 16 and 32-tray combiners using the 50-Ohm microstrip through line. The loss increases approximately with
58
f as expected. Again, we observed mismatch from the type-N connectors that leads
to small periodic dips in the loss. Higher-order modes are also observed at the higher frequencies and lead to dips in the S21 curve. However, those dips are still within the acceptable range and less of a concern comparing with the broad bandwidth of the system. Although the S21 measurement result is not the simple sum of the loss of input and output array, it allows us to roughly quantify the output array’s combining loss as half of the measured total loss.
Thus, the combining loss varies from
approximately 0.3dB at 4GHz to 1dB at 18GHz.
This translates to a potential
combining efficiency in excess of 75% over the entire band. 5
S11_16 Tray[dB] S11_32 Tray [dB]
0
-5
-10
-15
-20
0
6
12
18
Frequency [GHz]
Figure 3.10 Return loss for 16-tray and 32-tray combiners with 50 Ohm microstrip through-line in place of the active device.
Return loss measurement is shown in Figure 3.10. We observe stronger reflection than the measurement with 50-Ohm terminators. When we use through lines to connect input and output finline tapers, reflection from both the input and output
59
finline transition will add together at the input port. That will lead to about 3 dB increases in the reflection coefficient. It is important to note that the dissipative loss stays constant when we double the number of elements in the waveguide finline array, consistent with the observation in last chapter and [3] that dissipative loss is approximately independent of the number of finline transitions used. This characteristic makes the coaxial combiner promising for combing the power from large number of active devices. From the comparison of 16-tray and 32-tray configuration, we can see the 2 schemes have very similar results. The finline array is designed for 32-tray configuration; but when we only load 16-tray inside, the impedance transformation ratio for each finline taper is reduced from 9.6:1 to 4.8:1. Simulation with HFSS shows that 16-tray scheme has a little better impedance matching over 32-tray scheme. However, the parasitic effect of the bonding wires and other discontinuity overrules the small differences. Since the difference is smaller, we choose 16-tray configuration in the demonstration of the active amplifier.
3.2
Performance of the Active Combiner
For final testing of the combiner system, we use a set of 32 broadband MMIC amplifiers to build an active broadband amplifier. In this section, we use the term combiner referring to the broadband amplifier using the coaxial waveguide combiner. The Triquint TGA8349 TWA MMIC amplifiers are selected as the demonstration vehicles. Typical input SWR of TGA8349 is 1.2:1, and output SWR is 1.3:1. This 60
MMIC can generate 16dBm output power at 1 dB compression point. The small signal bandwidth of this MMIC amplifier is from DC to 14 GHz.
Bias Pad
Bias Capacitor
Input Taper
MMIC
Output Taper
Figure 3.11 The circuit tray with MMIC amplifiers.
A circuit tray with antennas and MMIC amplifiers is shown in Figure 3.11. Each tray carries a finline circuit card with 2 MMIC amplifiers that are soldered to the surface of the AlN circuit card. Since AlN has very high thermal conductivity, the heat generated by the low power MMIC amplifier can be effectively dissipated to the waveguide through the AlN board. A total of 32 MMIC amplifiers are integrated into the 16-tray system. The biasing pads are epoxyed at the side of the circuit tray. The dual-gate GaAs FET MMIC amplifier needs 4 bias voltages for gate, drain, second gate control and ground separately. Single layer capacitors are placed for filtering AC signals on DC bias line. Bonding wires are used for the DC and RF connection. Multiple wires are placed for RF connections to reduce parasitic inductance. The 10 mil thick AlN substrate sits on the notch of the aluminum circuit tray, carrying all of the finline tapers, MMIC amplifiers and bias capacitors. 61
Figure 3.12 Side view of the loaded section.
Input Waveguide Transition
Loaded Section Output Waveguide Transition
Bias Lines Circuit Tray
Figure 3.13 The overview of the active combiner.
An open view of the loaded center section is shown in Figure 3.12. The coaxial waveguide opening is formed when we stack all the circuit trays together.
Figure
3.13 shows the completely assembled combiner system including broadband coaxial tapers for feeding the loaded section. Bias lines connect the pins on the circuit tray to a bias board. 8-channel KEPCO power supply provides drain voltage and current. Each channel drives 2 circuit trays that have 4 MMIC amplifiers in total. The
62
separation of DC power supply maintains isolation between the bias lines that will help to protect other circuits when some MMIC amplifiers fail. 20
20
S21_Combiner [dB]
S21_MMIC [dB]
15
15
10
10
5
5
0
0 2
4
6
8 10 12 Frequency [GHz]
14
16
Figure 3.14 Measured small-signal gain of the active combiner and individual MMIC amplifier.
The system is very stable and no additional circuits or biasing capacitors are needed to suppress oscillations. Figure 3.14 and Figure 3.15 show the small-signal gain and reflection coefficient of the completed active combiner system. The broadband property is verified by both of the figures. The combiner has 10 to 11.5 dB gain over a broadband from 3.5 GHz to 14GHz. The passive combiner itself has a lower cut-off frequency. It has potential to operate up to 18 GHz. The upper end of the bandwidth of the combiner is limited by the frequency response of the MMIC. The loss of the system is consistent with the loss of the passive combiner. The total loss of the combiner including ohmic and mismatch losses is nearly a constant of 2 dB over the entire band.
This corresponds to 1dB output loss and hence 75%
combining efficiency. 63
0
0 S11 [dB]
S22 [dB]
-5
-5
-10
-10
-15
-15
-20
-20 5
10 Frequency [GHz]
15
Figure 3.15 Reflection coefficient of both input and output ports for the active combiner. 35
20 Gain [dB]
Output Power [dBm]
30
15
25
10
20
5
15
0 5
10
15 Input Power [dBm]
20
25
Figure 3.16 Output power and gain vs. input power.
Large-signal measurements were also recorded, using a TWT amplifier to drive the array into compression. Two directional couplers were connected at both the input and output port. Power sensors are connected to the couplers to measure the input and output power. The loss of this configuration is calibrated first. Then power
64
measurements are carried out at 10 GHz. Measurement setup will be introduced in more details in chapter 5. As shown in Figure 3.16, the 16-tray combiner system generated 1-Watt CW power at the 1 dB compression point. Using a nominal output power of 16dBm for each MMIC amplifier, this translates to a measured combing efficiency of 80% at this frequency, in good agreement with previous estimates based on system loss. The power-frequency response was characterized from 4 to 15 GHz using a fixed input power of 20 dBm. The result is shown in Figure 3.17. 29 to 30 dBm output power is reached in most of the band. It also follows the expected output power curve of the MMIC amplifier, indicating that the higher stopband of the combiner is determined by the MMIC amplifier. 35
15
30
10
25
5
20
0
-5
15 2
4
6
8 10 12 Frequency [GHz]
14
16
Figure 3.17 Output power and gain vs. frequency.
The 32-MMIC combiner, which has 16 circuit cards, has been demonstrated and 1-Watt output power is achieved. It verifies that the optimized broadband transition
65
designs based on SDM analysis are effective for large-scale power combining inside coaxial waveguide. The combiner system has a capacity of integrating as many as 64 MMIC power amplifiers. A good impedance match is achieved from 3.5 GHz up to 18 GHz. More characteristics such as residue phase noise will be covered in chapter 5. We will continue our work to optimize the system and demonstrate a higher power module in the next chapter.
References 1. 2. 3.
Pengcheng, J., et al., Multioctave spatial power combining in oversized coaxial waveguide. IEEE Transactions on Microwave Theory and Techniques, 2002. 50(5): p. 1355-60. Alexanian, A., Planar and Distributed Spatial Powe Combiners. ECE Technical Report, 1997. #97-20. Pengcheng, J., et al., Design of waveguide finline arrays for spatial power combining. IEEE Transactions on Microwave Theory and Techniques, 2001. 49(4, pt.1): p. 609-14.
66
4.
Design of High Power Amplifier Using the Coaxial Waveguide Combiner
CHAPTER 4 Design of High Power Amplifier Using the Compact Coaxial Waveguide Combiner In this chapter we report a high power amplifier design using enhanced broadband passive combiner structure. A significant reduction in size has been achieved while maintaining a 6-18GHz bandwidth and capacity for 32 MMIC amplifiers. A broadband slotline to microstrip line transition was developed and monolithically integrated with the finline antennas, to eliminate a troublesome bond-wire transition in earlier design and provide better compatibility with commercial MMIC amplifiers. The Spectral Domain Method (SDM) is applied to compute the field in the structure, and small reflection theory is applied again to synthesize the waveguide taper and optimize finline taper array.
67
4.1
Motivation
As shown in Figure 4.1, our previous design is undesirably large in both the diameter of the center section and the length of waveguide taper. A new compact combiner structure, which provides the same combining performance as the larger one, was developed for the high power module. N Connector
Center Section Previous Design
Waveguide Taper
New Design
Figure 4.1 Comparison between previous design and the new compact design.
For low power demonstrations, the MMIC amplifiers were die-attached to the AlN substrate on which the finline antennas were processed. However, junction thermal resistance between AlN substrate and metal carrier will lead to a high temperature step when the MMIC amplifiers generate a large amount of heat. In the high power module, MMIC amplifiers will need to be attached to the metal carrier for better thermal management. In Chapter 3, bonding wires were used to connect from MMIC amplifier’s input/output pads to the end of the finline. Since the MMIC amplifiers sit on the ground of the finline circuit, the bonding pads of MMIC amplifier are 100 um (MMIC thickness) higher than those on the finline circuit. The bonding wires have to go down a slope for connection. It increases the difficulty for
68
assembly, also leads to comparably longer boding wires. The parasitic inductance of the wire bonding deteriorates the impedance matching. Therefore it is necessary to keep the connection pads of the antennas and the commercial MMIC amplifiers in the same plane and minimize the length of bonding wires. A monolithic slotline to microstrip line transition was developed to serve this purpose. The new transition has a broad bandwidth and doesn’t limit the overall frequency response of the system. If needed, the transition design can be easily extended to higher frequencies. The new combiner’s passband was changed to 6-18 GHz since the broadband MMIC amplifiers we are going to use work exactly at this band. The waveguide housing and the finline antenna array were synthesized again to meet the changes on the bandwidth and dimensions. Same performances as the original design were maintained through the optimization process.
