Super-Regenerative Receiver For UWB-FM Rui - Hou - 2008 PDF

Super-regenerative Receiver for UWB-FM
Rui Hou
Department of Microelectronics
Delft University of Technology
September 20, 2008

Super-regenerative Receiver for UWB-FM
by
Rui Hou
A thesis submitted to the Department of Microelectronics, Faculty of Electrical
Engineering, Mathematics and Computer Science, Delft University of Technology,
in partial fulllment of the requirements for the degree of
Master of Science
in
Microelectronics
Supervisors:
Prof. John R. Long
Nitz Saputra
Thesis committee:
Prof. John R. Long
Dr. ing. Leo C. N. de Vreede
Dr. Ko A. A. Makinwa
Nitz Saputra
September 20, 2008

Abstract
UWB-FM is a low-complexity ultra-wideband (UWB) communication system de-
signed for short-range, low- and medium-data-rate wireless applications such as the
personal area network (PAN). These applications often require simple, integrated
receivers with low power consumption.
Most of previous work utilized delay-line demodulators for UWB-FM detection.
This coherent detection method oers the best performance in general but is not
necessarily power-ecient. Reported power consumption was around 20 mW for a
2 GHz RF bandwidth.
The goal of this research is to explore the possibility of reducing power consump-
tion of a UWB-FM receiver by exploiting the super-regeneration principle.
As a result, a fully integrated super-regenerative receiver in IBM 90-nm RF
CMOS technology is designed to detect 500 MHz bandwidth UWB-FM signals at
4.5 GHz. Circuit simulations show that a receiver sensitivity of -82.2 dB is attainable
for a 100 kbps baseband data-rate and 106 bit-error-rate. The whole receiver draws
an average of 2 mA from a 0.9 V supply.
This work is, according to the author's knowledge, the rst time the super-
regeneration principle is used for UWB-FM detection. The 1.8 mW power dissipation
is also by far the smallest among UWB-FM receivers reported in the literature.
In addition, the contribution of this work includes an innovative optimization of
the quenching waveform for WBFM detection and a novel low-power driven design
procedure for LNAs.
In conclusion, super-regenerative receivers are promising for short-range, low-
data-rate UWB-FM applications, due to their simplicity and low power-consumption.
i
ii
Acknowledgments
The work presented in this thesis could not have been done without the help and
inuence of many individuals.
Firstly, I would like to give my gratitude to my supervisor, Prof. John R. Long,
for his constant guidance and support. I have beneted greatly from his expertise,
inspiration, criticism and encouragement.
Additionally, I am very grateful to Ph.D student Nitz Saputra, who is also taken
care of my work from beginning to end. His broad knowledge has brought me out
of trouble quite a few times. Our discussions are always delightful, full of insights
and enlightenment. I thank him for his valuable feedback regarding my thesis.
I would also like to thank the committee members, Prof. Leo de Vreede and
Prof. Ko Makinwa for their time reading this thesis and attending the defense.
Working in the Electronics group has been quite a pleasant experience, due to
the kind and intelligent people there. In particular, I thank my college student
Yixiong Hu and Yousif Shamsa. We have been cooperating for two years, and it has
been a pleasure since then.
This two years studying in Delft University of Technology has been colorful and
full of adventure. Before my graduation, I would like to express my gratitude to all
the professors and teachers who gave me lectures and trainings. Evidently, I will
benet from the knowledge they imparted for the rest of my life. I thank all my
college students for their kindness and help.
My family is always supporting me, especially during my study in the Nether-
lands. I am very grateful for that.
Last but not least, I am deeply thankful for the love, encouragement and support
of my brilliant wife, Cheng Jiang.
Rui Hou
Delft, the Netherlands
iii
iv
Contents
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Purpose and Scope of the Research . . . . . . . . . . . . . . . . . . . 1
1.3 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Background 3
2.1 UWB-FM Modulation Scheme . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 Current Approach: Super-regeneration . . . . . . . . . . . . . . . . . 5
3 Super-regeneration 7
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Driven Parametric Oscillator Model . . . . . . . . . . . . . . . . . . 8
3.3 Solution of the ODE . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.3.1 General Solution . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.3.2 Particular Solution . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3.3 Complete Solution . . . . . . . . . . . . . . . . . . . . . . . . 15
3.4 Characteristics of an SRO . . . . . . . . . . . . . . . . . . . . . . . . 16
3.4.1 Sensitivity Curve . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.4.2 Oscillation Envelope . . . . . . . . . . . . . . . . . . . . . . . 17
3.4.3 Gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4.4 Frequency Response . . . . . . . . . . . . . . . . . . . . . . . 18
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4 Architecture Design 21
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2 System Specication . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2.1 RF Signal Characteristics . . . . . . . . . . . . . . . . . . . . 22
4.2.1.1 Operating Frequency . . . . . . . . . . . . . . . . . . 22
4.2.1.2 Transmitting Power and Propagation . . . . . . . . . 22
4.2.1.3 Noise Power and SNR . . . . . . . . . . . . . . . . . 23
v
vi CONTENTS
4.2.2 Sub-band Signal Characteristics . . . . . . . . . . . . . . . . . 23
4.2.2.1 Sub-band Signal Bandwidth . . . . . . . . . . . . . . 24
4.2.2.2 SNR Requirement . . . . . . . . . . . . . . . . . . . 24
4.2.3 Receiver Front-end Specication . . . . . . . . . . . . . . . . . 25
4.2.3.1 Noise Figure . . . . . . . . . . . . . . . . . . . . . . 25
4.2.3.2 Linearity . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2.3.3 Spurious Emission . . . . . . . . . . . . . . . . . . . 25
4.2.4 Specication Summary . . . . . . . . . . . . . . . . . . . . . 26
4.3 The FM Demodulation Scheme . . . . . . . . . . . . . . . . . . . . . 26
4.3.1 Indirect FM Detection . . . . . . . . . . . . . . . . . . . . . . 26
4.3.2 Limiter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.3.3 FM Discriminator . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3.4 AM Demodulator . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.4 Receiver Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.5 System-level Simulation . . . . . . . . . . . . . . . . . . . . . . . . . 33
5 Receiver Circuit Design 38

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.2 Super-regenerative Oscillator . . . . . . . . . . . . . . . . . . . . . . 38
5.2.1 Design Objective . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.2.2 Resonator Selection . . . . . . . . . . . . . . . . . . . . . . . . 39
5.2.3 The Schematic . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.2.4 Biasing Waveform . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.2.4.1 Frequency Response . . . . . . . . . . . . . . . . . . 41
5.2.4.2 Oscillation Envelope . . . . . . . . . . . . . . . . . . 44
5.2.4.3 Tank Conductance and Biasing Current Waveform . 44
5.2.5 Dynamic-range Power Trade-o . . . . . . . . . . . . . . . . . 47
5.2.6 The Inductor . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.2.6.1 Inductor Design Objective . . . . . . . . . . . . . . . 48
5.2.6.2 Inductor Selection . . . . . . . . . . . . . . . . . . . 49
5.2.7 Power Supply, Oscillation Amplitude and Oxide Integrity . . . 50
5.2.7.1 Power Supply Consideration . . . . . . . . . . . . . . 50
5.2.7.2 Oscillation Amplitude . . . . . . . . . . . . . . . . . 50
5.2.7.3 Dielectric Integrity Verication . . . . . . . . . . . . 51
5.2.8 Simulation Result . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.3 Peak Detector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.3.2 The Dierential Peak Detector . . . . . . . . . . . . . . . . . . 53
5.3.3 Biasing and Conversion Gain . . . . . . . . . . . . . . . . . . 54

CONTENTS vii
5.3.4 Parameter selection . . . . . . . . . . . . . . . . . . . . . . . . 57
5.3.5 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . 58
5.3.5.1 Transfer Characteristic . . . . . . . . . . . . . . . . . 58
5.4 Low-noise Amplier . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.4.1.1 Input Impedance . . . . . . . . . . . . . . . . . . . . 60
5.4.1.2 Gain and Noise Figure . . . . . . . . . . . . . . . . . 60
5.4.1.3 Reverse Isolation . . . . . . . . . . . . . . . . . . . . 61
5.4.1.4 Specication Summary . . . . . . . . . . . . . . . . . 61
5.4.2 The Schematic . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.4.3 Power-driven Design Procedure . . . . . . . . . . . . . . . . . 63
5.4.3.1 Minimum Ft . . . . . . . . . . . . . . . . . . . . . . 64
5.4.3.2 Minimum Current Density . . . . . . . . . . . . . . . 64
5.4.3.3 Minimum Power . . . . . . . . . . . . . . . . . . . . 66
5.4.4 Reverse Isolation . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.4.4.1 Unilateralization . . . . . . . . . . . . . . . . . . . . 67
5.4.4.2 Neutralization . . . . . . . . . . . . . . . . . . . . . 68
5.4.5 Input Impedance Matching . . . . . . . . . . . . . . . . . . . 68
5.4.5.1 Load Dependency of Input Impedance . . . . . . . . 69
5.4.5.2 Neutralization . . . . . . . . . . . . . . . . . . . . . 70
5.4.5.3 Low power, Low-Q Matching . . . . . . . . . . . . . 73
5.4.6 ESD Protection . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6 Auxiliary Circuits 80
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
6.2 Biasing Waveform Generator for SRO . . . . . . . . . . . . . . . . . 80
6.2.1 The Schematic . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.2.2 Saw-tooth Voltage Generator . . . . . . . . . . . . . . . . . . 81
6.2.3 Linear Transconductance . . . . . . . . . . . . . . . . . . . . . 81
6.2.4 OTA Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.2.4.1 Specication . . . . . . . . . . . . . . . . . . . . . . 82
6.2.4.2 The schematic . . . . . . . . . . . . . . . . . . . . . 84
6.3 Dynamic Biasing for LNA . . . . . . . . . . . . . . . . . . . . . . . . 86
6.3.1 LNA Minimum Start-up Time Analysis . . . . . . . . . . . . 86
6.3.2 Dynamic Biasing Circuit . . . . . . . . . . . . . . . . . . . . . 87
6.3.2.1 Comparator . . . . . . . . . . . . . . . . . . . . . . . 88
6.3.2.2 Logic Gates . . . . . . . . . . . . . . . . . . . . . . . 89

viii CONTENTS
6.4 Output Buer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.4.1 Device Testability . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.4.3 The schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.5 Current Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.5.1 The schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.5.2 Power Supply and Temperature Sensitivity . . . . . . . . . . . 97
6.5.3 Power-up Behavior . . . . . . . . . . . . . . . . . . . . . . . . 100
7 Receiver Performance 101

