Vous êtes sur la page 1sur 154

LINEARITY ENHANCEMENT TECHNIQUES FOR WIDEBAND RF FRONT-END RECEIVERS

by Kihwa Choi

A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Electrical and Computer Engineering Carnegie Mellon University

Pittsburgh, Pennsylvania June 2009

Keywords: RF front-end, receiver, low-noise amplifier, active balun, folded mixer, gain, linearity, noise figure, scattering parameters, intermodulation distortion, second-order input intercept point, third-order input intercept point, linearity enhancement techniques, Volterra series, Taylor series, Volterra kernel, frequency-dependent nonlinearity coefficients, derivative superposition, wideband derivative superposition, self-biasing current reuse technique, DC offset, impedance matching network.

Copyright Kihwa Choi, 2009 All rights reserved.

ii

To my wife, daughter, and son for their patience and support

iii

ACKNOWLEDGEMENTS
This thesis would have been impossible without support of many people. I would like to express my sincere gratitude to them, though their invaluable support deserves much more than this short note of appreciation. It is difficult to overstate my gratitude to my advisor Prof. Tamal Mukherjee and Jeyanandh Paramesh. Their enthusiastic and inspirational leadership in research helped me stay on the right track from the moment I changed research topic. Their extensive technical support was definitely far beyond their duty as a thesis advisor. Until then, Prof. C. Patrick Yue guided me to learn how to do research on RF circuit design. Also, I would like to express my gratitude to Prof. L. Richard Carley and Prof. Ramesh Harjani. It was great honor to have them as my thesis committee. Their constructive suggestions made the thesis sound great in many aspects and made the missing essential parts in the thesis filled. I am grateful to my company, Samsung Electronics, for providing me this wonderful chance to do research so that I can catch up with the latest research trend. I would like to really express my gratitude to my parents, brothers, and sister for their faithful support throughout my life. Also, I would like to thank all my colleague students, Abhishek Jajoo, Cheng-Yuan Wen, Gokce Keskin, Jaewon Choi, Jon Proesel, Sandipan Kundu, Shadi Saberi Ghouchani, Umut Arslan, and the other graduate students. Without them, my life and research at Carnegie Mellon University might have been monotonous and dry. Finally, I would like to thank my wife, HoKyoung Kim, for her patience and support, including dedication to our two sweethearts. I dedicate this thesis to her with my love.

iv

ABSTRACT
A multi-standard RF front-end wideband receiver should achieve not only wide bandwidth to support wideband operation, but also high linearity to minimize sensitivity degradation due to in-band intermodulation and cross modulation distortion due to coexistence of strong interference signals. This dissertation addresses both wideband circuit designs and linearity enhancement techniques. First of all, wideband circuit designs are described for RF front-end receivers including low-noise amplifier (LNA), active balun, and down-conversion mixer. The RF front-end receiver is designed for wideband operation, while accommodating low voltage and low power operation. Small supply voltage, needed for scaled CMOS where transistors can operate at multi-GHz frequencies, however, degrades linearity of the RF front-end receiver. The degradation of the second-order distortion is minimized by using fully differential topology in the mixer, but there is still the second-order distortion due to process variation and asymmetric circuit operation, resulting in DC offsets from different mechanisms. The DC offsets in a fully differential circuit are segmented into different categories depending contribution mechanisms. The relationship between DC offset and nonlinearity is derived and the measurement approach of nonlinearity from DC offset is demonstrated with simulated and measured results in the wideband receiver. Linearity requirements are more stringent in multi-standard RF front-end receivers since out-of-band interference signals fall into within operation bandwidth, requiring higher linearity since they cannot be filtered out any longer for multi-standard applications by an off-chip band selection filter. Wideband derivative superposition

(WBDS) and self-biasing current reuse (SBCR) techniques are combined to achieve high linearity even in a low-voltage operation for a wideband LNA. These linearity enhancements are analyzed by the newly introduced frequency-dependent nonlinearity coefficients and conceptual diagrams are shown to provide intuitive understanding about nonlinear behavior of a linearity-enhanced wideband LNA. The coefficients provide more accurate linearity estimation in an initial circuit design phase and enable us to reduce circuit optimization iterations since they can capture the memory effects of FETs such as parasitic capacitances. To confirm the proposed linearity enhancement techniques, the two prototype LNAs with Chebyshev bandpass filter (BPF) and transformer-based input matching networks are designed, and the simulated and measured results of IIP3 are presented over an operation bandwidth as well as with different frequency spacing of two sinusoidal testtone signals. Furthermore, the measured results of the second-order input intercept point (IIP2) of the two LNAs are shown to observe the linearity degradation due to strong inband interference signals in a multi-standard radio depending on frequency allocation and spacing, which have not been addressed in precedent publications.

vi

LIST OF CONTENTS

Acknowledgements .......................................................................................................... iv Abstract.............................................................................................................................. v Table of contents .............................................................................................................. ix List of Tables ................................................................................................................... xii Chapter 1 1.1 1.2 1.3 1.4 1.5 Chapter 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 Chapter 3 3.1 3.2 3.3 3.4 Chapter 4 4.1 4.2 4.3 4.4 Introduction .................................................................................................... 1 RF Front-End Receiver..................................................................................... 1 Issues in Multi-Standard RF Front-End Receivers ........................................... 4 Motivation ........................................................................................................ 5 Research Contributions..................................................................................... 8 Thesis Organization .......................................................................................... 9 Wideband RF Circuit Design Techniques .................................................... 11 Introduction .................................................................................................... 11 Wideband RF Front-End Receiver ................................................................. 13 Wideband Low Noise Amplifier .................................................................... 15 Active Balun with Compensation Circuits ..................................................... 22 Low-Voltage Folded Mixer ............................................................................ 23 Simulated and Measured Results of the Wideband Receiver ......................... 34 Summary......................................................................................................... 41 Wideband Linearity Enhancement Techniques for LNAs ........................... 43 Introduction .................................................................................................... 43 Theory for Nonlinearity Analysis ................................................................... 45 Linearity Enhancement Techniques ............................................................... 54 Summary......................................................................................................... 68 Proposed Highly-Linear Wideband LNA .................................................... 70 Introduction .................................................................................................... 71 Self-biasing Current Reuse (SBCR) Technique ............................................. 72 Wideband Derivative Superposition (WBDS) Method .................................. 77 Derivation of IIP3 Expression Using Volterra Series .................................... 81

vii

4.5 Chapter 5 5.1 5.2 5.3 Chapter 6

Summary......................................................................................................... 84 LNA Design, Simulated and Measured Results ........................................... 85 Circuit Design of the Proposed LNA ............................................................. 85 Simulated and Measured Results.................................................................... 93 Summary....................................................................................................... 106 Conclusions ................................................................................................ 107

Suggestions for Future Research ................................................................................. 110 Introduction to Volterra Series and Harmonic Input ................................................... 112 Volterra Analysis of the Derivative Superposition Topology ..................................... 116 Volterra Analysis of the Wideband Modified Derivative Superposition Topology ... 126 Matching Table of Frequency-dependent Nonlinearity Coefficients from HB Analysis ..................................................................................................................................... 136 Bibliography ................................................................................................................ 138

viii

TABLE OF CONTENTS

Number

Page

Figure 1-1. Direct-conversion receiver architecture. ...........................................................2 Figure 1-2. Architectures of RF front-end receivers for multi-standard applications. ........6 Figure 1-3. Frequency spectrum and state-of-art IIP3. ........................................................8 Figure 2-1. Receiver block diagram for link budget calculation. ......................................14 Figure 2-2. (a) Simplified input matching network of a transformer-based wideband LNA and (b) its equivalent input matching circuit. ...............................................17 Figure 2-3. Embedded Chebyshev BPF input matching network......................................18 Figure 2-4. Shunt peaking network: (a) simplified schematic and (b) its equivalent circuit. ....................................................................................................................20 Figure 2-5. Wideband LNA with transformer-based input matching network. .................22 Figure 2-6. A schematic of the active balun with compensation feedback circuit. ...........23 Figure 2-7. Simplified schematic of the folded Gilbert cell mixer. ...................................24 Figure 2-8. Simulated DC offset vs. LO device mismatch. ...............................................26 Figure 2-9. Chip micrograph and performance summary ..................................................27 Figure 2-10. Frequency response of CG, NF and IIP3. .....................................................28 Figure 2-11. Effect of supply voltage on CG. ....................................................................29 Figure 2-12. Effect of input stage current density on CG, NF, and IIP3. ..........................30 Figure 2-13. Effect of LO amplitude on CG, NF, and IIP3. ..............................................30 Figure 2-14. DC offsets due to different contributors in the mixer. ..................................33 Figure 2-15. (a) IIP2 and IIP3 extrapolation plot and (b) IF output spectrum...................34 Figure 2-16. The chip micrograph of the receiver with wirebond and COB. ....................35 Figure 2-17. Comparison of simulated and measured S-parameter S11. ..........................36 Figure 2-18. Frequency responses of CG, NF, and IIP3. ...................................................36 Figure 2-19. DC offsets due to different contributors in the receiver. ...............................38 Figure 2-20. DC offset voltage due to self-mixing with different LO amplitude. .............39 Figure 2-21. Differential circuit with input-referred offset voltage for relationship between DC offset and nonlinearity. .....................................................................39 Figure 2-22. Nonlinearity of simulated, measured, and calculated results. .......................41 Figure 3-1. (a) A MOSFET transistor, (b) its incremental model including back-gate effect, and (c) simplified equivalent circuit with the assumption that the nonlinearity is weak and memoryless. ...................................................................46 Figure 3-2. Flow chart for the derivation of an IIP3 derivation.........................................51 Figure 3-3. Line spectrum of the positive frequency terms with two-tone input signals. ...................................................................................................................52 Figure 3-4. (a) Simplified schematic of a CS amplifier, (b) its DC transfer characteristics, and (c) IIP3 plots. ..........................................................................57 Figure 3-5. IIP3 plots with different source degeneration inductances at 0.1 and 1 GHz. .......................................................................................................................59 Figure 3-6. (a) Schematic of the derivative superposition method, (b) conceptual diagram of DS and (c) small-signal equivalent circuit. .........................................60 Figure 3-7. Derivative superposition method. (a) 1st-, 2nd-, and 3rd-order power series coefficients of the auxiliary transistor, (b) 1st-, 2nd-, and 3rd-order ix

power series coefficients of the main transistor, (c) 3rd-order power series coefficients and superposition, (d) Calculated IIP3 dBV. .....................................61 Figure 3-8. IIP3 plots in different frequencies with 0.1 nH source inductance. ................63 Figure 3-9. (a) Schematic of the modified DS method and (b) conceptual vector diagram. .................................................................................................................65 Figure 3-10. (a) Schematic of the alternative DS method and (b) conceptual diagram.....67 Figure 4-1. (a) Simplified common-source amplifier and (b) IIP3 vs. Vds. .......................73 Figure 4-2. The effects of voltage headroom on IIP3: (a) simplified schematic and (a) IIP3 vs. Vds plot. ...................................................................................................74 Figure 4-3. LNA schematic with the self-biasing current reuse technique of the redcolored portion. ......................................................................................................75 Figure 4-4. (a) Node voltages and (b) currents in the LNA with the SBCR technique. ....76 Figure 4-5. Conceptual diagram of linearity behavior over an operating frequency (a) at the low-frequency optimized topology, (b) at the high-frequency optimized topology, and (c) the resulting IIP3 plot. ...............................................................78 Figure 4-6. LNA schematic with wideband derivative superposition (WBDS) method. ................................................................................................................................79 Figure 4-7. Polar plot of the frequency-dependent nonlinearity coefficients as a function of frequency. ............................................................................................80 Figure 4-8. Conceptual vector diagram with SBCR and WBDS techniques (a) at low frequency and (b) at high frequency. .....................................................................81 Figure 4-9. (a) Simplified LNA schematic with SBCR and WBDS techniques and (b) its equivalent circuit. ..............................................................................................82 Figure 5-1. Input matching network (a) Chebyshev BPF matching (b) transformerbased matching.......................................................................................................86 Figure 5-2. Distributed gate capacitance and channel resistance at high frequencies. ......88 Figure 5-3. Simplified schematic of linearity-enhanced wideband LNAs. .......................89 Figure 5-4. First-, second-, and third-order DC transfer coefficients of (a) the auxiliary and (b) main transistors, (c) third-order coefficients and superposition, (d) calculated IIP3 dBV in the proposed topology. ........................91 Figure 5-5. Chip micrographs of the proposed LNAs with (a) transformer-based and (b) Chebyshev BPF input matching networks. ......................................................92 Figure 5-6. Simulated and measured results of the designed LNA (a) with transformer-based matching network and (b) with Chebyshev BPF matching network. .................................................................................................................93 Figure 5-7. Parallel RC load impedance for Bode-Fano limit. ..........................................94 Figure 5-8. NF and NFopt of the LNA with Chebyshev matching network. ......................95 Figure 5-9. NF comparison of the LNA with transformer-based matching with or without the SBCR. .................................................................................................96 Figure 5-10. Noise summary of the devices in the LNA with Chebyshev BPF matching. ................................................................................................................97 Figure 5-11. IIP3 vs. gate bias voltage at 4GHz in the LNA with Chebyshev BPF input matching network. ........................................................................................98 Figure 5-12. IIP3 extrapolation plot at 4GHz in the LNA with Chebyshev BPF input matching network...................................................................................................99

Figure 5-13. IIP3 plot over frequency range in the LNA with transformer-based input matching network. ......................................................................................100 Figure 5-14. Simulated and measured IIP3 (a) vs. frequency range and (b) vs. frequency spacing plots in the wideband LNAs with transformer-based matching and with Chebyshev BPF matching. ....................................................101 Figure 5-15. IIP3 vs. frequency spacing plots of (a) transformer-based matching and (b) Chebyshev BPF matching LNAs at different reference frequencies. ............102 Figure 5-16. IIP2 vs. frequency plots of (a) transformer-based matching and (b) Chebyshev BPF matching LNAs at different 2nd-order IMD frequencies. ..........102 Figure 5-17. Mechanism of IMD2 tones depending on frequency spacing. ....................104 Figure 5-18. IIP2 vs. frequency spacing plots of transformer-based matching (a)-(c) and Chebyshev BPF matching (d)-(f) LNAs at different 2nd-order IMD frequencies with different reference frequencies. ................................................106 Figure 6-1. IIP3 comparison plot with narrowband and wideband LNAs. ......................108

xi

LIST OF TABLES

Number

Page

Table 1. IIP3 requirements of several standards ............................................................... 5 Table 2. Component values of Chebyshev BPF and transformer-based matching networks......................................................................................................... 86 Table 3. Component values of the designed wideband LNA prototypes ................... 89 Table 4. IIP3 comparison with narrow band and wideband LNAs ................................ 109

xii

Chapter 1

Equation Chapter 1 Section 1

Introduction

As wireless telecommunication services are proliferating across the world, more demands for both a new standard and multiple standards seem uprising rapidly to satisfy the end users who want to access an increasing number of services from a single handset regardless of geographical region [1]. This leads to ubiquitous wireless connectivity that supports multiple standards across multiple frequency bands [2]. Minimizing the number of external components and highly integrated solutions in low-cost CMOS technologies are the keys when the same mobile handset supports multi-standard services. Zero-IF or direct-conversion receiver architecture is most suitable for the high-level integration for multi-standard RF receivers. One of the key challenges for multi-standard RF receivers is how to achieve high linearity and low noise over a wide frequency range [3]. 1.1 RF Front-End Receiver

One of the key components for wireless telecommunication systems is the RF frontend receiver which receives both wanted and interference signals. It filters out the unwanted out-of-band signal out, and amplifies the received weak signal with low noise, and then downconverts the amplified signal into a baseband signal, with subsequent filtering, amplification and dizitization, as shown in the direct-conversion receiver 1

architecture of Figure 1-1. Gain, noise, and linearity should be carefully taken into account in each block of the receiver to deliver desired communication quality when the receiver is designed, while making a circuit implementation feasible in a given technology. The importance of these three design parameters is described below.

BPF

LNA

Balun

Mixer

LPF

VGA

ADC I

0 90

LO

Figure 1-1. Direct-conversion receiver architecture.

The received signal can be very weak since the transmitted signal will experience attenuation due to spatial separation between the transmitter and the receiver or due to objects located on the signal path. The signal has to be amplified such that it has large enough amplitude to be digitized correctly while providing a required signal-to-noise ratio (SNR) at the input of an analog-digital converter (ADC). Since the received signal is weak, the noise generated in the RF front-end receiver must be minimized the degradation of the receiver sensitivity. A low-noise amplifier is typically implemented to reduce noise caused by the losses of the input matching network and the noise of the amplifying devices. The received signal can include unwanted out-of-band signals with large amplitude. These unwanted signals can be filtered out by a band selection filter of an RF front-end receiver. Therefore, linearity requirements for the out-of-band interference signals can be 2

alleviated at the cost of using the band selection filter. Most of RF front-end receivers have dedicated hardware for multiple standards with multiple band selection filters. Accordingly, high linearity is not necessarily required in a conventional narrowband receiver unless the filtered out-of-band interference signals are so strong that they can deteriorate the linearity of the receiver. However, the co-existence of multiple standards in the same cell area creates a hostile jamming environment for multi-standard wideband receivers. These interference signals degrade the receiver sensitivity and thus can cause the handset to drop the call. Thus higher linearity is demanded to guarantee the required sensitivity in the hostile environment, compared to a single standard receiver. In general, a RF front-end receiver consists of a low-noise amplifier, balun, and mixer. In the receiver, a single-ended LNA topology is preferred since it can reduce I/O pins and power consumption while providing easy interconnection between the single-ended antenna or RF filter and LNA. On the other hand, a differential topology in the subsequent devices such as a mixer and a variable gain amplifier as shown in Figure 1-1 is preferred not only to minimize the second-order distortion but also to reject power supply and substrate noise [4]. Therefore, a balun is required to convert the single-ended signal of the LNA into the differential signal for the mixer in the receiver. For implementation of the balun, a passive or active balun can be chosen depending on the receiver link budget in the system level design. An active balun is preferred to alleviate the gain requirement in the subsequent stage as well as the NF requirement in the previous stage. To facilitate low-voltage operation, a folded Gilbert cell mixer [5] or a passive mixer is preferred to a conventional Gilbert cell mixer since it has three stacked transistors and one load resistor.

1.2

Issues in Multi-Standard RF Front-End Receivers

There are mainly two approaches to satisfy requirements of multi-standard RF frontend receivers: 1) parallel combination of a narrowband receiver and 2) a single tunable or wideband receiver. The single chip solution is more flexible and efficient in terms of area, power, and cost. The key components in achieving the RF requirements of multiple standards with a fully integrated single chip solution are an LNA with low noise, a mixer with a very high dynamic range, and a careful control of DC offset [6]. For the key components of multistandard RF front-end receivers, wide bandwidth, low noise, and high linearity are important design parameters to achieve RF requirements. As a CMOS technology scales down, the noise and bandwidth performance of RF front-end improves, but unfortunately the linearity performance degrades with supply voltage reduction and high-field mobility effects [7], [8]. On the contrary, the required linearity becomes higher due to the coexistence of adjacent blockers from multiple standards, which were filtered out by a band selection filter in a narrowband front-end receiver. In other words, since the co-existing blockers within an operating wide bandwidth experience intermodulation and crossmodulation without out-of-band filtering, higher linearity is inevitably required to support multiple standards in a single chip solution. Furthermore, the second-order intermodulation products are becoming a critical contributor of nonlinearity even in an amplifier before downconversion in a mixer since the IMD2 terms which fall out of the operating bandwidth in a narrowband amplifier fall within the in-band in a wideband amplifier and thus deteriorate linearity along with the third-order intermodulation terms depending on frequency spacing between interference signals.

To evaluate the required performance of RF blocks for multi-standard applications, the receiver (Rx) link budget calculation can be performed using cascade equations for gain, NF, and IIP3, based on the system-level performance requirements. The required IIP3 for several narrowband and wideband standards is summarized in Table 1. As described above, the IIP3 requirement for multi-standard front-end receivers is expected to be much higher than that for each receiver. For example, the IIP3 requirement for multi-standard receivers has to be at least greater than the highest IIP3 among the standards, i.e., 0 dBm and +12 dBm for LNA and mixer, respectively. Table 1. IIP3 requirements of RF blocks in several standards.

Standard UMTS [3] WLAN 802.11 a/b/g [3] WiMAX 802.16 e [9] UWB (Group 1) [10] Multiband receiver [3] * high gain mode

Receiver IIP3 [dBm] -4.6 -16* -18 -4.6

LNA IIP3 [dBm] 0 -5 -2.7 -6.7 0

Mixer IIP3 [dBm] +12 +5 +12

1.3

Motivation

As we can see in the upper figure of Figure 1-2, multiple-dedicated RF front-end circuits are necessary to support different standards. The single chip solution as shown in the lower figure of Figure 1-2 is preferred due to its compact size and possible reconfigurability between standards. On top of wideband circuit design, the key challenge in the design of the single-chip, multi-standard RF front-end is to achieve high linearity 5

without out-of-band filtering since multiple blockers from different communications coexist within an operation bandwidth and thus deteriorate linearity due to crossmodulation and intermodulation.

Figure 1-2. Architectures of RF front-end receivers for multi-standard applications. As the part of a low-voltage wideband RF front-end receiver, the wideband folded mixer was designed in [5] to facilitate low-voltage operation with 0.81.2-V supply. The designed mixer showed reasonably competitive performance over 37 GHz bandwidth even under 0.8-V power supply. Along with this folded wideband mixer, the wideband RF front-end receiver with an active balun was designed in [11] such that it achieved relatively wideband operation over 35 GHz bandwidth while achieving high secondorder linearity and low DC offset by adopting a fully differential topology after the

wideband LNA. However, the RF front-end receiver showed low linearity performance due to both low supply voltage and degradation of linearity by IMD2 contribution on the IIP3. According to the previous research [3] and system-level link budget calculation, the required IIP3 in a mixer should be much higher than that in a LNA since interference signals are amplified in the LNA, with deteriorating IMD terms further in the following stages. For example, for multi-standard receiver [3], the required IIP3 in the mixer as shown in Table 1 is 12 dB higher than that in the LNA. To understand why the wideband receiver in [11] has low IIP3 even though the mixer IIP3 is not as low as to degrade the receiver IIP3, further analysis and simulation are performed. It turned out that the linearity of the LNA and active balun was not so high due to low supply voltage as well as IMD2 contribution on the IIP3. To achieve high linearity in the wideband receiver, some survey has been performed including precedent narrowband linearity enhancement techniques for LNAs, as described below. To support multiple standards, wide bandwidth and high linearity are required for different frequency coverage as shown in the frequency spectrum plot of Figure 1-3 along with IIP3 versus frequency plot for state-of-art LNAs. The square marks represent IIP3 of narrow band LNAs and the circle marks represent IIP3 of wideband LNAs. The reference circled with red dotted line used linearity enhancement techniques to achieve such a high IIP3 for a narrow band LNA and wideband LNA. While numerous techniques have been proposed for increasing LNA bandwidth (e.g., [10], [11]), there have been relatively few studies of linearity enhancement in wideband LNAs. For example, [4] and [12] employ noise and distortion cancellation techniques to

achieve an IIP3 of about 0 dBm for small frequency spacing at 0.82.1 GHz and 0.25.2 GHz while consuming a large amount of power of 17.4 mW and 21 mW, respectively. In the design of highly-linear wideband RF front-end receivers, linearity enhancement and wideband circuit techniques should be incorporated and implemented simultaneously while minimizing power consumption for high mobility of handsets.

(1) J. Lee, MTT 2006 (2) A. Ismail, JSSC 2004 (3) A. Bevilacqua, JSSC 2004 (4) D. Mukherjee, RAWCON 2002 (5) V. Aparin, MTT 2005 (6) S. Ganesan, MTT 2006 (7) C. Kim, JSSC 2005 (8) S. Blaakmeer, JSSC 2008 (9) F. Agnelli, CAS 2006 (10)A. Amer, CAS 2007

Linearity Enhancement Techniques

Figure 1-3. Frequency spectrum and state-of-art IIP3.

1.4

Research Contributions

The contribution of this research is to provide the techniques to implement both a wideband RF front-end receiver and a linearity-enhanced wideband LNA, in conjunction with analyzing DC offset, nonlinearity behavior, and the relationship between DC offset and linearity. The details of contributions are described below: 1) Designed the wideband RF front-end receiver including a LNA, active balun, and mixer.