4.2
Coaxial Waveguide Design
To re-engineer the combiner and minimize the physical size of coaxial waveguide, we applied the small reflection theory to the coaxial waveguide taper again. We succeeded in reducing the diameter of the center section from 4 inches to 2.2 inches, and the length of each waveguide taper from 6.2 inches to 2.2 inches. The new compact waveguide housing were machined and tested. The simulation and measurement results of the unloaded coaxial waveguide are shown in Figure 4.2. Similar impedance matching and loss as the original waveguide housing are achieved for the new design. One improvement is that we don’t observe the same spike at 15 69
GHz as the previous result in chapter 3. The reduction in waveguide size helps to move the higher modes out of the 6-18 GHz band. Two N-to-SMA connectors were used in measurement, which is not considered in the simulation. The loss and reflection of the N-to-SMA connector are also shown in Figure 4.2 and contribute partly to the discrepancy between simulation and measurement. 0
0
-10
-2
-20
-4
-30
-6
-40
System S11 [dB] Simulation S11 [dB] Connector S11 [dB]
System S21 [dB] Simulation S21 [dB] Connector S21 [dB]
-8
-10
-50 0
5
10
15
Freqency [GHz]
Figure 4.2 S parameter of unloaded coaxial waveguide.
4.3
Synthesis of Waveguide Finline Array
Figure 4.3 Layout of the finline taper and the slotline to microstrip line transition.
70
Sixteen finline circuit cards were radially loaded inside the center waveguide opening. The symmetry and the thinness of the circuit substrate allow us to focus analytical attention on a unit cell. The unit cell was further approximated by a parallel-plate waveguide as described in chapter 3 and [1]. The Spectral Domain Method and the small reflection theory were used again to optimize the finline taper. The layout of the new finline taper is shown in Figure 4.3. Each circuit tray carries 2 finline tapers. To improve the linearity of the combiner, power should be distributed evenly to each taper. However, the field inside the waveguide is not radially uniform. So each of the finline taper on a single tray is designed with a different slot opening to equalize the power. When we put the finline array inside the waveguide, they will have the same outer radius to inner radius ratio.
4.4
Slotline to Microstrip Line Transition
In Nick Cheng’s design, a separate microstrip line transition was integrated to provide connection from the end of finline to MMICs. This approach needs a separate microstrip line circuitry and carefully bonding of the circuitry to finline substrate. In our previous design discussed in chapter 3, we attached MMIC to the finline circuit and directly connected the MMIC to the end of finline. This approach simplifies the circuitry; but since the connection pads are not in the same plane, it increases the difficulty for wire bonding. Moreover, this design is not effective for high power MMICs that will generate a lot of heat. To simplify the assembly and reduce parasitic
71
inductance, a monolithic slotline to microstrip transition was employed in the new design [2]. As shown in Figure 4.3, the finline taper was processed on the back of the AlN substrate, with a 90-degree slotline short stub at the end. A 90-degree microstrip line open stub is aligned to the slotline stub on the top of the substrate. The center of the 2 stubs are on the same line perpendicular to the surface, and their edges are parallel to each other. When put onto a metal carrier, the slotline becomes the ground of the microstrip line, which is in the same plane as the ground of the MMIC amplifiers. Due to the space limitation inside the compact structure, the stubs have to be bent 15degree inwards, and the microstrip line detours around the slotline stub in a small loop.
Zs
Zs
Zm
1:n jXom
Zm
jXss Ys
Ys Γin
jBs
jBm
Gm
Figure 4.4 Circuit model of slotline to microstrip line transition.
The transition in Figure 4.3 is modeled in Figure 4.4. The short slotline stub and open microstrip stub can be treated as a series of straight sections with various widths that are cascaded together [3]. To improve the accuracy in modeling the structure
72
enclosed in waveguide, we used 3D simulator Agilent HFSS to compute the reactance of the slotline stub jXss and microstrip stub jXom. Then we applied the values into the circuit model and optimized other parameters in the transition. In the circuit model, Zm and Zs is the characteristic impedance of microstrip line and slotline respectively, and n is the transformer ratio, n=−
1 b2 b Ey ( h) dy , Vo ∫− 2
Ey (h) = −
Gm =
2π u 2π u Vo (cos h − cot qo sin h), b λo λo
n 2 Zm , Zm 2 + Xom 2
Bm = −
n 2 Xom , Zm 2 + Xom 2
Ys =
1 , Zs
Bs = −
(4.1)
1 . Xss
Here, Vo is the voltage across the slot and Ey(h) is the electric field of the slotline on the other surface of the substrate. The details of the calculation of n are given in [4]. The reflection coefficient can be expressed as: Γin =
Ys − Gm − j ( Bs + Bm) . Ys + Gm + j ( Bs + Bm)
(4.2)
Our goal is to achieve bandwidth from 6 to 18GHz. Simulation shows that the lower band is more sensitive to the dimensions. So we chose 10GHz as the center frequency, then optimized the transition at this frequency to satisfy Ys=Gm, and Bs=Bm.
73
The impedence of the microstrip line is fixed to be 50 Ohm, corresponding to a 278 um strip width on a 254 um thick AlN substrate. The characteristic impedance of the slotline times n2 should be close to 50 Ohm. We chose the width of the slotline to be 40 um to match Ys with Gm. Due to the limitation of space inside the waveguide structure; the radius of the slotline stub was selected to be 2000 um. To realize Bs= Bm and minimize the reflection coefficient, the microstrip open stub should have a radius of 1500 um. Further optimization with Agilent HFSS showed that a microstrip stub with a 1600 um radius has a wider bandwidth. Simulation results shown in Figure 4.5, indicates that the slotline to microstrip transition can achieve a bandwidth of more than 12 GHz, from 6 to 18 GHz. If scaled down properly, the transition can also work at higher bands. The parasitic effect is much smaller than the bonding wire connection used in earlier work.
Figure 4.5 S parameters of slotline to microstrip line transition from Agilent HFSS simulation.
74
4.5
Compact Passive Structure of Coaxial Waveguide Combiner
(a)
(b) Figure 4.6 (a) Open view of the passive coaxial waveguide combiner, (b) circuit card with back-to-back finline antenna and transition to microstip line.
Figure 4.6(a) shows the open view of the combiner loaded with 16 circuit cards for loss measurement. Each circuit card has 2 transitions placed back to back as shown in Figure 4.6(b). The microstrip line dummy circuit used in last chapter was replaced by straight microstrip line since the integration of the new slotline to microstrip line transition. Slots are machined along the walls of the center flared coaxial waveguide, and circuit cards are slid into it as shown in Figure 4.6(a). The performance of the overall passive structure, which includes connectors, waveguide tapers, a divider and a combiner, is shown in Figure 4.7. 6-18 GHz bandwidth is observed from both the simulation and the measurement. The majority of the
75
discrepancy comes from the connectors, which introduce more than 1 dB of loss at the higher band.
Figure 4.7 Comparisons between simulation and measurement for passive coaxial waveguide combiner.
4.6
Leakage from Output to Input
When the MMIC amplifiers are integrated, metal trays similar to the one introduced in last chapter will be machined except that the size will be much smaller. The trays will be the carriers for antennas and MMIC amplifiers. When the wedge shaped metal trays are stacked together, the waveguide is partially blocked by the bridges that connect the inner and outer part of the metal trays. The dominant mode cannot propagate along the waveguide except through the finline circuits. While the metal bridges leave some room for MMIC amplifiers when stacked together, they will also provide passages for some of the higher waveguide modes. Since the circuit card
76
shown in Figure 4.6(b) can pass the energy through it, it is not viable to valuate the leakage of the waveguide from its S parameter measurement. Here we use the previous measurement results of the bigger combiner, which is loaded with a finline array that is terminated with 50-Ohm resistors. The measured S12 of that combiner is equivalent to the leakage from the output to the input through waveguide modes. From Figure 4.8 we can see when the waveguide is less loaded with only 16 tray compared to 32 tray, the leakage through higher modes is more severe. At higher frequencies, the 16-tray scheme has more leakage. The reason is the waveguide cell is larger for less loaded combiner and its size is not small enough to cut-off all the higher modes when frequency is higher than 15 GHz. The compact version introduced in this chapter will help to increase the cut off frequency of a waveguide cell to 18 GHz. The in band (6-18GHz) isolation can be improvement for the compact design. 0 -10
S12_16 Tray [dB] S12_32 Tray [dB]
-20 -30 -40 -50 -60 0
5
10 15 Frequency [GHz]
Figure 4.8 Waveguide leakage from output to input.
77
20
Since we use very high gain MMIC amplifiers, high leakage from the output to the input will cause instability problems when the output is not well matched. To account for this, we put Emerson EM absorbent at the bottom of the metal tray’s bridge. When the trays are stacked together, the absorbent will be right on the top of the MMIC and can increase the attenuation through that passage. Figure 4.9 shows a flipped circuit tray and the place to put the absorbent.
Apply thin layer of EM absorbent
.05" 0.36"
0.44"
Figure 4.9 Thin layer of EM absorbent on the back of the tray.
4.7
Uniformity
A waveguide cell (1/16th of the coaxial waveguide) and its electrical field distribution along the finline are shown on the left side of Figure 4.10. Since the field inside the coaxial waveguide is radially distributed, the finline pair is designed with the same outer radius to inner radius ratio to keep the impedance the same.