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
7.2 Test-bench Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . 101
7.2.1 Transient Noise Analysis . . . . . . . . . . . . . . . . . . . . . 101
7.2.2 The Test-bench . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.3 Global Power Reduction: A Step-controlled SRO . . . . . . . . . . . 104
7.3.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.3.2 The Adapted Design for Step-control . . . . . . . . . . . . . . 107
7.3.2.1 The Step-controlled SRO . . . . . . . . . . . . . . . 107
7.3.2.2 The Dynamic Biasing Circuit . . . . . . . . . . . . . 108
7.3.3 Test-bench Simulation Results . . . . . . . . . . . . . . . . . . 108
7.3.4 The Power Reduction of the Step-controlled Receiver . . . . . 108
8 Conclusions 114
8.1 Summary of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
8.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Bibliography 117
List of Figures
2.1 The time-domain waveform of the baseband data, subcarrier and
UWB signal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Power spectrum of the UWB-FM signal. . . . . . . . . . . . . . . . . 4
2.3 WBFM delay-line demodulator. . . . . . . . . . . . . . . . . . . . . 5
3.1 Super-regenerative oscillator. . . . . . . . . . . . . . . . . . . . . . . . 8
3.2 Slope-controlled state: the conductance G (t), sensitivity s (t) and
pulse envelope p (t). . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Step-controlled state: the conductance G (t), sensitivity s (t) and
pulse envelope p (t). . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.1 The top level block diagram of a UWB-FM receiver. . . . . . . . . . . 21
4.2 Block diagram of a generic FM-AM receiver. . . . . . . . . . . . . . . 27
4.3 The threshold eect of FM detection. . . . . . . . . . . . . . . . . . 28
4.4 Tuned and detuned FM-AM conversion. . . . . . . . . . . . . . . . . 30
4.5 Block diagram of the super-regenerative receiver for UWB-FM. . . . 32
4.6 Top-level simulation setup. . . . . . . . . . . . . . . . . . . . . . . . 34
4.7 The spectrum of the UWB-FM signal at the input of the receiver. . 34
4.8 The nonlinear time-varying model of the SRO. . . . . . . . . . . . . 35
4.9 The peak detector model. . . . . . . . . . . . . . . . . . . . . . . . . 35
4.10 The simulated baseband signal, oscillations, their envelopes and the
quenching waveform. . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.1 The schematic of the super-regenerative oscillator. . . . . . . . . . . 40
5.2 Frequency responses of slope- and step-controlled SROs. . . . . . . . 41
5.3 The optimization of slope-controlled frequency response. . . . . . . . 42
5.4 The optimization of step-controlled frequency response. . . . . . . . 43
5.5 The waveform of the total conductance and the tail current for the
SRO. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.6 Pole positions of an SRO. . . . . . . . . . . . . . . . . . . . . . . . . 46
5.7 Negative GM nonlinearity. . . . . . . . . . . . . . . . . . . . . . . . 47
5.8 The baseband signal, output oscillations and the SRO biasing current. 52
ix
x LIST OF FIGURES
5.9 The simplest peak detector. . . . . . . . . . . . . . . . . . . . . . . . 53
5.10 The dierential NMOS peak detector. . . . . . . . . . . . . . . . . . 54
5.11 Transfer characteristics of peak detectors. . . . . . . . . . . . . . . . 56
5.12 The simulated transfer characteristic. . . . . . . . . . . . . . . . . . 58
5.13 The oscillation and its envelope. . . . . . . . . . . . . . . . . . . . . 59
5.14 The schematic of the LNA. . . . . . . . . . . . . . . . . . . . . . . . 62
5.15 Simplied LNA input. . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.16 Transit Frequency vs. gm/Id. . . . . . . . . . . . . . . . . . . . . . . 65
5.17 Current density vs. gm/Id. . . . . . . . . . . . . . . . . . . . . . . . 65
5.18 Gate capacitance vs. gm/Id. . . . . . . . . . . . . . . . . . . . . . . 66
5.19 Unilateralization techniques. . . . . . . . . . . . . . . . . . . . . . . 67
5.20 Simplied model of inductive degenerated common-source. . . . . . . 69
5.21 The variation of load and input resistance of LNA. . . . . . . . . . . 71
5.22 Neutralization Currents. . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.23 Input resistance vs. source inductance. . . . . . . . . . . . . . . . . . 73
5.24 Matching network and its equivalent representation. . . . . . . . . . 74
5.25 The transient voltage and current of an ESD event. . . . . . . . . . . 76
5.26 Input port reection coecient, S11. . . . . . . . . . . . . . . . . . . 77
5.27 Forward power gain, S21. . . . . . . . . . . . . . . . . . . . . . . . . 78
5.28 Reverse power gain, S12. . . . . . . . . . . . . . . . . . . . . . . . . 78
5.29 The noise gure of the LNA. . . . . . . . . . . . . . . . . . . . . . . 79
6.1 The schematic of the biasing waveform generator. . . . . . . . . . . . 81
6.2 Amplitude spectrum of the biasing waveform. . . . . . . . . . . . . . 82
6.3 The schematic of the OTA. . . . . . . . . . . . . . . . . . . . . . . . 84
6.4 Transient simulation result of the biasing waveform generator. . . . . 85
6.5 The equivalent biasing network for LNA. . . . . . . . . . . . . . . . . 87
6.6 The schematic of the dynamic biasing network for LNA. . . . . . . . 88
6.7 Dynamic biasing waveforms. . . . . . . . . . . . . . . . . . . . . . . 89
6.8 The schematic of the comparator. . . . . . . . . . . . . . . . . . . . 90
6.9 The schematic of the NOT and NAND gate. . . . . . . . . . . . . . 90
6.10 The transient simulation result of the dynamic biasing network. . . . 92
6.11 The transient simulation result of the LNA under dynamic biasing. . 93
6.12 Output pulses and their power spectral density. . . . . . . . . . . . . 94
6.13 The schematic of the output voltage buer. . . . . . . . . . . . . . . 96
6.14 Bode plot after Miller compensation. . . . . . . . . . . . . . . . . . . 97
6.15 Voltage output of the peak detector and the buered version. . . . . 98
6.16 The schematic of the peaking current reference. . . . . . . . . . . . . 99
6.17 Power-supply and temperature sensitivity of the output current. . . 99

LIST OF FIGURES xi
6.18 Power-up behavior. . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
7.1 Block diagram of the receiver test-bench. . . . . . . . . . . . . . . . 102
7.2 Bandpass lter and its frequency response. . . . . . . . . . . . . . . 103
7.3 Receiver outputs under signal excitation. . . . . . . . . . . . . . . . 105
7.4 Receiver outputs without input signal. . . . . . . . . . . . . . . . . . 106
7.5 Conductance and Tail current waveform for step-control. . . . . . . . 107
7.6 The dynamic biasing circuits for the SRO and the LNA. . . . . . . . 109
7.7 Delay-line voltages and biasing currents of the SRO and the LNA. . 110
7.8 Receiver outputs under signal excitation. . . . . . . . . . . . . . . . 111
7.9 Receiver outputs without input signal. . . . . . . . . . . . . . . . . . 112

List of Tables
4.1 RF frequency and bandwidth specication. . . . . . . . . . . . . . . 22
4.2 Sub-band signal characteristics. . . . . . . . . . . . . . . . . . . . . . 24
4.3 System-level specication for the UWB-FM receiver front-end. . . . . 26
5.1 The comparison of FM-AM conversion gain of slope-, step-control and
the ideal case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.2 Inductor parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.3 Output SNR vs. LNA transconductance and noise gure budget. . . 60
5.4 LNA specication summary. . . . . . . . . . . . . . . . . . . . . . . 62
7.1 The comparison of current consumption between the slope- and step-
controlled receivers. . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
8.1 Performance comparison of UWB-FM receivers. . . . . . . . . . . . . 114
xii
Chapter 1
Introduction
1.1 Motivation
During the last 10 years, the rapidly advanced short-range communication tech-
nologies are creating new opportunities for interconnections such as personal area
networks (PAN) and wireless sensor networks (WSN). These networks have many
potential applications, ranging from clinical diagnosis to wildlife monitoring. The
network nodes are necessarily simple and highly integrated so they can be massively
produced to reduce the cost. Low power consumption is also an indispensable fea-
ture for these devices, since consumer products with bulky batteries are unattractive
and eld deployment of ephemeral sensors is unacceptable.
A wireless transceiver is a critical component providing the communication link
between distributed nodes. Reducing its cost and power consumption leads to global
cost and power saving, comfortably portable consumer products and extended life-
time of eld-deployed sensors.
1.2 Purpose and Scope of the Research

The goal of this research is to explore the possibility of reducing power consumption
of a UWB-FM receiver by exploiting the super-regeneration principle. UWB-FM [1]
is a low-power low-complexity ultra-wideband (UWB) modulation scheme designed
for short-range, low- and medium-data-rate (LDR and MDR) PAN and WSN ap-
plications. It utilizes the wideband frequency modulation (WBFM) to spread the
spectrum of a continuous carrier over a large bandwidth. Super-regeneration [2, 3]
is the process of amplifying radio frequency (RF) signals through the periodically
building-up oscillations. Super-regenerative receivers (SRR) are naturally suited for
low-power low-complexity applications, by virtue of their simple structure and low
power-consumption [4, 5, 6, 7, 8, 9, 10].
1
2 CHAPTER 1. INTRODUCTION
To demonstrate the feasibility of UWB-FM detection by super-regeneration and
its low-power potential, a receiver is implemented in IBM 90-nm RF CMOS tech-
nology. According to the requirements of short-range UWB-FM applications, this
receiver should have an operating frequency of 4.5 GHz, an RF bandwidth of 500
MHz, a data-rate of 100 kbps, a bit-error-rate of 106 and a line-of-sight communi-
cation range of 10 meters, with low power-consumption as the fundamental design
objective.
1.3 Thesis Organization