2) Demonstrated how to segment DC offsets caused by different DC offset mechanisms in a differential circuit, showed the relationship between DC offset and nonlinearity, and quantified the relationship by simulated, calculated, and measured results in the wideband RF front-end receiver. 3) Proposed a linearity-enhanced wideband LNA topology. 4) Analyzed how linearity over the wide frequency range is improved with a wideband derivative superposition method using newly introduced frequencydependent nonlinearity coefficients in the proposed LNA topology. 5) Designed linearity-enhanced wideband LNA prototypes and verified the effectiveness of the proposed linearity-enhanced wideband LNA topology.

1.5

Thesis Organization

Chapter 2 describes the design of a low-power wideband RF front-end receiver. First, receiver architectures are described and then the link budget calculation is explained to define the requirements of the receiver building blocks such as an LNA, active balun, and mixer. Second, wideband LNA topologies are introduced and then a proper topology for the wideband LNA is chosen. As a part of design for a wideband RF front-end receiver, matching networks are described for the design of the wideband LNA. Third, the design of the active balun and low-voltage folded mixer are described to complete the wideband RF front-end receiver. The detailed analysis is shown for the mixer with simulated and measured results. Fourth, the simulated and measured results of the wideband RF frontend receiver are shown with frequency responses, receiver DC offset, linearity, and the relationship between DC offset and linearity.

Chapter 3 describes theories for nonlinearity analysis including DC and frequencydomain theories. Then some limitations of precedent analysis approaches are addressed and the frequency-dependent nonlinearity coefficients are newly introduced to capture memory effects of a MOSFET. The state-of-art linearity enhancement techniques are presented to provide the understanding of distortion cancellation using Volterra series as well as to address the limitation of those narrow band techniques. Chapter 4 describes two proposed linearity enhancement approaches. One is the wideband derivative superposition (WBDS) which makes IMD3 terms cancelled with IMD2 terms at two frequencies such that IIP3 has two peaks over an operating frequency bandwidth. The other one is the self-bias current reuse (SBCR) technique which bleeds some amount of current directly to an RF input transistor with self-biasing such that the voltage headroom problem in a deeply-scaled CMOS technology can be alleviated even under low supply voltages. Conceptual diagrams of nonlinearity cancellation are shown with both derived IIP3 expression and newly introduced frequency-dependent nonlinearity coefficients to provide insight how two IIP3 peaks can be achieved over wide operating bandwidth in the proposed LNA topology. Chapter 5 describes the prototype implementation of two linearity-enhanced wideband LNAs using the proposed LNA topology. Two wideband LNAs are implemented in a 0.13 m CMOS technology and the simulated and measured results are presented to confirm the effectiveness of the proposed topology. Chapter 6 concludes with a summary as well as suggestions for future research.

10

Chapter 2

Equation Chapter (Next) Section 1

Wideband RF Circuit Design Techniques

2.1 Introduction Several wireless standards have been used for their own communication services with dedicated RF front-end receivers. The straight solution for multiple standards employs parallel narrowband receivers at the expense of die area. A key factor for successful design of multi-standard wideband systems is a low-power wideband RF front-end receiver across multiple frequency bands in a single chip while meeting RF performance requirements in a wideband LNA, an active balun, and a down-conversion mixer. To design a compact low-power wideband receiver, the design of a 35-GHz CMOS wideband RF front-end receiver is presented with an LNA utilizing a transformer matching network, an active balun with compensation feedback, and a low-voltage folded mixer. To achieve low-power operation and to realize a compact input matching network, the LNA utilizes transformer-based input matching. The active balun is adopted to convert single-ended signal into double-balanced one while alleviating gain and NF requirements in the following stage and in the previous stage, respectively. To facilitate low-voltage operation, the folded mixer is employed. For measurement of the wideband receiver, a chip is attached on a PCB board using Chip-On-Board (COB) with bond wires. 11

The relationship between DC offset, IIP2, and IIP3 is derived and confirmed by simulated, calculated, and measured results. With this approach, nonlinearity in mass production line can be estimated accurately and promptly without expensive RF measurement facilities. While a number of wideband RF front-ends have been reported, detailed circuit analysis and optimization have been limited to wideband LNA designs. The design of wideband mixers has not been studied extensively except in [13]. The distributed mixer achieves wideband performance at the expense of large die area and high power consumption. Another important requirement for the wideband mixer is low-voltage operation to facilitate integration of the RF transceiver and the baseband DSP using scaled CMOS processes. Low-voltage mixers using a folded Gilbert cell topology have been proposed in [14], [15]. However, these designs are limited to narrow-band operation owing to the use of LC-tank for biasing. To facilitate low-voltage operation, the mixer employs a folded Gilbert cell topology with PMOS devices for LO switches and utilizes on-chip broadband RF chokes for biasing. Detailed DC offset analysis in a mixer is shown in the following section [5] and receiver nonlinearity is calculated based on the measured DC offsets and then compared to the measured results. In Section 2.2, the wideband RF front-end receiver is described on top of receiver architectures and Rx link budget calculation. The implementation of the wideband lownoise amplifier is described in Section 2.3 along with matching networks, followed by the design of the active balun in Section 2.4 and the design of the low-voltage folded mixer in Section 2.5. The simulated and measured results of the wideband receiver are shown in

12

Section 2.6. Finally, the wideband circuit design techniques are summarized in Section 2.7. 2.2 Wideband RF Front-End Receiver 2.2.1 Receiver Architectures

The main categories of the RF front-end receiver architectures are heterodyne and homodyne [16]. The heterodyne receiver is widely used for current wireless applications since it achieves high performance requirements without limitations. However, it requires an external image rejection filter and thus more components are required, resulting in more area and power consumption. On the other hand, the homodyne receiver also called the direction conversion or zero IF receiver has no image problem and thus needs no external band selection filters. This architecture is more promising for high level integration. However, the direction conversion architecture has some drawbacks [17]: DC offset, 1/f noise, I/Q mismatch, even-order distortion, and LO leakage. In this Chapter, the direct conversion receiver architecture is chosen to implement a compact low-power wideband RF front-end receiver. 2.2.2 Receiver Link Budget Calculation

When a RF front-end receiver is designed with several RF building blocks, an Rx link budget calculation is mandatory in the receiver design such that each block can reasonably share the performance requirements by determining the feasibility of any given blocks. The link budget calculation is also an excellent means for anyone to begin to understand the various factors which must be traded off to realize a given area, cost, feasibility, and level of reliability for a communications link.

13

In the RF front-end receiver, the total gain, NF, and IIP3 of the receiver (Gtotal, NFtotal, and IIP3total, respectively) are calculated by the cascade equations of (2.1)-(2.4) where g1, g2, , gn are the linear gain of each stage, nf1, nf2, , and nfn are the noise factor of each stage, and iip3,1, iip3,2, , iip3,n are the linear IIP3 of each stage, as shown in Figure 2-1.

Gtotal [dB] = 10log ( g1 g2 L gn )


nf n 1 nf 1 nf 3 1 + +L+ NFtotal [dB ] = 10 log nf1 + 2 g1 g1 g 2 g1 g 2 L g n 1

(2.1) (2.2)

1 g g L gn1 g gg + 1 + 1 2 +L+ 1 2 IIP2total [dBm] 10log iip iip2,n 2,1 iip2,2 iip2,3
1 g 12 g 22 L g n21 g 12 g 12 g 22 IIP3total [dBm] 10 log 2 + 2 + 2 + L + 2 iip3,1 iip3,2 iip3,3 iip3, n

(2.3)

(2.4)

g1 nf1 iip2,1 iip3,1

g2 nf 2 iip2,2 iip3,2

g3 nf3 iip2,3 iip3,3

g4 nf 4 iip2,4 iip3,4

g5 nf5 iip2,5 iip3,5

Figure 2-1. Receiver block diagram for link budget calculation.

2.2.3

Wideband RF Front-End Receiver Design

The design of a 35-GHz CMOS wideband RF front-end receiver is performed with an LNA utilizing a transformer matching network, an active balun with compensation feedback, and a low-voltage folded mixer. To achieve low-power operation as well as to realize a compact input matching network, the LNA utilizes transformer-based input 14

matching. The active balun is adopted to convert single-ended signal into doublebalanced one while alleviating gain and NF requirements in the subsequent stage and in the previous stage, respectively. To facilitate low-voltage operation, the mixer employs a folded Gilbert cell topology with PMOS devices for LO switches and utilizes on-chip broadband RF chokes for biasing. For measurement, a chip is attached on a PCB board using Chip-On-Board (COB) with bond wires. The RF front-end consumes 19.3 mW from a 1.2-V supply. The receiver achieves a CG of 28.532.3 dB, a single-sideband NF of 5.59.9 dB, an IIP2 of 12.319.2 dBm, and an IIP3 of 28.3 to 23.7 dBm between 2 5 GHz. Relationship between DC offset, IIP2, and IIP3 is derived and confirmed by simulated, measured, and calculated results. The calculated IIP2 and IIP3 based on measured DC offsets have relatively better match with measured IIP2 and IIP3 results, compared to the simulated ones. With this approach, nonlinearity in mass production line can be estimated accurately and promptly without expensive RF measurement facilities. 2.3 Wideband Low Noise Amplifier 2.3.1 Wideband LNA Topologies

Among wideband LNAs, the distributed amplifiers [18] absorb all circuit parasitic capacitances by incorporating on-chip transmission lines and provide wide bandwidth at the expense of delay. These LNAs demand high-quality transmission lines, making them less attractive to low-cost on-chip solutions due to the large chip area. The resistive feedback amplifiers [19] can achieve wideband input matching, reducing the NF by the local feedback with a feedback resistance and high voltage gain. However, large power consumption is required to obtain a high loop gain in a single stage due to the inherently low transconductance of a CMOS transistor, while stability will be caused with multiple 15

stages. For moderate chip solutions in terms of area and power consumption, the ladder network [20] and Chebyshev BPF network [10] amplifiers are proposed with multisection reactive networks so that the overall input reactance is resonated out over a wide bandwidth. Another benefit of the LNA employing the Chebyshev BPF network is that the topology can incorporate with linearity enhancement techniques addressed in Section 3.3. 2.3.2 Matching Networks

In this sub-section, a transformer-based input matching and Chebyshev BPF input matching networks are analyzed to show how a wide bandwidth can be achieved with a multi-section reactive network and a transformer. In addition, the shunt-peaking output load network widely used in a wideband LNA is explained to address its limitation on linearity enhancement techniques in Section 4.2. A. Transformer-based Input Matching Network [11] For a moderately wide bandwidth with a relatively small chip area, the transformerbased input matching network is proposed. Figure 2-2 (a) shows the simplified input matching network of the transformer-based wideband LNA and Figure 2-2 (b) represents its equivalent circuit.

16

(a) LNA and (b) its equivalent input matching circuit.

(b)

Figure 2-2. (a) Simplified input matching network of a transformer-based wideband

With the assumption that k = M / L1 L2 is as close to 1 as to neglect (1 k 2 ) L1 in Figure 2-2 (b), the input impedance of the LNA is given by

sk2 L1(1+ sT CT Ls + s2CT Ls ) Zin (s) = 1+ sT CT Ls + s2 (CT Ls + n2k2CT L1) + s3k2T CT Cp Ls L1 + s4k2CT Cp Ls L1
where
n

(2.5)

is the turn ratio of the primary and secondary coils, k the coupling coefficient,

CT the total gate-source capacitance of the LNA input transistor, and T the unity-gain
frequency of the input transistor. As written in Eq. (2.5), the input matching networks have two complex poles at the frequency smaller than the lower frequency of the target bandwidth and two complex poles at the frequency greater than the higher frequency of the target bandwidth, respectively. In addition, it has one zero at DC and one complex
1 1 4 / T 2 CT Ls / 4 , such zeros at the center frequency of the target bandwidth, T

that the LNA has wideband input matching by arranging four complex poles and one complex zero. B. Input Matching Network Using LP-to-BP Filter Transformations

17

The input matching network using LP-to-BP filter transformations shown in Figure 2-3 expands the use of an inductively degenerated common-source amplifier by embedding the input network in a multi-section reactive network so that the overall input matching network is resonated over a wide frequency range [10]. The parasitics of the input device are embedded as the part of input matching networks.

Figure 2-3. Embedded Chebyshev BPF input matching network.

The input impedance of the MOS amplifier with a source degeneration inductor is written as in Eq. (2.6)
Z1 ( s ) = s ( Lg + LS ) + 1 + T LS sCT

(2.6)

where CT is the equivalent gate-source capacitance, Lg the gate inductance, LS the source degeneration inductance, and T the unity-gain frequency of the NMOS amplifier. The gate-drain capacitance Cgd is not taken into account due to the complexity of the equation. The -network topology in Figure 2-3 is chosen to achieve sharp out-of-band cutoff characteristic by employing Chebyshev LP-to-BP transformations. For filter termination and input impedance matching, the real part of the input impedance Z1(s) is ideally determined to be equal to the source resistance, that is, T LS=RS. Ideally the power loss in the filter passband is 0 dB with a ripple. Based on the bandwidth and in-band ripple, 18

the reactive components of the filter, C1C3 and L1L3, are determined. The parasitics of input devices are embedded as the part of input matching networks. The input impedance of the amplifier including the -network filter structure is derived as shown in Eq. (2.7).
Z in ( s ) =

2 4 6 s ( p1 + p2 s 2 + p3 s 4 ) + 1 + p4 s + p5 s + p6 s Z1 ( s)

(z
,

2 4 + z2 s 2 ) s 2 + 1 + z3 s + z4 s sL1 Z1 ( s )

(2.7) ,
z 4 = C2 C3 L2 L3

where

z1 = L1 L3

z2 = C2 L1 L2 L3

z3 = C2 L2 + C2 L3 + C3 L3

p1 = L3

, ,

p2 = C1 L1 L3 + C2 L1 L3 + C2 L2 L3

p3 = C1C2 L1 L2 L3

p 4 = ( C1 + C 2 ) L1 + C 2 L2 + ( C 2 + C 3 ) L3

p5 = C1C 2 L1 ( L2 + L3 ) + C1C 3 L1 L3 + C 2 C 3 L3 ( L1 + L2 )

, and p6 = C1C2C3 L1 L2 L3 . Ideally with the

assumption that the real part of the Z1 (s) is equal to RS , the input matching network has one zero at DC, four complex zeroes, and 6 poles such that it provides wideband input matching by arranging the poles and zeros properly at the expense of an in-band ripple. C. Wideband Output Load Network: Shunt Peaking In addition to a wideband input matching network, the output load network should be designed properly to achieve flat gain over a wide frequency range. One of the most popular peaking techniques, the shunt-peaking technique as shown in Figure 2-4 is applied. By adding an inductance in series with a load resistor, the impedance looking into the load introduces a zero at z = Rd / Ld and thus increases with frequency. The increasing impedance compensates the offset of the decreasing impedance due to the parasitic capacitance of the following stage, resulting in constant load impedance over a wide frequency range. The behavior of the shunt-peaking load can be interpreted with the time-domain view. Since the inductor delays current flow through the resistor, the

19

capacitor can be charged fast due to more available current. Faster charging time means a shorter rising time, that is, can be interpreted the wider bandwidth.

(a) circuit.

(b)

Figure 2-4. Shunt peaking network: (a) simplified schematic and (b) its equivalent

The output impedance Z out ( s) in Figure 2-4 (b) is written by


Z out ( s ) = Rd s ( Ld / Rd ) + 1 . 2 s Ld C par + sRd C par + 1

(2.8)

To obtain an optimum inductance value, a factor m is introduced and defined as the ratio of the Rd C par and Ld / Rd time constants [21]:
m= Rd C par Ld / Rd

.
m

(2.9) should be approximately 1.41, resulting in extending the

To maximize the bandwidth,

bandwidth to 1.85 times wider than the uncompensated bandwidth. On the other hand, the shunt-peaking inductance Ld is determined by two opposite requirements [10]: the

20

inductance Ld should be large to have large gain, and it should be small to resonate with the load parasitic capacitance added by the parasitic of the following stage such as a mixer in a receiver and a buffer amplifier for test purpose out of band. The load resistance Rd is determined so that the zero frequency is located closely to the lowest operating frequency to compensate the gain decreased abruptly at the low frequencies. Usually the largest load resistance is chosen to achieve high gain, but it limits the voltage headroom, resulting in degrading linearity that will be addressed in detail in Section 4.2. 2.3.3 Implementation of LNA with Transformer-Based Matching Network

Figure 2-5 shows the simplified schematic of the LNA with transformer-based input matching network of the previous sub-section 2.3.2. To achieve high gain and large reverse isolation, a cascode topology is employed with the source degeneration inductor Ls to realize 50- real input impedance. The external gate-source capacitance Cd is added to reduce the gate inductance value required for input match since the gate-source capacitance of the transistor M1 is usually small. The use of the degeneration inductor poses inherent narrowband input match due to narrowband resonance in the equivalent series RLC resonant circuit. These series RLC circuits are embedded as the part of an input matching network with the transformer (L1, L2) and shunt capacitor (Cp). The detailed analysis of the input matching network is described in the sub-section 2.3.3.

21

VDD

Rd Ld LNAout VBS1 Vin k Rs Cp Vs L1 L2 Cd Ls Cc M2 128 / 0.12 M1 256 / 0.12

Figure 2-5. Wideband LNA with transformer-based input matching network.

2.4 Active Balun with Compensation Circuits Fully differential circuits are preferred in mixed-signal chip design as balanced circuits have advantages of keeping common mode substrate noise from high speed digital circuits at a minimum level. The direct conversion receiver suffers from the second-order distortion like low IF. To overcome those issues, fully differential RF circuit techniques have been employed in the following stages after the single-ended LNA. Baluns are basic elements required in RF components such as balanced mixers and phase splitters to convert single-ended input signal into differential output signal [22]. There are two different types of baluns: passive baluns by passive LC networks and active baluns by differential amplifiers. Since passive baluns occupy large chip area and are lossy, an active balun is adopted in this design while alleviating gain requirement in the LNA and NF requirement in the mixer. If a differential amplifier has infinite impedance at the drain

22

node of the tail current source M7 in Figure 2-6, differential amplifier can provide equal amplitude and 180 phase difference. Due to finite impedance and parasitics at higher frequencies, good gain and phase balance are not achievable. In order to compensate gain and phase imbalance, a fraction of the single-ended signal is fed back to the input transistor M4 through a series RLC network as shown in Figure 2-6. The detailed design is shown in the schematic of Figure 2-6.
VDD

150 RF+ M5 64 / 0.12 RF-

150 M6 64 / 0.12

88 LNAout M3 32 / 0.12 VBS2

0.8pF Rbias 20K

3nH M4 32 / 0.12

M7 320 / 0.5

Figure 2-6. A schematic of the active balun with compensation feedback circuit.

2.5 Low-Voltage Folded Mixer This section presents the design and analysis of a low-voltage downconversion mixer in 0.13-m CMOS for wideband applications between 37 GHz. To facilitate low-voltage operation with 0.81.2-V supply, the mixer employs a folded Gilbert cell topology with PMOS devices for LO switches and utilizes on-chip broadband RF chokes for biasing. The folded topology allows the transconductance and LO stages to have different bias

23

current. By setting the bias current in the PMOS switches near zero, the mixer DC offset due to device mismatch is greatly reduced. The effect of supply voltage on the mixer performance is studied. The wideband frequency responses of the mixer performance under different supply voltages are studied in detail. To achieve high performance with low power consumption, DC bias current density and LO amplitude are optimized based on experimental data. DC offsets due to different sources are measured methodically to analyze their relative importance. 2.5.1 Folded Mixer Circuit Design

The schematic of the folded, double-balanced mixer is shown in Figure 2-7. The NMOS differential pair, M1 and M2, forms the input transconductance stage (gm-stage). The PMOS LO switches, M3 through M6, are folded with respect to the gm-stage.
VDD L1=5.4 nH Rs=18.4 L2=5.4 nH Rs=18.4

LORF+ M1 M2 16 / 0.12 M7 320 / 0.5 Rbias Rbias RFLO+ M3 M4 32 / 0.12 IF+ RL 480 Rbias IFRL 480 M5 M6 32 / 0.12 LO+

Vbias

Rbias

RFdc

LOdc

Figure 2-7. Simplified schematic of the folded Gilbert cell mixer.

24

This technique is effective in 0.13-m technology because PMOS devices with moderate W/L are sufficiently fast to completely steer the current from the gm-stage to the LO switches with reasonable LO amplitudes. The folded topology offers a key advantage over the standard stacked topology for allowing independent settings of the bias currents through the gm-stage and LO switches. The bias current for the gm-stage should be high enough to achieve the desired CG, NF, and IIP3. However, the bias current through the LO switches should be minimized to suppress DC offset, thermal and 1/f noise. The Vgs of the LO switches is set near Vt to achieve a low bias current (~50A) and at the same time ensure that the required LO amplitude remains at a reasonable level (~300 mVpp) for complete current commutation. The small bias current in the LO switches also allows the usage of large load resistances (RL = 480 ) to increase the CG without consuming large IR drop from the limited voltage headroom. The RF chokes, L1 and L2, present a high impedance from 3 to 7 GHz such that the output AC currents of the gm-stage will flow into the LO switches. The RF chokes are realized using two inductors rather than one differential inductor to achieve higher self-resonance frequency (SRF) and hence wider operating bandwidth. The series inductance and resistance of the RF choke are 5.4 nH and 18.4 , respectively. The RF choke has a SRF of 10.8 GHz due to its parasitic shunt capacitance which is 40 fF. The DC offset in mixers is a critical parameter for direct conversion receivers since most of the gain occurs after the downconversion of the input signal and the receiver can be saturated if the offset is too large. Static DC offset can be caused by device mismatch, LO self-mixing due to LO-to-RF leakage, and secondary nonlinearity. Mismatch in the LO switches and load resistances is usually a major contributor to DC offset in fully

25

balanced mixers. Figure 2-8 shows the simulated DC offset voltage at the IF output port versus the LO device mismatch percentage for both the proposed folded mixer and a standard Gilbert cell mixer. The folded mixer is simulated with two LO bias current levels at ~0 and 44 A/m. As expected, the DC offset voltage due to mismatch decreases with the low bias current level. Since the folded mixer can have nearly zero LO bias current and still functions properly, the IF output DC offset is suppressed to 2 mV even with a 20% device mismatch. In contrast, a standard Gilbert cell mixer with the same device mismatch exhibits a DC offset of more than 40 mV. The higher DC offset voltages of the folded mixer at the same current density is caused by larger load resistance value. For example, the load resistance is around 150 in the Gilbert cell mixer, but 480 in the folded mixer, causing larger voltage drop across the load resistor. The detailed analysis of the different sources for static and dynamic DC offsets will be presented with measured data in the next section.
50 40 DC Offset Voltage [mV] 30 20 10 0 -10 -20 -30 -40 -50 -25 -20 -15 -10 -5 0 5 10 15 Device Mismatch Percentage [%] 20 25

Figure 2-8. Simulated DC offset vs. LO device mismatch.

26

2.5.2

Simulated and Measured Results of the Folded Mixer

The chip micrograph and performance summary are shown in Figure 2-9. The active area is 360x380 m2. The layout of the mixer uses pre-characterized components from an in-house RF parameterized cell (P-cell) library for accurate device and interconnect model [23]. The layout is fully symmetrical in order to reject common-mode noise and to minimize phase and amplitude imbalance in the differential signal paths which can degrade CG, linearity, and port-to-port isolation. All signal paths are shielded from each other to improve port-to-port isolation. The measurements are performed using an Agilent E4440A spectrum analyzer. Cascade SGS probes with external 180 hybrids are used for supplying the RF and LO signals. A high-impedance differential active probe (Agilent N1025A) is used for measuring the IF signals.

Technology 0.13-m CMOS Active area 360 m380 m Power 5.8 mW at 1.2 V 3.0 7.0 GHz Frequency 5.3 8.2 dB CG 9.6 13.5 dB NF 37.3 43.4 dBm IIP2 3.2 0.3 dBm IIP3 2.4 4.3 mV DC offset

Figure 2-9. Chip micrograph and performance summary

27

A. Frequency Response The measured frequency responses of the mixer CG, NF, and IIP3 under a 1.2-V supply are shown in Figure 2-10 along with simulation results. The dotted line and solid line represent the simulated and measured performance, respectively. Good agreement between simulation and measurement is achieved due to the accurate modeling of the device and layout parasitics as well as the test setup including the off-chip hybrids. Both CG and NF exhibit the best performance, 7.8 dB and 9.7 dB, respectively, near 5.2 GHz where the effective choke impedance reaches its peak value. The IIP3 is recorded between 3.2 to 0.3 dBm and does not show a strong frequency dependency since it is predominately determined by the input transconsductance which does not vary significantly with frequency.
14 12 10 8 CG [dB] NF [dB] IIP3 [dBm] 6 4 2 0 -2 -4 -6 2.5 3 3.5 4 4.5 5 5.5 6 Frequency [GHz] 6.5 7 7.5
IIP3 CG

NF

Figure 2-10. Frequency response of CG, NF and IIP3.