78
Mode1 Out-of-Phase
Mode2 In-Phase
Figure 4.10 In-phase and out-of-phase modes at output port (the microstrip line end)
0 -5 -10
S21_Mode2 [dB] S21_Mode1 [dB]
-15 -20 -25 -30
4
6
8
10
12
14
16
18
20
Frequency [GHz] Figure 4.11 Strength of mode 1 and mode 2 at output port
The electrical filed distribution at the output port (microstrip end) is shown on the right side of Figure 4.10. In this design, each microstrip line couples energy from the complimentary side of each finline. The microstrip lines at output port will output
79
out-of-phase signals with the same strength that is mode 1 at the output port. But the signal strength at the two lines can’t be identical as ideally because of coupling between the two transitions. The effect can be as attributed to the combination of the out-of-phase mode (mode 1) and the in-phase mode (mode 2). Figure 4.11 shows the amplitude of the 2 modes. Result shows that mode 1 is the dominant mode, and mode 2 is not neglectable. The combination of 2 modes will lead to different amplitude and phase at the microstrip lines. Let vs1 and vs 2 represent the signals at each line, their amplitude and phase are vs1 = Am1 cos ω t + Am 2 cos(ω t + ∆θ m ) | vs1 |= ( Am1 + Am 2 cos ∆θ m ) 2 + ( Am 2 sin ∆θ m ) 2
φs1 = arcsin
(4.3)
Am 2 sin ∆θ m | vs1 |
vs 2 = Am1 cos ω t − Am 2 cos(ω t + ∆θ m ) | vs 2 |= ( Am1 − Am 2 cos ∆θ m ) 2 + ( Am 2 sin ∆θ m ) 2
φs 2 = arcsin
(4.4)
Am 2 sin ∆θ m | vs 2 |
where Am1 & Am2 are the amplitude of mode 1 & 2 respectively, and ∆θ m is the phase difference of mode 1 & 2. As shown in Figure 4.12, the undesired in-phase mode (mode 2) causes a maximum of 8% amplitude difference and maximum of 10-degree phase error
80
between the inner and outer finline. Hence, the 16 inner channels will have the same phase and amplitude difference as the 16 outer channels. 1.3
200 Slot1_Amp Slot2_Amp
1.2
Phase Difference [degree]
180
1.1
160
1
140
0.9
120
0.8
4
6
8
10
12
14
16
18
100 20
Frequency [GHz] Figure 4.12 Phase and amplitude difference at the 2 microstrip line.
Figure 4.13 Generic combiner system.
The variation in phase and amplitude causes imperfect summation and leads to reduction in combining efficiency. We need to calculate the output power in the form
81
of the sum of N channel signals to quantitatively analyze the change in efficiency due to non-uniformity. We assume the input signal is Ain = A cos(ω t )
(4.5).
Considering the phase and gain variation in each channel of the splitter, then the input and output signal of each amplifier will respectively be A(1 + δ Ai ) cos(ω t + δφi ) N A(1 + δ Ai )G (1 + δ Gi ) = cos(ω t + δφi + δϕ i ) N
ain ,i = bout ,i
(4.6)
where G is the nominal voltage gain, δ Ai and δφi are the amplitude and phase variation of each channel in the splitter, δ Gi and δϕ i are the gain and phase errors of each amplifier. If we assume the combiner has the same characteristic as the splitter, the sum of the N channel signals can be expressed in phasor form as 1 N AG N bout ,i = ri (1 + δ Ai ) 2 (1 + δ Gi ) cos(ω t + 2δφi + δϕ i ) ∑ ∑ N i =1 N i =1 N AG ri (1 + δ Ai ) 2 (1 + δ Gi )e j (2δφi +δϕi )e jω t = ∑ N i =1
Bout =
(4.7)
where we assume the combiner has the same phase and amplitude variation as the splitter. Here we also considered the statistical failure of the MMIC that is represented by ri .
82
The details of the analysis of phase and gain error introduced by amplifiers are covered in [5]. Here we add the effect of the phase and amplitude variation introduced by the splitter and combiner. 2
The output power is proportional to P = Bout . If we denote the “no-error” output power as Po = ( AG ) 2 , then the relative change in the presence of error is P 1 = 2 Po N
N
N
∑∑ r r (1 + δ A ) (1 + δ A ) (1 + δ G )(1 + δ G )e i =1 j =1
2
i j
i
2
j
i
j [2(δφi −δφ j ) + (δϕi −δϕ j )]
j
(4.8).
If we neglect the phase and amplitude error from the combiner and splitter, after taking the ensemble average and assuming the individual amplitude and phase error of each amplifier have the same variance, we have 2 2 2 P 1 2 − δϕ 2 − δϕ 2 − δϕ ] ≈ Pe e = Pe e + [ Pe (1 + δ G 2 ) − Pe e Po N
(4.9)
where Pe is the survival rate of devices and Pe = r . If we only consider the phase and amplitude from the combiner and splitter and neglect the error from amplifiers, the N channels will be divided into 2 groups, the inner channel group and the outer channel group. N/2 inner channels have the same amplitude and phase, and we assume the amplitude and phase error is δ A1 and δφ1 . Here we assume N is an odd number. The N/2 outer channels also have the same amplitude and phase, and we assume the amplitude and phase errors are δ A2 and
83
δφ2 . We must notice that the inner group and outer group are intrinsically out of phase, so δφ2 is the phase error relative to 180 degree phase. Then we have P 1 = 2 Po N
N
N
∑∑ (1 + δ A ) (1 + δ A ) e 2
2
i
i =1 j =1
j [2(δφi −δφ j )]
j
1 = [(1 + δ A1 )4 + (1 + δ A2 )4 + 2(1 + δ A1 ) 2 (1 + δ A2 ) 2 cos 2δφ ] 4
(4.10)
where δφ = δφ2 − δφ1 . Since δφ2 is the phase error relative to 180 degree phase, δφ is also the phase error relative to 180 degree phase. Using the amplitude and phase error values from Figure 4.12, we can calculate the combining network’s efficiency with equation (4.10). The efficiency is expressed as the output power over “no-error” output power Po. Figure 4.14 shows the efficiency over the frequency band from 5 to18 GHz. 1 0.9 0.8
P/Po
0.7 0.6 0.5
4
6
8
10
12
14
16
18
20
Frequency [GHz]
Figure 4.14 Efficiency of the combining network.
We can see that the non-uniformity in the inner and outer channel leads to reduction in combining efficiency. Asymmetric the input array and the output array 84
can compensate for the amplitude and phase error. It will improve the uniformity related combining efficiency to 100%.
4.8
Fabrication Procedure
The copper carriers were machined by the UCSB engineering machine shop. An Electric Discharge Machine (EDM) was used to achieve accuracy of 3 mils or less. Cost and fabrication time will drop dramatically if we use a die-cast process. The metal trays were then plated with high purity gold using an electrolyte-plating process. The gold layer will protect the copper from oxidation and reduce RF loss.
Lower Right
Left Upper
Left Triquint MMIC AuSn CuMo Subcarrier AuGe
Cu Carrier (Gold plated)
Right
Figure 4.15 Assembly procedure.
The MMIC amplifiers were next mounted onto the metal trays with Cu/Mo subcarriers by a eutectic solder. The coefficient of thermal expansion of GaAs is 85
much different from that of Cu, which will cause problems in both assembly and lifetime if they are directly mounted together. To increase the reliability, a Cu/Mo subcarrier is necessary since it has similar thermal expansion coefficient as the GaAs substrate. A small cavity with same size as the Cu/Mo subcarrier was machined on the metal tray for better alignment. Since the GaAs MMIC is destroyed when the temperature is higher than 320 oC, the Cu/Mo subcarrier was bonded to the metal tray first with a Au/Ge eutectic solder at 360 oC. Then the GaAs MMIC amplifier was bonded to the Cu/Mo subcarrier with a Au/Sn eutectic solder at 280 oC. The procedure is shown in Figure 4.15. Some GaAs MMIC assembly guidelines are:
• AuSn (80/20) solder with limited exposure to temperatures at or above 300 oC • Use an alloy station or conveyor furnace with a reducing atmosphere • No fluxes should be utilized • Coefficient of thermal expansion matching is critical for long-term reliability • Storage in dry nitrogen atmosphere The component placement and adhesive attachment assembly notes are:
• Vacuum pencils and/or vacuum collets are the preferred method of pick up • Air bridges during placement should be avoided.
86
• Force impact is critical during auto placement • Curing should be done in a convection oven; proper exhaust is a safety concern • Microwave or radiant curing should not be used because of differential heating • Coefficient of thermal expansion matching is critical
4.9
Circuit Tray & Bias Line
Figure 4.16 Circuit tray of combiner.
The assembled circuit tray is shown in Figure 4.16. The 2-channel MMIC amplifier sits on the bridge that connects inner and outer sections. Input and output antennas were epoxyed on both sides. Bonding wires connect the end of microstrip line to the input and output pads of the MMIC amplifier. Bias pins were epoxyed at the outer side of the tray. DC currents are input to MMIC amplifiers through biasing lines and bonding wires. The overall DC impedance from pins to the MMIC amplifier’s pads is from 0.2 Ohm to 0.3 Ohm.
87
4.10
Efficiency, Reliability and Thermal Analysis
Amplifier Efficiency
Solid state power amplifiers (SSPA) are superior to tube amplifiers in size and scalability. However, a great challenge for SSPAs is efficiency. Three definitions of efficiency are commonly used. Drain efficiency is defined as the ratio of RF-output power to dc-input power, i.e., η = Po / PDC . Power added efficiency (PAE) incorporates the RF-drive power by subtracting it from the output power, i.e., η = ( Po − Pi ) / PDC . PAE gives a reasonable indication of power amplifier (PA) performance when gain is high; however, it can become negative for low gains. An overall efficiency such as η = Po /( PDC + Pi ) is usable in all situations. Class A, B, AB and C amplifiers are widely used in PA designs, but their drain efficiency only ranges from 50% to around 85% theoretically. Innovative class D, E, and F amplifiers can improve the drain efficiency up to unit ideally. Recent achievements from Raytheon and TRW have shown class-E amplifiers with a power added efficiency of over 60% using a pHEMT and DHBT respectively at X band[6, 7]. Although the new class-E amplifier design has shown a wide bandwidth over 1 GHz, it is still mandatory to use a class A amplifier for broader bandwidth applications. In the lossless situation, class A amplifiers have a drain efficiency of 50%. However, considering the lossy mechanism inside the devices and the matching circuit, the power added efficiency is only around 30%.