This thesis focuses on the design and implementation of a super-regenerative receiver
for UWB-FM applications. Chapter 2 provides the background about the UWB-FM
modulation scheme, previous work and the current approach. Chapter 3 analyzes
mathematically the super-regeneration theory, which forms a solid foundation for
the following design practice. In Chapter 4, the receiver specication is derived
based on its application background; an architecture is selected and the system-level
simulation result is presented to demonstrate the feasibility of the receiver. Chapter
5 concentrates on the circuit design of the RF part of the receiver, including a super-
regenerative oscillator (SRO), a peak detector and a low-noise amplier (LNA).
This chapter also presents an innovative optimization of the quenching waveform
for WBFM detection and a novel low-power driven design procedure for the LNA.
Chapter 6 focuses on the circuit design of the analog part of the receiver, including a
current saw-tooth waveform generator for the SRO, a dynamic biasing circuit for the
LNA, an output buer for testability and a current reference for biasing. In Chapter
7, the simulation set-up of the complete receiver is introduced and the simulation
result is presented. This chapter also details a global power reduction attempt and
an iteration of the design process. Chapter 8 concludes the thesis with a summary
of research results and discussion of future work directions.

Chapter 2
Background
2.1 UWB-FM Modulation Scheme

UWB-FM is a low-power low-complexity UWB modulation scheme targeting short-
range, low- and medium-data-rate wireless applications [1]. This scheme involves
double frequency modulations (FM), a low modulation-index frequency shift keying
(FSK) followed by a high modulation-index analog FM.
The double FM scheme is illustrated in Fig. 2.1. As shown in the graph, a
digital baseband signal, d (t), having a data rate of 20, 40 or 100 kbps, modulates
a triangular subcarrier of 1-2 MHz, m (t), using FSK with a modulation index of
1. The subcarrier, m (t), then modulates the RF carrier of 3-5 GHz, v (t), using
FM, to spread its spectrum to a bandwidth of 500 MHz. The modulation indexes
and the RF carrier frequency shown in the graph have been modied for the sake of
visibility.
The power spectrum of a UWB-FM signal centering at 4.5 GHz, being modu-
lated by a 1 MHz triangular wave, is shown in Fig. 2.2. The low-cost low-power
demodulation of this signal to reconstruct the subcarrier is the problem studied in
this research.
2.2 Previous Work

Most of previous work utilized delay-line demodulators for the UWB-FM detection
[11, 12]. The block diagram of such a demodulator is shown in Fig. 2.3. A delay
element rst converts the input FM signal into a PM signal. Then a multiplier,
operating as a phase detector, recovers the modulation signal. This coherent de-
tection scheme provides the best dynamic range in principle [13]. However, since a
mixer and a high gain amplier are used at the RF frequency, UWB-FM receivers
incorporating delay-line demodulators are not necessarily power-ecient.
3
4 CHAPTER 2. BACKGROUND
Baseband Signal d(t)

1.5
1
Amplitude (V)
0.5
0
0.5
1
1.5
0 1 2 3 4 5 6
Time (us)
Subcarrier m(t)
1.5
1
Amplitude (V)
0.5
0
0.5
1
1.5
0 1 2 3 4 5 6
Time (us)
RF Carrier v(t)
1.5
1
Amplitude (V)
0.5
0
0.5
1
1.5
0 1 2 3 4 5 6
Time (us)
Figure 2.1: The time-domain waveform of the baseband data, subcarrier and UWB
signal.
Power Spectrum of a UWBFM Signal

35
40
Power Spectral Density (dBm/MHz)
45
50
55
60
65
70
4 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Frequency (GHz)
Figure 2.2: Power spectrum of the UWB-FM signal.

2.3. CURRENT APPROACH: SUPER-REGENERATION 5
Multiplier
Vrf Vdemod
Amplifier
frf fsub
Delay Element
Figure 2.3: WBFM delay-line demodulator.
In the literature, power consumption of 2 GHz bandwidth receivers is reported.
One implementation excluding LNA, fabricated in 0.18 um Si-Ge BiCMOS technol-
ogy, consumes 9.6 mW power [11]. Another one including an LNA, implemented in
0.18 um CMOS process, consumes 19.4 mW [12].
2.3 Current Approach: Super-regeneration

This research explores the possibility of reducing power consumption of UWB-FM
demodulation by exploiting the super-regeneration principle.
Super-regenerative receivers were invented by Edwin H. Armstrong in 1922 [2].
Employing positive feedback, these receivers operated in an intermittent oscillatory
condition. Oscillations were periodically built up from weak RF excitation and then
quenched to avoid amplication saturation. In doing so, a huge amount of gain
could be obtained by a single stage. The amplication was so remarkable that a
single vacuum tube could amplify noise to an audible level.
However, super-regenerative receivers have never been in a dominant position,
since their drawbacks are also conspicuous. Firstly, super-regenerators are inherently
also an oscillator. The spurious emission of these receivers can easily cause inter-
ference. Secondly, classical super-regenerative receivers suer from poor frequency
selectivity. Thirdly, super-regenerative receivers are inherently frequency unstable.
Due to these drawbacks, super-regenerative receivers have limited narrow-band ap-
plications, such as garage-door openers, toys, and low cost walkie-talkies.
The literature has shown revived interests of super-regenerative receivers for
the last 10 years. Short-range low-data-rate wireless applications have a growing
demand for low-power low-cost receivers. Super-regenerative receivers, due to their
simplicity and huge gain, are naturally a candidate architecture. Typically, this
type of receivers have been designed for simple narrow-band modulation schemes
such as on-o keying (OOK) [4, 5, 6, 7, 8, 9, 10], providing that proper measures
[14, 15] are taken to improve the poor selectivity and inherent frequency instability.
Occasionally, they were also used to detect spread spectrum [16, 17] and pulse based
6 CHAPTER 2. BACKGROUND
UWB signals [18, 19, 20], beneting from the alleviated selectivity and frequency-
stability requirements of these applications.
The classical way of applying super-regeneration principle to FM detection is
based on the slope demodulation process [21]. In essence, a super-regenerator is
detuned from the RF center frequency so as to generate variable gain for frequency
variation. Other techniques [22, 23] are also seen in the literature.
Chapter 3
Super-regeneration
3.1 Introduction
Super-regenerative receivers, despite their structural simplicity , are quasi-periodical
nonlinear time-varying dynamical systems. The diculty in modeling their behavior
results the commonly used cut and try design methodology. In fact, the underlying
theory [3, 24] is so complex that it has never been understood by more than a handful
of people at a given time [25]. However, many trade-os occurring in this design
practice rely on the underlying mathematics, as will be presented in the rest of the
thesis.
Super-regenerative receivers have two modes of operations, the linear and the log-
arithmic mode. In the linear operation mode, growing-up oscillations are quenched
before the amplitude saturation shows its eect. In such a way, the peaking oscilla-
tion amplitude has a linear relationship with the initial RF excitation amplitude. In
contrast, the logarithmic operation mode quenches the oscillations after the ampli-
tude saturation. In doing so, all the pulses have the same amplitude. The dierent
RF power is distinguished by the dierent duration of oscillations, since large RF
excitation yields faster oscillation building-up, and vice versa. The duration of os-
cillations is logarithmically proportional to the initial RF excitation.
A logarithmic-mode super-regenerative receiver is modeled by a nonlinear ordi-
nary dierential equation with variable coecients, whose analytical solution does
not necessarily exist. Therefore, this chapter focuses on the linear-mode response
of the system. Since the properties of the two modes dier only at the end of an
oscillation build-up cycle, many results obtained in this analysis are also relevant to
the logarithmic mode.
The dierential equation modeling super-regenerative oscillators are presented
in Section 3.2 and solved in Section 3.3. Based on this solution, the characteristics
of super-regenerators are further analyzed in Section 3.4.
7
8 CHAPTER 3. SUPER-REGENERATION
i(t) -G1(t) G0 L C v(t)
Quenching
Oscillator
Figure 3.1: Super-regenerative oscillator.
3.2 Driven Parametric Oscillator Model