B. Performance vs. Biasing and LO Amplitude Next, the effects of supply voltage on the mixer performance are studied. The frequency responses of the CG at 0.8-V, 1-V, and 1.2-V supply are compared in Figure 28

2-11. As the supply voltage reduces from 1.2 V to 0.8 V, the CG is degraded by an average of 2.5 dB. However, the NF and IIP3 remain relatively the same with less than 1 dB of degradation on average over the entire frequency range of 37 GHz. This illustrates that the folded mixer is a robust topology for low-voltage operation.
10 9 8 7 CG [dB] 6 5 4 3 2 1 0 2.5 3 3.5 4 CG Measured at VDD=1.2V CG Measured at VDD=1.0V CG Measured at VDD=0.8V 4.5 5 5.5 6 Frequency [GHz] 6.5 7 7.5

Figure 2-11. Effect of supply voltage on CG.

The biasing conditions for the input gm-stage have a major impact on the mixer performance and therefore must be optimized. Figure 2-12 shows that the CG, NF, and IIP3 improve with increasing bias current density and then saturate as the input transconductance starts to degrade due to velocity saturation at high current density. For this design, the optimal bias current density is 115 A/m and further increases merely consume more power without any performance improvements. The effect of LO amplitude on the mixer performance is also examined. Figure 2-13 shows that the CG, NF, and IIP3 degrade noticeably when the LO amplitude is below 300 mVpp because of insufficient voltage swing to completely steer the LO current.

29

16 14 12 10 CG [dB] NF [dB] IIP3 [dBm] 8 6 4 2 0 -2 -4 -6 0 20 40 60 80 100 120 140 160 180 Current Density [uA/um] CG NF IIP3

Figure 2-12. Effect of input stage current density on CG, NF, and IIP3.

20 15 10 CG [dB] NF [dB] IIP3 [dBm] 5 0 -5 -10 -15 0 0.1 0.2 0.3 0.4 Single-ended LO Amplitude [Vpp] 0.5 CG NF IIP3

Figure 2-13. Effect of LO amplitude on CG, NF, and IIP3.

C. DC Offset To measure the mixer DC offset due to the different mechanisms including device mismatch, LO self-mixing, and second-order intermodulation product, the testing

30

procedure described in [24] is adopted. In the first measurement, the RF and LO inputs are supplied with DC bias only and are terminated with precision 50- terminations. As a result, the measured DC offset is due to device mismatch only (Vos, mismatch). In the second measurement, the LO signal is applied whereas the RF input is still with DC bias only. In this case, the measured DC offset (Vos, LO) includes the effect of both device mismatch and LO self-mixing owing to LO-to-RF leakage. The DC offset due to LO self-mixing (Vos, self-mixing) can be determined by taking the difference between Vos, LO and Vos, mismatch from the first measurement:
Vos , self mixing = Vos , LO Vos ,mismatch .

(2.10)

In the third measurement, two-tone signals at 3.99975 GHz and 4.00025 GHz are applied to the RF input while the LO frequency is set to 3.990 GHz. The two-tone RF signal strength is set at 20 dBm to model low-power interferers. It should be pointed out that the RF frequencies are chosen to be about 10 MHz away from the LO frequency so as to separate the effects of LO self-mixing and second-order nonlinearity. With this setup, the total static DC offset (Vos, total) is measured, which includes the effect of device mismatch, LO self-mixing, and second-order nonlinearity. Consequently, the contribution of second-order nonlinearity (Vos, IM2) to the total DC offset can be extracted using Vos, LO from the second measurement as follows:
Vos , IM 2 = Vos ,total Vos , LO .

(2.11)

As an attempt to estimate the dynamic DC offset which can be caused by strong inband interferers, the fourth measurement is performed with the two-tone RF signals increased from 20 dBm to 0 dBm. The excess DC offset due to the high-power

31

interferers (Vos, strong IM2) can then be determined from the measured DC offset (Vos, system) as

Vos , strongIM 2 = Vos , system Vos , IM 2 Vos , LO .

(2.12)

The breakdown of DC offsets due to the different sources is summarized in Figure 2-14 based on the measurement results from 10 different samples. The measured offset ranges from 2.4 to 4.3 mV when LO amplitude of 450 mVpp is applied. LO self-mixing is the main contributor at 51 %. Device mismatch is responsible for 37 % of the offset. Typically, device mismatch is the dominant cause for mixer DC offset. The superior performance obtained is due to the folded topology which allows the LO bias current to be set at near zero. Even when the strong RF interferers at 0 dBm are applied, the DC offset due to second-order intermodulation contributes only 12 % to the total DC offset. However, further measurements reveal that when the LO amplitude is reduced to optimal level at about 300 mVpp, the DC offset reduces to between 1.73.6 mV. The device mismatch becomes the main contributor at about 45 % whereas the LO self-mixing and high-power intermodulation have almost the same contributions at 29 % and 26 %, respectively. It is observed that the offset due to high-power intermodulation is higher at the lower LO amplitude. This result confirms that excessive LO amplitudes should be avoided for the proposed folded mixer to minimize DC offset due to LO self-mixing. Furthermore, within the bandwidth limitation, the LO switching PMOS devices should be made as wide as possible to reduce both mismatch and the required LO amplitude.

32

7
Device Mismatch

6 DC Offset Voltage [mV] 5 4 3 2 1 0 0 1 2


Self-mixing Low-power Intermodulation High-power Intermodulation

4 5 6 7 8 Number of Samples

10 11

Figure 2-14. DC offsets due to different contributors in the mixer.

D. Linearity Figure 2-15 shows an extrapolation plot of IIP2 and IIP3 based on a two-tone test with RF inputs at 5.20 GHz and 5.2005 GHz and the IF output spectrum centered at 10.25 MHz. Over the 37 GHz wideband, the mixer achieves an IIP2 of 37.3 to 43.4 dBm, and an IIP3 of 3.2 to 0.3 dBm.
80 60 IF Output Power [dBm] 40 20 0 -20 -40 -60 -80 -100 -120 -40 -30 -20 -10 0 10 20 30 RF Input Power [dBm] 40 50
IIP3=-2.02dBm IIP2=38.63dBm

(a)

33

(b) Figure 2-15. (a) IIP2 and IIP3 extrapolation plot and (b) IF output spectrum.

2.6 Simulated and Measured Results of the Wideband Receiver The wideband front-end receiver including the LNA, active balun, and folded mixer is designed in a 0.13-m CMOS process and fabricated as shown in Figure 2-16. The active area is 1200x400 m2. For measurement, the RF front-end portion is diced and attached on the designed PCB board with bondwire connectivity for signal and power supply. To convert a differential mixer output signal into a single-ended signal, an external receiver [25] is used on the PCB board.

34

Figure 2-16. The chip micrograph of the receiver with wirebond and COB.

A. Conversion Gain, NF, and Nonlinearity The simulated and measured S11 are compared in Figure 2-17. The measured S11 including bondwire and PCB parasitics is less than 10 dB over 2.55.5 GHz without external matching networks on the PCB board. The measured frequency responses of the CG, NF, and IIP3 under a 1.2-V supply are shown in Figure 2-18 along with simulation results. The line with circles and the line with triangles represent the simulated and measured performance, respectively.

35

0 S11 Measured S11 Simulated -5

S11 [dB]

-10

-15

-20

-25

1.5

2.5

3.5 4 4.5 5 Frequency [GHz]

5.5

6.5

Figure 2-17. Comparison of simulated and measured S-parameter S11.

40 30 CG [dB], NF [dB], IIP3 [dBm] 20 10 0 -10 -20 -30 -40 1.5 2 2.5 3 3.5 4 4.5 Frequency [GHz]

CG

NF

IIP3

5.5

6.5

Figure 2-18. Frequency responses of CG, NF, and IIP3.

All the measured results include parasitics from the COB and PCB without external input matching networks. Both CG and NF exhibit the best performance, 32.3 dB and 5.5 dB, respectively, near 3 GHz where the S11 has deep notch and the output load impedance of the LNA has peak impedance since the active balun and mixer are designed to have

36

relatively flat gain characteristic by having opposite gain slope over the frequency range. The IIP3 is recorded between 28.3 to 23.7 dBm over 25 GHz frequency. The lower nonlinearity comes from the load resistance of 150 in the active balun since it causes around 0.3 V voltage drop across the load resistors. If the load resistors are replaced by inductors, linearity will be improved. The CG and IIP3 are de-embedded to get performance of the receiver itself, but the NF includes noise contribution of the external receiver.

B. DC Offset and Nonlinearity The DC offset is a critical parameter for direct conversion receivers since most of the gain occurs after the downconversion of the input signal and the receiver can be saturated if the offset is too large. Static DC offset is caused by device mismatch, LO self-mixing due to LO-to-RF leakage, and secondary nonlinearity. To measure the receiver DC offsets due to the different mechanisms including device mismatch, LO self-mixing, and second-order intermodulation product, the testing procedure described in [5] is adopted. The measured DC offsets due to the different mechanisms are isolated and plotted in Figure 2-19 over the frequency range. The main contributor of the DC offset in the receiver is the self-mixing due to LO feedback to the input of the LNA, active balun, and mixer. The big change in the plot compared to the previous DC offset plot of the mixer itself is the increment of the DC offset due to selfmixing. It was less than 3mV in the mixer, but it increases a few tens of mV to a few hundreds of mV amplitude. Since the LO amplitude coupled at the LNA and active balun input is amplified as much as the gain of each block or the cascaded blocks. For example,

37

if the cascade gain of the LNA and active balun is greater than 20 dB, the DC offset due to the self-mixing will be amplified by the amount of the gain as well. Abrupt change in the DC offset due to self-mixing over the operating bandwidth comes from amplitude and phase imbalance of the external hybrid and cable assembly to make differential LO signals. The maximum imbalance of amplitude and phase are 0.5 dB and 14, respectively. As shown in Figure 2-19, the main contributor to the DC offset is LO selfmixing. In order to understand the effects of LO amplitude on the DC offset due to the self-mixing, LO signal is swept and DC offset due to the self-mixing term is measured and shown in Figure 2-20. The LO amplitude should be minimized to reduce DC offset from the self-mixing as long as complete commutation is achieved in the LO switching stage of the mixer.

350 DC Offset Voltage [mV] 300 250 200 150 100 50 0 2 2.5 3

Device Mismatch Intermodulation Self-mixing

3.5 4 4.5 Frequency [GHz]

5.5

Figure 2-19. DC offsets due to different contributors in the receiver.

38

50 40 30 Vos,SM [mV] 20 10 0 -10 -20

-15

-10 -5 LO Power [dBm]

Figure 2-20. DC offset voltage due to self-mixing with different LO amplitude.

RF linearity measurement requires expensive test facilities such as network analyzer, spectrum analyzer, signal generator, and so on. For nonlinearity measurement of IIP2 and IIP3, two signal generators and one spectrum analyzer are mandatory equipments. Furthermore, it takes time since input signal amplitude needs to be swept by 1 dB step to pick appropriate extrapolation point up for accurate measurement. As an alternative way, DC offset can be utilized to get relatively accurate IIP2 and IIP3 as described below. Suppose that there is a static offset voltage in differential circuit arising from device mismatch or bias asymmetry, the offset voltage can be referred to the input of the differential circuit to model all internal offsets as shown in Figure 2-21 [26].
Vos,in Vin

Differential Circuit
(Av,diff or Gdiff)

Oscilloscope Vout Zeq Vos,I M2 or Multimeter

Figure 2-21. Differential circuit with input-referred offset voltage for relationship between DC offset and nonlinearity.

39

The DC offset due to the second-order intermodulation (Vos,IM2) can be employed to evaluate the IIP2. Because Vos,IM2 represents the amplitude of the second-order intermodulation product, IIP2 can be calculated as in Eq.(2.13)
IIP 2 = 2 Pin P2,os + CG

(2.13)

where Pin is the input power and P2,os is the amplitude of the second-order intermodulation product (Vos, IM2) in dBm, and CG is the conversion gain of the receiver in dB. When P2,os is calculated, the input impedance (Zeq) of the multimeter or oscilloscope which measures the DC offset voltage as shown in Figure 2-21 should be taken into account. In this measurement, LeCroy 9354L oscilloscope is used along with AP020 probe having 1-M input impedance. If the input-referred offset voltage (Vos,,in) of the differential circuit and the calculated IIP2 from the output DC offset voltage are known, IIP3 can be calculated based on the following equation which is derived using power series Vout = a1Vin + a2Vin 2 + a3Vin 3 +L as shown in [7] with two input tones including the input-referred offset voltage, Vin = Vos ,in + A cos( 1t ) + A cos( 2 t ) .
IIP3 = 10 log 2Vos ,in (VIIP 2 2Vos ,in ) / 50 + 30

(2.14)

where VIIP2 is the input-referred second-order intercept voltage, VIIP3 the input-referred third-order intercept voltage. To calculate IIP3 based on the measured DC offset and calculated IIP2, the measured output DC offset voltage should be referred to the input using the following equation
Vos ,in = Vos , IM 2 Av ,diff

(2.15)

40

where Av,diff is the gain of the differential circuit as shown Figure 2-21. The IIP2 and IIP3 are calculated with Eq. (2.14) and Eq.(2.15), respectively and compared to the simulated and measured results in Figure 2-22. The line with circles represents the simulated results, the line with squares the calculated ones, and the line with triangles the measured ones. The measured results show relatively better matching with the calculated results from the measured DC offset voltage rather than the simulated results. This comparison proves promising usefulness of the DC offset in differential circuits to estimate nonlinearity by simply measuring DC offsets since the calculated IIP3 is placed in between the simulated and measured results.
30 25 20 15 IIP2 & IIP3 [dBm] 10 5 0 -5 -10 -15 -20 -25 -30 1.5 2 2.5 3 3.5 4 4.5 Frequency [GHz] 5 5.5 6 6.5

IIP2

IIP3

Figure 2-22. Nonlinearity of simulated, measured, and calculated results.

2.7 Summary The low-power wideband receiver with active balun is realized using 0.13-m CMOS technology. The compact LNA with transformer-based input matching is realized, followed by the active balun to provide fully differential circuit for the folded mixer. The

41

folded mixer topology utilizing PMOS devices in the switching stage and broadband RF chokes for biasing is shown to be an effective technique for both low-voltage and wideband operation. The key sources to the receiver DC offset are measured systematically using a multi-step procedure under different excitations. The usefulness of DC offset in the receiver is proven to estimate nonlinearity without measuring IIP2 and IIP3 which requires expensive RF equipments. The low-power, high-performance wideband down-conversion mixer is realized to suppress the impact of device mismatch on DC offset. The key sources to the mixer DC offset are measured systematically using a multi-step procedure under different excitations.

42

Chapter 3

Equation Chapter (Next) Section 1

Wideband Linearity Enhancement Techniques for LNAs

3.1 Introduction The linearity of an LNA using a Si bipolar junction transistor [27] was improved by low-frequency low-impedance base termination without degrading gain or noise figure (NF). This technique is not effective for FET amplifiers if a complementary metal oxidesemiconductor (CMOS) is biased in a strong inversion region. The feed-forward linearization technique [28] was used to achieve very high linearity in a CMOS LNA. However, this technique is sensitive to device mismatch between the main and auxiliary devices as well as error in the signal scaling. Inherently an FET has a high IIP3 peak called a sweet spot in a moderate inversion region [29], [30]. This sweet spot can be reached by biasing the transistor gate such that the third-order derivative of its DC transfer characteristic crosses zero. The sweet-spot is utilized only for low-power mobile devices of short-range, less-sensitivity standards such as cordless phones and Bluetooth due to a low unity-gain frequency. The IIP3 peak is very sensitive to a gate bias voltage, making it difficult to bias accurately to achieve a high IIP3 due to bias variations. The linearity enhancement technique called a derivative superposition (DS) [31] was proposed to reduce the IIP3 sensitivity to the gate bias voltage by extending zero crossing 43

points. This DS method was applied for an FET amplifier superposing two transistors in parallel, resulting in 10dB improvement of the IIP3 in [32]. This linearity improvement was limited by the IMD2 products and their feedbacks to the input, which were decreased substantially by reducing the source-degeneration inductance as well as cascoding the output of the linearizing amplifiers. The small degeneration inductance hinders a simultaneous power-noise input match resulting in a higher NF. To achieve a high linearity without degrading NF, two modified derivative superposition techniques [33], [34] are proposed such that the composite vector sum of the IMD2 and IMD3 is cancelled out each other. However, all the linearity enhancement techniques addressed before are proposed for narrow-band amplifiers. Highly linear broadband LNA employing noise and distortion cancellation techniques [12] is recently reported with an inductorless topology. It shows high linearity in both the 900 MHz and 2 GHz bands but low gain even with two cascade stages while consuming 17.4 mW. There are several techniques to implement a wideband LNA by employing silicon transistors, in particular, in CMOS technology. At the expense of high power consumption and large area, distributed amplifiers [35], [36], [37], [18] achieve very wideband performance by absorbing circuit parasitics and these are suboptimal for mobile applications. Wideband inductor-less LNAs are proposed in [19], [38], based on resistive feedback. These LNAs achieve NF less than 3 dB and other performance comparable to their narrowband counterparts by utilizing deeply-scaled nanoscale CMOS transistors. Thus they have extremely small die area but consume a large amount of power. As an alternative wideband topology, the common-gate amplifier was employed and low NF was achieved with noise cancellation scheme in [39], [40]. For moderate die

44

area and power consumption, two wideband LNAs with a source-degeneration inductor [10], [20] are presented based on Chebyshev BPF input matching and ladder input matching networks, respectively. Both of topologies use a shunt-peaking technique [41] for a wideband output load network and show low linearity performance since the large peaking resistor reduces voltage headroom. In addition, a wideband LNA with a transformer-based input matching network [11] is proposed as an alternative solution to implement the input matching network of compact size but still suffers from low linearity due to low voltage headroom caused by a large peaking resistor. While numerous techniques have been proposed for increasing LNA bandwidth (e.g., [10], [11]), there have been relatively few studies of linearity enhancement in wideband LNAs. For example, [12] and [4] employ noise and distortion cancellation techniques to achieve an IIP3 of about 0 dBm for small frequency spacing at 0.82.1 GHz and 0.25.2 GHz while consuming 17.4 mW and 21 mW, respectively. For highly-linear wideband LNAs, linearity enhancement and wideband circuit techniques must be used simultaneously. In Section 3.2, nonlinearity analysis methods are introduced in order to provide some intuitive tools through the extracted and derived nonlinearity coefficients described later. In Section 3.3, the precedent linearity enhancement techniques are addressed in order to make the most of them for wideband linearity enhancement techniques described in Chapter 4. 3.2 Theory for Nonlinearity Analysis Several linearity enhancement techniques have been introduced in recent years: derivative superposition (DS) [32], [42], modified derivative superposition (MDS) [33],

45

and alternative derivative superposition [34]. The details of linearity enhancement techniques including the inherent IIP3 peak of a CMOS transistor are addressed below with conceptual diagrams and simulated results after defining several terminologies and expressions. 3.2.1 DC Theory for Nonlinearity Analysis

For RF applications, a MOSFET transistor in Figure 3-1 (a) is usually biased in saturation region and its small-signal equivalent circuit can be represented as shown in Figure 3-1 (b).

(a)

(b)

(c) Figure 3-1. (a) A MOSFET transistor, (b) its incremental model including back-gate effect, and (c) simplified equivalent circuit with the assumption that the nonlinearity is weak and memoryless. 46

One contributor to the distortion of the transistor arises from the nonlinearity of its transconductance in saturation. When the nonlinearity of the transistor is assumed to be weak and memoryless1, its small-signal equivalent circuit gets simplified as shown in Figure 3-1 (c) by neglecting the other MOSFET model parameters such Cgd, gmb, and so on. The nonlinearity of the transconductance is segmented into three nonlinear DC transfer functions, with the assumption that the three nonlinearity terms are sufficient. The drain current can be expressed in terms of its small-signal gate-source bias voltage vgs by a power series around the DC operating point
id ( v gs ) = g 1 v gs + g 2 v gs 2 + g 3 v gs 3 + L

(3.1)

where g1 is the small-signal transconductance, and g2 and g3 the higher order nonlinearities as defined in Eq. (3.2).
g1 = I D VGS g2 = 1 2 ID 2 VGS 2 g3 = 1 3 I D 6 VGS 3

(3.2)

In particular, g3 should be carefully considered since it represents the IMD3 and thus determines the IIP3. The g3 depending on a gate-source voltage VGS changes polarity from positive to negative when a device transitions from the weak and moderate inversion regions to the strong inversion region. The input amplitude at the IIP3 point is given by the following equation [43].
AIIP3 = 4 g1 3 g3

(3.3)

If the circuit under consideration contains no capacitors or inductors or the frequencies of interest are so low that capacitors and inductors do not play a role yet, it is referred to as a memoryless circuit.
1

47

As we can see in Eq. (3.3), the AIIP3 depends on only the DC transfer characteristic of the transistor itself without including its own circuit parasitics and external circuit components such as source-degeneration inductance. It implies that the nonlinearity expression based on the DC transfer characteristic cannot provide a good IIP3 estimation since it does not account for the high frequency effects such as feedback and coupling. Therefore, a new approach that includes the effects of frequency and memory is inevitable to be introduced for RF applications. 3.2.2 Frequency-Domain Theory for Nonlinearity Analysis

The DC theory for the nonlinear analysis cannot capture memory effects in the frequency domain. As an alternative, the Volterra series2 [44] is introduced with the harmonic input method [45] not only to capture the memory effect of devices such as capacitors and inductors, but also to provide insight in a circuit design by derived analytical expressions before a numerical iteration in simulation is performed. The details of the Volterra series and harmonic input method are introduced in Appendix A, in addition to the convolution and Taylor series. Some details of the Volterra series will be revisited below both to provide fundamental understandings of the Volterra series briefly and to define a new Volterra coefficient as a consolidated parameter in the frequency domain. If a nonlinear device is in weakly nonlinear region, the Volterra series can be used to evaluate its nonlinearity. First of all, as used in the DC theoretical analysis, the power series in Eq. (3.1) for the output current holds. On the other hand, the gate-source voltage

The Volterra series and Volterra theorem were developed in 1887 by Vito Volterra, an Italian mathematician and physicist.

48

of each nonlinear device is modeled by truncated Volterra series in terms of an excitation voltage v s
vgs = A1 ( s ) o vs + A2 ( s1 , s2 ) o vs2 + A3 ( s1 , s2 , s3 ) o vs3

(3.4)

where A1 , A2 , A3 are the first-, second-, and third-order Volterra kernels in the frequency domain and is a Volterra operator. Thus the output current of a FET amplifier can be represented by the Volterra series in terms of excitation v s in the time domain:
i( vs ) = C1( s ) o vs + C2 ( s1 , s2 ) o vs 2 + C3 ( s1 , s2 , s3 ) o vs 3

(3.5)

where C n ( s1 , s 2 ,K , s n ) represents the frequency domain form of the nth-order Volterra kernel, s is the Laplace variable, and the operator represents the magnitude and phase change of the input spectral components of v s by the nth-order Volterra kernel,
C n ( s1 , s 2 ,K , s n ) , when multi-tone sinusoidal signals are excited as an input signal.

With two-tone excitation,


v s = A[cos( a t ) + cos( b t )] ,

(3.6)

the input amplitude AIIP3 at the IIP3 point is given by the following form in the frequency domain:
AIIP 3 = C1 ( sa ) 4 3 C3 ( sb , sb , sa )

(3.7)

where C1 ( s a ) represents the Volterra kernel of the g1 at the frequency of a when only a single tone of the frequency a is excited and C3 ( sb , sb , sa ) does the Volterra kernel of the g 3 at the frequency of 2b a when three tones of the frequencies a , b , and

b are excited theoretically. In practical simulation or measurement, only two-tone

49

sinusoidal signals, a and b , need to be excited to create the second-order IMD product. The IIP3 expression in Eq. (3.8) represents the available power of the signal generator at the third-order intercept point with reasonable input impedance match.
2 AIIP C1 ( sa ) 1 3 = 8 Re [ Z1 ( sa )] 6 Re [ Z1 ( sa )] C3 ( sb , sb , sa )

IIP3 =

(3.8)

To extract the Volterra kernels C1 and C3 for the IIP3 calculation in the analytical expression of Eq. (3.8), we need not only to model nonlinearities associated with the drain current of the FET amplifier by its gate-source voltage as written in Eq. (3.1), but also to capture the effects of memory and frequency given by the expression of Eq. (3.4) through the input matching network. Then using the harmonic input method, the wanted Volterra kernels can be derived with the equivalent circuit of the FET amplifier. The whole IIP3 derivation procedure is summarized in the flow chart of Figure 3-2. The expressions of the current and voltage in the chart refer specifically to the derivative superposition topology. However, the general procedure can be applied to any other topologies. Even though the IIP3 expression can take into account the memory and frequency effects of a capacitor and inductor in the circuit, it cannot capture those effects of a transistor itself since the transistor is modeled with only the gate-source capacitance and dependent drain current source. It means that when Eq. (3.2) is introduced for the IIP3 expression we have the following assumptions: 1) The body effect is negligible, g m b 0 . 2) Except the gate-source capacitance
C gs

, all parasitic capacitances are zero.