88
If the combiner has an output combining efficiency of 75%, the amplifier’s overall PAE is only a little more than 20%. It means 4 times the output power is wasted in the form of heat. When there is no input signal to the amplifier, 5 times the rated power is converted into heat. A 50-watt output power rated amplifier must have the ability to dissipate more than 250 watt of heat effectively. The pressure will be much alleviated if narrowband high efficiency class E amplifiers are integrated into the combiner. Another modification is to use Class B push pull amplifiers that will increase the efficiency decently while maintaining broad bandwidth.
Amplifier Reliability
The combiner system integrates a large quantity of MMIC amplifiers. Although GaN and SiC amplifiers have been demonstrated with promising performance in research labs, there are still no mature commercially available products for frequencies higher than C band in high power applications. GaAs is still the dominant material for MMIC power amplifiers. The reliability of GaAs devices is the key parameter for a high power combiner system. GaAs device reliability involves probability statistics, time, and a definition of failure. Given the failure criteria, the most direct way to determine reliability is to submit a large number of samples to actual use conditions and monitor their performance against the failure criteria over time. Since most applications require device life times of many years, this approach is unfeasible. To acquire reliability
89
data in a shorter time, an acceleration factor must be used to quicken the failure process. In most cases this acceleration factor is high temperature. The rationale behind high temperature life tests is that most physical and chemical processes are accelerated by temperature. The rate of acceleration for each failure mechanism is a constant called the activation energy. Most GaAs semiconductor failure mechanisms follow the Arrhenius equation that relates the rate of failure to temperature, time and activation energy. The Arrhenius Equation and Activation Energy is expressed as:
Tf 2 = Tf1 e
Ea ( 1 - 1 ) k T2 T1
(4.11)
where Tf = time to failure, Ea = activation energy in electron-volts (eV), k = 8.6142 E-5 (eV/°K), T = absolute temperature in Kelvin (°C +273).
Figure 4.17 Typical TriQuint Texas MESFET, HFET, and PHEMT median life time data.
90
To predict lifetimes at normal operating temperatures, multiple high temperature lifetime tests were performed by TriQuint Semiconductor. The median life from each of the tests is plotted as a point on an Arrhenius graph. All points are fit on a line. The slope of the line is the activation energy. Median life at any temperature can then be determined. Typically three temperatures are used. This data is available for MESFET, HFET, and PHEMTs under typical bias conditions. The failure criteria commonly used is 1 dB RF output power degradation. The ¼ um PHEMT process of TriQuint has shown a median lifetime of 2e6 hours at channel temperature of 140 oC with failure criteria being 1 dB degradation of RF output power. The TriQuint MMIC amplifiers used in our combiner system were strictly tested. In the test process, one and a half mil thick 80%Au/20%Sn eutectic solder was used to mount the MMIC amplifier on a 20 mil thick Cu/Mo Carrier, then placed on a hot plate at a temperature of 70 oC. The worst case would be that no is RF applied and 100% of the DC power is dissipated. Under the condition of Vd = 8 V, Id = 2.4 A, Pdiss = 19.2 W and channel temperature =145 oC, the lifetime is tested to be 1.6 E+6 hours, which is equal to 180 yrs.
Thermal Analysis
As shown in last section, the lifetime of the GaAs MMIC amplifier is over 180 years if the hot plate temperature is kept below 70 oC. If we want the solid state amplifier to last 50 years which is much longer than TWTA’s 15 year lifetime, the plate temperature should be no higher than 85 oC as indicated in Equation (4.11). If
91
the temperature of the combiner’s metal tray is controlled to be lower than 85 oC, the amplifier will be able to serve most communication systems. Generally, the ambient temperature is 25 oC. Heat transfer occurs as a result of the temperature gradient between the amplifiers and the environment. The heat generated by the MMIC amplifiers is dissipated into the air by 2 modes: conduction and convection. Conduction is the transmission of heat through a substance without perceptible motion of the substance itself. Heat is conducted by the copper tray to its outside surface. Convection is the term applied to heat transfer due to the bulk movement of a fluid. When a fan blows air over the outside surface of the carriers, the air absorbs heat from the carriers by convection. The one dimensional heat conduction equation is
qx = − kA
where qx is the heat flow in the x direction,
dT dx
(4.12)
dT is the temperature gradient or slope dx
of the temperature curve, A is the area normal to the heat flow direction and k is the thermal conductivity of the material. The numerical value of thermal conductivity is an indication of how fast heat is conducted through a material and is a macroscopic representation of all the molecular effects that contribute to the conduction of heat through a material. Based on equation
92
(4.12), measurements can be made to determine the thermal conductivity for various substances. As shown in Figure 4.18, copper is only inferior to silver in thermal conductivity at room temperature and is 1.6 times better than aluminum. So we chose copper as the material to be used for the metal carriers in the high power combiner instead of aluminum that was used in the medium power combiner design. The temperature gradient is reduced at the price of higher weight and cost. Since the high power system is mostly used in base stations, the performance is more important than the weight and price.
Figure 4.18 Variation of thermal temperature for various metals[8].
93
conductivity
with
A Copper/Molybdenum subcarrier is used between MMIC amplifiers and copper tray. It can help avoid expansion problems with temperature changes. To minimize thermal resistance from MMIC to the outside surface, a eutectic solder is used for die bonding instead of epoxy.
Thermal Simulation
In a structure similar to that shown in Figure 4.19, heat is transferred from the plate to the air. The mechanism of heat transfer at the wall is conduction because the fluid velocity at the wall is zero. However, the rate of heat transfer depends on the slope of the T vs. y curve at the wall---dT/dy at y=0. A steeper slope indicates a greater temperature difference and is highly dependent on the flow velocity. The heat transferred by convection is found to be proportional to the temperature difference. It is qc = hc A(Tw − T∞ )
(4.13)
in which hc is called the average convection heat transfer coefficient or the film conductance. This coefficient accounts for the overall effects embodied in the process of convection heat transfer. The overbar notation indicates that the film conductance defined in Equation (4.13) is an average that is conventionally assumed to be constant over the length of the plate. Typical values of hc are shown in Table 4.1.
94
Figure 4.19 Uniform air flow past a heated plate[8]. Table 4.1 Typical values of the convective heat-transfer coefficient hc for various fluids[8].
Fluid and condition Air in natural convection
hc
W/(m2 K) 5-25
Air in forced convection
30-300
Oil in forced convection
60-1,800
Water in forced convection
300-6,000
The wedge tray has a limited surface area due to its small radius. To help dissipate heat, fins were machined into the outside surface. The entire irregular configuration makes the heat transfer of the structure hard to be calculated analytically. As a result, a mechanical software package, SDRC’s Ideas 8.0, was chosen to simulate the heat transfer. The TMG (Thermal Management) and ESC (Electrical System Cooling) functions are well designed, making them good choices for our purpose. In the thermal model, we only simulated 1/16th of the waveguide structure due to the
95
symmetry. Since all the MMIC amplifiers generate the same amount of heat, there is no heat transfer between metal trays. Thus it is safe to assume an insulation layer on the interface between trays. The heat is conducted to the fins on the outside surface and convectively dissipated into the air. On each metal tray, the MMIC amplifier has two channels, each of which consumes 1.2A current at 8V DC bias. The worst case is analyzed when all the energy is dissipated in the form of heat and the total power consumed is around 20 watts. In the thermal model, the Cu/Mo subcarrier and eutectic solder layers are all included. Those parts are the same as those used in TriQuint’s hot plate lifetime test process. A 20-watt heat source is applied to the surface of the subcarrier. The metal tray’s material is copper with thermal conductivity of 400 W/(m K). Air flows through the fins and has an average heat transfer coefficient of 200 W/(m2 K). Simulations will show the temperature of the metal tray. The wedge shaped tray shown in Figure 4.20(a) has an outside surface area of 640 mm2. The outside surface temperature is 211 oC, while the temperature of the metal tray under the subcarrier is 220 oC. In Figure 4.20(b), 3 fins are added to the outside surface and increase the surface area to about 4 times of that shown in Figure 4.20(a). Then, it can be seen that the temperature drops dramatically, ranging from 82 oC under the subcarrier to 71 oC on the outside surface. Using the relationship between the hot plate temperature and lifetime, we conclude that the MMIC amplifiers can work over 50 years since the hot plate temperature is only 82 oC. 96
No thermal conduction on inner faces due to symmetry
Temperature Range: From 211 oC to 226 oC
Forced Air Convection on outer face (200 W/m2C2)
(a)
No thermal conduction on inner faces due to symmetry
Forced Air Convection on outer face(200 W/m2C)
Temperature Range: From 71 oC to 84 oC
(b) Figure 4.20 Simulation results with Ideas 8.0.
97
Another advantage of the combiner system is its graceful degradation. The MMIC amplifier’s failure means 1-dB degradation in output power. The amplifier degrades favorably and only reaches to 1-dB degradation after several MMIC amplifiers have failed, which will help to increase the lifetime of the amplifier greatly. The design related issues are well discussed in this chapter including waveguide and antenna design, fabrication, uniformity, efficiency, reliability and thermal analysis. The performance of the high power amplifier system will be covered in the next chapter.
References 1. 2. 3. 4. 5. 6. 7. 8.