Super-regenerative oscillators are externally stimulated oscillators with time-varying
damping coecients. A circuit representation of such an oscillator is shown in Fig.
3.1, where a periodically time-varying negative conductance modies the damping
of a parallel RLC resonant tank. SROs originated from delay-line oscillators can be
better modeled as a variable-gain amplier in a positive feedback loop [24]. Since
they are mathematically equivalent, the rest of this chapter will use Fig. 3.1 as the
SRO model without losing generality.
The dierential equation characterizing the time varying dynamical system shown
in Fig 3.1 can be derived using Kirchho 's current law

dv
(t) 1
C + G (t) v (t) + v (t) dt = i (t) . (3.1)
dt L
Dierentiating both sides yields
" #
G (t) 1 G (t) i (t)
v (t) + v (t) + + v (t) = , (3.2)
C LC C C
where
G (t) = G0 G1 (t) (3.3)
is the time varying conductance of the resonant tank.
The waveforms of the varying conductance are shown in Fig. 3.2 and 3.3. A
quenching cycle starts from time 0. The total conductance of the tank is positive
during the period from time 0 to t1 , and negative during the period from t1 to t2 .
If the transition of conductance at t1 and t2 are slow, such that the regeneration
period includes many oscillation cycles, as shown in Fig. 3.2, the SRO is called
to be working in a slope-controlled state. On the other hand, if the transition of
conductance at t1 and t2 are fast compared with oscillation cycles, as shown in Fig.
3.2. DRIVEN PARAMETRIC OSCILLATOR MODEL 9
G0
G(t) 1111111111
0000000000
0000000000
1111111111
0000000000
1111111111
0000000000
1111111111
0000000000
1111111111
A+
0000000000t111111
1111111111 00000
00000
11111
00000
11111
0 A t2 t
G1 00000
11111
Damping Regeneration Super Damping
regeneration
1 s(t)
0 t1 t2 t
1 p(t) 11111111111111111111
00000000000000000000
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111
0 t111111111111111111111
00000000000000000000
t2 t
Figure 3.2: Slope-controlled state: the conductance G (t), sensitivity s (t) and pulse
envelope p (t).
G0
G(t) 1111111111
0000000000
0000000000
1111111111
0000000000
1111111111
0000000000
1111111111
0000000000
1111111111
A+
0000000000t111111
1111111111 00000
0 00000
11111
00000
11111
A t2 t
G1 00000
11111
Damping Super Damping
regeneration
1 s(t)
0 t1 t2 t
p(t) 1111111111111111111
0000000000000000000
0000000000000000000
1111111111111111111
1
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0 t1
0000000000000000000
1111111111111111111
t2 t
Figure 3.3: Step-controlled state: the conductance G (t), sensitivity s (t) and pulse
envelope p (t).
3.3, the SRO is called to be working in a step-controlled state.
The division of receiver operation into two states is historical and articial. In
the following analysis, it will be shown that the generic solution suitable for all types
of waveforms is too general to be of any practical use. The two operation states are
the two extreme situations, whose solutions in closed analytical forms exist. The
operation using any other conductance waveforms, e.g. a sinusoidal wave, can be
approximated by the combination of the two extreme cases with accuracy.
Equ. 3.2 is a second-order linear ordinary dierential equation (ODE) with
variable coecients. The next section derives the solution of this equation.
3.3 Solution of the ODE

The complete set of solutions of a second-order linear ordinary dierential equation
can be represented as
" #
h i c1
v (t) = vz1 (t) vz2 (t) + vp (t) (3.4)
c2
where the rst term is the general solution and the second term, the particular
solution. The general solution is the natural response, or zero-input response, of
the dynamical system, when no stimulus is applied. The particular solution, on the
other hand, is the forced response, or zero-state response, of the dynamical system,
when a particular stimulus is applied to a system initiating from the relaxed state.
In following subsections, the fundamental system of the general solution, the
vector containing vz1 (t) and vz2 (t), is rst derived from the corresponding homo-
geneous equation. Then the particular solution, vp (t), is derived using the method
of variation of the parameters. At last, the complete solution is determined and
discussed.
3.3.1 General Solution

Assuming zero input, we obtain the corresponding homogeneous equation of Equ.
3.2
h i
v (t) + 2 (t) 0 v (t) + 02 + 20 (t) v (t) = 0 , (3.5)
where
G (t)
(t) = (3.6)
2C0
is the time-varying damping factor, and
1
0 = (3.7)
LC
3.3. SOLUTION OF THE ODE 11
is the natural frequency. Equ. 3.5 represents an undriven parametric oscillator. The
standard procedure to solve this equation is to transform it into Hill equation by
eliminating the damping term.
This transformation is accomplished by a change of variables
t
v (t) = x (t) exp 0 (t) dt . (3.8)
0
Substituting Equ. 3.8 and
t
v (t) = [x (t) 0 (t) x (t)] exp 0 (t) dt (3.9)
0
n h io t
2 2
v (t) = x (t) 20 (t) x (t) + x (t) 0 (t) 0 (t) exp 0 (t) dt
0
(3.10)
1
into Equ. 3.5 yields
h i
x (t) + 02 02 2 (t) + 0 (t) x (t) = 0 . (3.11)
Equ. 3.11 can be further simplied into the equation of free oscillation
x (t) + 02 x (t) 0, (3.12)
on condition that
2 (t) 1 (3.13)
and

(t) 0 . (3.14)

We shall briey examine these conditions before going on. Equ. 3.13 implies
under-damping in both positive and negative conductance periods. More specically,
in order to validate the free oscillation simplication, the instantaneous quality factor
of the resonant tank should always satisfy
0 C 1
Q (t) = = 0.5 . (3.15)
G (t) 2 (t)
Equ. 3.14 implies a slow conductance variation compared with the oscillation fre-
quency. In other words, the system is assumed to be quasi-static to the oscillation
when we use the free oscillation approximation.
Given that conditions 3.13 and 3.14 are satised, Equ. 3.11 degenerates into
1 It is worth mentioning here that the Hill equation,

3.11, characterizes a special type of ampliers
parametric ampliers, which was a popular choice for high-performance LNAs back in the 1970s
[26].
Equ. 3.12, which is a linear constant-coecient ODE. Applying Laplace transform
to both sides yields
s2 X (s) sX 0 X 0 + 02 X (s) = 0 ,

(3.16)
which has the solution in complex frequency domain
sX (0 ) + X (0 )
X (s) = . (3.17)
s2 + 02
Performing inverse Laplace transform gives us the time-domain solution to Equ.
3.12 " #

X (0 )
x (t) = X 0 cos 0 t +

sin 0 t u (t) (3.18)
0
or in the vector form of Equ. 3.4
" #
h i Vi
x(t) = cos (0 t) sin (0 t) u (t) , (3.19)
Vq
where u (t) is the Heaviside step function and Vi and Vq are real constants represent-
ing the inphase and quadrature amplitude of the free oscillation to be determined by
boundary conditions. Substituting Equ. 3.19 into Equ. 3.8, we obtain the general
solution to the ODE of super-regenerative oscillators
t h "
i V
#
i
vz (t) = exp 0 (t) dt cos (0 t) sin (0 t) u (t) . (3.20)
0 Vq
Equ. 3.20 is the natural response, or zero-input response, of the super-regenerative
oscillator. We can observe from this equation that when no input is applied, the
output of an SRO is an exponentially growing or decaying sinusoidal signal with
frequency determined by the resonant tank. Whether the envelope is growing or
decaying depends on the integration of the damping function, or the accumulative
conductance. When the accumulative conductance is positive, the envelope of the
output is decaying, and vice versa.
3.3.2 Particular Solution

The particular solution of the inhomogeneous linear ODE 3.2 is based on the same
fundamental system as the one derived from its corresponding homogeneous equation
3.5. So it should be of the form
t h "
i v (t)
#
i
vp (t) = exp 0 (t) dt cos (0 t) sin (0 t) (3.21)
0 vq (t)
where vi (t) and vq (t) are the time-varying inphase and quadrature amplitudes to
be determined in the following part of this subsection.
If Equ. 3.21 satises the homogeneous equation, vi (t) and vq (t) need to be
constants, in other words,
t h "
i v (t)
#
i
exp 0 (t) dt cos (0 t) sin (0 t) =0. (3.22)
0 vq (t)
Substituting Equ. 3.21 into Equ. 3.2 and applying Equ. 3.22, we obtain
t h "
i v (t)
#
d i i (t)
exp 0 (t) dt cos (0 t) sin (0 t) = . (3.23)
dt 0 vq (t) C
Combining Equ. 3.22 and 3.23, we can solve vi (t) and vq (t) using Crammer's rule,
resulting

0 y (t) sin (0 t)

i(t)

C
sin (0 t) y (t) + 0 cos (0 t) y (t) sin (0 t) i (t)
vi (t) = = (3.24)
W (t) y (t) 0 C

y (t) cos (0 t) 0

cos (0 t) y (t) 0 sin (0 t) y (t) i(t)
C cos (0 t) i (t)
vq (t) = = , (3.25)
W (t) y (t) 0 C
where t
y (t) = exp 0 (t) dt (3.26)
0
and the Wronskian determinant

y (t) cos (0 t) y (t) sin (0 t)
W (t) = .

cos (0 t) y (t) 0 sin (0 t) y (t) sin (0 t) y (t) + 0 cos (0 t) y (t)
(3.27)
Integrating Equ. 3.24 and 3.25 and substituting the result back into Equ. 3.21
yields the particular solution
t
y (t) i ( )
vp (t) = [sin (0 t) cos (0 ) cos (0 t) sin (0 )] d . (3.28)
0 C 0 y ( )
We further simplify the solution by applying the angle sum identity and substituting
Equ. 3.26 into 3.28
t t
1
vp (t) = exp 0 (t) dt i ( ) exp 0 (t) dt sin 0 (t ) d .
0 C 0 0 0
(3.29)
We split the second exponential integral interval into [0, t1 ] and [t1, ]. The for-
mer one becomes independent and can be moved out, combined with the rst
exponential integral, yielding
t t
1
vp (t) = exp 0 (t) dt i ( ) exp 0 (t) dt sin 0 (t ) d .
0 C t1 0 t1
(3.30)
Equ. 3.30 can also be expressed in a compact form
t
1
vp (t) = Ks p (t) i ( ) s ( ) sin 0 (t ) d (3.31)
0 C 0
where t2
Ks = exp 0 (t) dt (3.32)
t1
is the super-regenerative gain,
t
p (t) = exp 0 (t) dt (3.33)
t2
is the normalized oscillation envelope, and
t
s (t) = exp 0 (t) dt (3.34)
t1
is the sensitivity curve.
Equ. 3.31 is the forced response or zero-state response of super-regenerative
oscillators under arbitrary stimulus. In this project, we are more interested in the
SRO response of a constant-envelope frequency-modulated signal
i (t) = A cos (t + ) , (3.35)
where A is the constant envelope, is the constant phase and is a slowly varying
frequency. The derivative of this FM signal is

d
i (t) = A sin (t + ) + t . (3.36)
dt
We further assume that the variation of the frequency in one quenching cycle is
negligible comparing with the instantaneous frequency, i.e.
d
t when t [0, Tq ] (3.37)
dt
which simplies Equ. 3.36 into
i (t) A sin (t + ) . (3.38)
Substituting Equ. 3.38 into Equ. 3.31 yields
t
A
vp (t) = Ks p (t) s ( ) sin (t + ) sin 0 (t ) d . (3.39)
0 C 0
Applying trigonometric product-to-sum identity, we have
t
A
vp (t) = Ks p (t) s ( )
20 C 0
{cos [ ( + 0 ) 0 t + ] cos [ ( 0 ) + 0 t + ]} d . (3.40)
Assuming 0 , the integral of the rst cosine term is nearly zero because of its
high frequency. As a result, Equ. 3.40 can be approximated by
t
A
vp (t) Ks p (t) s ( ) cos [ ( 0 ) + 0 t + ] d. (3.41)
20 C 0
Equ. 3.41 is the forced-response to the excitation given by Equ. 3.35.
3.3.3 Complete Solution