50

3) The

C gs

is bias-independent.

4) The gate and source resistance of a transistor are ignored. 5) No channel length modulation, i.e., g o .

Figure 3-2. Flow chart for the derivation of an IIP3 derivation.

51

3.2.3

Frequency-Dependent Nonlinearity Coefficients

To capture those effects excluded in the previous IIP3 derivation, a frequencydependent nonlinearity coefficient 3 is newly introduced and defined. Before the frequency-dependent nonlinearity coefficient is defined in the frequency domain, the line spectrum due to the harmonic and intermodulation distortion is shown in Figure 3-3 when two sinusoidal signals are excited. When two sinusoidal signals are applied at the input of a nonlinear circuit, there are two, six, and ten different mixes, respectively, contained in the first-order, second-order, and third-order outputs [30]. The corresponding positive output spectrum is shown in Figure 3-3 with the order number.
(1) (1) (3) (3) (2) (2) (3) (3) (2) (2) (2) (3) (3) (3) (3)

22 + 1

2 1

2 1 2

2 2 1

2 1

1 +2

2 2

3 1

21 +2

3 2

Frequency

Figure 3-3. Line spectrum of the positive frequency terms with two-tone input signals.

Even though the Volterra kernels can capture the memory effects of external components such as a capacitor and inductor, they basically depend on the DC transfer characteristics of the nonlinear device, g1 , g2 , and g3 defined in Eq. (3.2). The frequency dependence of the device nonlinearity itself is not taken into account for the IIP3 calculation, resulting from neglecting the other device parasitic shown in Figure 3-1

The notation and relationship between the Volterra kernel and frequency-dependent nonlinearity coefficient are summarized in the Appendix D.
3

52

(b) except Cgs and gmvgs. To include the frequency and memory effects on the IIP3, a new RF transfer function called the frequency-dependent nonlinearity coefficients of a nonlinear device itself is introduced below in conjunction with a Harmonic Balance (HB) analysis. Based on the given spectrum components in Figure 3-3, we obtain the phasor forms of the drain current and gate-source voltage in each spectrum of the nonlinear device. As an example, the frequency-dependent nonlinearity coefficients of the FET transistor MB in Figure 3-1 (a) will be first derived below with ac voltages and currents in the frequency domain. The ac gate-source voltage of the MB is defined in Eq. (3.9)
vgsB , HB1, f 1 = vgB , HB1, f 1 vsB , HB1, f 1

(3.9)

where the subscripts represent the node (g:gate, s:source) and device name (B:MB), the order number of the HB ( HB1: 1st order), and the chosen frequency. Likewise, the firstorder, second-order, and third-order frequency-dependent nonlinearity coefficients 4 are defined in a ratio of the ac drain current to the ac gate-source voltage at each output frequency drawn in with the assumption that f1 and f 2 are very closely located.
idB , HB1, f 1 vgsB , HB1, f 1 idB , HB1, f 2 vgsB , HB1, f 1 idB , HB 2,2 f 1 vgsB , HB1, f 1 idB , HB 2,2 f 2 vgsB , HB1, f 1

g1B , HB1, f 1 =

, g1B , HB1, f 2 =

(3.10)

g 2 B , HB 2, f 2 f 1 =

idB , HB 2, f 2 f 1 vgsB , HB1, f 1 idB , HB 2, f 1+ f 2 vgsB , HB1, f 1

, g 2 B , HB 2,2 f 1 =

(3.11)

g 2 B , HB 2, f 1+ f 2 =

, g 2 B , HB 2,2 f 2 =

(3.12)

Frequency-dependent nonlinearity coefficient and Volterra kernel are used differently here to make difference between kernels for a device itself and for a circuit, respectively. The frequency-dependent nonlinearity coefficient is written in the form of g1, g2, and g3 with some added subscripts. The Volterra kernel is written in the form of A1, B2, C3, and so on with some added subscripts as shown in Appendix B, C, and D.
4

53

g 3 B , HB 3,2 f 1 f 2 =

idB , HB 3,2 f 1 f 2 vgsB , HB1, f 1 idB , HB 3,2 f 1+ f 2 vgsB , HB1, f 1

, g 3 B , HB 3,2 f 2 f 1 =

idB , HB 3,2 f 2 f 1 vgsB , HB1, f 1 idB , HB 3,2 f 2 + f 1 vgsB , HB1, f 1

, g 3 B , HB 3,3 f 1 =

idB , HB 3,3 f 1 vgsB , HB1, f 1 idB , HB 3,3 f 2 vgsB , HB1, f 1

(3.13)

g 3 B , HB 3,2 f 1+ f 2 =

, g 3 B , HB 3,2 f 2 + f 1 =

, g 3 B , HB 3,3 f 2 =

(3.14)

Using all the extracted frequency-dependent nonlinearity coefficients, the first-, second-, and third-order composite frequency-dependent nonlinearity coefficients are calculated by summing all the same order coefficients and dividing by the number of the coefficients as in Eq. (3.15)-(3.17).
g1B , HB = ( g1B, HB , f 1 + g1B , HB , f 2 ) / 2
g 2, BHB = 1 ( g1B, HB, f 2 f 1 + g1B, HB,2 f 1 + g1B, HB, f 1+ f 2 + g1B, HB,2 f 2 ) 4

(3.15)

(3.16) (3.17)

g3, HB = ( g3, HB,2 f 1 f 2 + g3, HB,2 f 2 f 1 ) / 2

With the newly defined frequency-dependent nonlinearity coefficients, the Eq. (3.1) is re-defined for the frequency-domain theory.
2 3 id , HB (vgs ) = g1, HB ( s ) o vgs + g 2, HB ( s1 , s2 ) o vgs + g3, HB ( s1 , s2 , s3 ) o vgs

(3.18)

The frequency-dependent nonlinearity coefficients are used in the following section to show the frequency and memory effects on the IIP3 in dBV, in comparison with the IIP3 in dBV from the DC transfer characteristics. Lets look at linearity enhancement techniques and evaluate them based on the introduced frequency-dependent nonlinearity coefficients. 3.3 Linearity Enhancement Techniques

54

To implement a linearity-enhanced wideband LNA, both linearity enhancement techniques and wideband circuit design techniques need to be incorporated relevantly at the same time. Since the precedent linearity enhancement techniques are intended for a narrow-band LNA, two different approaches are developed in this proposal to achieve high linearity over a wide frequency range. In the first technique, a self-biasing currentreuse technique is proposed to overcome the limitation due to the peaking resistor in shunt-peaking techniques [10], [21] which are generally utilized in a wideband output load network with relatively large resistance. The large resistor limits the voltage headroom in the main and cascode transistors. Thus the small voltage headroom does not only prevent achieving an IIP3 peak point, but also making the IIP3 very sensitive to process variation as well as bias variation. Using the proposed current reuse technique, a PMOS in conjunction with a RF choke inductor broadens the operating gate bias range which guarantees high linearity without impacting the voltage headroom. In the second technique, two IIP3 peaks can be realized by merging and modifying two different linearity enhancement techniques [33], [34]. Each topology is optimized at the lower and upper frequency boundaries to get relatively flat linearity response over the designed frequency range. The proposed linearity enhancement technique is analyzed by using a Volterra series and verified by simulated results. To achieve a wide bandwidth in conjunction with enhanced linearity, a commonsource LNA topology with a source-degeneration inductor is preferred with proper input and output matching networks as the source-degeneration inductor should be used to cancel out the second- and third-order intermodulation distortion products with the

55

derivative superposition technique while providing the required real part of the input impedance at around 50 . 3.3.1 Sweet Spot

A MOSFET has a sweet-spot of the IIP3 when it is in a transition condition from a weak and moderate inversion region to a strong inversion region. Since the drain current of a MOSFET depends exponentially on the gate and drain voltage in the weak and moderate inversion regions, these regions are not preferred for highly-linear circuits [46]. Furthermore, the cut-off frequency fT in the moderate inversion region is often too low to achieve high frequency operation. Interestingly, the third-order intercept point IIP3 of the drain current shows a significant peaking or sweet-spot in the moderate inversion region of a MOSFET amplifier where the third-order derivative term denoted as g3 in Eq. (3.2) crosses zero. This peaking occurs as the dominant mechanism of the drain current changes from diffusion, a largely exponential behavior, to drift, a slightly less than square-law behavior [47]. The simplified schematic of a common-source (CS) amplifier with a sourcedegeneration inductor is shown in Figure 3-4 (a) and the first-, second-, and third-order DC transfer characteristics are shown in Figure 3-4 (b).

56

1 0.8 0.6 gm1, gm2, gm3 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 0.3 0.35
g

gm1 gm2 gm3

0.4 0.45 0.5 Gate Bias Voltage [V]

0.55

0.6

(a)
30 25 20 15 IIP3 [dBm or dBV] 10 5 0 -5 -10 -15 -20 0.3 0.35

g Simulated IIP3HB w/ Cgd Calculated IIP3DC w/ Cgd Calculated IIP3HB w/ Cgd Calculated IIP3HB w/o Cgd

(b)

0.4 0.45 0.5 Gate Bias Voltage [V]

0.55

0.6

(c) Figure 3-4. (a) Simplified schematic of a CS amplifier, (b) its DC transfer characteristics, and (c) IIP3 plots.

The calculated IIP3 in dBV shows the peaking at the zero-crossing point of the thirdorder DC transfer characteristic, gm. The simulated IIP3 peak at 1 GHz is reduced and shifted due to the memory effects caused by the device parasitic and degeneration inductance. Using the DC transfer characteristics has difficulty in determining the transistor size and estimating the IIP3 since the simulated IIP3 in the circuit is deviated from the IIP3 calculated by the DC parameters.

57

To show the memory effect of the gate-drain capacitance Cgd, the model files are modified, in other words, the Cgd card is intentionally turned off. As we can see in Figure 3-4 (c), the simulated IIP3 curve with Cgd matches well the calculated IIP3 from frequency-dependent nonlinearity coefficients extracted by using HB simulation. However, the calculated IIP3 curve without Cgd doesnt match the simulated IIP3 curve, implying that the Cgd affects the nonlinearity performance of the transistor itself, with respect to the gate bias voltage. However, the IIP3 peak without the Cgd doesnt deviate that much from the IIP3 peak with the Cgd. To see the effect of the source degeneration inductance on the IIP3, the inductance is changed and the IIP3 values are calculated with the DC and frequency-domain transfer characteristics as shown in Figure 3-5. The black line represents the IIP3 calculated using the DC transfer characteristic and the blue and red lines do the IIP3 calculated using the frequency-dependent nonlinearity coefficients defined in Eq. (3.9)-(3.17) at 0.1 GHz and 1 GHz, respectively. The IIP3 degradation is becoming critical due to larger feedback of intermodulation products as both inductance and frequency increase, resulting in no IIP3 peak at 1 GHz with 1 nH inductance as shown in the right-bottom graph of Figure 3-5. On the other hand, the IIP3 peak value in this moderate inversion region is very sensitive to bias variations and furthermore the IIP3 peaking can be achieved over a very narrow bias range, while limiting an operating frequency due to the low unity-gain frequency in the moderate inversion region.

58

IIP3HB of CS Amplifier w/o Ls 25 20 15 IIP3 [dBV] 5 0 -5 -10 -15 -20 0.4 0.45 0.5 0.55 Gate Bias Voltage VgMT [V] 0.6 IIP3 [dBV] 10 25 20 15 10 5 0 -5 -10 -15

IIP3HB of CS Amplifier w/ Ls=0.1nH DC 0.1GHz 1GHz

-20 0.4

0.45 0.5 0.55 Gate Bias Voltage VgMT [V]

0.6

IIP3HB of CS Amplifier w/ Ls=0.5nH 25 20 15 IIP3 [dBV] 5 0 -5 -10 -15 -20 0.4 0.45 0.5 0.55 Gate Bias Voltage VgMT [V] 0.6 IIP3 [dBV] 10 25 20 15 10 5 0 -5 -10 -15

IIP3HB of CS Amplifier w/ Ls=1nH

-20 0.4

0.45 0.5 0.55 Gate Bias Voltage VgMT [V]

0.6

Figure 3-5. IIP3 plots with different source degeneration inductances at 0.1 and 1 GHz.

3.3.2

Derivative Superposition

To reduce the IIP3 sensitivity to bias variations as well as obtain the high unity-gain frequency fT of the FET amplifier, the derivative superposition method was proposed in [31] as shown in Figure 3-6 (a). The DS topology utilizes the fact that the polarity of the third-order derivative term g 3 changes from positive to negative when the gate bias voltage VGS changes from the weak and moderate inversion regions to the strong inversion region. If two common-source FETs are configured in parallel such that with proper offset voltage the positive and negative polarity regions are aligned to cancel out each other at a certain gate bias range, the corresponding g 3 will be driven to zero. In other words, the third-order nonlinearity term becomes zero and the IIP3 does infinite. 59

IN

OUT MA MB

LB VgST VgMT

(a)

(b)
Vs Z1(s) + VgA + CA VgB iB (VgB) gBVgB V2 LB Zin(s) iA (VgA) gAVgA G Vg i(Vs)=iT D

CB

(c) Figure 3-6. (a) Schematic of the derivative superposition method, (b) conceptual diagram of DS and (c) small-signal equivalent circuit.

60

At the IIP3 peak, the FET MA is in the weak inversion region with the positive g3 A and the FET MB is in the strong inversion region with the negative g 3 B such that the composite g 3 should be near zero as shown in Figure 3-7 (b). The improvement of IIP3 due to expanded zero-crossing range of the composite g 3 can be obtained as shown in Figure 3-7 (d). Practically this IIP3 peak happens only at low frequencies where the parasitic doesnt contribute to nonlinearity significantly. At high frequencies, the source degeneration inductance provides a feedback path for the total drain current iT to the input gate-source voltage, resulting in additional high-order intermodulation products.

Figure 3-7. Derivative superposition method. (a) 1st-, 2nd-, and 3rd-order power series coefficients of the auxiliary transistor, (b) 1st-, 2nd-, and 3rd-order power series coefficients of the main transistor, (c) 3rd-order power series coefficients and superposition, (d) Calculated IIP3 dBV.

61

To illustrate imperfect cancellation of the composite third-order intermodulation products, the conceptual diagram is drawn in Figure 3-6 (b). There is still the IMD2 component which contributes to the IMD3 through feedback to the input even though the IMD3 components cancel out each other perfectly. The analytical explanation for an imperfect cancellation is given in the Appendix B using Volterra series analysis and the derived IIP3 expressions are presented in Eq. (3.19). In the derived IIP3 expression, the IIP3 represents the available power of the signal generator at the third-order intercept point with reasonable input impedance matching condition Z1* ( s ) = Z in ( s )
IIP3 =
2 AIIP 1 C1 ( sa ) 3 = 8 Re [ Z1 ( sa ) ] 6 Re [ Z1 ( sa ) ] C3 ( sb , sb , sa )

(3.19)

where

Zin ( s) = sLB +

1 g L + 1B B . s ( C A + CB ) C A + CB

(3.20)

The IIP3 of the LNA in Figure 3-6 (a), obtained from the Volterra analysis of the equivalent circuit of Figure 3-6 (c), is given by
IIP3 = 4 2 g12B LB ( C A + CB ) 3

(3.21)

where
2 2 g2 B 3 + j 2 ( CA + CB ) + Z1 ( j 2 )

= g3 A + g3 B

g1B +

1 j 2 LB

CA + CB LB

g3 A + g3 B

2 2 g2 B 3g1B 1 +

1 1 j 2 g1B LB

(3.22) As shown in Eq. (3.21) and (3.22) , the IIP3 doesnt become infinite due to the contribution of the second-order IMD products in Eq. (3.22) even though the composite 62

third-order term g3 = g3 A + g3 B of the transconductance becomes zero. Using the DS linearity enhancement technique, the IIP3 is not able to be zero due to both the amplitude contribution of the second-order term and the phase change of the source degeneration inductor shown in the third-term of Eq. (3.22). The higher frequency

and the larger

inductance LB , the smaller IIP3, as the IIP3 is inversely proportional to both of the parameters. As shown in Figure 3-8, the DS method does not provide IIP3 peaking at higher frequencies which the effect of the circuit reactance is not negligible as frequency increases.
IIP3HB of DS Amplifier w/ Ls=0.1nH 25 20 15 10 IIP3 [dBV] 5 0 -5 -10 -15 -20 0.4 0.42 0.44 0.46 0.48 0.5 0.52 0.54 0.56 Gate Bias Voltage VgMT [V] 0.58 0.6 DC 0.1GHz 1GHz 2GHz 4GHz 7GHz

Figure 3-8. IIP3 plots in different frequencies with 0.1 nH source inductance.

The calculated IIP3 curve using the DC transfer characteristic doesnt capture any memory effect on the IIP3. On the other hand, the calculated IIP3 curves using the frequency-dependent nonlinearity coefficients extracted from the HB analysis show the IIP3 degradation in the same circuit as the frequencies increase. As a result, the DS

63

method doesnt achieve a high IIP3 value at the high frequencies. Therefore, another linearity enhancement technique is inevitable. 3.3.3 Modified Derivative Superposition [33]

The source degeneration inductance in a common-source amplifier provides a feedback path for the drain current to the gate-source voltage. The feedback becomes stronger with increasing frequency. The second-order harmonics are fed back to the input and mix with the fundamental input signal. Eventually the feedback of the second-order harmonics generates additional third-order intermodulation products. From Eq. (3.22), it is evident that Z1 (2 ) should be increased to reduce the secondorder contribution to IMD3. In order to enhance the IIP3 at higher frequencies, the second-order contribution to IMD3 can be utilized so that it makes the same amplitude and opposite phase with the third-order IMD contribution. To accommodate this concept, the modified derivative superposition (MDS) method was proposed as shown in Figure 3-9 (a) [33]. The sources in two parallel FETs are connected to different nodes of the source inductors to adjust the magnitude and phase of the composite third-order contributors including the second-order contributor. The source of the FET MA are connected to the common node of the two inductors L A and LB to adjust the magnitude and phase of its third-order term g 3 A contribution to the IMD3 with respect to the g 2 B and g 3 B contributions of the FET MB. Graphically, the conceptual vector diagram is plotted in Figure 3-9 (b) to show how the composite third-order IMD terms cancel out each other.

64

IN

OUT MA MB LB

LA VgST VgMT

(a)
Composite 3rd-order contribution Im g3AVgA3 g3BVgB3 =tan-1(LBg1B) Re g2BVgB2

(b) Figure 3-9. (a) Schematic of the modified DS method and (b) conceptual vector diagram.

As we can see in the first term of Eq. (3.24), the g 3 A magnitude and phase are mainly tuned by LB and g1B . The second-order contribution is adjusted by L A , LB , and g1B as shown in the denominator of the third term in Eq. (3.24)
IIP3
2 2 4g1 B [L A (C A + C B ) + L B C B ] 3

(3.23)

65

= g 3 A (1 + jLB g 1B ) 1 + (LB g 1B )2 1 +

2 LB C B 2g2 B + g 3B L A (C A + C B ) + L B C B 3 g 1B

1 1+ j 2 (L A + L B )g 1B 1

(3.24) where the input impedance has the following expression while assuming a conjugate impedance matching,

Zin (s) = sLA +

s ( CA + CB + sg1BCA LB + s 2CACB LB )

1 + sg1B ( LA + LB ) + s 2CB LB

(3.25)

The modified superposition method basically intends to be used for a narrow-band LNA. In order to evaluate this technique for a wideband low-noise amplifier, the wideband LNA is designed and the IIP3 is simulated over the frequency range. It shows higher IIP3 values at the specific frequency but poor IIP3 values at other frequencies. Furthermore, the MDS shows that the IIP3 is very sensitive to the gate-source bias voltage as well as the power supply voltage since the big resistance in shunt-peaking technique limits voltage headroom. 3.3.4 Alternative Derivative Superposition [34]

The two main drawbacks of the MDS method are pointed out in [34]. The most important drawback is that the additional weak inversion transistor added in parallel to achieve linearity degrades NF due to its high gate induced current noise which is added to the input [33]. The other one is that the auxiliary NMOS adds more input parasitic capacitance and as a result reduces the operating frequency range. Furthermore, it changes the input matching.

66

As an alternative to minimize the side effects of the auxiliary NMOS, the sourcetapped auxiliary transistor is added to adjust the composite IMD3 term equal in magnitude and opposite in phase as shown in Figure 3-10 (a).

(a)
ID

Ld

Vout Vin Vs (ID+IM3) LS f(Vs)=IM3 G D

(b) Figure 3-10. (a) Schematic of the alternative DS method and (b) conceptual diagram.

The second-order nonlinearity coefficient g 2 B appears in Eq. (3.27) due to the feedback effect. The effect of g 2 B on IIP3 has become independent of any circuit parameters, thus resulting in a constant value. The third-order derivative of the DC

67

transfer characteristic of the auxiliary transistor MA is adjusted to achieve higher IIP3 by choosing appropriate inductance values of L A and LB .
IIP3 = 1 6 Re (Z s ( s )) A1 ( s )
2

g1B

(3.26)

= g 3B

2 2 g2 2 2 + s L AC A B + g 3 A n ( s ) n( s ) 3 g 1B 2 1 + s 2 L AC A

(3.27)

n( s ) =

sL B (g 1 B + sC B ) 1 + sC A (sL A + sL B )
1 + s 2 C A (L A + L B ) 2s 2 LB C B 1 + s 2 L AC A

(3.28)

A1 ( s ) =

(3.29)

This linearity enhancement technique requires very large source inductance values to obtain higher IIP3 value to provide the same magnitude and opposite phase with the composite IMD3 products. The inductances of 5 nH and 1.05 nH are utilized for this topology at the 950 MHz operating frequency. These kinds of linearity enhancement techniques addressed in the section of 3.3 work only for a narrow band amplifier as the same amplitude and opposite phase can be achieved at a certain frequency. For highly linear wideband amplifiers, multiple IIP3 peaks over a wideband frequency range are inevitable for multi-standards communications due to different linearity requirements. 3.4 Summary In the modified DS technique of [33], an IIP3 peak occurs only at the operating frequency because the 2nd-order Volterra kernel of a main transistor has the same amplitude and opposite phase with respect to the composite 3rd-order Volterra kernel of

68

the main and auxiliary transistors at that frequency. The linearity enhancement technique of [34] taps an input signal of an auxiliary transistor from the source of the main transistor not only to achieve high linearity, but also to reduce NF degradation due to a gate induced current noise. These two linearity enhancement techniques [33], [34] are proposed for the narrow band LNAs. Highly linear broadband LNA employing noise and distortion cancellation techniques [4], [12] achieve an IIP3 of about 0 dBm for small frequency spacing at 0.8 2.1 GHz and 0.25.2 GHz while consuming a large amount of power of 17.4 mW and 21 mW, respectively. In the next chapter, low-power wideband linearity enhancement techniques are proposed with nonlinearity analysis by the Volterra series and newly introduced frequency-dependent nonlinearity coefficients.