Pengcheng, J., et al. Analysis of a passive spatial combiner using tapered slotline array in oversized coaxial waveguide. 2000. Zinieris, M.M., R. Sloan, and L.E. Davis, A broadband microstrip-to-slot-line transition. Microwave and Optical Technology Letters, 1998. 18(5): p. 339-42. Shuppert, B., Microstrip/slotline transitions: modeling and experimental investigation. IEEE Transactions on Microwave Theory and Techniques, 1988. 36(8): p. 1272-82. Gupta, K.C., R. Garg, and I.J. Bahl, Microstrip Lines and Slotlines. 1979: Artech House, Norwood, MA. York, R.A., Some considerations for optimal efficiency and low noise in large power combiners. IEEE Transactions on Microwave Theory and Techniques, 2001. 49(8): p. 147782. Tayrani, R. A monolithic X-band class-E power amplifier. 2001. Quach, T., et al. Ultra-efficient X-band and linear-efficient Ka-band power amplifiers using indium phosphide double heterojunction bipolar transistors. 2001. Janna, W.S., Engineering Heat Transfer. Second ed. 2000: CRC Press.
98
5.
Performance of High Power Amplifier Using Compact Coaxial Waveguide Combiner
CHAPTER 5 Performance of High Power Amplifier Using the Compact Coaxial Waveguide Combiner In the last chapter, we explained the design and fabrication procedure for the compact coaxial waveguide combiner for high power applications. In this chapter the measurement of the power and bandwidth performance of the combiner system will be presented. Since the motivation for the spatial power combiner is its power handling capacity, the design emphasis on power capacity neglect several key parameters such as linearity, noise figure, phase noise in previous work. For commercial power amplifiers, those parameters are also important parameters in determining the applicability of the amplifiers to certain systems. These issues are thoroughly discussed and the performance of a high power amplifier using the compact coaxial waveguide combiner is measured in this chapter. For clarity’s sake,
99
we use the term combiner to refer to the high power amplifier using the compact coaxial waveguide combiner.
5.1
Measurement System
Measurement Setup
LabView Program
Signal Source
Power Meter HP438A
Power Supply
TWTA
Power Supply 8 Channel Supply
Coupler #2
Coupler #1
Power Sensor A
Power Sensor B
Gate Bias
Drain Bias
Spectrum Analyzer
Attenuator 20dB / 100
DUT
Figure 5.1 Power measurement setup.
Figure 5.1 shows the measurement setup, which is similar to the setup used by Nick Cheng[1]. A traveling wave tube amplifier (TWTA) follows the signal source to
100
generate the broadband CW signal with a power level between 20 dBm and 43 dBm. The high noise floor of TWTA means that the output power of the TWTA must be greater than 20 dBm for an adequate signal to noise ratio. The output power of the TWTA is between 20 dBm to 35 dBm in its linear region of operation. Power sensors record the input and output power levels. Instead of using 3.5 mm connectors as an Agilent8485A does, the output high power sensor, an Agilent 8481B, uses type N connectors. With an integrated 30 dB attenuator, its dynamic range spans from 0 dBm to 44dBm. The input power sensor, an Agilent 8485A, only has a dynamic range from –30 dBm to 20 dBm. It is very important to consider the sensitivity, settling time and protection when choosing an appropriate power level at the power sensors. A 20 dB power attenuator is used to reduce the output power to a value within the safe range of the power sensor. A spectrum analyzer monitors the output signal for oscillation through the output coupler. The combiner requires two different power supplies to drive the GaAs pHEMTs. The gate bias is provided by an Agilent 3631. Since the drain bias requires high current and good stability, we chose KEPCO RA55 8 channel power supply. Each channel of the RA55 is independent with a capacity of up to 10 Volts and 12 Amps. The average current for each MMIC amplifier is around 1.2 Amp at 8 Volts; each channel of the RA55 provides drain current for 4 MMIC amplifiers. Voltage drops on the bias line lower the drain voltage, resulting in reduced output power. With a remote-sensing capability, the KEPCO RA55 power supply can properly compensate the voltage drop between the power supply and the DC bias 101
board, but thick bias lines and gold plated DC pins are still needed for the connections between the DC bias board and combiner.
Automatic Control Program
Figure 5.2 Virtual instrument front panel of Labview Program.
The signal source, power supply, power sensor and spectrum analyzer are all controlled by a laptop computer through the GPIB bus. Using National Instrument’s Labview, a Microsoft Windows based program was developed for instrument configuration, calibration and data acquisition. National Instruments supplies virtual instrument (VI) libraries for a variety of popular test equipments. The Labview program displays virtual front panels of the instruments on the computer. Users can execute operations on the virtual front panel, 102
which sends commands to the appropriate instrument via GPIB connection. The Labview program is able to automatically initialize instruments and carry out very complex calibration and measurement procedures, which are normally tedious to be done manually. Figure 5.2 shows the virtual front panels of our measurement system. The panel at the bottom of the screen is the calibration section. Since the power sensors use different calibration factors at different frequencies, calibration factor tables for both power sensors are stored into the calibration panel. The calibration program automatically loads the data and outputs the calibration results in the calibration data table on this panel. Since the amplifier measurement system involves many instruments, computer automation improves measurement efficiency. Additionally, the Labview program can add overload protection in the event of a MMIC amplifier failure during a measuremnt, ttherebyand avoiding the failure of other MMIC amplifiers due to excess current flow.
Calibration Procedure
Calibration is a necessary procedure for all RF measurements. The goal is to deembed the parameters of the input and output network, and then move the reference plane directly to the input and output port of the DUT.
103
Power Sensor A SMA to N Adapter
Coupler #1
Power Sensor B
(a) Power Sensor A
Power Sensor B SMA to N Adapter
Coupler #1
Attenuator 20dB / 100 Watt
Coupler #2 Spectrum Analyzer
(b) Figure 5.3 Calibration procedures.
Unlike the new vector network analyzers, which have four directional couplers and vector receivers, our measurement setup has only two directional couplers and scalar power sensors. Loss of the input/output coupler, cable and connectors are calculated by the calibration procedure describe in Figure 1.3. In step (a), the difference between coupler #1’s coupling port and the type N port is measured with two power sensors. Using the data from step (a), the power difference between the coupler #2’s output port and the input of the attenuator are calculated in step (b). All of the calibration data is stored in the Labview program and used for post measurement data correction. Equal length SMA to type N adaptors with different genders are used for connection to the DUT. Even though different gender adaptors are used, the calibration data is still sufficiently accurate since we choose the adaptors from the same manufacturer.
104
5.2
MMIC Amplifier Characterization
The MMIC amplifier we chose is a TGA9092 EPU (Engineering Prototype Unit), manufactured by TriQuint Semiconductor. The TriQuint TGA9092 is a dual-channel, three-stage wide band HPA MMIC designed using TriQuint’s proven 0.25 µm power pHEMT process which supports a variety of high performance applications, including military EW programs, VSAT, and other applications requiring wideband high power performance. Each amplifier channel consists of one 1200 µm input device driving a 2400 µm intermediate stage, which in turn drives a 4800 µm output stage.
Figure 5.4 Layout of TriQuint TGA9092 EPU.
105
Figure 5.5 2 dB compression point output power of TGA9092 (from TriQuint data sheet). 30
30
25
25
20
20 MMIC1 MMIC2 MMIC3 MMIC4
15
MMIC5 MMIC6 MMIC7 MMIC8
15
10
10
5
5
0
0 6
8
10
12
14
16
18
Freq [GHz] Figure 5.6 Small signal S21 data of 8 TGA9092 from the same wafer (from TriQuint Semiconductor).
A 2 dB compression point output power measurement result is shown in Figure 5.5. The P2dB power measurement, measured at different input power levels, exhibits the power capacity of the MMIC. The small signal S parameters of eight TGA9092
106
MMIC amplifiers are shown in Figure 5.6. All of the MMIC amplifiers are biased with a 8V drain voltage and 0.31 V gate voltage. The average drain current is 1.2 A with +/- 5% variation. Since the MMIC amplifiers are from the same wafer, the maximum gain variation is less than 2.5 dB. To minimize the efficiency reduction due to gain variation, it is recommended to use MMIC amplifiers from the same wafer, or at least the same process batch, for the combiner system.
5.3
Output Power
Small Signal Modeling
As explained in the previous chapters, the performance of the waveguide structure and finline transition is simulated by HFSS, a 3D FEM simulator. Since HFSS can only simulate 3D passive structures, the overall active amplifier’s performance is not directly available from HFSS simulations. We exported the S parameter results from HFSS to S2p files, then imported them into Agilent Advance Design System (ADS). The ADS small signal circuit model of the combiner is shown in Figure 5.7. Waveguide and Slotline Transistion
Lossy Matching Network
MMIC
Waveguide and Slotline Transistion
Figure 5.7 Schematic for small signal modeling of the combiner.
107
Due to the complexity of the finline transition, the HFSS simulation requires tremendous memory resources. Even the most advanced computer in our research lab, with 1 GB of memory, is only useful for simulations of 1/16th of the whole finline array. Using symmetry, we equally divided the combiner into 16 parallel identical sections from the input end to output end. The method is viable because of the symmetry of the combiner. In the ADS model, we only simulated one of the 16 sections. The section included the input/output waveguide tapers, finline transitions, a lossy matching network and a MMIC amplifier. The one section simulation is accurate enough to represent the overall amplifier performance because the spatial power combining theory proved that the power is evenly distributed to and combined from each channel and the overall gain is the same as for a single channel. Slotline Transition
MMIC Input R R2 R R=60 Ohm R1 R=60 Ohm
WIRE Wire2 D=1.0 mil L=300 um Rho=1.0
WIRE Wire1 D=1.0 mil L=300 um Rho=1.0
Figure 5.8 Circuit schematic of lossy matching network.