The complete solution of the linear varying-coecient ODE 3.2 is the sum of its
general solution given by Equ. 3.20 and its particular solution given by Equ. 3.41.
Generally, to solve dierential equations,

" we still
# need to apply boundary condi-
Vi
tions to nd out the unknown constants in the general solution, Equ. 3.20.
Vq
However, practical applications of super-regenerative oscillators make this nal step
unnecessary.
In SRO applications, we are more interested in the super-regeneration period. We
often make natural responses of SROs negligible compared with magnitudes of input
signals in that period, because forced responses provide amplication and natural
responses are input independent. The suppression of the natural response is achieved
by periodical quenching of oscillations. When the response of an SRO is primarily
determined by its input, the SRO is called to be working in the noncoherent state.
Otherwise, when the build-up of oscillations are not only triggered by the input
signal, but also initiated by the residue oscillation from a previous quenching cycle,
it is called hang-over and the SRO is called to be working in the coherent state.
The coherent state of operation is undesirable because of its reduced sensitivity
to incoming signals. Therefore, for properly designed super-regenerative oscillators

with negligible hang-over, their complete response in the super-regeneration phase
should be dominated by their forced response given by Equ. 3.41.
3.4 Characteristics of an SRO

The previous section derives the response of an SRO under a constant envelope FM
excitation. Based on this result, this section analyzes an SRO with respect to its
sensitivity period, oscillation envelope, gain and frequency response.
3.4.1 Sensitivity Curve

The sensitivity curve is derived in Equ. 3.34 and rewritten here as
t
s (t) = exp 0 (t) dt . (3.42)
t1
At t = t1 , it shows the maximum value of 1. When t > t1 , the damping is negative, so

the sensitivity decays with time. When t < t1 , the damping is positive but its integral
is negative, so the sensitivity also decays as t is getting away from t1 . Because of
its exponential dependence on time, this curve is normally sharp. Practically, it can
be seen as a sampling operation of the incoming signal at t1 . Because the value of
s (t) is close to zero when t is far away from t1 , we can change the integral interval
of Equ. 3.41 from [0, t] to [0, t2 ] or even [, +] when t is out of the sensitivity
period. We can then write the Equ. 3.41 as
t2
A
vp (t) Ks p (t) s ( ) cos [ ( 0 ) + 0 t + ] d (3.43)
20 C 0
or even
A
vp (t) Ks p (t) s ( ) cos [ ( 0 ) + 0 t + ] d. (3.44)
20 C
We now separately study the two types of operations, namely the slope-controlled
state and the step-controlled state. As shown in Fig. 3.2, the slope-controlled state
has gradual damping transition from positive to negative values. Equ. 3.6 in this
case is specied as
(t t1 )
sl (t) = , (3.45)
2C0
where denotes the absolute slope of conductance. Applying Equ. 3.42, we have
the sensitivity curve in slope-controlled state
" #
(t t1 )2
ssl (t) = exp . (3.46)
4C
3.4. CHARACTERISTICS OF AN SRO 17
It has a shape of Gaussian functions, as shown in Fig. 3.2.
In the step-controlled state, performing the same procedure gives us

G0

2C0
when t < t1
st (t) = , (3.47)
G1 when t t1
2C0
and
exp G0 (t1 t) when t < t1
2C
sst (t) = . (3.48)
exp G1 (tt1 ) when t t1
2C
The sensitivity curve has a shape of a double-sided decaying exponential function,
as shown in Fig. 3.3.
3.4.2 Oscillation Envelope
The oscillation envelope is derived in Equ. 3.33 and rewritten here as
t
p (t) = exp 0 (t) dt . (3.49)
t2
At t = t2 , it shows the maximum value of 1. When t > t2 , the damping and its
integral are both positive, so the sensitivity decays with time. When t < t2 , the
damping is negative but its integral is positive, so the sensitivity also decays as t is
getting away from t2 . Similar to sensitivity, because of its exponential dependence
on time, this curve is sharp. Practically, the oscillation envelope is a sharp pulse at
t2 .
Following the same procedure that we use to derive the sensitivity curves, we
can also obtain the oscillation envelope in slope-controlled state, in the form of a
Gaussian function " #

(t t2 )2
psl (t) = exp , (3.50)
4C
as shown in Fig. 3.2, and the one in step-controlled state, in the form of a double-
sided decaying exponential function

exp G0 (t2 t) when t < t2
2C
pst (t) = . (3.51)
G (tt2 )
exp 1
2C
when t t2
as shown in Fig. 3.3.

3.4.3 Gain
In the literature, the gain of a super-regenerative oscillator is divided into 3 parts,
namely the passive gain, the regenerative gain and the super-regenerative gain.
The passive gain originates from the passive resonant tank. It is dened as
1
K0 = , (3.52)
G0
which trivially converts an input current into an output voltage.
The regenerative gain quanties the amplication eect in the regeneration pe-
riod. It is dened as t2
G0
Kr = s (t) dt , (3.53)
2C 0
which is determined by the area under the sensitivity curve. A wide sensitivity
window yields high regenerative gain.
The super-regenerative gain originates from exponentially growing envelope in
the negative-conductance period. It is derived from Equ. 3.32 and rewritten here as
t2 t2
1
Ks = exp 0 (t) dt = exp G (t) dt . (3.54)
t1 2C t1
It is exponentially proportional to the area under the negative conductance curve.
Substituting the three expressions of gain into Equ. 3.43 yields
t2
A 1
vp (t) = K0 Kr Ks p (t) t2 s ( ) cos [ ( 0 ) + 0 t + ] d. (3.55)
0 s (t) dt 0
0
The meaning of Equ. 3.55 becomes clear if we assume = 0 . In this situation,
when the receiver is tuned to the frequency of the input signal, Equ. 3.55 becomes
vp (t) = AK0 Kr Ks p (t) cos (0 t + ) , (3.56)
which indicates that the response of an SRO to a tuned signal is an oscillation with
a gain of K0 Kr Ks and a pulse shape of p (t).
3.4.4 Frequency Response

Applying Euler's formula to Equ. 3.44 yields

ej (0 ) ej(0 t+) ej (0 ) ej(0 t+)

A
vp (t) = Ks p (t) s ( ) + d.
20 C 2 2
(3.57)
3.5. CONCLUSION 19
If we dene the Fourier transform of the sensitivity function as

() = s (t) exp (jt) dt = F {s (t)} , (3.58)

then Equ. 3.57 can be expressed as
A
Ks p (t) ( 0 ) ej(0 t+) + ( 0 ) ej(0 t+)

vp (t) = (3.59)
40 C
A
Ks p (t) Re ( 0 ) ej(0 t+)

= (3.60)
20 C
A
= Ks p (t) | ( 0 )| cos [0 t + + ( 0 )] (3.61)
20 C
where * stands for complex conjugate.
Equ. 3.61 implies that when the frequency deviation is small compared with the
resonant frequency, the frequency response of an SRO is approximately the complex
conjugate of the Fourier transform of the sensitivity curve.
Performing Fourier transform
2 1 2
eat e 4a (3.62)
2a
to Equ. 3.46 yields

r
2C C (0 )2
|sl ( 0 )| = e (3.63)

Performing Fourier transform to Equ. 3.48 yields
0 +
G0 G1
t jt t jt
|st ( 0 )| = e 2C e dt + e 2C e dt (3.64)
q 0
2 2
+ ( 0 )2 + ( 0)
(G0 G1 )2
G G
1 0
4C 2 4C 2
= h 2 ih 2 i (3.65)
G0 2 G1 2
4C 2
+ ( 0 ) 4C 2
+ ( 0 )
3.5 Conclusion
In this chapter, the characteristics of super-regenerative oscillators are mathemat-
ically examined. First, a driven parametric oscillator model is presented, and de-
scribed as a second-order linear ordinary dierential equation with varying coe-
cients. Then, the general solution and a particular solution under an excitation of a
constant-envelope frequency-modulated signal are derived. After that, the complete
response of the dynamical system in the super-regenerative period is approximated
by its forced response. At last, characteristics of SROs, namely sensitivity, oscilla-
tion envelope, gain and frequency response are derived for SROs working in slope-
and step-controlled state.
The summarized SRO characteristics and even the simplication assumptions
in the equation solving process given in this chapter are instrumental to the ar-
chitecture and circuit design presented in the rest of the thesis. For example, in
a later chapter, we will argue that the combination of a slope-controlled start-up
and a step-controlled quenching yields the desirable receiver behavior for UWB-FM
reception. In this stage, we can already predict by using Equ. 3.51 and 3.63 that the
resulting frequency response has a Gaussian shape and the oscillation pulses have
shapes of a double-sided decaying exponential function.