69

Chapter 4

Equation Chapter (Next) Section 1

Proposed Highly-Linear Wideband LNA

To achieve a wide bandwidth in conjunction with high linearity, a common-source LNA topology with a source-degeneration inductor is chosen with proper input and output matching networks since the source-degeneration inductor should be used to cancel out the IMD2 and IMD3 products with the DS technique. To implement a linearityenhanced wideband LNA, both linearity enhancement techniques and wideband circuit design techniques need to be incorporated relevantly at the same time. Since the precedent linearity enhancement techniques are intended for a narrow-band LNA, two approaches (WBDS and SBCR) are proposed to achieve high linearity over a wide frequency range. First of all, two IIP3 peaks can be realized by merging and modifying two linearity enhancement techniques [33], [34]. Linearity of each topology is optimized at the lower and upper frequency boundaries to get relatively flat linearity response over the designed frequency range, respectively. Second, the SBCR technique is proposed to overcome the limitation due to the peaking resistor in shunt-peaking techniques [41] which are generally utilized in a wideband output load network with relatively large

70

resistance. The large resistor limits the voltage headroom in the main and cascode transistors. The proposed linearity-enhanced wideband LNA topology is analyzed by the Volterra series with the newly introduced frequency-dependent nonlinearity coefficients. Two LNAs using the proposed topology are designed in a 0.13 um CMOS process and demonstrated with simulated and measured results to confirm the effectiveness of the proposed topology. 4.1 Introduction As a benchmark to implement a linearity-enhanced wideband amplifier, both wideband design techniques and linearity enhancement techniques are addressed in detail in the preceding chapter. Based on the limitations of the precedent designs, two different techniques are proposed and analyzed to achieve the linearity-enhanced wideband LNA in this chapter. For highly-linear wideband LNAs, linearity enhancement and wideband circuit techniques must be used simultaneously. This chapter starts with a source-degenerated topology with input and output matching networks. Two techniques, namely the WBDS and SBCR techniques are proposed to achieve high linearity over a wide frequency range. In the WBDS, two IIP3 peaks are realized by merging and modifying two linearity enhancement techniques [33], [34]. Then linearity of each is optimized at the bandwidth boundaries to obtain a flat linearity response over the designed frequency range. The SBCR is then used to increase the headroom in a shunt-peaking load. The proposed linearity-enhanced wideband LNA topology is analyzed using a Volterra series and its distortion cancellation is visualized by frequency-dependent nonlinearity

71

coefficients, leading to a 26 GHz design in 0.13m CMOS. Measurement results show linearity is enhanced. 4.2 Self-biasing Current Reuse (SBCR) Technique For a linearity-enhanced topology, the cascode LNA with a source-degeneration inductor is preferred as the second- and third-order nonlinear terms cancel out each other through the degeneration inductors as shown in Figure 3-9 and Figure 3-10. In the LNA topologies with wideband input matching networks, a shunt-peaking load network is preferred in many cases [10], [20] to provide a wideband output matching. Due to the large shunt-peaking resistance, there is no enough voltage headroom to guarantee high linearity as supply voltage scales down. In these two cases mentioned above, around 400450 mV across the peaking resistors is necessary to provide moderate RF performance such as gain and IIP3, regardless of the supply voltages. In a 1.2-V power supply, only 0.75-0.8V voltages can be used to bias the main transistor and cascode transistor and thus they are not high enough to keep two stacked transistors in a highly linear region when a gate bias voltage is swept to control linearity depending on required IIP3 specifications in a certain operating condition.

72

I OUT

C VIN
Vgs=0.47V

M
W=192um L=120nm

Figure 4-1. (a) Simplified common-source amplifier and (b) IIP3 vs. Vds.

To verify the effect of limited voltage headroom itself on the linearity, a commonsource amplifier in Figure 4-1 (a) is designed and simulated without a sourcedegeneration inductor. The blue line with circle marks in Figure 4-1 (b) represents the calculated IIP3 in dBV at 0.1 GHz and the red line with triangle marks do the IIP3 in dBV at 1 GHz. As the drain-source voltage increases, the IIP3 is improved continuously and then starts saturated beyond around 0.65 V due to velocity saturation and hot carrier effect even though more current is burning. Furthermore, in order to verify the effects of the limited voltage headroom in a practical circuit, the common-source amplifier with the linearity-enhanced topology shown in Figure 3-9 (a) is designed as shown in Figure 4-2 (a) and simulated. The effect of the drain-source voltage on the IIP3 in dBm is shown in Figure 4-2 (b). The current consumption is plotted with the straight red line. As we can see, approximately a 0.6-V drain-source voltage is necessary to obtain the IIP3 peaking. Therefore the techniques which can alleviate the supply requirement need to be implemented for a linearity-enhanced LNA.

73

(a) IIP3 vs. Vds plot.

(b)

Figure 4-2. The effects of voltage headroom on IIP3: (a) simplified schematic and (a)

A current-reuse technique [48] is used for a mixer to enable a low-voltage operation and obtain better performance in comparison with a conventional Gilbert cell mixer [49]. In the current-reuse technique, some portion of the driver-stage bias current is bled through an additional PFET transistor to reduce the voltage drop across the load resistor, while keeping or increasing the bias current to get better performance under the same supply voltage as the IIP3 and conversion gain of the mixer are proportional to the square root of the bias current [50]. For a wideband LNA design, the current-reuse technique shown in the red-color portion of Figure 4-3 is employed to provide large enough voltage headroom without limiting the choice of load resistance by injecting approximately half the total current of the main input transistor through the PFET transistor MP. Furthermore, the current flowed through the device MP is controlled at the same changing rate as the total current due to

74

the bias voltage feedback from the drain node of the cascode transistor MC, called a selfbiasing current reuse technique. The appropriate amount of current through MP can be adjusted by the resistance ratio of two resistors in the voltage divider, Ra and Rb.

Figure 4-3. LNA schematic with the self-biasing current reuse technique of the redcolored portion.

When the gate bias voltage of the main transistor MB is increased without the SBCR portion, the drain current iT of the MB is accordingly increased and then the increased whole current goes through the load resistor Rd and cascode transistor MC. The larger load current increases the voltage drop across the load resistor Rd and results in decreasing the drain voltages of the MB and MC. This causes difficulties in keeping MB and MC in saturation and thus degrades RF performance of the LNA. With the proposed SBCR technique, we can avoid a voltage headroom limitation since some portion of the drain current flows through the MP without degrading the behavior of the transistor MB. 75

Furthermore, the less current through the cascode transistor MC improves the performance of gain and NF by reducing parasitic capacitances at the drain of the transistor MB since the smaller size of the cascode transistor is required to handle less current. To prevent an ac current from flowing into the MP, the inductor Lpm is employed as an RF choke. Another advantage of this SBCR circuit is that the fed-back voltage smoothes the current increment rate through the load resistor to hold the similar voltage at the output node and thus makes the IIP3 peaks less sensitive to the gate-source bias voltage due to the small change of the drain-source voltages of MB and MC.
1.2 1.1 1 0.9
24 22 20 18

w/o SBCR w/ SBCR

Current IpT, Iss [mA] [mA]

Voltage [V] Vd, Vo [V]

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0.4 0.45

Vo

16 14 12 10 8 6 4 2 0 0.4 0.45

IT

Vd
0.5 0.55 0.6 0.65 Gate Bias Voltage VgMT [V] 0.7

IP
0.5 0.55 0.6 0.65 Gate Bias Voltage VgMT [V] 0.7

(a)

(b)

Figure 4-4. (a) Node voltages and (b) currents in the LNA with the SBCR technique.

The simulated output voltage Vo and drain voltage Vd are plotted in Figure 4-4 (a) with the total current iT and the current iP from the PMOS current source in Figure 4-4 (b). The solid line represents the voltage and current with the SBCR technique and the dotted line does those without the SBCR. The variation of the voltage Vo with the SBCR is about 0.1 V within the main gate-bias point of 0.5-0.6 V and thus 0.28 V less than that without the SBCR. It means that the nonlinearity effect due to the drain-source voltage variation of

76

the devices of MB and MC is reduced since the transconductance is less modulated by the voltage difference between the gate and the drain, predicted by the following expression.

gm =

2nCox W

1 + VDS

( L) I

(4.1)

4.3 Wideband Derivative Superposition (WBDS) Method Only one IIP3 peak at the operating frequency can be achieved with narrow band topologies [31]-[42], [33], [34] shown in Figure 3-6, Figure 3-9, and Figure 3-10 since the IIP3 peaks occurs only when the third-order Volterra kernel of the drain current of the auxiliary transistor has the same amplitude and opposite phase in the main transistor with respect to the composite Volterra kernel of its drain current. Since this cancellation occurs at a single frequency, these linearity enhancement techniques are effective for only a narrow-band LNA. Assuming that two narrow-band linearity-enhanced topologies [33], [34] are optimized independently to obtain IIP3 peak at the low frequency corner in Figure 4-5 (a) and to obtain IIP3 peak at high frequency corner in Figure 4-5 (b) and then are merged together, two independent IIP3 peaks over the wide frequency range can be achieved as shown in Figure 4-5 (c).

77

IIP3

IIP3

IIP3

+ fL f fH f

= fL fH f

(a) Aparins topology

(b) Ganesans topology

(c) Proposed topology

Figure 4-5. Conceptual diagram of linearity behavior over an operating frequency (a) at the low-frequency optimized topology, (b) at the high-frequency optimized topology, and (c) the resulting IIP3 plot.

Two different cancellation techniques [33], [34] are adopted and modified to achieve a linearity-enhanced wideband LNA as shown in Figure 4-6, with reducing the number of an inductor simultaneously. The MDS technique is employed as shown inside the black dotted box (A) for the lower frequency IIP3 optimization at the operating frequency range and the alternative DS technique is modified and added up as shown inside the blue dotted box (B) for the higher frequency IIP3 optimization. To minimize the number of source inductors, LA is shared for two different topologies while providing the functionality of adjusting the IMD2 term out of phase at two different frequencies. The circuit functionality is analyzed by the Volterra series to show how high linearity can be achieved over the wide operating frequency. The frequency-dependent nonlinearity coefficients from the HB analysis are introduced below based on the definitions of the HB frequency-dependent nonlinearity coefficients given in Eq. (3.9)-(3.17).

78

VDD VIN Input Matching Network Rd Ld VOUT MC MA MB LB LA (B)

MF

(A)

Figure 4-6. LNA schematic with wideband derivative superposition (WBDS) method.

As defined in Eq. (3.9)-(3.17), using the extracted frequency-dependent nonlinearity coefficients of each device in Figure 4-6, the first-, second-, and third-order composite frequency-dependent nonlinearity coefficients are derived by summing all the frequencydependent nonlinearity coefficients of all the devices MA, MB, and MF as in Eq. (4.2)-(4.4).
g1, HB1 = g1 A, HB1 + g1B , HB1 + g1F , HB1 g 2, HB 2 = g 2 A, HB 2 + g 2 B , HB 2 + g 2 F , HB 2 g 3, HB 3 = g 3 A, HB 3 + g 3 B , HB 3 + g 3 F , HB 3

(4.2) (4.3) (4.4)

To observe the frequency-dependent nonlinearity coefficients as a function of frequency, the size of the main and auxiliary transistors is first determined. The gate bias voltage and offset voltages are applied properly by evaluating the DC transconductance characteristics so that the third-order derivatives of the DC transconductance sum up to zero within a certain gate bias voltage. Then the frequency-dependent nonlinearity coefficients of the MA and MB in the HB simulation are extracted and shown in Figure 4-7

79

from 2 GHz to 7 GHz. In this HB simulation, the gate bias voltage of the main device MB is 560 mV and that of the auxiliary device is 428 mV, with the size 192 um/120 nm and 64 um/120 nm, respectively.
1st-order g1B_HB 2nd-order g2B_HB 3rd-order g3B_HB

g1A_HB

g2A_HB

g3A_HB

Figure 4-7. Polar plot of the frequency-dependent nonlinearity coefficients as a function of frequency.

The first-, second-, and third-order frequency-dependent nonlinearity coefficients revolve to the clockwise direction in the polar plot while the amplitude is shrinking as frequency increases. As observed in Figure 4-7, the amplitude of the third-order Volterra coefficient is much less than one of the second-order frequency-dependent nonlinearity coefficients because most of the IMD3 terms are cancelled out with the proper device sizing and biasing, based on the DC transfer characteristics of the input transistors. As

80

expected, the IMD2 terms become the main contributions on the IMD3 due to feedback to the input side and mixing up with the two first-order terms. The mechanism of the frequency-dependent nonlinearity coefficients depending on frequency can be redrawn in the conceptual vector diagram as shown in Figure 4-8. If we can use the fact that all the Volterra kernels revolve to the clockwise direction as frequency increases, two independent topologies having the IMD3 cancellation techniques can be employed to provide two IIP3 peak points at two different frequencies. The fL and fH in Figure 4-8 (a)-(b) represent the lower frequency and higher frequency with respect to the center frequency in a wide operating frequency range.
Composite 3rd-order contribution of MA & MB Im g3B,HB VgsB3 @ fL g2B,HB VgsB
2@ f L

Composite 3rd-order contribution of MF & MB Im g3B,HB VgsB3 @ fH g2B,HB VgsB2 @ fH Shrunk and rotated due to frequency Re g3F,HB VgsF3 @ fH

g3A,HB VgsA3 @ fL Re

MB & MA

MB & MF

(a) frequency and (b) at high frequency.

(b)

Figure 4-8. Conceptual vector diagram with SBCR and WBDS techniques (a) at low

4.4 Derivation of IIP3 Expression Using Volterra Series The simplified LNA schematic with the SBCR technique and WBDS method is shown in Figure 4-9 (a). To verify how the proposed topology can achieve wideband linearity at high frequencies, the detailed analysis is done in Appendix C using the Volterra series with the equivalent circuit in Figure 4-9 (b).

81

VDD VIN Input Matching Network Rd Ld VOUT MC MA MB LB LA (B)

MF

(A)

(a)

(b) Figure 4-9. (a) Simplified LNA schematic with SBCR and WBDS techniques and (b) its equivalent circuit.

In the IIP3 derivation using the Volterra series, we have the assumptions described below: 1) The body effect is negligible, g bm 0 . 2) Except the gate-source capacitance 3) The
C gs C gs ,

all parasitic capacitances are zero.

is bias-independent.

82

4) The gate and source resistance of a transistor are ignored. 5) The series resistance of an inductor is ignored as well. 6) No channel length modulation, i.e., g o . 7) Very week input signals so that higher-order IMD terms than third-order ones are negligible. 8) a b , i.e., 0 . 9) The input is conjugately matched. 10) g 1 A 0 , g 2 A 0 , g 1F 0 , g 2 F 0 . The final IIP3 expression derived by the Volterra series is given by Eq. (4.5)-(4.6). As shown in the simplified expression of Eq. (4.6), the IMD2 term becomes independent of the other circuit components. Instead, the two IMD3 terms can be adjusted by changing the two source inductances L A & L B , parasitic capacitances CA & CB , transconductance
g1B , and frequency .
IIP3 = 1 C1( sa ) 6 Re[Z 1( sa )] C3 ( sb , sb , sa )

4 g1 B 2 w 2 [LA (C A + C B ) + C B LB ] = 3
2 g 2B 2 L B ( C B + j g1B C A L A ) g 3A 1 + ( g1B L B ) [1 + j g1B L B ] + 1 + 2 2 2 3g1B LA ( CB + CB ) + CB LB + g g g L + j g1B L B 3A 1B 3F B

(4.5)

= g 3B
= g 3B

g 2B 2 + g 3A [ m( ,g1B ,C A ,C B ,L A ,L B ) + j n ( ,g1B ,C A ,C B ,L A ,L B ) g1B ] 3g1B + g 3F [ q( ,g1B ,C A ,C B ,L A ,L B ) + j r ( ,g1B ,C A ,C B ,L A ,L B ) g1B ]

(4.6)

where the input impedance has the following expression while assuming a conjugate impedance matching,
Z in ( s ) = sL A + 1 + [g 1A + g 1B + (g 1A + g 1F )(g 1B + sC B )Z m ]sL A + (g 1B + sC B ) Z m s [C A + C B + C A (g 1B + sC B )Z m ]

(4.7)

83

In

Eq.

(4.5),

the

functions,

m( , g1B , C A , CB , LA , LB )

n( , g1B , C A , CB , LA , LB )

q ( , g1B , C A , C B , LA , LB ) , and r ( , g1B , C A , CB , LA , LB ) , represent the simplified functions of the

circuit components including parasitic capacitances and DC transfer characteristics. By choosing the DC transfer characteristics and two inductances properly, the nonlinear coefficient

becomes close to zero so that the IIP3 gets larger.

4.5 Summary For highly-linear wideband LNAs, linearity enhancement and wideband circuit techniques must be used simultaneously. The basic LNA topology starts with a sourcedegenerated topology with input and output matching networks. Two techniques, namely the WBDS and SBCR techniques are proposed to achieve high linearity over a wide frequency range. In the WBDS, two IIP3 peaks are realized by merging and modifying two linearity enhancement techniques [33], [34]. Then linearity of each is optimized at the bandwidth boundaries to obtain a flat linearity response over the designed frequency range. SBCR is then used to increase the headroom in a shunt-peaking load. The proposed linearity-enhanced wideband LNA topology is analyzed using a Volterra series and its distortion cancellation is visualized by frequency-dependent nonlinearity coefficients. The proposed linearity enhancement techniques can be applied to the folded mixer to achieve high linearity such that an overall IIP3 meets a system-level linearity requirement in a multi-standard front-end receiver. For this, the tail current source of the folded mixer must be removed and replaced by two source-degeneration inductors along with two additional auxiliary transistors. Thanks to the folded mixer topology, the SBCR technique is not necessarily required to accommodate the large voltage headroom. 84

Chapter 5

Equation Chapter 5 Section 1

LNA Design, Simulated and Measured Results

5.1 Circuit Design of the Proposed LNA The LNAs with SBCR technique and WBDS method are designed to verify the effects of the proposed topology on the IIP3 over a wide frequency range in a 0.13 m CMOS process. The LNAs using the aforementioned linearity enhancement techniques are designed with two different input matching networks: one with a Chebyshev BPF input matching network and the other with a transformer-based input matching network. 5.1.1 Input Matching Network

In Figure 5-1 (a), the embedded third-order Chebyshev BPF input matching is used to resonate out the reactive part of the input impedance over a wide frequency range. On the other hand in the transformer-based matching network whose equivalent circuit is shown in Figure 5-1 (b), additional resonant tanks are used to extend the operating bandwidth. The component values of the two matching networks are summarized in Table 2.

85

(a) based matching.

(b)

Figure 5-1. Input matching network (a) Chebyshev BPF matching (b) transformer-

Table 2. Component values of the Chebyshev BPF and transformer-based matching networks. Chebyshev BPF 535 fF 775 fF 502 fF 1.69 nH Transformerbased 526 fF 601 fF 750 fF 1.91 nH Chebyshev BPF 0.51 nH 1.47 nH 1.39 nH Transformerbased 1.91 nH 2.13 nH 3.21 nH 1.26 nH

C1 C2 C3 L1

L2 L3 or LP LS Lg

5.1.2

Noise Models

One of the key design goals is to minimize the noise figure while maximizing the gain and linearity of an LNA. Since the noise performance of the LNA is so critical, an accurate noise model in a MOS transistor is essential, which implies that a long-channel MOS noise model doesnt hold any longer as process scales down. More accurate noise models have been proposed for short-channel MOS devices: one includes high-field effects [51] and the other one includes induced gate current noise [51], [52]. The most significant noise sources of a MOS transistor at high frequencies are the drain current noise and the induced gate noise. The traditional MOS noise model is modified to include an increased drain current noise resulting from high-field effects in short-channel devices. In this case, the drain noise current is expressed as in (5.1) 86

2 ind = 4kT f g d 0

(5.1)

where is a bias-dependent noise coefficient, k is the Boltzmanns constant, T is the absolute temperature, f is the bandwidth, and gd0 is the zero-bias drain conductance of the device. The coefficient is equal to 2/3 for long-channel devices in strong inversion, but is 2 to 3 for short-channel devices due to hot electron effects [51]. The drain conductance gd0 for short-channel devices is given by

g d 0 = eff Cox

W (VGS Vth ) . Leff

(5.2)

where eff is the effective electron mobility, Cox is the gate-oxide capacitance per unit area, W is the channel width, Leff is the effective channel length, VGS is the gate-source bias voltage, and Vth is the device threshold voltage. At high frequencies, an induced gate noise becomes significant, which arises from the distributed gate capacitance and channel resistance of the device as shown in Figure 5-2. The induced gate noise current is expressed
2 ing = 4 kT f g g

(5.3)

gg =

4 2 ( CoxWLeff 45 g d 0

(5.4)

where is the bias-dependent noise coefficient which has value of 4/3 in long-channel devices and increases in short channel devices and at high gate-source and drain-source voltages. Since the induced gate noise is partially correlated with the drain channel noise current, a correlation coefficient is defined as shown in (5.5) [51].
c=
* ing ind 2 2 ing ind

(5.5)

87

c can be predicted theoretically as j0.395 in long-channel devices and is purely imaginary, reflecting the capacitive coupling between the drain channel and induced gate noise sources.

Figure 5-2. Distributed gate capacitance and channel resistance at high frequencies.

The noise model in [51] can be extended for an MOS transistor in weakly inversion (WI) region [33], resulting in =1/2, =45/16, c=j0.707, and

gd 0,WI =

ID ( kT / q )

(5.6)

where ID is the drain saturation current of the weakly inversion transistor and q is the electron charge. From the equations of (5.3)-(5.6), we can get the following equation (5.7) and make an interesting observation. While the weakly inversion transistor draws a small amount drain current, its induced gate noise becomes significant since it is inversely proportional to the drain current.

i =
2 ng

16kT f 2 ( CoxWLeff 45I D

) ( kT / q )
2

(5.7)

5.1.3

Circuit Design

Two LNA prototypes employing the WBDS and SBCR techniques are designed in a 0.13 m CMOS process to verify the effectiveness of the proposed topology. One LNA 88

uses a Chebyshev matching network while the other uses a transformer-based matching network. The simplified LNA is shown in Figure 5-3 with a buffer amplifier added for test purposes. Shunt-peaking is used to widen the output bandwidth of the load network and a buffer amplifier is added for test purposes. The component values are summarized in Table 3.

Figure 5-3. Simplified schematic of linearity-enhanced wideband LNAs.

Table 3. Component values of the designed wideband LNA prototypes. MA MB MC MF MP M1 M2 Common Wu/L=2.0um/0.12um, m=32 Wu/L=2.0um/0.12um, m=96 Wu/L=2.0um/0.12um, m=32 Wu/L=2.0um/0.12um, m=32 Wu/L=2.0um/0.12um, m=480 Wu/L=2.0um/0.12um, m=32 Wu/L=2.0um/0.24um, m=96 LA LB Lpm Rd Ld Ra Rb Chebyshev BPF 0.41 nH 0.69 nH 2.90 nH 49 2.43 nH 10 K 40 K Transformer-based 0.50 nH 0.69 nH 2.90 nH 54 2.43 nH 10 K 40 K

89

In the first design step, the width of the three transistors MA, MB, and MF is determined to achieve the highest IIP3 in dBV based on the DC transfer characteristics with a 1.2V supply voltage so that the composite 3rd-order DC characteristic is set to near zero as done in the narrow band DS topologies. The IIP3 in dBV with the size of the determined transistors is simulated in DC analysis option and the IIP3 curve shows three peaks around at the gate voltage of 530 mV as shown in Figure 5-4 (d). But the coupling and frequency effects are not taken into account in the IIP3 curve. To estimate a gate bias voltage and two inductances more accurately, the frequency-dependent nonlinearity coefficients extracted from the frequency-domain analysis are used. The two inductance values of LA and LB are optimized with the lab-customized MATLAB program using the frequency-dependent nonlinearity coefficients of the transistors themselves extracted from the HB analysis. In the proposed topology, the size of the cascode transistor MC is one third of the main transistor MB since the smaller device reduces parasitic capacitance limiting bandwidth and minimizes nonlinearity of the transconductance of the cascode device MC. It can be achieved due to the SBCR technique without sacrificing gain. The PMOS device MP and two voltage-divider resistors Ra and Rb are chosen to make approximately half the total current flow through the main device MB and two auxiliary devices MA and MF, while maintaining the drain voltage of the main device at around 0.6 V. The RF choke inductance Lpm prevents an ac current from flowing reversely through the PMOS device, resulting in reducing the gain.

90

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 0.4

0.45 0.5 0.55 0.6 0.65 Gate Bias Voltage VgMT [V]

0.7

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 0.4

g1A/F, g2A/F, g3A/F

g3A, g3B, g3F, g3

0.45 0.5 0.55 0.6 0.65 Gate Bias Voltage VgMT [V]

0.7

(a)
0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 0.4 40 35 30 25 20 15 10 5 0 -5 -10 -15 -20 0.4

(c)

g1B, g2B, g3B

0.45 0.5 0.55 0.6 0.65 Gate Bias Voltage VgMT [V]

0.7

IIP3 [dBV]

0.45 0.5 0.55 0.6 0.65 Gate Bias Voltage VgMT [V]

0.7

(b)

(d)

Figure 5-4. First-, second-, and third-order DC transfer coefficients of (a) the auxiliary and (b) main transistors, (c) third-order coefficients and superposition, (d) calculated IIP3 dBV in the proposed topology.