The TGA9092 MMIC amplifier has a gain in excess of 25 dB which very easily cause oscillation problems in a packaged waveguide environment when the output to input isolation is only slightly higher than 20 dB at some frequency, as shown in chapter 4. The circuit becomes stable when the overall gain is reduced within the 20
108
dB limitation due to the insertion of the lossy matching network. Bonding wires were included in the lossy matching network circuit model to improve the accuracy of the simulation, the results of which are as shown in Figure 5.8. The good agreement between the measurement and the simulation, which is shown in Figure 5.9, verifies the effectiveness of the modeling. There is 8-dB difference in the gain between the MMIC amplifier and combiner, which arise from the lossy matching network. 30
Simulation & Measurment of Combiner
30
20
20
10
10
0
Measurement_S21 [dB] Simulation_S21 [dB]
MMIC_S21 [dB]
0
-10
-10
-20 5
10
15
20
-20
Frequency [GHz]
Figure 5.9 Comparison of simulation and measurement of the combiner and measurement of the MMIC amplifier.
The lossy matching network reduces the feedback loop gain and stabilizes the circuit. In the mean while, the lossy matching network improves input impedance match of the MMIC amplifier, which is initially very bad because the MMIC amplifier is optimized for wide bandwidth and high power. The comparison of S11 and S22 between the MMIC amplifier and the combiner is shown in Figure 5.10. Aside from the reduction in S11 for a combiner relative to a MMIC amplifier, we also 109
observe ripples in the S11 and S22 curve of the combiner, which are caused by the cancellation of the reflected signal from different end of the waveguide taper. 0
0
-10
-10
-20
-20 S11_MMIC [dB] S11_Combiner [dB]
S22 _MMIC [dB] S22 _Combiner [dB]
-30 5
10
15
20
-30
Frequency [GHz]
Figure 5.10 S11 & S22 of the MMIC amplifier and the combiner.
Power Measurement
Figure 5.11 Combiner with bias lines.
Figure 5.11 shows the assembly of the combiner system. The bias lines were connected from a biasing board to the 16 individual circuit trays. The KEPCO 8
110
channel power supply and Agilent power supply were connected to the biasing board. The calibration procedure was performed before the measurement and the calibration data were stored in the Labview measurement program. The measurement setup has been illustrated in section 5.1.
50
100
Pout=46.4 dBm 40
80
Pout [dBm]
30
Efficiency [%]
Gain [dB]
60
20
40
10
20
0
0 6
8
10
12
14
16
18
Freq [GHz] Figure 5.12 Frequency sweep at 30 dBm input power.
The input power level was chosen to be 30 dBm. A frequency sweep measurement result is shown in Figure 5.12. A maximum power of 44 Watts is obtained at 10 GHz. The gain curve followed a similar shape to the small signal gain curve in Figure 5.9, with the exception that gain compression may occur differently over the band. The 3 dB bandwidth is from 6 to 17 GHz. The output power curve varies slightly from the 2 dB compression output power curves in Figure 5.5 that were measured at different input power levels to reach 2 dB compression. We noted
111
that two MMIC amplifiers of the 32-MMIC combiner were nonfunctional during the measurement. The combiner’s output power was measured with 30 working MMIC amplifiers, which is 88% of the 32-MMIC combiner’s output power basing on the graceful degradation theory[2]. The KEPCO 8 channel power supply’s output currents were recorded through the Labview program and the data are shown in Table 5.1. Each power supply channel provided current for four MMIC amplifiers. The amplifier works in a Class AB state, close to Class A. When the input power increases, the current also increases to some extent due to waveform cutoff. The channels that are adjacent to the broken channel have a larger increase in current as a result of a more severe overdrive. This deteriorates intermodulation distortion. The best linearity can be achieved if all MMIC amplifiers are biased in class A with 1.2 amps biasing currents. Table 5.1 Current changes
Current (Amps) @Vds=8V,Vg=-0.4V Small signal before 2 MMICs are broken Small signal after 2 MMICs are broken Pin = 30 dBm after 2 MMICs are broken
I1
I2
I3
I4
I5
I6
I7
I8
Itotal
Iavg/ MMIC
3.95
4.11
3.77
3.93
3.73
4.52
3.48
3.52
31.00
0.97
4.02
4.23
3.78
1.84
3.78
4.58
3.53
3.58
29.34
0.98
4.21
4.61
3.88
2.42
4.71
5.05
3.62
3.45
31.95
1.07
PAE =
Pout − Pin Pout − Pin P −P = = out 8 in PDC VDS I D ,total VDS ∑ I D ,i i =1
112
(4.14)
The power added efficiency (PAE) is calculated from equation (4.14). The PAE is measured and shown both in Figure 5.12 and Figure 5.13. At 10 GHz, the small signal gain of the combiner is 18.2 dB, which is about 0.4 dB smaller than the small signal gain of the combiner when all MMIC amplifiers are functional. At 30 dBm input power, the gain is compressed by 1.8 dB. The power added efficiency was about 17% at an output power of 44-Watts. 50
50
40
40
Pout [dBm] Gain [dB]
30
Efficiency [%]
30
20
20
10
10
0
0 20
22
24
26
28
30
Input Power [dBm] Figure 5.13 Power sweep at 10 GHz.
5.4
Linearity
Linearity is important for broadband communication systems. A two-tone intermodulation distortion (IMD) measurement is used to evaluate the linearity of the amplifiers. The IMD is a ratio of the strength of the third order component produced by two adjacent fundamental signals to the strength of one of the fundamental signals. 113
The extrapolated cross point of the fundamental and the third order intermodulation component is known as the third order intercept point (IP3). Although the power level of the fundamental carrier can never be equal to that of the third order intermodulation component because of (you used too many “due to”) saturation, it is reflective of the amplifier’s linearity.
IP3
POutput
PSat
P1dB
3rd
5th
PInput Figure 5.14 Output power and harmonics.
Compared to TWTAs that work in the saturation mode, solid-state amplifiers offer better linearity by operating at P1dB point. To reach an IMD level of –25 dBc, a typical TWTA needs to back off more than 7 dB from the rated single carrier output power. A solid-state amplifier only needs to back off around 2 to 3 dB from P1dB to reach the same IMD level. The output signal’s voltage has following relationship with the input voltage: Vout =a 0 + a1v + a2 v 2 + a3v 3
114
(4.15)
For one tone signal, the amplitude at the 1 dB compression point vic is given by 3 3 a3vic 4 = 1 − 10−0.05 a1vic
(4.16)
Table 5.2 Two-tone distortion products
a1v
a3v 3
One tone: ω
1
3 4
2 tone: ω1 , ω 2
1
9 4
25 4
3 4
25 8
2 tone third order distortion product: 2ω1 ± ω 2 , 2ω 2 ± ω1
a5v 5
We now consider the same amplifier but with a two-tone signal applied to its input, both tones having equal amplitude. The amplitude of each carrier at the IP3, vip 3 , is by definition, a1vip 3 =
3 3 a3vip 3 . 4
(4.17)
The amplitudes of each IM3 product[3] is determined using Table 5.2. Note that the definition of IP3 used here relates the amplitude of a single IM carrier to the amplitude of a single input carrier. Combining equation (4.16) and (4.17) yields 2
vip 3 1 = −0.05 vic 1 − 10
115
(4.18)
which corresponds to a ratio of 9.2, or about 9.6 dB. We can conclude that if only the third order component is present, the IMD3 at the 1 dB compression point will be 18.2 dBc. The other phenomenon we observed is that the sum of the two tone’s signal output power at the 1 dB compression point in the two tone measurement is smaller than the P1dB power in one tone measurement. The relationship of P1dB for one-tone and two-tone signals can be proved using similar approaches. In the two-tone case, if the 1 dB compression amplitude of each carrier is vic 2 , the amplitude of either fundamental carrier will be 9 3 voc = a1vic 2 − a3vic 2 . 4
(4.19)
So that the two tone P1dB compression point is given by 2
vic =
4a1 (1 − 10−0.05 ) . 9a3
(4.20)
By comparison with (4.16), the ratio between the single-tone and two-tone signal amplitude at 1 dB compression point is 2
vic1 =3 vic 2
(4.21)
The 1.7 dB difference explains the discrepancy in the P1dB power in one-tone and two-tone measurements.
116
If only the third order component considered, the third-order distortion trace would be linear in input power. When the power is backed off from the IP3 point, the third-order products would fall at a rate of 3 dB for every decibel of output or input power reduction, or 2 dBc relative to carrier. So normally, an amplifier needs to back off about 3 dB from its one tone P1dB to achieve an IMD3 level of –25 dBc. However, the third order product is not simply linear for GaAs amplifiers. In practice, the nonlinearity is beneficial in that it allows for a larger IMD3 near P1dB than would be possible without other higher order contributions. IMD Measurement of MMIC 50 40
OIP3 k=1
30 20 10 0
k=3
-10 -20 -30 -20
-15
-10
-5
0
5
10
15
20
Pin [dBm]
Figure 5.15 Two-tone measurement of the MMIC amplifier.
The gm curve of GaAs MMIC amplifier under investigation is shown in Figure 5.16. It is measured from a small test device on the MMIC chip. A strong fifth order component is introduced from the nonlinearity of the gm curve which causes the concavity of the third order product in Figure 5.15. To analyze the linearity, we fit a
117
polynomial curve with the gm data. The fitted curve is the thinner trace in Figure 5.16. 40
30
20 gm [ms]
10
0
-10 -1.5
-1
-0.5
0
0.5
Vgs [V]
Figure 5.16 Gm Curve of the Triquint GaAs HEMT.
The expression for the fitted gm curve is: 2
3
4
5
g m (Vgs ) = a0 + a1Vgs + a2Vgs + a3Vgs + a4Vgs + a5Vgs + a6Vgs
6
(4.22)
Table 5.3 Polynomial coefficient for gm curve fitting
Polynomial Coefficients
a0
a1
a2
a3
a4
a5
a6
Value
36.8
-15
-6.3
35.5
-15
-20
-0.1
I ds (Vgs ) = ∫ g m (Vgs )dVgs
(4.23)
Since the gate is biased at –0.4V, we have Vgs = −0.4 + vin and 2
3
4
I ds (−0.4 + vin ) = c0 + c1vin + c2 vin + c3vin + c4 vin + c5vin
118
5
(4.24)
Table 5.4 Polynomial coefficient for Ids
Polynomial coefficients Value
c1
c3
c5
39.3
-16.8
5
The third order product can be expressed as 3 25 3 5 v2ω1 ±ω2 = ( c3vin + c5vin ) ⋅ Fscale ⋅ RL 4 8
(4.25)
where Fscale is the scaling factor between the real device in the amplifier and the testing device used for gm curve measurement, and RL is the load impedance. -15
-20
-25
-30
-20
-18
-16
-14
-12
-10
Figure 5.17 Third order component with contribution of fifth order component.