Chapter 4
Architecture Design
4.1 Introduction
In this chapter, the design issues in the architecture level are discussed. Section 4.2
derives the system specication based on the application background. The whole
design process tries to satisfy the specication given at the end of this section. From
the aspect of FM demodulation, Section 4.3 determines necessary blocks and their
characteristics to be implemented in this receiver. The resulting block diagram
of the UWB-FM receiver front-end is presented and explained in Section 4.4 and
simulated in Section 4.5 to verify its feasibility in the system level.
4.2 System Specication

A typical UWB-FM receiver consists of an LNA, a WBFM demodulator and a sub-
band FSK demodulator. The former 2 components constitute the receiver front-end
to be implemented in this project, as shown in Fig. 4.1. The specication for com-
plete UWB-FM receivers has been proposed in My Personal Adaptive Global NET
(MAGNET) project [27]. This section discusses the corresponding requirements to
UWBFM Receiver Frontend
WBFM FSK
LNA
Demodulator Demodulator
Figure 4.1: The top level block diagram of a UWB-FM receiver.
21
22 CHAPTER 4. ARCHITECTURE DESIGN
Parameter Value
RF center frequency 4.5 GHz

RF bandwidth (-10 dB) 500 MHz
Table 4.1: RF frequency and bandwidth specication.
the receiver front-end.
Subsection 4.2.1 and 4.2.2 outline the input RF signal and the output subband
signal characteristics. The receiver specication is calculated in Subsection 4.2.3
and concluded in 4.2.4.
4.2.1 RF Signal Characteristics

This subsection calculates the receiver input signal-to-noise ratio by considering the
transmitting power, channel propagation, antenna gain and noise bandwidth.
4.2.1.1 Operating Frequency

The RF signal frequency and bandwidth are listed in Tab. 4.1.
UWB-FM systems utilize the 3-5 GHz frequency band with several multiple
access schemes available. The two which have inuence to the receiver architecture
are listed below.
RF FDMA, 500 MHz RF bandwidth, 3 RF bands from 3 to 5 GHz [27].
Sub-carrier FDMA, 2 GHz RF bandwidth, 1 RF band from 3 to 5 GHz [28].
The reception of multiple subcarriers requires a certain linearity of a receiver front-
end, because nonlinear distortion intermodulates subcarriers. The RF bandwidth
for this project is 500 MHz, and only 1 subcarrier is carried in each RF channel.
Therefore, the linearity requirement of our receiver front-end is relaxed.
4.2.1.2 Transmitting Power and Propagation

The transmitting power of an indoor UWB transmitter is limited by FCC regula-
tion [29] to -41.3 dBm/MHz EIRP (Equivalent Isotropic Radiated Power), in the
frequency range between 3.1 and 10.6 GHz. This rule sets a maximum transmitting
carrier power of
PT KT = 41dBm/MHz + 10 log (500MHz) = 14 dBm, (4.1)
where PT is the transmitter power and KT is the antenna gain at the transmitter
side.
4.2. SYSTEM SPECIFICATION 23
The path loss of the communication channel is dened as
n
4d
P L (d) = (4.2)

where d, and n are the distance, carrier wavelength and propagation exponent,
respectively. For short-range, line-of-sight UWB applications, the propagation ex-
ponent seldom exceeds 2 [30]. In other words, transmission loss is often proportional
to less than square of the distance. Assuming free space propagation, which has a
propagation exponent of 2, we have the signal power at the receiver input of
P T KT KR
PR = (4.3)
(4d/)2
where KR is the antenna gain at the receiver side.
UWB-FM receivers do not have particular antenna requirements. Commonly
used omni-directional antennas provide a few dB gain (2.3 dB for a dipole antenna).
But imperfect matching could cause several dB of loss. So, it is still sensible to as-
sume the use of isotropic antennas at both ends of the channel. For a communication
range of 10 meters, we have the signal power at the receiver input of
4 10m
PR = 14dbm 20 log( 3108 m/s
) = 80dBm. (4.4)
4.5109 Hz
4.2.1.3 Noise Power and SNR
The noise power at the receiver input is
NR = 174dbm + 10 log(500MHz) = 87dBm. (4.5)
The signal-to-noise ratio at the input is thus
SN Ri = PR /NR = 80dbm + 87dbm = 7dB. (4.6)
4.2.2 Sub-band Signal Characteristics

UWB-FM reception involves an FSK demodulation of subband signals, which re-
quires a certain SNR at the output of the front-end to work properly.
According to [27], the subband signal is specied in Tab. 4.2. In this project,
we implement the maximum data-rate of 100 kbps.

Parameter Value
Sub-carrier frequency 1 MHz

Sub-carrier modulation FSK
Modulation index F SK = 1
Data rate 20, 40, 100 kbps
Table 4.2: Sub-band signal characteristics.
4.2.2.1 Sub-band Signal Bandwidth
The modulation index of FSK is dened as the ratio of the maximum frequency
deviation to the modulation frequency
24f
F SK = . (4.7)
fm
The maximum frequency deviation and bandwidth of subband can be calculated as
1
4fF SK = F SK fm = 50 kHz (4.8)
2
and
BWF SK = 4fF SK = 200 kHz. (4.9)
4.2.2.2 SNR Requirement
For orthogonal BFSK modulation, and optimum detection (with a matched lter)
we have the bit error rate r

1 Eb
Pe = erf c , (4.10)
2 2N0
where Eb /N0 is the energy per bit to noise power density ratio. According to ap-
plication analysis, a maximum bit error rate of 106 is required for low-data-rate
(LDR) physical layer [31]. In that case, we need a Eb /N0 of nearly 14 dB.
Finally, the link spectral eciency can be calculated by
R
= 0.5, (4.11)
BWF SK
and the signal-to-noise ratio needed from the output of WBFM demodulator is then
Eb R
SN Ro = = 14dB 3dB = 11dB. (4.12)
N0 BW
4.2. SYSTEM SPECIFICATION 25
4.2.3 Receiver Front-end Specication

4.2.3.1 Noise Figure
From Equ. 4.6 and 4.12, the noise gure of the receiver front-end is
N Ftot = SN Ri SN Ro = 4dB. (4.13)
A negative noise gure is indeed possible in UWB-FM receivers, since the subcar-
rier bandwidth is signicantly smaller than the RF bandwidth. Frequency-domain
ltering can be applied after the RF front-end to distinguish signal from noise. The
processing gain illustrating this property is
BWRF 500MHz
Gp = = = 34dB. (4.14)
BWF SK 200kHz
Therefore, noise-gure headroom of 30 dB exists for LNA and FM demodulator
when processing gain is taken into account.
4.2.3.2 Linearity
Non-linearity distortion jeopardizes the amplitude of a signal. For FM signals, since
information is carried only on the frequency, they are immune to amplitude distor-
tion. In practical FM systems, high-eciency power ampliers in transmitters and
amplitude limiting device in receivers are commonly used although they do cause
severe amplitude distortions.
Furthermore, the subcarrier in UWB-FM is FSK modulated, which does not
require linearity either. Therefore, even the frequency discriminator in UWB-FM
front-ends can be made nonlinear. This permits the use of integrators and super-
regenerative oscillators as frequency-to-amplitude converters. FM-discriminators are
discussed in Section 4.3.3.
4.2.3.3 Spurious Emission

This project involves the use of a super-regenerative oscillator as a detector which
could produce considerable power in RF frequencies and has the potential to violate
FCC or European power emission regulations. So an isolation specication should
be imposed to LNA to prevent the oscillation from coupling back to antenna.
According to [32], narrow-band spurious emissions dened in eective isotropic
radiated power (EIRP) for receivers should not exceed -47 dBm in frequency range
from 1 GHz to 12.75 GHz.

Parameter Value
RF center frequency 4.5 GHz

RF bandwidth (-10 dB) 500 MHz
Receiver Sensitivity -80 dBm
Noise Figure 30 dB
Spurious Emissions (EIRP) -47 dBm
Table 4.3: System-level specication for the UWB-FM receiver front-end.
4.2.4 Specication Summary

The system-level specication is summarized in Tab. 4.3.
4.3 The FM Demodulation Scheme

4.3.1 Indirect FM Detection
Information carried in frequency cannot be recovered directly [13]. An intermediate
transformation has to be performed to convert an FM signal into either an AM or
a PM signal. Then, the base-band signal can be reconstructed by a corresponding
amplitude or phase demodulation.
The choice between the FM-AM and the FM-PM intermediate transformation
involves a performance-complexity trade-o. All phase detection methods are syn-
chronous. In other words, phase detection unavoidably involves a comparison be-
tween the incoming PM signal and a reference signal. Coherent detection oers the
best performance at the cost of complexity and power consumption. On the other
hand, amplitude demodulation can be either synchronous or asynchronous. Nonco-
herent amplitude detectors, such as square-law and peak detectors are commonly
used for low-complexity receivers. The penalty paid for noncoherent detection is the
lack of phase selectivity and thus a degraded dynamic range.
UWB-FM receivers target on low-complexity low-power applications. The non-
coherent amplitude detection is clearly favored as long as the system specications
derived in Section 4.2 are satised. If the performance of the noncoherent detection
is not adequate, the coherent amplitude detection can still be performed. Therefore,
the FM-AM conversion scheme is chosen for this project.
The ideal FM-AM conversion is the dierentiation of the FM signal to time. A
frequency modulated sinusoidal carrier can be expressed as
s (t) = A cos [ (t) t + ] (4.15)
where A is the constant amplitude, is the constant initial phase and (t) is the
instantaneous angular frequency as a function of time. The derivative of this FM

4.3. THE FM DEMODULATION SCHEME 27
Limiter FMAM Converter AM Demodulator

Vo A Vo
LNA
Vi f Vi
Figure 4.2: Block diagram of a generic FM-AM receiver.
signal is
s (t) = A (t) sin [ (t) t + ] , (4.16)
which has an amplitude variation being proportional to the instantaneous frequency.
A general FM-AM receiver structure is shown in Fig. 4.2. The 3 building blocks,
namely, the limiter, frequency discriminator and AM detector are discussed in detail
in the following subsections.
4.3.2 Limiter
An ideal FM signal has a constant amplitude. In reality, however, channel propa-
gation induces amplitude noise. The eect of this noise can be derived by adapting
Equ. 4.15 and 4.16 into
sn (t) = A (t) cos [ (t) t + ] (4.17)
and
sn (t) = A (t) cos [ (t) t + ] A (t) (t) sin [ (t) t + ] . (4.18)
The rst term in Equ. 4.18 is called the radial component, which carries no informa-
tion but interference. The second term, named the tangential component, carries the
information, but is contaminated by the amplitude noise. In order to eliminate the
amplitude noise, amplitude limiters are frequently used before FM-AM converters
to regulate the amplitude.
The consequences of amplitude limiting are controversial. On one hand, a limiter
reduces amplitude noise. The total noise is partially discriminated and removed.
On the other hand, the nonlinearity of the limiter reduces the dynamic range of the
receiver as well. In other words, input signals are suppressed more than the noise.
Thus, the net eect of amplitude limiting depends on the input SNR.
The relationship between the output and input SNR of an ideal UWB-FM de-
modulator is shown in Fig. 4.3. The performance of another demodulator used for
commercial FM (20 kHz wide baseband and 75 kHz maximum frequency deviation)
is also plotted as a comparison. From the gure, it can be observed that FM demod-
Output SNR vs. Input SNR