The source-follower buffer amplifier is designed to drive an external 50- load and biased by a current mirror independently to de-embed the power consumption. The tail current source of the buffer amplifier has twice the unit length, i.e., 240 nm, to provide high output impedance. In nominal condition, the gate bias voltage of the main transistor MB, VgMT, is 560 mV and those of two auxiliary transistors, VgST, VgFT, are 400 mV, 160 mV smaller than VgMT. This offset voltage of 160 mV drives the main device in saturation and the auxiliary devices in weak inversion, resulting in the opposite polarities of the IMD3 terms.

91

The micrographs of two designed LNA prototypes are shown in Figure 5-5: (a) LNA with transformer-based input matching network and (b) LNA with Chebyshev BPF input matching network. In the layout, ground-guard ring and substrate taps are used to reduce substrate noise coupling. All signal grounds are connected with a thick top metal to reduce connection resistance such that traces have a low return resistance. The same layout strategy is applied to all power grounds as well.

(a)

(b) Figure 5-5. Chip micrographs of the proposed LNAs with (a) transformer-based and (b) Chebyshev BPF input matching networks.

92

5.2 Simulated and Measured Results A. Frequency Response The simulated and measured results of gain, S11, and NF are shown in Figure 5-6 (a) and Figure 5-6 (b) show the simulated and measured S11, gain, and NF of the two prototypes. The dotted and solid lines represent the simulated and measured performance, respectively. The LNA with the Chebyshev matching network shows the 3 dB bandwidth from 2.3 GHz to 6 GHz with the maximum gain of 12.7 dB and the minimum NF of 4.8 dB. The LNA with the transformer-based matching network shows the 3 dB bandwidth from 2 GHz to 5.3 GHz with the maximum gain of 12.4 dB and the minimum NF of 4.9 dB. Both LNAs show S11>-10dB at some frequency bands. The poor S11 comes from the complexity of the linearity-enhanced topology and design procedure to achieve high linearity, i.e., first the size of the three transistors is determined and then two degeneration inductors are chosen. More analytical analysis is necessary to address the input matching issue.
20 15 10 S11, Gain, NF [dB] 5 0 -5 -10 -15 -20 -25 -30 2 2.5 3 3.5 4 4.5 Frequency [GHz] 5 5.5 6

20

Gain
S11, Gain, NF [dB]

15

Gain NF

NF S11

10 5 0 -5 -10 -15 -20 -25 -30 2 2.5 3

S11

3.5 4 4.5 Frequency [GHz]

5.5

(a)

(b)

Figure 5-6. Simulated and measured results of the designed LNA (a) with transformerbased matching network and (b) with Chebyshev BPF matching network.

93

The overall stage transconductance Gm does not benefit from deeply scaled CMOS technologies due to matching constraint. Bode-Fano limit [53] for a parallel RC network as shown in Figure 5-7 is

ln ( ) d RC
0

(5.8)

where () is input reflection coefficient, R input parallel resistance, and C input parallel capacitance. The in-band reflection coefficient is represented by
( ) inband e 1/ 2 fRC

(5.9)

where f is the bandwidth of the lossless input matching network. In the proposed linearity-enhanced LNA topologies, the parallel R and C are fixed for linearity enhancement. Under the given bandwidth of the input matching network, the input reflection coefficient is determined.

Figure 5-7. Parallel RC load impedance for Bode-Fano limit. Furthermore, since the equivalent Gm is inversely proportional to frequency, large NF at high frequencies is inevitable as frequency increases in the LNA with filter-type input matching network. Some amount of the NF degradation is caused by the input matching. For example, the Chebyshev input matching network has around 2.5 dB insertion loss over the operating bandwidth, directly resulting in the NF degradation. To observe how much NF degradation is caused by an input matching network, the optimum NF (NFopt) is simulated without the matching network and compared to the measured NF in the LNA 94

with the Chebyshev input matching network as shown in Figure 5-8. With some limitation on the input matching due to the linearity-enhanced topology, if a linearityenhanced wideband LNA is properly designed, NF less than 4 dB over 26 GHz bandwidth can be achieved.

20 18 16 14 NF [dB] 12 10 8 6 4 2 0 2 2.5 3 3.5 4 4.5 Frequency [GHz] 5 5.5 6 Measured NF Simulated NFopt

Figure 5-8. NF and NFopt of the LNA with Chebyshev matching network.

The NF performance of the LNA with transformer-based matching network is compared with and without the SBCR technique in Figure 5-9. Without the SBCR, the LNA shows a little better NF. In order to observe noise contributions from the circuit devices, the noise summary plots are shown in Figure 5-10. Figure 5-10 (a) shows the percentage of the noise contribution and Figure 5-10 (b) shows the absolute amount of the noise contribution in V2. Without the SBCR, larger current flows through the cascode device MC causing more noise contribution. However, the overall noise spot noise

95

without the SBCR is a little smaller than that with the SBCR, the same as the NF simulation results as shown in Figure 5-9.

12 10 8 NF [dB] 6 4 2 0 2 Measured w/ CR Measured w/o CR

2.5

3.5 4 4.5 Frequency [GHz]

5.5

Figure 5-9. NF comparison of the LNA with transformer-based matching with or without the SBCR.

NoisePercentagew/CR@Chebyshev NoisePercentagew/oCR@Chebyshev 40 35 30 25 20 15 10 5 0

(a)

96

NoiseVoltageV^2w/CR@Chebyshev NoiseVoltageV^2w/o CR@Chebyshev 25 20 15 10 5 0

(b) Figure 5-10. Noise summary of the devices in the LNA with Chebyshev BPF matching.

B. Linearity: IIP3 The IIP3 of two linearity-enhanced wideband LNA prototypes is simulated and measured in a 0.13 m CMOS process. For IIP3 measurement, two tone signals are applied as an input signal after combining two sinusoidal signals from two signal generators with an external wideband power combiner. The output fundamental tones and third-order tones are monitored using the Agilent Spectrum Analyzer E4440A with the settings of 1 KHz SPAN, 9.1 Hz RBW, and 910 Hz VBW to measure their amplitude more accurately. The IIP3 is measured with 500 KHz frequency spacing in two input tones for most of the cases, as done in the simulation. To minimize the linearity degradation due to the buffer amplifier, large current of around 14 mA is consumed. The IIP3 of the LNA circuit itself is de-embedded by system level simulation with the measured stand-alone results of the buffer amplifier such as gain and linearity. 97

First of all, to verify the IIP3 peak versus the gate bias voltage of the main transistor, the LNA with Chebyshev BPF input matching network is measured at 4GHz which has the highest IIP3 peak over the frequency range. As we can see in Figure 5-11, the IIP3 has the peak value of +21.5 dBm at the gate bias voltage of 560 mV which has the offset voltage of 160 mV with respect to two auxiliary transistors. The simulated offset voltage was 132 mV to obtain the highest IIP3 peak in this LNA at 4 GHz. Figure 5-12 shows the IIP3 extrapolation plot at 4 GHz in the same LNA with Chebyshev BPF matching network. As shown in Figure 5-12, the input power range having a 3-dB slope in the IMD3 output signal is relatively narrow, compared to the IIP3 extrapolation plot of the LNAs without linearity enhancement techniques. More research needs to be performed in the future to analyze why this happens.

25

20 IIP3 [dBm]

15

10

5 0.35

0.4

0.45

0.5 0.55 0.6 Gate Voltage [V]

0.65

0.7

0.75

Figure 5-11. IIP3 vs. gate bias voltage at 4GHz in the LNA with Chebyshev BPF input matching network.

98

60 30 0 IIP3 [dBm] -30 -60 -90

-120 -35 -30 -25 -20 -15 -10 -5 0 Pin [dBm]

10 15 20 25

Figure 5-12. IIP3 extrapolation plot at 4GHz in the LNA with Chebyshev BPF input matching network.

On the other hand, to show the extent of IIP3 improvement with the SBCR technique, the IIP3 versus the frequency range is simulated and measured as shown in Figure 5-13 in the LNA with transformer-based input matching network. The blue line represents the IIP3 curve with the SBCR technique and the black line represents the IIP3 curve without the SBCR. The dotted and solid lines show the simulated and measured results, respectively. With the SBCR, the LNA achieves around 10 dB greater IIP3 value under the same gate bias condition.

99

30 25 20 IIP3 [dBm] 15 10 5 0 2

2.5

3.5 4 4.5 Frequency [GHz]

5.5

Figure 5-13. IIP3 plot over frequency range in the LNA with transformer-based input matching network.

The IIP3 is simulated and measured over the operating frequency range to confirm the effectiveness of the proposed linearity-enhanced wideband topology with two LNAs. The simulated and measured results in Figure 5-14 show high linearity over the wide frequency range. The LNA with the transformer-based matching network shows the maximum IIP3 of +15.6 dBm and relatively flat IIP3 response of +12 dBm on average from 2 GHz to 5.3 GHz. The LNA with the Chebyshev matching network has the maximum IIP3 of +21.5 dBm at 4 GHz and shows the IIP3 of +10.6 dBm on average from 2.3 GHz to 6 GHz.

100

40 35 30 IIP3 [dBm]

25
Meas.@Chebyshev Sim. @Chebyshev Meas.@Xfmr Sim. @Xfmr

20 15 10 5 0 4 10

20 15 10 5 0 2 2.5 3 3.5 4 4.5 Frequency [GHz] 5 5.5 6

IIP3 [dBm]

25

10

10 10 10 Frequency Spacing [Hz]

10

(a)

(b)

Figure 5-14. Simulated and measured IIP3 (a) vs. frequency range and (b) vs. frequency spacing plots in the wideband LNAs with transformer-based matching and with Chebyshev BPF matching.

An interesting issue in a wideband LNA is the IIP3 performance when differently spaced interferers exist within the multiple standard bands. To observe this important measure for a multi-standard wideband LNA, two test tones with different frequency spacing are applied. The resulting IIP3 performance is shown in Figure 5-15. The IIP3 holds steady for small frequency spacing and starts to degrade for larger frequency spacing since the linearity enhancement techniques lose their effectiveness. This explains that the highest IIP3 can be obtained in the proposed linearity enhancement technique when two test tones are closely placed. For multi-standard wideband LNAs, the IIP3 measurement setup should be taken into account carefully depending on a field environment having multiple coexisting in-band interferers which are not out-of-band interferers any longer.

101

40 35 30 IIP3 [dBm] 25 20 15 10 5 0 4 10 10
5

40
Meas.@4G Meas.@3G Meas.@5G

y p

q Meas.@4G Meas.@3G Meas.@5G

35 30 IIP3 [dBm] 25 20 15 10 5

10 10 10 Frequency Spacing [Hz]

10

0 4 10

10

10 10 10 Frequency Spacing [Hz]

10

(a)

(b)

Figure 5-15. IIP3 vs. frequency spacing plots of (a) transformer-based matching and (b) Chebyshev BPF matching LNAs at different reference frequencies.

C. Linearity: IIP2 The IIP2 performance of two LNA prototypes is measured at four different IMD2 frequencies as shown in Figure 5-16 (a)-(b). The IIP2 curve at the lowest IMD2 frequency f2-f1 shows the highest value, which is caused by high-pass filtering effects of ac coupling capacitors at the LNA and buffer outputs.
100 90 80 70 50 40 30 20 10 0 2 2.5 3 3.5 4 4.5 Frequency [GHz] 5 5.5 6 IIP2 [dBm] IIP2 [dBm] 60 Meas.@f2-f1 Meas.@2f1 Meas.@f1+f2 Meas.@2f2 100 90 80 70 60 50 40 30 20 10 0 2 2.5 3 3.5 4 4.5 Frequency [GHz] 5 5.5 6 Meas.@f2-f1 Meas.@2f1 Meas.@f1+f2 Meas.@2f2

(a)

(b)

Figure 5-16. IIP2 vs. frequency plots of (a) transformer-based matching and (b) Chebyshev BPF matching LNAs at different 2nd-order IMD frequencies.

102

For multi-standard RF front-end receivers, some issues should be brought up to address IIP2 measurement setup and falling frequency shift by changing frequency spacing within a wide bandwidth. In general, when the IIP2 is measured in a narrow-band or wideband LNA, the frequency spacing of two input tone signals are small so that two tones experience the similar gain and nonlinear behavior. Furthermore, even the frequency spacing between interference signals in wideband applications is not far away, compared to the multi-standard applications. On the other hand, the spacing for multistandard applications can be larger than the bandwidth of one of the multiple standards. To emulate the effects of the second-order intermodulation distortion in multi-standard applications, different IIP3 measurement setups are visualized in Figure 5-17. As we can see in Figure 5-17 (a), when two tones are assigned at the low-edge frequency and are put closely, we have three IMD2 tones at 2f1, f1+f2, and 2f2 which can deteriorate IIP3 performance due to the feedback to the input and re-mixing with the fundamental signals. However, when we address and measure the IIP2, most of the IMD2 tones fall out of band, but they fall in band, deteriorating nonlinearity of the receiver. As the frequency f2 increases with the f1 fixed, the 2f2 falls out of band but the f2-f1 is about to fall in band as shown in Figure 5-17 (b). Finally, the f2-f1 comes in band and three higher IMD2 tones are going out of band as shown in Figure 5-17 (c), with further increasing the spacing. The similar nonlinear behavior will occur when two tones are assigned at the center or at the high-edge frequency.

103

(a)

(b)

(c) Figure 5-17. Mechanism of IMD2 tones depending on frequency spacing.

To observe the effects of IMD2 on the IIP2, three different measurements are set up: 1) The f1 is fixed at 3 GHz and the f2 is shifted up. 2) The f1 and f2 are centered at the center frequency of the bandwidth (i.e. 4 GHz in this case), and each is shifted down and up. 3) The f1 is shifted down and the f2 is fixed at 5 GHz. With these three measurement setups, the IIP2 of two LNA prototypes is measured at four different IMD2 frequencies to show the amount of IIP2 degradation as shown in

104

Figure 5-18 (a)-(f). As the frequency spacing increases, the IIP2 value at f2-f1 is degraded rapidly since the attenuation of the f2-f1 term by ac coupling capacitors is decreasing. On the other hand, device and layout parasitic starts to get involved in the IIP2 degradation as the frequency f2-f1 increases.

100 90 80 70 IIP2 [dBm] 60 50 40 30 20 10 0 4 10 10


5

y p

q
100

y p

q Meas.@3G f2-f1 Meas.@3G 2f1 Meas.@3G f1+f2 Meas.@3G 2f2

Meas.@3G f2-f1 Meas.@3G 2f1 Meas.@3G f1+f2 Meas.@3G 2f2


IIP2 [dBm]
6 7 8 9

90 80 70 60 50 40 30 20 10 0 4 10

10 10 10 Frequency Spacing [Hz]

10

10

10 10 10 Frequency Spacing [Hz]

10

(a)
100 90 80 70 IIP2 [dBm] 60 50 40 30 20 10 0 4 10 10
5

(d)
100
Meas.@4G f2-f1 Meas.@4G 2f1 Meas.@4G f1+f2 Meas.@4G 2f2

90 80 70 IIP2 [dBm] 60 50 40 30 20 10 0 4 10

Meas.@4G f2-f1 Meas.@4G 2f1 Meas.@4G f1+f2 Meas.@4G 2f2

10 10 10 Frequency Spacing [Hz]

10

10

10 10 10 Frequency Spacing [Hz]

10

(b)

(e)

105

100 90 80 70 IIP2 [dBm] 60 50 40 30 20 10 0 4 10 10


5

Meas.@5G f2-f1 Meas.@5G 2f1 Meas.@5G f1+f2 Meas.@5G 2f2 IIP2 [dBm]

100 90 80 70 60 50 40 30 20 10 0 4 10 Meas.@5G f2-f1 Meas.@5G 2f1 Meas.@5G f1+f2 Meas.@5G 2f2

10 10 10 Frequency Spacing [Hz]

10

10

10 10 10 Frequency Spacing [Hz]

10

(c)

(f)

Figure 5-18. IIP2 vs. frequency spacing plots of transformer-based matching (a)-(c) and Chebyshev BPF matching (d)-(f) LNAs at different 2nd-order IMD frequencies with different reference frequencies.

5.3 Summary Two LNA prototypes are designed and fabricated in a 0.13 m CMOS process. The circuit design procedure for the proposed LNA topologies are described to achieve high linearity. Their simulated and measured frequency responses are demonstrated and show good match. The IIP2 and IIP3 are measured with several measurement setups to address the issues brought up for multi-standard applications in this research. This new approach can help what kind of performance and measurement setup should be carefully taken into account in future research for multi-standard applications or software-defined radios. These approaches emphasize the importance of the second-order linearity in the wideband receiver under the co-existence environment of in-band strong interference signals.

106

Chapter 6
Conclusions

The goal of this dissertation was to implement both a wideband RF front-end receiver and a linearity-enhanced wideband LNA. The details of the research scope were: 1) 2) Designed a wideband RF front-end receiver. Demonstrated DC offset mechanisms, partitioned different DC offsets depending on the mechanisms, and showed the relationship between DC offset and nonlinearity. 3) 4) Proposed and analyzed a linearity-enhanced wideband LNA topology. Designed linearity-enhanced wideband LNA prototypes and verified the effectiveness of the proposed topology.

As a part of research goals, the wideband RF front-end receiver including the LNA, active balun, and folded mixer is designed. It shows reasonable performance except linearity. DC offsets are segmented by different mechanisms and quantified by the measured results. Furthermore, the relationship between DC offset and linearity is derived and demonstrated by simulated, calculated and measured results in the wideband RF front-end receiver which has a differential topology from the active balun to the mixer. 107

To achieve high linearity in a wideband LNA, two linearity enhancement techniques are proposed: the wideband derivative superposition method to obtain two IIP3 peaks over a wide frequency range and the self-biasing current reuse technique to provide sufficient voltage headroom. The effectiveness of the proposed topology is analyzed by the Volterra series and then supported by conceptual diagram inspired from the behavior of the extracted frequency-dependent nonlinearity coefficients. The simulated and measured results in two LNA topologies show high IIP3 over the wide frequency range. Furthermore, the IIP3 performance as a function of frequency spacing of two test tones is demonstrated to estimate IIP3 performance variation when differently spaced interferers are received.

Figure 6-1. IIP3 comparison plot with narrowband and wideband LNAs.

The IIP3 performance depending on the frequency spacing is characterized to emulate blocker environment for multi-standard applications. The measured results are shown. Furthermore, the IIP2 performance is also characterized to observe the effects of the second-order intermodulation distortion terms on linearity.

108

The overall performance is summarized in Figure 6-1 and Table 4 and compared to the current state-of-art designs of linearity-enhanced narrow band LNA topologies [33], [34], wideband LNA topology [10], and wideband LNA topologies with noise and distortion cancellation [4], [12]. Both of linearity-enhanced wideband LNAs [4], [12] show around 0 dBm IIP3 at small frequency spacing, which is much less than the IIP3 of the proposed LNA topologies.

Table 4. IIP3 comparison with narrow band and wideband LNAs. Ref. V. Aparin [33] S. Ganesan [34] A. Bevilacqua [10] W. H. Chen [12] S. C. Blaakmeer [4] This work (Chebyshev) This work (Transformer) Freq. [GHz] 0.9 0.9 2.39.2 0.82.1 0.25.2 2.36.0 2.05.3 Gain [dB] 15.5 18.5 9.3 NF [dB] 1.65 1.76 4.0 2.6 < 3.5 4.8 4.9 IIP3 [dBm] +22 +21 -6.7 16 ~0 10.6 21.5 12.0
2) 3)

1) 1)

Supply Power Tech. [V] [mW] 2.6 24.2 0.25um 2.5 1.8 1.5 1.2 22.5 9 17.4 21 8.3 0.35um 0.18um 0.13um 65 nm 0.13um 0.13um

1)

14.5 15.6 12.7 12.4

>0
4) 5) 4)

1.2

1.2 8.3 5) 15.6 1) 2) 3) 4) 5) S21 at large freq. spacing at small freq. spacing average peak

In summary, the proposed wideband linearity enhancement techniques show high linearity performance in the wideband LNA designs. These techniques can be applied to any other circuits including an active balun, a mixer, and a variable gain amplifier, resulting in the high linearity of a wideband RF front-end receiver.

109

Suggestions for Future Research

The linearity enhancement techniques show high linearity in the wideband LNAs. However, the S11 and NF performance are relatively poor due to the design procedure and circuit complexity to achieve high linearity. Further research to obtain good noise and power match simultaneously needs to be done in the future. Some amount of the NF degradation is caused by the poor input matching. For example, the Chebyshev input matching network has around 2.5 dB insertion loss over the operating bandwidth, directly resulting in the NF degradation. For the future improvement of both parameters, the impedance looking into the main core of linearity enhanced topology needs to be conjugate with respect to the impedance looking back into the input matching network. In addition, the proposed linearity enhancement techniques can be applied to the folded mixer to achieve higher linearity. Using the proposed techniques and future works described above, a very high linear wideband RF receiver can be implemented for multistandard applications. Even higher linearity in the wideband receiver can be obtained along with the MEMS based mixer [54]. The IIP2 and IIP3 performance versus the frequency spacing are characterized in this thesis, along with different measurement setups. More research to achieve high IIP2 and IIP3 at the same time will be an interesting topic because both IIP2 and IIP3 are equally important in the wideband receiver for multi-standard applications. One of the solutions to achieve high IIP2 and IIP3 simultaneously is to design an LNA with a differential 110

topology with the proposed linearity enhancement techniques since the differential topology helps to reduce both common-mode signals and even-order IMD signals, and the linearity enhancement techniques improves the IMD3. Furthermore, with the differential structure, the substrate noise coupled from digital signals in a mixed-mode circuit can be reduced.

111

Appendix A

Introduction to Volterra Series and Harmonic Input

Several theories such as convolution, Taylor series, and Volterra series have been used to analyze a linear or nonlinear system. As known, a linear, causal system with memory is described by the convolution representation [55]
y (t ) = h(t ) x ( t ) dt

(A.1)

where x(t ) is the input, y (t ) the output, and h(t ) the impulse response of the system. A nonlinear system without memory is represented with Taylor series
y (t ) = an x n (t )
n =1

(A.2)

where x(t ) is the input, y (t ) the output, and a n the Taylor series coefficients. Volterra series were introduced by the Spanish mathematician Vito Volterra in 1959 [44] and have been used for distortion analysis [30], [56], [57], [58], [59] and have also been implemented in SPICE [60]. Like the combination of the two representations defined above, the Volterra series can describe a weakly nonlinear system with memory effects

112

y (t ) =

n 1 du L du c u L u ( , , ) 1 n n 1 n k =1 x(t uk ) n =1 n ! 1 = du1 c1 (u1 ) x(t u1 ) 1! 1 + du1 du2 c2 (u1 , u2 ) x(t u1 ) x(t u2 ) 2! 1 + du1 du2 du3 c3 (u1 , u2 , u2 ) x(t u1 ) x(t u2 ) x(t u3 ) 3! +L

(A.3)

where x(t ) is the input, y (t ) the output, and cn (u1 ,L, un ) the Volterra kernels of the system in the time domain [61]. In the representation, the underlying assumption is that the kernels cn (u1 ,L, un ) are symmetric, which means that cn (u1 ,L, un ) must have the same value regardless of the permutation of u1 , u2 ,L, un . The Voltterra series describes the output of a nonlinear system with memory as the sum of responses of the first-, second-, and higher-order operators which are described in the time or frequency domain with a transfer function called a Volterra kernel. These operators are shown in the blockdiagram representation of Fig. A1. The block diagram with the Volterra kernel C1 represents the linearized system. When a nonlinear circuit is excited by an input signal x(t) with very low amplitude, then the output signal y(t) can be described by only the firstorder operator C1 which has the linear behavior. This operator or kernel is characterized by the transfer function of the linearized circuit in the frequency domain. As the amplitude of the input signal increases, the output signal starts to show substantial nonlinear effects including C2, C3, and so on. As long as the input amplitude is sufficiently low, the nonlinear effects can be described accurately by taking into account only C2 and C3. The total output signal is the sum of the outputs of all the operators as shown in Fig. A1.

113

Fig. A1 Block-diagram representation of a system or circuit characterized by a Volterra series.