As shown in Figure 5.17, the third order product begins to decrease when the input power is strong and the contribution of fifth order component takes effect. The third order component increases again when the input signal’s envelope starts to saturate as shown in Figure 5.15.
119
10 8 6 4 2 -18
- 16
- 14
-12
- 10
Figure 5.18 Fundamental component with and without the contribution from fifth order products.
In Figure 5.18, the dashed black line is a linear curve; the red line and the blue dashed line represent the fundamental component with and without the contribution of fifth order products respectively. The fifth order products compensate the saturation of the fundamental component due to contribution to the third order products when the input power is close to P1dB, and improves the linearity as shown in the figure. For this reason, we expect our GaAs amplifier to have better linearity when the output power is close to the P1dB point. As shown in Figure 5.15, the GaAs amplifier only needs to back off less than 2 dB from the P1dB point to reach a IMD3 of –25 dBc. To evaluate the change of the IP3 point in power combining, we need to compare the third order intermodulation component (IM3) of a MMIC amplifier and the combiner. For a MMIC amplifier, we can express the fundamental and IM3 output power as Pout = Gm Pin IM 3 = A Pin
120
3
(4.26)
where Gm is the gain of a MMIC amplifier, and A is the coefficient for IM 3 . For MMIC:
Gm
Pin
Pout IM 3
For Combiner:
Li
Lm Lossy Matching Network
Pin
Li
Gm
Pin , e IM 3 ,e
Pout
N
IM 3 Lossy reflection Matching N Way Combiner
N Way Divider 2
Overall Gain G c = G m Li L m *We assume divider and combiner are identical Figure 5.19 Linearity analysis for the MMIC amplifier and the combiner.
At the IP3 point, the linearly extrapolated fundamental output power is equal to the IM3. The OIP3 is the output power at IP3 point where Pout =IM 3 . The OIP3 of a MMIC amplifier is 3
Gm 12 ) . OIP3m = ( A
(4.27)
For a combiner, we have 2
Gc = Gm Li Lm 2
Pout = Gc Pin = Gm Li Lm Pin .
121
(4.28)
For each MMIC amplifier in the combiner, we have Pin ,e = IM 3,e
Pin Li Lm N 3 = A Pin ,e
(4.29)
where N is the number of channels in the combiner and Lm is the loss of the lossy matching network. We assume the divider and combiner have the same loss Li . The IM 3,e from each MMIC amplifier are added in the same way as the fundamental signal. The sum of the IM 3,e at the output port is expressed in IM 3 as 3
IM 3 = N IM 3,e Li = N A Pin ,e Li
(4.30)
Then, we have Pout = IM 3 = N A ( 3
Pin Li Lm 3 ) Li n
G OIP3c = N Li ( m ) A
1 2
(4.31)
where OIP3c is the OIP3 of the combiner. Comparing equations (4.27) and (4.31), we conclude that OIP3c = N Li OIP3m.
(4.32)
For a 32-channel combiner with a Li of 1dB, the combiner will have a factor of 14 dB improvement in OIP3 over a MMIC amplifier. We note that the OIP3 has no relationship with the lossy matching network. We will observe the 14 dB improvement no matter whether we use the lossy matching network or not. The
122
relationship between the fundamental component and third order component remains the same for the combiner and a MMIC amplifier. The intermodulation distortion was measured by two tones at 10 GHz with a separation of 1 MHz in spectrum. The output signal was input to a spectrum analyzer. The spectrum analyzer attenuates its input signal with its internal attenuators to minimize the intermodulation by the mixers inside the spectrum analyzer. The measurement setup is shown in Figure 5.20. The system was calibrated with the procedure shown in Figure 1.3. After the DUT was added in the measurement setup, the fundamental and third order intermodulation components were read from the spectrum analyzer, and then corrected using the calibration data. Power Sensor A Source 1
SMA to N Adapter
Coupler #1
Source 2
DUT
Power Combiner Power Sensor B
Attenuator 20dB / 100 Watt
Coupler #2 Spectrum Analyzer
Figure 5.20 Intermodulation distortion measurement setup.
The IMD measurement result of the combiner with 30 working MMIC amplifiers is shown in Figure 5.21. There is no obvious concaved curve in this figure. The change of the IM3 curve is due to the unequally driving of the MMIC amplifiers because of the nonfunctioning of 2 MMIC amplifiers. The mismatch introduced by 123
the broken MMIC amplifiers causes the adjacent MMIC amplifiers to be driven by a larger than average power. Those adjacent MMIC amplifiers saturate in amplitude faster when the input signal increases. The amplitude saturation will offset the improvement due to the fifth order products shown in Figure 5.15. Since the IP3 point is determined by the linear part of the fundamental and IM3 traces, it will not be changed when some of the MMIC amplifiers are overdriven. 60 OIP3_Combiner
40
Combiner Single Tone k=1
OIP3_MMIC MMIC Single Tone
20
k=1
0 IMD3 IMD3 -20
-20
-10
0
10
20
30
40
Pin [dBm]
Figure 5.21 Comparison of IMD between the MMIC amplifier and the combiner.
At 10 GHz, the output IP3 (OIP3) is 52 dBm compared to 38 dBm of a single MMIC, which corresponds to a 14 dB improvement over a single MMIC amplifier. Figure 5.21 consolidates the conclusion from equation (4.32).
124
5.5
Noise Figure For MMIC:
N add N o = Gm N i + N add
Ni Gm
For Combiner: Li
Li
Lm N add Ni ,m
Lossy Matching Network
N o,m
Gm
Ni
No
N
Lossy Matching Network
N Way Combiner
N Way Divider 2
Gc = Gm Li Lm
Overall Gain
*We assume divider and combiner are identical Figure 5.22 Noise Figure of a MMIC amplifier and a combiner.
For a single MMIC, the noise figure is Fm =
Si / N i No G ⋅ N + N add = = m i So / N o Gm ⋅ N i Gm ⋅ N i
(4.33)
For a combiner, we assume the input and output matching networks are the same and the combiner’s gain is represented by Gc as shown in Figure 5.22.
125
The added noise from each channel is uncorrelated and can be treated as a random variable. When N uncorrelated random variables sum up at the output port, the total noise power will be the same as the noise power of a single channel. When the loss of the combiner Li is taken into consideration, the overall added noise for the combiner is N add ⋅ Li , where N add is the added noise of each MMIC amplifier. Since the input noise is correlated when distributed to each channel, the input noise from each channel is summed in phase at the output of the combiner. The overall noise figure Fc is then expressed as: Fc =
Gc ⋅ N i + N add ⋅ Li . Gc ⋅ N i
(4.34)
2
For a combiner system without the lossy matching network, Gc = Gm Li . In that case the combiner’s noise figure can be expressed as Fc = 1 + ( Fm − 1)
where Li =
1 Li
(4.35)
Gc −1 . If we assume Li ≈ 1 , Fc ≈ Fm Li . Gm
The measured noise figure Fc and calculated noise figure Fc * , are shown in Figure 5.23, along with the noise figure of a single MMIC amplifier for comparison. Symbol NF represents the noise figure F in dB. The measurement is based on a 2 x 2 array rectangular waveguide combiner. Good agreements verify the conclusion that when no lossy matching network is 126
integrated in this combiner system and the loss of the combiner is very low; the noise figure of the combiner and the MMIC amplifier are very close to each other. 25
25
20
20
15
NFm(dB) NFc(dB) NFc*(dB)
Gm(dB) Gc(dB)
15
10
10
5
5
0
0 8
8.5
9
9.5
10
10.5
11
Freq(GHz)
Figure 5.23 Measurement of noise figure of a single MMIC amplifier and a combiner and calculation from equation (4.35). 2
For a combiner system with an integrated lossy matching network, Gc = Gm Li Lm , and we have Fc = 1 + ( Fm − 1)
1 . Li Lm
(4.36)
If we assume Li ≈ 1 , Lm 1, then equation (4.36) can be approximated as −1
−1
Fc ≈ Fm Li Lm .
127
(4.37)
5.6
Spurious-Free Dynamic Range
Given the IP3 and noise figure, one can define another important property of the \amplifiers called the spurious-free dynamic range (SFDR). The SFDR represents the ability of a system to detect or boost signals in the presence of noise and other strong signals, and is important in several system applications such as transponding the multiple carriers that routinely pass through terrestrial base stations operating under code-division multiple access (CDMA) technology. Another example is the detection of a frequency-chirped radar return signal in the presence of strong clutter. The lower limit of the SFDR occurs when the input signal power equals the input band limited Rx noise power. To account for multiple input signals, the upper limit of the SFDR is often defined as the power when an IM3 tone (for two equal input tones) equals the output receiver noise power. This yields the expression 2 OIP3 SFDR = ( )3 F ·G·k B ·T0 · ∆f
(4.38)
where F is the receiver noise factor, G is the gain, ∆f is the instantaneous bandwidth, k B is Boltzman's constant and T0 is the ambient temperature[4].
From this expression, the scaling behavior of the SFDR in an amplifier is obvious. For a combiner system without an integrated lossy matching network, G and F don’t 2 3
scale, only OIP3 scales with N. The SFDR is proportional to N . This is shown in
128
Gc = Gm Li Fc ≈ Fm Li
2
−1
OIP3c = NLi OIP3m
(4.39)
2
SFDRc = N 3 SFDRm
where subscript c refers to combiner and m refers to MMIC amplifier. For a combiner with lossy matching network, we have 2
Gc = Gm Li Lm −1
Fc ≈ Fm Li Lm
−1
OIP3c = NLi OIP3m
(4.40)
2
SFDRc = N 3 SFDRm .