100
UWBFM
90 Commercial FM
80
70
60
SNRo (dB)
50
40
30
20
10
0
0 5 10 15 20 25 30 35 40 45 50
SNRi * Gp (dB)
Figure 4.3: The threshold eect of FM detection.
ulators exhibit thresholds, below which the output SNRs degrade dramatically as
the input SNRs decrease. If an FM receiver operates above the threshold, amplitude
limiting yields improvement of output SNR. Conversely, if the input SNR is lower
than the threshold, limiting the amplitude leads to a degradation of output SNR
[13].
The UWB-FM demodulator to be designed has an input SNR of at least 7 dB
and a processing gain of 34 dB, as derived in Equ. 4.6 and 4.14. This leads to an
output SNR of around 90 dB. As shown in Fig. 4.3, the receiver is operating 1 dB
above the threshold of the UWB-FM system. Therefore, for a UWB-FM receiver,
the advantage and drawback of the amplitude limiter countervail each other. The
net improvement or degradation of the output SNR is limited, if there is any.
According to the preceding analysis of the amplitude limiter, two design decisions
are made. Firstly, the amplitude limiter is not going to be implemented in this
receiver, since it is not eective for UWB-FM. Secondly, the nonlinearity of the
LNA can be tolerated, since it has the same eect as an amplitude limiter. This
nonlinearity is commonly considered to be harmful to the dynamic range of the
receiver. However, for FM receivers, this distortion also suppresses the amplitude
noise.
4.3.3 FM Discriminator
FM discriminators convert FM signals into AM-FM ones. Any circuit having a non-
constant amplitude-frequency response can be used for this conversion. Frequency
domain lters are the most common FM discriminators.

Discriminator Types A rst order dierentiator has a transfer function of
H(j) = j (4.19)
which is an ideal FM discriminator in the sense that it converts frequency into
amplitude linearly. Furthermore, when additive white Gaussian noise (AWGN) with
a power spectral density of N0 is applied to the signal, the output noise power is
Nout = |H(j)|2 N0 = 2 N0 (4.20)
which quadratically shapes the noise in such a way that low frequencies have low
noise level. This noise shaping is the fundamental reason of the SNR improvement
of FM systems [13].
A rst order integrator has a transfer function of
1
H(j) = (4.21)
j
which converts frequency nonlinearly. For AWGN with a power spectral density of
N0 , the output noise power is
N0
Nout = |H(j)|2 N0 = (4.22)
2
which shapes the noise in a wrong way that the noise power is concentrated in the
low-frequency band. Although it has nonlinearity and inferior SNR performance, it
is used more often than dierentiators because it requires fewer components. An
LC tank is a typical integrator being biquadratically transformed from low-pass to
band-pass form.
Super-regenerative oscillators themselves have nonuniform frequency responses.
Slope-controlled SROs has a frequency response shaping as a Gaussian curve, as
shown in Equ. 3.63. The frequency response of the step-controlled SROs is similar
to the response of a pair of loosely coupled tuned circuits [3], as shown in Equ. 3.65.
These gain variations as a function of frequency can be exploited to do the FM-AM
conversion. However, both of the shapes are nonlinear and the noise is also shaped
in a wrong way.
Since a super-regenerative oscillator is going to be used for amplication, it is
straightforward to exploit its frequency response to convert FM into AM without
extra components, unless its inferior noise performance is intolerable.
Tuned Filters Vs. Detuned Filters Tuned dierentiators and integrators con-
vert FM signals into double-sideband (DSB) AM without a carrier. Tuned super-

Frequency Response Envelope of the AM Output

|A| |A|
1 1
0
0 0.5 1 1.5 2 t (us)
4.25 4.5 4.75 f (GHz)
0
Instantaneous Frequency of the FM Input
0.5
1.5
2
t (us)
(a) Tuned FM-AM conversion.

Frequency Response Envelope of the AM Output
|A| |A|
1 1
0
0 0.5 1 1.5 2 t (us)
4.25 4.5 4.75 f (GHz)
0
Instantaneous Frequency of the FM Input
0.5
1.5
2
t (us)
(b) Detuned FM-AM conversion.
Figure 4.4: Tuned and detuned FM-AM conversion.

regenerative oscillators act as FM-AM converters followed by full-wave rectiers,
since their frequency responses are not monotonic within the FM bandwidth. The
FM-AM conversion eect of a tuned lter is illustrated in Fig. 4.4 (a). If the fol-
lowing stage is a noncoherent AM detector, none of these signals can be correctly
demodulated.
The common remedy to tackle this problem is to detune a lter (or an SRO)
away from the center frequency of an FM signal in such a way that the amplitude
response of the lter is monotonic within the frequency band of interests, as shown
in Fig. 4.4 (b). The penalty paid is that half of the lter bandwidth is opened to
no signal but noise.
In this design, we exploit a special property of the UWB-FM modulation scheme,
so that a tuned lter (or SRO) and a noncoherent AM detector can be used together.
Since the signal to be recovered is an FSK modulated signal which carries informa-
tion only on its frequency, not on its amplitude, the information carried in the
frequency can be recovered, although a full-wave rectier distorts the amplitude
heavily and doubles the frequency. More specically, as shown in Fig. 4.4 (a), a 1
MHz triangular subcarrier is recovered as a 2 MHz triangular wave; a 0.9-1.1 MHz
FSK modulated signal is recovered as a 1.8-2.2 MHz FSK modulated signal.
4.3.4 AM Demodulator
AM detection can be synchronous or asynchronous. Asynchronous, or noncoherent,
detectors retrieve the amplitude modulus of the AM signal. The modulus of the
FM-AM-converted signal (Equ. 4.18) has an amplitude modulus of
q
Vmod = ksn (t)k = A2 (t) + A2 (t) 2 (t). (4.23)
Synchronous, or coherent, detectors produce the projection of the AM signal on a
reference signal. Dening the reference signal as
r (t) = B sin (t + ) , (4.24)
the projection of the AM signal on this reference signal is then
Vprj = sn (t) r (t) + snq (t) rq (t) (4.25)
where snq (t) and rq (t) are the quadrature of sn (t) and r (t), dened as
snq (t) = A (t) sin [ (t) t + ] + A (t) (t) cos [ (t) t + ] (4.26)
rq (t) = B cos (t + ) . (4.27)

UWBFM Receiver Frontend
LNA SRO Peak Output

Detector Buffer
Output
RF Circuits
Dynamic Waveform Current Auxiliary

Biasing Generator Reference Circuits
Delay Clock Amplitude Frequency
Figure 4.5: Block diagram of the super-regenerative receiver for UWB-FM.
Substituting Equ. 4.18, 4.24, 4.26 and 4.27 into Equ. 4.25 yields

Vprj = A cos A sin B sin + A sin +A cos B cos (4.28)
= AB sin cos +AB sin2 +AB sin cos +AB cos2 (4.29)
= AB (4.30)
Comparing Equ. 4.23 and 4.30, coherent detection can distinguish the ampli-
tude noise from signal while the noncoherent detection cannot. This drawback for
noncoherent detectors is intrinsic since they do not have a phase reference.
Coherent detectors, although provide the best achievable performance [13], are
power consuming since at least one multiplier is necessary for the projection calcu-
lation. Synchronization dissipates extra power if a local oscillator is used to produce
the reference signal.
In this design, the AM demodulation is performed by the simplest form of non-
coherent detectors, a peak detector, for its structural simplicity and low power con-
sumption. Its inferior noise performance is partially made up by the high gain of
the SRO in front of it.
4.4 Receiver Architecture

The block diagram of the super-regenerative receiver is shown in Fig. 4.5. The RF
part of the receiver, consisting of an LNA, an SRO and a peak detector, originates
4.5. SYSTEM-LEVEL SIMULATION 33
from the generic FM receiver plotted in Fig. 4.2. The LNA suppresses the noise
of following stages, matches the impedance of the antenna for maximum power
transmission and shields the high-power oscillation of SRO from coupling back into
the antenna. The SRO provides most of the receiver gain and converts the incoming
FM signal into AM. The peak detector extracts the envelope of the periodically
building-up oscillations as the output signal.
The auxiliary circuits consist of a waveform generator, a dynamic biasing cir-
cuit, an output buer and a current reference. The waveform generator produces
a certain biasing waveform to bias the SRO for a certain frequency response. The
dynamical biasing circuit shut the LNA down in the quenching period of SRO to
save power. The output buer drives the measurement instrumentation. And the
current reference provides biasing for all on-chip circuits.
The receiver is designed to be exibly congurable. Four external control signals
are accepted. The clock and the amplitude signals adjust the time-constants
and amplitude of the quenching waveform. The delay signal controls the warm-up
time of the LNA before the SRO starts to work. The frequency signal tunes the
oscillation frequency of the SRO.
4.5 System-level Simulation