Based on the Volterra series representation of Fig. A1, the nonlinear transfer function of order n is constructed in a simple form from all of the lower-order nonlinear transfer functions as described in Eq. (A.4).
y (t ) = C1 ( s ) o x (t ) + C2 ( s1 , s2 ) o x (t ) 2 + C3 ( s1 , s2 , s3 ) o x (t )3

(A.4)

The operator performs a transformation of the input signal into the output signal, depending on the order of nonlinearity. It has to be remarked that the first-order Volterra operator is a convolution of the input signal with the impulse response of the linear system. The nth-order Volterra kernels in the frequency domain are defined by
C n ( f1 , L , f n ) =

du1 L du n c n ( u1 , L , u n ) e s1u1 L sn u n

(A.5)

where si = j 2 fi . The Volterra kernel C n is referred to as the nth-order transfer function as well. To determine the Volterra kernels C n in the frequency domain, usually the harmonic input method [45] is employed since it is an easy way to extract a nonlinear

114

transfer function when higher-order IMD components are negligible. In addition, it gives an insight about how nonlinear components behave and interact with each other. The fundamental idea of the harmonic input method is that when a one-tone, two-tone, and three-tone sinusoidal signals are applied as an input signal, the first-, second-, and thirdorder nonlinear transfer functions are yielded separately. For example, when a singletone sinusoidal signal is excited, the first-order kernel can be determined by comparing the first-order coefficients between the output of a circuit and the output of the Volterra series. When the input is

x(t ) = es1t +s2t +L+snt


where si = j 2 fi , i = 1,L, n and the si s are incommensurable, then
Cn ( f1 , L , f n ) = coefficients of e s1t +L+ snt .

(A.6)

(A.7)

115

Appendix B

Volterra Analysis of the Derivative Superposition Topology

If a nonlinear device is in weakly nonlinear region, the Volterra series can be used to evaluate its nonlinearity. The output current of a FET amplifier can be represented by the Volterra series in terms of excitation
vs

in the time domain: (B.1)

i( vs ) = C1( s ) o vs + C2 ( s1 , s2 ) o vs 2 + C3 ( s1 , s2 , s3 ) o vs 3

where C n ( s1 , s 2 ,K , s n ) represents the frequency domain form of the nth-order Volterra kernel, s is the Laplace variable, and the operator represents the magnitude and phase change of the input spectral components of v s by the nth-order Volterra kernel,
C n ( s1 , s 2 ,K , s n ) , when multi-tones are excited as an input signal. With two-tone excitation, v s = A[cos( a t ) + cos( b t )]

(B.2)

the input amplitude at the third-order intercept point is given by two different forms as follows:
4 g1 3 g3

AIIP3 =

(B.3a)

116

AIIP 3 =

C1 ( s a ) 4 3 C 3 ( sb , sb , s a )

(B.3b)

where C1 ( s a ) represents the Volterra kernel of the g 1 at the frequency of a when only a single tone of the frequency a is excited and C3 ( sb , sb , s a ) the Volterra kernel of the g 3 at the frequency of 2b a when two tones of the frequencies a , b are excited. Equation (B.3a) is derived with only DC characteristics of the transistor itself and (B.3b) with DC characteristics of the transistor as well as the other passive components around the transistor using the harmonic input method. To take into account the effects of the passive components with respect to amplitude and phase change, the equation (A3b) will be used. In the following derivations, IIP3 represents the available power of the signal generator at the third-order intercept point with reasonable input impedance match.
2 AIIP C1 ( sa ) 1 3 = 8 Re [Z 1 ( s a )] 6 Re [Z 1 ( s a )] C 3 ( s b , s b , s a )

IIP3 =

(B.4)

Vs Z1(s)

Vg

i(Vs)=iT D

a+ b

Cgs V1 S L

i1 (Vgs) gV

i2 (Vgs) gV
2

i3 (Vgs) gV
3

Vgs

2nd-order IMD products

Zin(s)

117

(a)
i(Vs)=iT D + VgA + CA VgB iB (VgB) g V V2 L Zin(s) iA (VgA) g V

Vs Z1(s)

Vg

CB

(b) Fig. B1. (a) Small-signal nonlinear equivalent circuit showing the 2nd-order IMD contribution on the 3rd-order IMD and (b) its equivalent circuit for IIP3 expression derivation

In the following IIP3 derivation using Volterra series, we have the assumptions described below: 1) The body effect is negligible, g bm 0 . 2) Except the gate-source capacitance 3) The
C gs C gs

, all parasitic capacitances are zero.

is bias-independent.

4) The gate and source resistance of a transistor are ignored. 5) The series resistance of an inductor is ignored as well. 6) No channel length modulation, i.e., g o .

118

7) Very week input signals so that higher-order IMD terms than third-order ones are negligible. 8) a b , i.e., 0 . 9) The input is conjugately matched. 10) g 1 A 0 , g 1B 0 .

In order to derive the Volterra kernels of C1 ( s a ) and C3 ( sb , sb , s a ) , we will build the equations of the output drain currents in terms of transconductance and gate bias voltages of the transistor.
i A ( v gA ) = g 3 A v gA 3

(B.5a)

iB ( v gB ) = g1B vgB + g 2 B vgB 2 + g 3B vgB3


i( v s ) = i A ( v gA ) + i B ( v gB ) = iT = g 3 A v gA 3 + g 1B v gB + g 2 B v gB 2 + g 3 B v gB 3

(B.5b)

(B.5c)

The contribution of g 1 A and g 2 A are ignored in the expressions above since they do not effect on the third-order intermodulation distortion in any way. The gate and source voltages can be modeled by the following truncated Volterra series with respect to the source voltage v s .
v gA = A1 ( s ) o vs + A2 ( s1 , s2 ) o vs 2 + A3 ( s1 , s2 , s3 ) o vs 3

(B.6a)

v gB = B1 ( s ) o vs + B2 ( s1 , s2 ) o vs 2 + B3 ( s1 , s2 , s3 ) o vs 3

(B.6b)

119

Substituting (B.6) into (B.5c) and solving the corresponding results with (B.1) using a symbolic equation solver such as Maple and Mathematica, we can get the following Volterra kernel expressions.
C1 ( s ) = g 1B B1 ( s )

(B.7a)

C3 ( s1 , s2 , s3 ) = g3 A A1 ( s1 ) A1 ( s2 ) A1 ( s3 ) + g1B B3 ( s1 , s2 , s3 ) + 2 g 2 B B1 ( s1 ) B2 ( s1 , s2 ) + g3 B B1 ( s1 ) B1 ( s2 ) B1 ( s3 )

(B.7b)

B1 ( s1 )B2 ( s1 , s 2 ) =

1 [B1 ( s1 )B2 ( s 2 , s3 ) + B1 ( s2 )B2 ( s1 , s3 ) + B1 ( s3 )B2 ( s1 , s 2 )] 3

(B.7c)

As we can see in the equations above, first the subsets of the Volterra kernel, A1 ( s ) ,
B1 ( s ) , B2 ( s1 , s 2 ) , and B3 ( s1 , s 2 , s 3 ) should be found. In order to derive relationship

between the DC parameters in Eq. (B.5) and the Volterra kernels in Eq. (B.6), the Kirchhoffs current law equations for each node of the circuit in Fig. B1 are written below.
vs vg Z1( s ) + v 2 v g (sC A + sC B ) = 0

(B.8a)

i A + i B + v g v 2 (sC A + sC B )

v2 =0 sL

(B.8b)

Solving these simultaneous equations for solutions into


v gA = v gB = v g v 2 = v gs

vg

and v 2 , and then substituting the

, we get the node voltages using the symbolic

equation solvers such as Mathematica and Maple.


v s [1 (sC A + sC B )Z 2 ] + (sC A + sC B ) (i A + i B )Z 1 Z 2 1 + (sC A + sC B ) (Z 1 Z 2 )

vg =

(B.9a)

120

v2 =

v s (sC A + sC B ) + [1 + (sC A + sC B )Z 1 ] (i A + i B ) Z2 1 + (sC A + sC B ) (Z 1 Z 2 ) v s (i A + i B )Z 2 1 + (sC A + sC B ) (Z 1 Z 2 )

(B.9b)

v gs =

(B.9c)

a( s ) = 1 + s 2 L (C A + C B )

(B.9d) (B.9e) (B.9f) (B.9g)

b( s ) = sL c( s ) = sZ 1 ( s )(C A + C B ) i A + i A = iT

Using Eq. (B.9d)-(B.9g), the equations of (B.9a)-( B.9c) can be rewritten in a simple form to make the following derivation easier.
a( s )v s + (i A + i B )b( s )c( s ) a( s ) + c( s ) v s (a( s ) 1) + (i A + i B )(1 + c( s )) a( s ) + c( s ) v s (i A + i B )b( s ) a( s ) + c( s )

vg =

(B.10a)

v2 =

(B.10b)

v gs =

(B.10c)

Using (B.10a)-(B.10c), the linear transfer functions A1 ( s ) and B1 ( s ) can be found with the harmonic input method. The harmonic input method enables us to stimulate the circuit with a single-tone and two-tone sinusoidal signals so that the output signal produces all the harmonic outputs and intermodulation products. First, in order to find the linear transfer functions A1 ( s ) and B1 ( s ) , only a single-tone sinusoidal signal
v s = e st

will

be applied as an input signal. Substituting (B.5a)-(B.5b) and (B.6a)-(B.6b) into (B.10c)

121

and then equating the coefficients of e st to solve the simultaneous equations, we can get the linear transfer functions A1(s) and B1(s).
A1 ( s ) = 1 sg 1 B LB 1 ( s ) a( s ) + c( s )

(B.11a)

B1 ( s ) =

1 a( s ) + c( s ) + sg 1 B L

(B.11b)

Simplifying further two equations above, (B.11a) and (B.11b), we get the following simple forms.
1 a( s ) + c( s ) + g 1 B b( s )

A1 ( s ) = B 1 ( s ) =

(B.11c)

The second-order transfer function B 2 ( s 1 , s 2 ) can be found by exciting the circuit with two-tone sinusoidal signals,
v s = e s1t + e s 2 t

. As it was done in the single-tone excitation,

the second-order transfer function B 2 ( s 1 , s 2 ) is derived by substituting (B.5a)-(B.5b) and (B.6a)-(B.6b) into (B.10c) and then equating the coefficients of e ( s1 + s2 )t to solve the simultaneous equations. Eventually we get the following transfer functions.
g 2 B b( s 1 + s 2 )B1 ( s 1 )B1 ( s 2 ) a( s 1 + s 2 ) + c( s 1 + s 2 ) + g 1B b( s 1 + s 2 )

A2 ( s 1 , s 2 ) =

(B.12a)

B2 ( s1 , s 2 ) =

g 2 B b( s 1 + s 2 )B1 ( s 1 )B1 ( s 2 ) a( s 1 + s 2 ) + c( s 1 + s 2 ) + g 1B b( s 1 + s 2 )

(B.12b)

= g 2 B b( s 1 + s 2 )B1 ( s 1 + s 2 )B1 ( s 1 )B1 ( s 2 )

Similarly, the third-order transfer function

B3 ( s1 , s 2 , s 3 )

is derived by substituting

(A5a)-(A5b) and (B.6a)-(B.6b) into (B.10c) and then equating the coefficients of

122

e ( s 1 + s 2 + s 3 )t

to solve the simultaneous equations. Eventually we get the following transfer

functions.
B3 ( s1 , s 2 , s3 ) = b( s1 + s 2 + s3 )B1 ( s1 + s2 + s3 ) g 3 B B1 ( s1 )B1 ( s2 )B1 ( s3 ) + 2 g 2 B B1 ( s1 )B2 ( s 2 , s3 ) + g 3 A A1 ( s1 ) A1 ( s2 ) A1 ( s3 )

(B.13) For the input excitation given by (B.2), the third-order intermodulation product (IMD3) at
2 b a is found by setting s1 = s 2 = sb and s3 = s a . Assuming that frequencies are

closely spaced, that is, s a sb s , and all Volterra kernels are complex, we can simplify (B.13) and (B.7c) as follows:
2 2 2 B3 ( s b , s b , s a ) = b( s )B1 ( s ) g 3 B B1 ( s ) B1 ( s ) + g 2 B [2 B1 ( s )B 2 ( s , s ) + B1 ( s )B 2 ( s , s )] + g 3 A A1 ( s ) A1 ( s ) . 3

(B.14) Substituting (B.11c) and (B.12b) into (B.14), we get


2 b( 2s )B1 ( s ) B1 ( s ) 2 2 2 B3 ( sb , sb ,s a ) = b( s )B1 ( s )(g 3 A + g 3 B )B1 ( s ) B1 ( s ) g 2 B . 3 a( 2 s ) + c( 2 s ) + g 1B b( 2 s )

(B.15)

Rewriting (B.7b) with s a and s b to see the IMD3 term, we get


C 3 ( sb , sb , s a ) = g 3 A A1 ( s ) A1 ( s ) + g 1B B3 ( sb , sb , s a ) + = (g 3 A 2 2 g 2 [2 B1 ( s )B2 ( s , s ) + B1 ( s )B2 ( s , s )] + g 3 B B1 ( s ) B1 ( s ) 3 B 2 2 + g 3 B )B1 ( s ) B1 ( s ) + g 1B B3 ( sb , sb , s a ) + g 2 B [2 B1 ( s )B2 ( s , s ) + B1 ( s )B2 ( s , s )] 3
2

(B.16) Substituting (B.12b) and (B.15) into (B.16) and simplifying further, we get
2 2 b( 2 s )(1 g 1 B b( s )B1 ( s )) 2 C 3 ( s b , s b , s a ) = B1 ( s ) B1 ( s ) (g 3 A + g 3 B )(1 g 1 B b( s )B1 ( s )) + g 2 (B.17) B 3 a( 2 s ) + c( 2 s ) + g 1 B b( 2 s )

123

2 2 2 C 3 ( s b , s b , s a ) = B1 ( s ) B1 ( s ) (g 3 A + g 3 B )(1 g 1 B b( s )B1 ( s )) + g 2 B b( 2 s )(1 g 1 B b( s ) B1 ( s ))B1 ( 2 s ) 3

(B.18)

C 3 ( s b , s b , s a ) = B1 ( s ) B1 ( s )

2 2 (1 g 1 B b( s )B1 ( s )) g 3 A + g 3 B g 2 B b( 2 s )B1 ( 2 s ) 3

(B.19)

To simplify (B.19) further, we will take into account the case of conjugately matched input at the fundamental frequency, 1.e., Z 1 ( s ) = Z in ( s ) , where Z in ( s ) is the input impedance of the circuit and Z 1 ( s ) the Thevenin equivalent source impedance. Using the small-signal equivalent circuit shown in Fig. B1 (b), the input impedance is derived as follows.
g 1B L 1 + s (C A + C B ) C A + C B

Z in ( s ) = sL +

(B.20)

In this case, due to the impedance matching condition,


1 g 1 B b( s ) B 1 ( s ) = 1 2

(B.21)

Substituting (B.21) into (B.19), we get the following equations.


C 3 ( s b , s b , s a ) = 1 2 B1 ( s ) B1 ( s ) 2

(B.22a)

= g 3 A + g 3B

2 2 g 2 B b( 2 s ) B 1 ( 2 s ) 3

(B.22b)

Eventually we can derive the IIP3 expression in terms of circuit parameters in the equivalent circuit since the required Volterra kernels are derived. Substituting (B.22) and (B.7a) into (B.4), we get the following equation.

124

IIP3 =

C1 ( s a ) g 1B 1 = 6 Re[Z 1 ( s a )] C3 ( sb , sb , s a ) 3 Re[Z 1 ( s a )] B1 ( s ) 2

(B.23)

Calculating Re[Z 1 ( sa )] B1 ( s ) with (B.20) and (B.11c), we get the following equation.
2

Re [Z 1 ( s a )] B1 ( s )

1 4 g 1 B L (C A + C B )

(B.24)

Substituting (B.24) into (B.23), we get the following IIP3 equation


IIP3 =
2 4 2 g 1 B L (C A + C B ) 3

(B.25)

where
2 2 g 2B 3

= g 3 A + g 3B

g 1B +
2 2g2 B

C + CB 1 + j 2 (C A + C B ) + Z 1 ( j 2 ) A j 2L L 1 1 j 2g 1 B L

(B.26)

g 3 A + g 3B

3 g 1B

1+

125

Appendix C

Volterra Analysis of the Wideband Modified Derivative Superposition Topology

If a nonlinear device is in weakly nonlinear region, the Volterra series can be used to evaluate its nonlinearity. The output current of a FET amplifier can be represented by the Volterra series in terms of excitation
vs

in the time domain: (C.1)

i( vs ) = C1( s ) o vs + C2 ( s1 , s2 ) o vs 2 + C3 ( s1 , s2 , s3 ) o vs 3

where C n ( s1 , s 2 ,K , s n ) represents the frequency domain form of the nth-order Volterra kernel, s is the Laplace variable, and the operator represents the magnitude and phase change of the input spectral components of v s by the nth-order Volterra kernel,
C n ( s1 , s 2 ,K , s n ) , when multi-tones are excited as an input signal. With two-tone excitation, v s = A[cos( a t ) + cos( b t )]

(C.2)

the input amplitude at the third-order intercept point is given by two different forms as follows:

126

AIIP3 =

4 g1 3 g3

(C.3a)

AIIP 3 =

C1 ( s a ) 4 3 C 3 ( sb , sb , s a )

(C.3b)

where C1 ( s a ) represents the Volterra kernel of the g 1 at the frequency of a when only a single tone of the frequency a is excited and C3 ( sb , sb ,s a ) the Volterra kernel of the g 3 at the frequency of 2b a when two tones of the frequencies a , b are excited. Equation (C.3a) is derived with only DC characteristics of the transistor itself and (C.3b) with DC characteristics of the transistor as well as the other passive components around the transistor using the harmonic input method. To take into account the effects of the passive components with respect to amplitude and phase change, the equation (C.3b) will be used. In the following derivations, IIP3 represents the available power of the signal generator at the third-order intercept point with reasonable input impedance match.
2 AIIP C1 ( sa ) 1 3 = 8 Re [Z 1 ( s a )] 6 Re [Z 1 ( s a )] C 3 ( s b , s b , s a )

IIP3 =

(C.4)

127

Vs

Z1(s)

Vg + VgB + VgA CA + Zm LB V1 LA VgA CF CB V2 iB (VgB) iF (VgF)

i(Vs)=iT

iA (VgA)

Zin(s)

Fig. C1. The small-signal nonlinear equivalent circuit for IIP3 expression derivation

In the following IIP3 derivation using Volterra series, we have the assumptions described below: 1) The body effect is negligible, g bm 0 . 2) Except the gate-source capacitance 3) The
C gs C gs

, all parasitic capacitances are zero.

is bias-independent.

4) The gate and source resistance of a transistor are ignored. 5) The series resistance of an inductor is ignored as well. 6) No channel length modulation, i.e., g o . 7) Very week input signals so that higher-order IMD terms than third-order ones are negligible. 8) a b , i.e., 0 . 9) The input is conjugately matched.

128

10) g 1 A 0 , g 2 A 0 , g 1F 0 , g 2 F 0 .

In order to derive the Volterra kernels of C1 ( s a ) and C3 ( sb , sb ,s a ) , we will build the equations of the output drain currents in terms of transconductance and gate bias voltages of the transistor.
i A ( v gA ) = g 3 A v gA 3

(C.5a)

iB ( v gB ) = g1B vgB + g 2 B vgB 2 + g 3B vgB3

(C.5b)

iF ( vgF ) = g 3F vgF 3
i( v s ) = i A ( v gA ) + i B ( v gB ) + i F ( v gF ) = iT = g 3 A v gA 3 + g 1B v gB + g 2 B v gB 2 + g 3 B v gB 3 + g 3 F v gF 3

(C.5c)

(C.5d)

The contribution of g 1 A and g 2 A are ignored in the expressions above since they do not effect on the third-order intermodulation distortion in any way. The gate and source voltages can be modeled by the following truncated Volterra series with respect to the source voltage v s .
v gA = A1 ( s ) o vs + A2 ( s1 , s2 ) o vs 2 + A3 ( s1 , s2 , s3 ) o vs 3

(C.6a)

v gB = B1 ( s ) o vs + B2 ( s1 , s2 ) o vs 2 + B3 ( s1 , s2 , s3 ) o vs 3

(C.6b)

v gF = F1 ( s ) o vs + F2 ( s1 , s2 ) o vs 2 + F3 ( s1 , s2 , s3 ) o vs 3

(B6c)

129

Substituting (C.6) into (C.5d) and solving the corresponding results with (C.1) using a symbolic equation solver such as Maple and Mathematica, we can get the following Volterra kernel expressions.
C1 ( s ) = g 1B B1 ( s )

(C.7a)

C 3 ( s1 , s 2 , s 3 ) = g 3 A A1 ( s1 ) A1 ( s 2 ) A1 ( s 3 ) + g 1B B3 ( s1 , s 2 , s 3 ) + 2 g 2 B B1 ( s1 )B2 ( s1 , s 2 ) + g 3 B B1 ( s1 )B1 ( s 2 )B1 ( s 3 ) + g 3 F F1 ( s1 )F1 ( s 2 )F1 ( s 3 ) = g 1B B3 ( s1 , s 2 , s 3 ) + 2 g 2 B B1 ( s1 )B2 ( s1 , s 2 ) + g 3 A A1 ( s1 ) + g 3 B B1 ( s1 ) + g 3 F F1 ( s1 )

(C.7b)

B1 ( s1 )B2 ( s1 , s 2 ) =

1 [B1 ( s1 )B2 ( s 2 , s3 ) + B1 ( s2 )B2 ( s1 , s3 ) + B1 ( s3 )B2 ( s1 , s 2 )] 3

(C.7c)

A1 ( s1 ) = A1 ( s1 ) A1 ( s 2 ) A1 ( s 3 )

(C.7d) (C.7e) (C.7f)

B1 ( s1 ) = B1 ( s1 )B1 ( s 2 )B1 ( s 3 )

F1 ( s1 ) = F1 ( s1 )F1 ( s 2 )F1 ( s 3 )

As we can see in the equations above, first the subsets of the Volterra kernel, A1 ( s ) ,
B1 ( s ) , B2 ( s1 , s 2 ) , B3 ( s1 , s 2 , s 3 ) , and F1 ( s ) should be found. In order to derive relationship

between the DC parameters in Eq. (C.5) and the Volterra kernels in Eq. (C.6), the Kirchhoffs current law equations for each node of the circuit in Fig. C1 are written below.
vs v g Z1( s )

+ v1 v g sC A + v 2 v g sC B = 0

(C.8a)

i A + i F + v g v 1 sC A + (v 2 v 1 )Z m

v1 =0 sL A

(C.8b)

130

(v g v 2 )sC B + i B (v 2Z v1 ) = 0
m

(C.8c)

Solving these simultaneous equations for solutions into

vg

, v1 , and v 2 , and then substituting the we get the node voltages using the

v gA = v g v 1 , v gB = v g v 2 , v gF = v 2 v 1 ,

symbolic equation solvers such as Mathematica and Maple.


(i A + i B + i F )sL A + v s + sC B [ (i A + i F )sL A + v s + i B Z 1 ] 1 + sC B (sL A + Z 1 + Z m ) + sC A (sL A + Z 1 )(1 + sC B Z m ) (i A + i B + i F )sL A + v s i B [1 + sC A (sL A + Z 1 )]Z m 1 + sC B (sL A + Z 1 + Z m ) + sC A (sL A + Z 1 )(1 + sC B Z m )

v gA =

(C.9a)

v gB =

(C.9b)

v gF =

{sC B [ (i A + i F )sL A + v s ] + i B [1 + sC B Z 1 + sC A (sL A + Z 1 )]}Z m 1 + sC B (sL A + Z 1 + Z m ) + sC A (sL A + Z 1 )(1 + sC B Z m )

(C.9c)

l=

sL A sL A = Z m ( s ) b( s )

(C.9d) (C.9e) (C.9f) (C.9g) (C.9h) (C.9h)

a( s ) = 1 + sC B Z m ( s )
b( s ) = Z m ( s ) = sL B 1 + s 2 C F LB

c( s ) = 1 + sC A (Z 1 + sL A ) d ( s ) = 1 + l + sC A (Z 1 + sL A ) e( s ) = sC B (Z 1 + sL A )

n( s ) = 1 + (g 1B + sC B )Z m ( s ) = a( s ) + g 1B b( s ) = 1+ s 2 C B LB 1 + s C F LB
2

g 1B sLB 1 + s 2 C F LB

(C.9i)

1 + g 1B sLB

131

iT = i A + i B + i F

(C.9j)

Using Eq. (C.9d)-(C.9j), the equations of (C.9a)-(C.9c) can be rewritten in a simple form to make the following derivation easier.
a( s )(v s iT lb( s )) + i B b( s )e( s ) a( s )c( s ) + e( s )

v gA =

(C.10a)

v gB =

v s iT lb( s ) i B b( s )c( s ) a( s )c( s ) + e( s )

(C.10b)

v gF =

(a( s ) 1)(v s iT

lb( s )) + i B b( s )(c( s ) + e( s )) a( s )c( s ) + e( s )

(C.10c)

Using (C.10a)-(C.10c), the linear transfer functions, A1 ( s ) , B1 ( s ) , and F1 ( s ) , can be found with the harmonic input method. The harmonic input method enables us to stimulate the circuit with a single-tone and two-tone sinusoidal signals so that the output signal produces all the harmonic outputs and intermodulation products. First, in order to find the linear transfer functions, A1 ( s ) , B1 ( s ) , and F1 ( s ) , only a single-tone sinusoidal signal
v s = e st

will be applied as an input signal. Substituting (C.5a)(C.5c) and (C.6a)e st to solve the

(C.6c) into (C.10a)-(C.10c) and then equating the coefficients of

simultaneous equations, we can get the linear transfer functions, A1 ( s ) , B1 ( s ) , and F1 ( s ) .