From equation (4.39) and (4.40), we conclude that the combiner will have a 10 dB improvement in SFDR over a single MMIC amplifier.
5.7
Phase Noise of Combiner
Residual Phase Noise
Phase noise is the fluctuation of the phase due to a resistor’s thermal noise, an active devices’ 1/f noise and shot noise. Residue phase noise is the added noise to a signal’s phase when the signal is processed by a 2-port device. It is commonly used to evaluate the phase noise characteristic of the 2-port device. Residual phase noise includes two basic noise mechanisms: additive noise and multiplicative noise. Additive noise is generated by the device and added linearly to
129
the signal. Multiplicative noise is the noise that modulates RF signal by the multiplication of baseband noise with the signal. The mixing is due to the nonlinearities in the 2-port network. The baseband noise may be generated by the active devices of the internal network, or it may come from low-frequency noise in the signal line or power supply.
Figure 5.24 Multiplicative residual noise model.
To measure the residual phase noise, we demodulate the RF signal and then analyze the noise at baseband. If we assume φpeak is ensemble average
(4.43)
δθ is Fourier transform Assuming that input noise and amplifier noise contributions are uncorrelated:
δθ in∗δϕi = 0 for all i, δϕi∗δϕ j = 0 for i ≠ j Excess noise from each amplifier is δϕ
δθ out
2
= δθ in
2
, making the output noise
2
+
1 2 δϕ N
(4.44)
(4.45)
The second part in equation (4.45) is the residual phase noise. It will be improved by a factor of N if all the added noise is uncorrelated. If the input phase noise is small compared to the residual phase noise, we will also see approximately a N times reduction in the output phase noise.
Residual Phase Noise Measurement
The residual phase noise is measured with a HP3048 phase noise system. As shown in Figure 5.26, the input signal is divided into two identical signals by a power divider. The phase of the two signals are adjusted into quadrature by a phase shifter. The HP11848 phase noise interface has a double balanced mixer as phase detector. As shown in Figure 5.26, HP 11848 demodulates the RF signal to base band noise signal and the base band signal’s spectrum is computed by HP3561A dynamic signal
132
analyzer. If the peak phase deviation is smaller than 0.2 radians, the RF signal phase noise spectrum is one half of the base band noise spectrum. 20 dB Attenuator 11848 Phase Noise Interface
DUT
+8dBm R Port
G=18dB
Source
V1Sin(2π f 0t + φ (t ))
π
L Port
Power Spliter
Kφ ⋅ φ (t ) = Vn (t )
V2 Sin(2π f 0t + φ (t ) + ) 2
+10dBm
Phase Shifter GPIB Control S (f) Sφ = n 2 Kφ L( f ) =
Sφ ( f ) 2
=
Sn ( f ) 2 2 Kφ
0.01Hz
100KHz
3561A Dynamic Signal Analyzer
Figure 5.26 HP3048 Phase noise measurement system. 0
0 L(f)_MMIC [dBc]
-20
L(f)_Combiner [dBc]
-20
-40
-40
-60
-60
-80
-80
-100
-100
-120
-120
-140
-140 -10 dB/decade
-160
-160 1
10
100 1000 Offset Freq [Hz]
10
4
Figure 5.27 Residual phase noise of a MMIC amplifier and the combiner for 1 Watt medium power combiner system.
133
The residual phase noise measurement results of a MMIC amplifier and the combiner in the 1-Watt medium power combiner system are shown in Figure 5.27. An average of 15 dB improvements in residual phase noise is observed. The residual phase noise is lower than –150 dBc at a 10KHz offset from the carrier frequency. In the MMIC amplifier’s measurement, some spikes are observed which are the spurs from the DC bias lines. The residual phase noise spectrum of the medium power combiner follows a 10 dB/decade curve from 1Hz to 1KHz offset from carrier frequency which is the characteristic of 1/f noise. In the medium power combiner, the MMIC amplifier is chosen to be a low noise amplifier. 1/f noise is the dominant source of phase noise. The 1/f noise is from the imperfect material of each GaAs device and is uncorrelated for each MMIC amplifier. From equation (4.45), 15 dB reductions in residue phase noise are expected which agrees well with the measurement results. The residual phase noise measurement results of a MMIC amplifier and the combiner in the high power combiner are shown in Figure 5.28. The residual phase noise level is around –140 dBc at a 10 KHz offset from the carrier frequency, which is about 10 dB higher than the measured value for the medium power combiner since the MMIC amplifiers were designed for high power applications. However, instead of 15 dB, only 5 to 6 dB residual phase noise reduction is observed from the phase noise spectrum. The reason of this lower reduction in phase noise is the partial correlation of the phase noise sources in each channel.
134
0
0
-10
-10 L(f)_Combiner [dBc]
L(f)_MMIC [dBc]
-20
-20
-30
-30
-40
-40
-50
-50
-60
-60
-70
-70
-80
-80
-90 -100
-90 -20 dB/decade
-100
-110
-110
-120
-120 -20 dB/decade
-130
-130
-140 -150 0.01
-140 -150
0.1
1 10 100 Offset Freq [Hz]
1000
10
4
Figure 5.28 Phase noise measurement of a single MMIC and the high power combiner.
KEPCO 8 Channel Power Supply
Figure 5.29 Power supply of the combiner system.
Since the 32 MMIC amplifiers are driven by KEPCO 8-channel DC power supply as shown in Figure 5.29, multiplicative noise from the power supply is partially
135
correlated. As shown in the figure, every four MMIC amplifiers have the same noise from power apply and there are other correlations from the sharing of the same ground line. And since the current is very high, the multiplicative DC noise is dominant in all the noise sources. We didn’t observe a 1/f curve from 1Hz to 1KHz offset from carrier frequency because the 1/f noise is inferior compared with the DC line noise. In stead, a low-pass characteristic curve is observed in the phase noise spectrum. That is because the capacitors used in the DC bias line forms low pass filters and the low pass filtered multiplicative noise spectrum is transferred to the carrier. If we integrate voltage regulator for each MMIC amplifier, asides from adding voltage protection feature to each MMIC amplifier, we will decorrelate the DC line noise. We will be able to achieve the same phase noise reduction as the medium power amplifier does.
5.8
Summary
The amplifier using the compact coaxial waveguide combiner shows the 3-dB bandwidth from 6 to 17 GHz with a maximum power of 44 Watt. We maintain the combiner’s linearity similar to that of a MMIC amplifier, while improving the OIP3 of the combiner to 52 dBm, which is 14 dB higher than that of a single MMIC amplifier used in the combiner. The SFDR range is also increase by 10 dB and residual phase noise is reduced by 5 to 7 dB. The residual phase noise floor is –140 dBc at a 10 KHz offset from carrier frequency. All these features enable this amplifier 136
a good candidate for high power amplifiers in wireless and satellite communication base stations.
References
1. 2. 3. 4.
Cheng, N.-S., Waveguide-Based Spatial Power Combiners, in Dept. of Electrical and Computer Engineering. June 1999, University of California: Santa Barbara, CA 93106. Rutledge, D.B., et al., Failures in power-combining arrays. IEEE Transactions on Microwave Theory and Techniques, 1999. 47(7, pt.1): p. 1077-82. Cripps, S.C., RF Power Amplifiers for Wireless Communications. 1999: Artech House. Brown, E.R. and J.F. Harvey, System characteristics of quasi-optical power amplifiers. IEEE Circuits and Systems Magazine, 2001. 1(4): p. 22-36.
137
6.
Conclusion and Future Works
CHAPTER 6 Conclusions and Future Works In this thesis, we discussed the modeling and fabrication technique of coaxial waveguide power combiner, and successfully demonstrated a low-noise medium power amplifier and a high power broadband amplifier with the combining technique, each integrating 32 MMIC amplifiers. This thesis proves that the power combining technique using coaxial waveguide structure is the most effective approach to integrate a large quantity of devices over a broadband width with high power combining efficiency. The high power broadband amplifier design will enable the power amplifier industry with a quick shift from traveling tube amplifiers to the solid-state amplifiers. The low-noise and high dynamic range properties in the medium power amplifier also show good applications in receiver design. Some challenges are still remained in the high power amplifier design. Preliminary thermal simulations are conducted in the design. Since the combiner
138
structure is very compact, the viscosity of the air can slow down the air flow speed inside metal fins. Longer fins with wider spacing are suggested to get better thermal dissipation. More extensive thermal analysis, which includes the cooling fans, is needed for a more reliable amplifier design. Oscillation is one of the biggest threats to high power amplifiers. We use lossy matching network to reduce the feedback loop gain to keep the amplifier stable in the high power combiner. But the medium power combiner, which uses much lower gain MMIC amplifiers, is oscillation-free. So choosing MMIC amplifier with proper gain is a key issue in the design phase. For application in communication systems, besides power, bandwidth, linearity and noise, efficiency is always combined together with those parameters in evaluating an amplifier system. Switch-type amplifiers are well investigated to achieve higher efficiency. But those types of amplifiers will have problems in linearity since the amplifiers are working at switching state. A class B push pull amplifier doesn’t have as high efficiency as Class D, E and F amplifiers, but its efficiency is about 28% higher than that of a class A amplifier. Furthermore, it is as well as suited for broadband application as a class A amplifier. The two outputs on each circuit tray in our combiner are designed with 180 degree phase difference which enable this architecture a perfect fit for class B push pull design. The challenge is to design the broadband balun which functions as a broadband transformer to combine the half period signal from each of the two amplifiers together. 139
The 2 amplifiers in the push pull amplifier work alternatively. It means during each half cycle one amplifier is loaded, while the other amplifier is open. The broadband balun needs to match the amplifiers with the load in either half cycle.
140