A1 ( s ) = n( s )B1 ( s )

(C.11a) (C.11b)

B1 ( s ) =

1 a( s )c( s ) + e( s ) + g 1 B b( s )d ( s )

F1 ( s ) = (n( s ) 1)B1 ( s )

(C.11c)

132

The second-order transfer function two-tone sinusoidal signals,

B 2 ( s1 , s 2 )

can be found by exciting the circuit with

v s = e s1t + e s 2 t .
B 2 ( s1 , s 2 )

As it was done in the single-tone excitation, the is derived by substituting (C.5a)-(C.5c) and
e ( s 1 + s 2 )t

second-order transfer function

(C.6a)-(C.6c) into (C.10a)-(C.10c) and then equating the coefficients of

to solve

the simultaneous equations. Eventually we get the following transfer functions.


B 2 ( s 1 , s 2 ) = g 2 B b( s 1 + s 2 )[l + c( s 1 + s 2 )]B 1 ( s 1 ) B 1 ( s 2 ) B 1 ( s 1 + s 2 )

(C.12) is derived by substituting

Similarly, the third-order transfer function

B3 ( s1 , s 2 , s 3 )

(C.5a)-(C.5c) and (C.6a)-(C.6c) into (C.10a)-(C.10c) and then equating the coefficients of
e ( s 1 + s 2 + s 3 )t

to solve the simultaneous equations. Eventually we get the following transfer

functions.
l 2 g B ( s )B ( s , s ) + g A ( s ) + g B ( s ) + g F ( s ) 3A 1 1 3B 1 1 3F 1 1 2B 1 1 2 2 3 B3 ( s 1 , s 2 , s 3 ) = b( s 1 + s 2 + s 3 )B1 ( s 1 + s 2 + s 3 ) + 2 g 2 B B1 ( s 1 )B 2 ( s 2 , s 3 ) + g 3 B B1 ( s 1 ) c( s 1 + s 2 + s 3 ) = b( s 1 + s 2 + s 3 )B1 ( s 1 + s 2 + s 3 ) AF3 ( s 1 , s 2 , s 3 ) + BB3 ( s 1 , s 2 , s 3 )

(C.13a) where,
AF 3 ( s 1 , s 2 , s 3 ) = l 2 g 2 B B 1 ( s 1 ) B 2 ( s 2 , s 3 ) + g 3 A A1 ( s 1 ) + g 3 B B 1 ( s 1 ) + g 3 F F1 ( s 1 )

(C.13b) (C.13c)

BB 3 ( s 1 , s 2 , s 3 ) = 2 g 2 B B 1 ( s 1 ) B 2 ( s 2 , s 3 ) + g 3 B B 1 ( s 1

) c ( s 1 + s 2 + s 3 ) .

For the input excitation given by (C.2), the third-order intermodulation product at
2 b a

is found by setting

s1 = s 2 = sb

and

s3 = sa

. Assuming that frequencies are

closely spaced, that is, (C.13) as follows:

s a sb s ,

and all Volterra kernels are complex, we can simplify

133

B3 ( sb , sb , sa ) = b( s) B1 ( s) AF3 ( sb , sb , sa ) + BB3 ( sb , sb , sa )

}
(C.14a)

2 2 = b( s) B1 ( s)2 B1 ( s) l [ g3 A N ( s) + g3 F N1 ( s)] + ( l + c( s) ) g3 B g 2 B 2b(2s) [l + c(2s)] B1 (2s ) 3

where,
2 2 2 2 AF3 ( sb , sb , sa ) = lB1 ( s ) B1 ( s) g3 B + g3 A n( s) n( s ) + g3 F ( n( s ) 1) ( n( s) 1) g 2 B 2 b(2s) [l + c(2s) ] B1 (2s ) 3 2 2 = lB1 ( s ) B1 ( s) g3 B + g3 A N ( s) + g3 F N1 ( s) g 2 B 2 b(2s ) [l + c(2s )] B1 (2s) 3

(C.14b)
2 2 BB3 ( sb , sb , sa ) = c( s) B1 ( s) B1 ( s) g3B g 2 B 2 b(2s) [l + c(2s)] B1 (2s) 3
N (s) = n(s) n(s)
N 1 ( s ) = ( n ( s ) 1)
2

(C.14c) (C.14d)

( n ( s ) 1) .
sa

(C.14e) and
sb

Rewriting (C.7b) with

to simplify the third-order IMD term further, we get

C3 (sb , sb , sa ) = AF ( sb , sb , sa ) / l g1B b( s) B1 (s) AF3 (sb , sb , sa ) + BB3 ( sb , sb , sa )

}
. (C.15)

2 2 2 = B1 ( s) B1 (s) [1 g1B lb( s) B1 ( s)] [ g3 A N ( s) + g3F N1 ( s)] + 1 g1B b(s) B1 ( s) ( l + c( s) ) g3B 3 g2 B b(2s) [l + c(2s)] B1 (2s)

Using the small-signal equivalent circuit shown in Fig. C1, the input impedance is derived as follows.
Z in ( s ) = sL A + 1 + [g 1A + g 1B + (g 1A + g 1F )(g 1B + sC B )Z m ]sL A + (g 1B + sC B ) Z m s [C A + C B + C A (g 1B + sC B )Z m ]

(C.16)

In this case, due to the impedance matching condition,


1 g 1 B b( s )(l + c( s ) )B 1 ( s ) = 1 2

(C.17)

134

After tedious approximation with following relationship.


Re[Z 1 ( s a )] B1 ( s ) =
2

g 1 A 0 , g 2 A 0 , g 1F 0 , g 2 F 0

, we can get the

4g 1B C B (L A + L B ) + C A L A 1 - 2 C B L B
2 2

)]
(C.18)

4g 1B [L A (C A + C B ) + C B L B ]
2 2

Substituting (C.15), (C.17), and (C.18) into (C.4), eventually we can derive the IIP3 expression in terms of circuit parameters in the equivalent circuit since the required Volterra kernels are derived.
IIP3 = C1 ( s a ) 1 6 Re[Z 1 ( s a )] C 3 ( s b , s b , s a ) g 1B 1 2 2 B1 ( s ) [1 g 1B lb( s )B1 ( s )] [g 3 A N ( s ) + g 3 F N 1 ( s )] + g 3 B g 2 B 2 b( 2 s )[l + c( 2 s )]B1 ( 2 s ) 2 3 8 Re[Z 1 ( s a )] B1 ( s )
2

4 3 =

1 6 Re[Z 1 ( s a )] B1 ( s )
2

g 1B 1 2 2 [1 g 1B lb( s )B1 ( s )] [g 3 A N ( s ) + g 3 F N 1 ( s )] + g 3 B g 2 B b( 2 s )[l + c( 2 s )]B1 ( 2 s ) 2 3

4 g 1B 2 2 [L A (C A + C B ) + C B L B ] 3

( C.19a) where,
= 1 +
LB ( CB + jg1B CA LA ) g 2B2 2 2 2 2 + + 1 + (g1B LB ) [1 + jg1B LB ] + jg1B LB g g g L g g3A 3A 1B 3F B 3B L A ( C A + CB ) + CB L B 3g1B

(C.19b)

135

Appendix D

Matching Table of Frequency-dependent Nonlinearity Coefficients from HB Analysis

# of Order 1 1 2 2 2 2 2 2 3 3 3 3

HB Volterra Coefficient

Frequency of Response

Amplitude of Response

Type of Response linear linear 2 -order IM products 2nd-order IM products 2nd harmonics 2nd harmonics DC shift DC shift 3rd-order IM products 3rd-order IM products 3rd-order IM products 3rd-order IM products
nd

g1, HB ,1
g1, HB , 2 g 2, HB ,1+ 2 g 2, HB , 21 g 2, HB ,21

1 2 1 + 2
1 2
21 22
0 0

A1 C1 ( j1 )
A2 C1 ( j2 )

A1 A2 C2 ( j1 , j2 )
A1 A2 C2 ( j1 , j2 )
1 2 A1 C2 ( j1 , j1 ) 2 1 2 A2 C2 ( j2 , j2 ) 2 1 2 A1 C2 ( j1 , j1 ) 2 1 2 A2 C2 ( j2 , j2 ) 2 3 2 A1 A2 C3 ( j1 , j1 , j2 ) 4 3 2 A1 A2 C3 ( j1 , j1 , j2 ) 4 3 2 A1 A2 C3 ( j1 , j2 , j2 ) 4 3 2 A1 A2 C3 ( j1 , j2 , j2 ) 4

g 2, HB ,2 2 g 2, HB ,11 g 2, HB , 2 2
g3, HB,21+ 2 g3, HB,21 2

21 + 2
21 2

g3, HB,1+2 2 g3, HB,2 21

1 + 22
1 22

136

3 3 3 3 3 3

g3, HB,1+ 2 2 g3, HB ,11+ 2 g3, HB ,211 g 3, HB ,2 2 2

1 + 2 2 = 1 1 1 + 2 = 2
21 1 = 1 22 2 = 2 31 32

3 A1 A22 C3 ( j1 , j2 , j2 ) 2 3 2 A1 A2 C3 ( j1 , j1 , j2 ) 2 3 3 A1 C3 ( j1 , j1 , j1 ) 4 3 3 A2 C3 ( j2 , j2 , j2 ) 4 1 3 A1 C3 ( j1 , j1 , j1 ) 4 1 3 A2 C3 ( j2 , j2 , j2 ) 4

3rd-order desensitization 3rd-order desensitization 3rd-order compression or expansion 3rd-order compression or expansion 3rd harmonics 3rd harmonics

g 3, HB ,3 1 g3, HB ,3 2

137

Bibliography
[1] R. Bagheri, A. Mirzaei, S. Chehrazi, M. Heidari, M. Lee, M. Mikhemar, T. Wai, and A. A. Abidi , "An 800-MHz-6-GHz software-defined wireless receiver in 90-nm CMOS," IEEE J. Solid-State Circuits, vol. 41, no. 12, pp. 2860-2876, Dec. 2006. [2] T. B. Zahariadis, "Migration toward 4G wireless communications," IEEE Wireless Commun., vol. 11, no. 3, pp. 6-7, Jun. 2004. [3] M. Brandolini, P. Rossi, D. Manstretta, and F. Svelto, "Toward multistandard mobile terminals - fully integrated receivers requirements and architectures," IEEE Trans. Microw. Theory Tech., vol. 53, no. 3, p. 10261038, Mar. 2005. [4] Stephan C. Blaakmeer, Eric A. M. Klumperink, Domine M. W. Leenaerts, and B. Nauta "Wideband balun-LNA with simultaneous output balancing, noise-cancelling and distortion-cancelling," Journal of Solid-State Circuits, vol. 43, no. 6, pp. 13411350, June 2008. [5] K. Choi, D. H. Shin, and C. P. Yue, "A 1.2-V, 5.8-mW, Ultra-Wideband Folded Mixer in 0.13-um CMOS," in IEEE RFIC Conference, Hawaii, 2007, pp. 489-492. [6] F. Agnelli, G. Albasini, I. Bietti, A. Gnudi, A. Lacaita, D. Manstretta, R. Rovatti, E. Sacchi, P. Savazzi, F. Svelto, E. Temporiti, S. Vitali, and R. Castello, "Wireless Multi-Standard Terminals: System Analysis and Design of a Reconfigurable RF Front-end," in IEEE CIRCUITS AND SYSTEMS MAGAZINE, 2006, p. 3859. [7] P. H.Woerlee, M. J. Knitel, R. van Langelde, D. B. M. Klaassen, L. F. Tiemeijer, A. J. Scholten, and T. A. Z. Duijnhoven, "RF-CMOS performance trends," IEEE Trans. Electron Devices, vol. 48, no. 8, p. 17761782, Aug. 2001. [8] K. Lee, I. Nam, I. Kwon, J. Gil, K. Han, S. Park, and B. Seo, "The impact of semiconductor technology scaling on CMOS RF and digital circuits for wireless application," IEEE Trans. Electron Devices, vol. 52, no. 7, p. 14151422, Jul. 2005. [9] A. Amer, E. Hegazi, and H. Ragai, "A low-power wideband CMOS LNA for WiMAX," IEEE Trans. Circuits and Systems, p. 48, 2007. [10] A. Bevilacqua and A. M. Niknejad, "An ultrawideband CMOS low-noise amplifier for 3.110.6-GHz wireless receivers," IEEE J. Solid-State Circuits, vol. 39, p. 2259 2268, Dec 2004. [11] K. Choi, D. H. Shin, and C. P. Yue, "An ultra-wideband RF front-end receiver with an active balun in a 0.13-m CMOS process," Carnegie Mellon University, Pittsburgh, Technical Report 2008. [12] W. -H. Chen, G. Liu, B. Zdravko, and A. M. Niknejad, "A highly linear broadband CMOS LNA employing noise and distortion cancellation," IEEE J. Solid-State Circuits, vol. 43, no. 5, pp. 1164-1176, May 2008. [13] Amin Q. Safarian, A. Yazdi, and P. Heydari, "Design and Analysis of an UltrawideBand Distributed CMOS Mixer," IEEE Trans. on VLSI Systems, vol. 13, no. 5, pp. 618-629, May 2005. [14] V. Vidojkovic, J. van der Tang, A. Leeuwenburgh, and A. H. M. van Roermund, "A low-voltage folded-switching mixer in 0.18-m CMOS," IEEE J. Solid-State 138

Circuits, vol. 40, no. 6, pp. 1259-1264, Jun. 2005. [15] E. Abou-Allam, J. J. Nisbet, and M. C. Maliepaard, "Low-voltage 1.9-GHz FrontEnd Receiver in 0.5-m CMOS Technology," IEEE J. Solid-State Circuits, vol. 36, no. 10, pp. 1434-1443, Oct. 2001. [16] J. Laskar, B. Matinpour, and S. Chakraborty, Modern Receiver Front-Ends. New York: Wiley, 2004. [17] B. Razavi, RF Microelectronics. pper Saddle River, NJ: Prentice-Hall, 1998. [18] P. Heydari, "Design and analysis of a performance-optimized CMOS UWB distributed LNA," IEEE J. Solid-State Circuits, vol. 42, no. 9, pp. 1892-1905, Sep 2007. [19] J-H. C. Zhan and S. Taylor, "A 5GHz Resistive-Feedback CMOS LNA for LowCost Multi-Standard Application," in IEEE ISSCC Dig. Tech., 2006, pp. 200-201. [20] A. Ismail and A. A. Abidi, "A 310 GHz low-noise amplifier with wideband LCladder matching network," IEEE J. Solid-State Circuits, vol. 39, no. 12, p. 2269 2277, Dec 2004. [21] T. H. Lee, The Design of CMOS Radio-Frequency Integrated Circuits, 2nd ed.: Cambridge, 2004. [22] H. Ma, S. J. Fang, F. Lin, and H. Nakamura, "Novel Active Differential Phase Splitters in RFIC for Wireless Applications," IEEE Trans. Microwave Theory and Techniques, vol. 46, no. 12, pp. 2597-2603, Dec. 1998. [23] D. H. Shin and C. P. Yue, "A Unified Modeling and Design Methodology for RFICs Using Parameterized Sub-Circuit Cells," in IEEE RFIC Symposium Digest, 2006, pp. 415-418. [24] R. Svitek and S. Raman, "DC offsets in direct-conversion receivers: characterization and implications," IEEE Microwave Magazine, vol. 6, no. 3, pp. 76-86, Sep. 2005. [25] [Online]. http://datasheets.maxim-ic.com/en/ds/MAX4144-MAX4146.pdf [26] A. A. Abidi, "General Relationships Between IP2, IP3, and Offsets in Differential Circuits and the Effects of Feedback," IEEE Trans. Microwave Theory and Techniques, vol. 51, no. 5, pp. 1610-1612, May 2003. [27] V. Aparin and L. E. Larson, "Linearization of monolithic LNAs using lowfrequency low-impedance input termination," in Eur. Solid-State Circuits Conf., 2003, pp. 137-140. [28] Y. Ding and R. Harjani, "A +18dBm IIP3 LNA in 0.35um CMOS," in IEEE ISSCC, 2001, pp. 162-163. [29] B. Toole, C. Plett, and M. Cloutier, "RF circuit implications of moderate inversion enhanced linear region in MOSFETs," IEEE Trans. Circuits and Systems, vol. 51, p. 319328, Feb 2004. [30] P. Wambacq and W. Sansen, Distortion Analysis of Analog Intergrated Circuits. Boston, MA: Kluwer, 1998. [31] D. R. Webster, D. G. Haigh, J. B. Scott, and A. E. Parker, "Derivative superposition A linearization technique for ultra broadband systems," in IEE Wideband Circuits, Modeling, Technology Colloq., 1996, p. 3/13/14.

139

[32] T. Kim, B. Kim, and K. Lee, "Highly linear receiver front-end adopting MOSFET transconductance linearization by multiple gated transistors," IEEE J. Solid-State Circuits, pp. 223-229, Jan 2004. [33] V. Aparin and L. E. Larson, "Modified derivative superposition method for linearizing FET low-noise amplifiers," IEEE Trans. Microwave Theory and Techniques, pp. 571-581, Feb 2005. [34] S. Ganesan, E. Sanchez-Sinencio, and J. Silva-Martinez, "A highly linear low-noise amplifier," IEEE Trans. Microwave Theory and Techniques, pp. 4079-4085, Dec 2006. [35] H.-T. Ahn and D. J. Allstot, "A 0.58.5-GHz fully differential CMOS distributed amplifier," IEEE J. Solid-State Circuits, vol. 37, p. 985993, Aug 2002. [36] R.-C. Liu, K.-L. Deng, and H.Wang, "A 0.622-GHz broadband CMOS distributed amplifier," in IEEE Radio Frequency Integrated Circuits Symp. Digest, 2003, p. 103106. [37] B. M. Ballweber, R. Gupta, and D. J. Allstot, "A fully integrated 0.55.5-GHz CMOS distributed amplifier," IEEE J. Solid-State Circuits, vol. 35, p. 231239, Feb 2000. [38] H. Adiseno, H. Maqnusson, and H. Olsson, "A 1.8-V Wide-Band CMOS LNA for Multiband Multistandard Front-End Receiver," in ESSCIR Dig. Tech., 2003, pp. 141-144. [39] E. Duvivier, G. Puccio, S. Cipriani, L. Carpineto, P. Cusinato, B. Bisanti, F. Galant, F. Chalet, F. Coppola, S. Cercelaru, N. Vallespin, J. Jiquet, and G. Sirna, "A Fully Integrated Zero-IF Transceiver for GSM-GPRS Quad-Band Application," IEEE J. Solid-State Circuits, vol. 38, pp. 2249-2257, Dec 2003. [40] C.-F. Liao and S. I. Liu, "A broadband noise-canceling MOS LNA for 3.110.6GHz UWB receiver," IEEE J. Solid-State Circuits, vol. 42, no. 2, p. 329339, Feb 2007. [41] B. Hofer, "Amplifier frequency and transient response (AFTR) notes," Tektronix Inc., Portland, 1982. [42] V. Aparin, G. Brown, and L. E. Larson, "Linearization of CMOS LNA's via optimum gate biasing," in IEEE Int. Circuits Systems Symp., 2004, pp. 748-751. [43] T. T. Ha, Solid-State Microwave Amplifier Design. New York, NY: Wiley, 1981. [44] V. Volterra, Theory of Functionals and Integro and Integro-Differential Equations. New York, USA: Dover, 1959. [45] D. D. Weiner and J. F. Spina, Sinusoidal Analysis And Modeling of Weakly Nonlinear Circuits. New York: Van Nostrand Reinhold, 1980. [46] R. Van Langevelde and F. M. Klassen, "Effect of gate-field dependent mobility degradation on distortion analysis in MOSFETs," IEEE Trans. Electron. Devices, vol. 44, p. 20442052, Nov 1997. [47] J. C. Pedro and J. Perez, "Accurate simulation of GaAs MESFETs intermodulation distortion using a new drainsource current model," IEEE Trans. Microwave Theory Tech., vol. 42, p. 2533, Jan 1994. [48] S. -G. Lee and J. -K. Choi, "Current-reuse bleeding mixer," Electronics Letters, vol. 140

36, no. 8, pp. 696-697, Apr 2000. [49] P. J. Sullivan, B. A. Xavier, and W. H. Ku, "Low voltage performance of a microwave CMOS Gilbert cell mixer," IEEE J. Solid-State Circuits, vol. 32, no. 7, pp. 1151-1155, Jul 1997. [50] L. A. Maceachern and T. Manku, "A charge-injection method for Gilbert cell biasing," in IEEE Canadian Conf. Electrical and Computer Engineering, 1998, pp. 365-368. [51] A. van der Ziel, Noise in Solid Satate Devices and Circuits. New York: John Wiley, 1986. [52] S. Wolf, Silicon Processing for the VLSI Era, Volume 3: The Submicron MOSFET. Sunset Beach, California: Lattice Press, 1995. [53] R. M. Fano, "Theoretical limitations on the broadband matching of arbitrary impedances," Massachusetts Institute of Technology, Boston, Technical Report No. 41, 1948. [54] F. Chen, J. Brotz, U. Arslan, C.-C. Lo, T. Mukherjee and G. K. Fedder, CMOSMEMS resonant RF mixer-filters, IEEE MEMS, Miami Beach, FL, Jan.-Feb. 2005, pp. 24-7. [55] W. J. Rugh, Nonlinear System Theory: The Volterra/Weiner Approach. Baltimore: John Hopkins University Press, 1981. [56] N. Wiener, Nonlinear Problems in Random Theory. Cambridge, Mass.: MIT Press, 1958. [57] J. F. Barrett, "The Use of Functionals in the Analysis of Non-linear Physical Systems," International Journal of Electronics, vol. 15, no. 6, pp. 567 - 615 , Dec 1963. [58] S. Maas, Nonlinear Microwave Circuits. Norwood, MA: Artech House, 1988. [59] J. Vuolevi, "Analysis, measurement and cancellation of the bandwidth and amplitude dependence of intermodulation distortion in RF power amplifiers," University of Oulu, Oulu, Dessertation ISBN 951-42-6514-9, 2001. [60] "HSPICE Users Manual Release 96.1," Meta-Software Inc., 1996. [61] E. Bedrosian, S.O. Rice, "The output properties of Volterra systems (nonlinear systems with memory) driven by harmonic and Gaussian inputs," in Proceedings of the IEEE, 1971, pp. 1688- 1707. [62] "First Report and Order, Revision of Part 15 of the Commissions Rules Regarding Ultra Wideband Transmission Systems," FCC, Washington, DC, 2002. [63] D. H. Shin, J. Park, and C. P. Yue, "A low-power, 3-5-GHz CMOS UWB LNA using transformer matching technique," in IEEE Asian Solid-State Circuits Conference, Jeju, Korea, 2007, pp. 95-98. [64] R. Van Langevelde and D. B. Klassen, "Accurate drain conductance modeling for distortion analysis in MOSFETs," in Proc. IEDM97, 1997, p. 313316. [65] R. Van Langevelde, L. F. Tiemeijer, R. J. Havens, M. J. Knitel, R. F. Roes, P. H. Woerlee, and D. B. Klassen, "RF-distortion in deep-submicron CMOS technologies," in Proc. IEDM00, 2000, p. 807811.

141

[66] B. Kim, J.-S. Ko, and K. Lee, "A new linearization technique for MOSFET RF amplifier using multiple gated transistors," IEEE Microw. Guided Wave Lett., vol. 10, no. 9, pp. 371-373, Sep. 2000. [67] A. Amer, E. Hegazi, and H. Ragai, "A low-power wideband CMOS LNA for WiMAX," IEEE Trans. Circuits and Systems, p. 48, 2007. [68] W. Y. Ali-Ahmad, "RF system issues related to CDMA receiver specifications," RF Design, pp. 22-32, Sep. 1999.

142

Vous aimerez peut-être aussi