The objective of the system-level simulation is to verify the theory discussed in
Chapter 3 and the feasibility of the UWB-FM super-regenerative receiver discussed
in this chapter.
The simulation method is the numerical integration of a nonlinear time-varying
dynamical model of the super-regenerative receiver stimulated by an UWB-FM sig-
nal. Mathworks Simulink is used to perform these simulations.
The top-level simulation setup is shown in Fig. 4.6. A 4.5 GHz, 500 MHz band-
width WBFM signal modulated by a 1 MHz triangular subcarrier is the excitation
of the receiver. The Channel block representing a 10-meter propagation attenuates
the signal to -80 dBm at 100 ohm resistance as specied in Equ. 4.4. The spectrum
of the signal at the input of the receiver is plotted in Fig. 4.7.
The three blocks shown at the bottom of Fig. 4.6, namely the GM, sro and
PeakDetector, model the LNA, SRO and peak detector in a receiver. Block Ga
provides a rectangular quenching waveform. The purpose of the feedback from the
output of the peak detector to the input amp of the SRO is to model the amplitude
limiting nonlinearity of the SRO.
The model of the SRO is plotted in Fig. 4.8. It is described by the second
order time-varying dierential equation 3.2 modeling the circuit shown in Fig. 3.1.
In addition, the nonlinearity is also modeled by the look-up table, so reduction of

FM K
i
Repeating ZeroOrder Channel
Sequence Hold FM
Modulator
Passband
1 2
Vbase Vrf
K i
GM Quench v Vi Vo 5
Venv
amp
PeakDetector
sro
Ga
3 4
Vquench Vosc
Figure 4.6: Top-level simulation setup.
Power Spectrum of the UWBFM Signal (Vrf)

100
105
Power Spectral Density (dBm/MHz)
110
115
120
125
130
135
140
4 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Frequency (GHz)
Figure 4.7: The spectrum of the UWB-FM signal at the input of the receiver.
1 du/dt K v 1 v 1 v
i 1
s s
Derivative invC1 v
Sum Integrator Integrator1
invC
Product
K
3
invLC
amp 1e3
Lookup Table G
G0
Subtract
2
Quench Product1
Figure 4.8: The nonlinear time-varying model of the SRO.
1
Vi
K u K
e 1
K
s
1 invnVt Math I0 1
Subtract1 invC Integrator Vo
Function
Vbias Subtract4
Subtract
Ve
0
1e5
Vref
Itail
K u K
e 1
0.4 K Vrefout
s
invnVt1 Math I1
Vth Subtract3 invC1 Integrator1 To Workspace
Function1
Subtract2 Ve1
Figure 4.9: The peak detector model.
negative conductance can be modeled when the oscillation amplitude increases.
The peak detector block, shown in Fig. 4.9, models the simplest peak detector,
formed by a diode, a capacitor and a current source. Despite its structural simplicity,
the correct operation relies on the dynamic nonlinearity of the circuit. In essence,
this block implements the implicit equation
t
1 VB + Vi VE
VE = I0 exp Itail dt. (4.31)
C 0 VT
In every integration step, VE is calculated by this equation and the output of the
peak detector is given by
Vo = VE VE1 . (4.32)
The simulation result is shown in Fig. 4.10. As shown in the graph, the restored
subcarrier is a pulse-amplitude-modulated version of the actual signal. Its frequency
(and presumably the FSK modulation index) is doubled as predicted in Subsection
4.3.3.
Intensive simulations are performed to verify that the gain, output pulse shape
and frequency response of the receiver are adjustable by tuning the amplitude, duty-
cycle, frequency and the shape of the quenching waveform.

Subcarrier (Vbase)
1
Amplitude (V)
0.5
0.5
1
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time (ns)
Oscillation Amplitude (Vosc)
300
200
Amplitude (mV)
100
0
100
200
300
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time (ns)
Oscillation Envelope (Venv)
250
200
Amplitude (mV)
150
100
50
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time (ns)
Conductance Waveform (Vquench)
1
Conductance (mS)
0.5
0.5
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time (ns)
Figure 4.10: The simulated baseband signal, oscillations, their envelopes and the
quenching waveform.
Chapter 5
Receiver Circuit Design
5.1 Introduction
In Chapter 4, receiver specications are derived and a block-level receiver structure
is proposed. The design and implementation of the RF building blocks, namely the
super-regenerative oscillator, the peak detector and the low-noise amplier, utilizing
IBM 90-nm CMOS technology are present in this chapter.
Since a detailed specication for each circuit block is not available at the begin-
ning, we follow a specic design procedure. First, the super-regenerative oscillator
and the peak detector are designed to realize the receiver functionality with minimal
power dissipation. Then, we use the noise and swing information of these stages to
determine the LNA specications, namely the gain, noise gure and reverse isola-
tion. After the LNA is worked out, quenching generators for SRO and LNA are
designed without inuencing the receiver performance. Finally, we implement the
output buer stage and the on-chip current reference.
The super-regenerative oscillator, peak detector and low-noise amplier are in-
troduced in Section 5.2, 5.3 and 5.4, respectively. The other building blocks, namely
the biasing waveform generator for SRO, the dynamic biasing circuit for LNA, the
output buer and the current reference are introduced in Chapter 6.
5.2 Super-regenerative Oscillator

5.2.1 Design Objective
The design goals for this super-regenerative oscillator are to provide maximal gain in
RF frequency, correct frequency response to convert FM into AM and to consume
minimal power. Maximal gain of SRO is desirable because rstly, it shields the
receiver back-end from jeopardizing system sensitivity, and secondly, the gain of
this stage is cheaper than that of an LNA.
38
5.2. SUPER-REGENERATIVE OSCILLATOR 39
5.2.2 Resonator Selection

Resonators are frequency selective elements used in oscillators. The quality of res-
onators directly inuences the noise and power consumption of SROs.
Generally speaking, o-chip discrete passives oer considerably better perfor-
mance. Nevertheless, we avoid using them as resonators for the following reasons.
First of all, the use of o-chip components increases the component count and de-
creases the integration level. These lead to an increased production cost which is
unfavorable for low-complexity low-cost UWB-FM applications. Secondly, o-chip
components are susceptible to parasitics caused by packaging and mounting. These
parasitics can be prohibitively large and even worse, dicult to control. Therefore,
in this project, we prefer integrated passives.
Commonly used resonators in the literature include LC tanks, distributed ele-
ments, quartz crystals, SAW devices and recently, BAW devices.
Quartz crystals and SAW devices have been the choice for high-performance
narrow-band SROs since their high quality factor solves the inherent low-selectivity
and frequency-instability problems of super-regenerative receivers. However, these
devices are incompatible with silicon technology and normally incapable of operating
at frequencies upper than 2 GHz.
BAW devices also have high Q factors and they can work at UWB frequencies.
Nonetheless, they need special MEMS fabrication processes or sophisticated wafer-
bonding technologies, which are commonly unavailable.
Distributed elements can be used as resonators such as quarter-wave lines, which
are commonly used in microwave oscillators. With high permittivity dielectrics,
transmission lines are also high-Q devices. However, physical dimensions of trans-
mission lines are normally unfavorable for integration. For instance, 4.5 GHz signals
have a wavelength of 67 mm in vacuum. A quarter-wave line with pure silicon as

the dielectric would have a length of 67/ rsillicon /4 = 4.9 mm.
LC tanks are the oldest type of lters used in SROs, originating from Armstrong's
patent in 1922. They have been the most popular choice because of the abundance of
discrete inductors and capacitors. However, integrated passives, especially inductors,
suer from the low Q caused by the resistance of metals and substrate coupling.
In conclusion, despite their high-loss, integrated lumped components are the best
choice with respect to fabrication cost and technology.
5.2.3 The Schematic

The schematic of the super-regenerative oscillator is plotted in Fig. 5.1. Its topology
is identical with a classical NMOS negative-GM oscillator. Transistors M1 and M2
are biased by the tail current source M3 to generate a negative conductance in

40 CHAPTER 5. RECEIVER CIRCUIT DESIGN
Vdd
Vfrq
M1 M2
Vbias M3
Figure 5.1: The schematic of the super-regenerative oscillator.
parallel with the LC resonation tank. A periodic biasing current yields a periodically
varying negative conductance, which causes the super-regeneration, mathematically
interpreted in Chapter 3.
A dierential oscillator structure is chosen for 3 reasons. First of all, it does not
consume more power than a single-ended one, for a given voltage swing. However,
it is capable to oscillate at an amplitude twice as much as a single-ended version,
for a given voltage supply. This characteristic is especially favorable to low-power
designs because supply voltage can be lowered without inuencing the voltage swing.
Secondly, a symmetrically driven inductor demonstrates a higher Q factor. As will be
discussed in Subsection 5.2.6, enhanced Q factor yields power and noise reduction of
oscillators. The latter one further leads to LNA power reduction. Thirdly, common-
mode noise can be rejected.
5.2.4 Biasing Waveform

According to Chapter 3, biasing waveforms control the oscillation envelope and fre-
quency response of super-regenerative oscillators. Being more specic, the positive
to negative conductance transitions decide the shape of frequency response and the
negative to positive ones control the shape of oscillation envelope.
Moreover, the operation states of an SRO can be categorized into two types,
determined by its biasing waveform. The transitions of conductance polarity are
gradual for a slope-control and sharp for a step-control, as shown in Fig. 3.2 and
3.3.
In this subsection, we examine the alternative frequency responses and pulse
shapes, select the most favorable combination for the low-power UWB-FM demod-
ulation and design the favored biasing waveform for the SRO.
Frequency response
1
Slopecontrolled
0.9 Stepcontrolled
0.8
0.7
Normalized gain
0.6
0.5
0.4
0.3
0.2
0.1
0
4.25 4.3 4.35 4.4 4.45 4.5 4.55 4.6 4.65 4.7 4.75
Frequency (GHz)
Figure 5.2: Frequency responses of slope- and step-controlled SROs.
5.2.4.1 Frequency Response

When the conductance transitions from positive to negative are slope-controlled,
the sensitivity curve of an SRO has a Gaussian shape (Equ. 3.46), whose Fourier
transform is also Gaussian shaped (Equ. 3.63). On the other hand, step-controlled
conductance transitions from positive to negative yield sensitivity curves as double-
sided decaying exponential functions (3.48), whose Fourier transform is shown in
Equ. 3.65. The frequency responses of these two types of quenching are shown in
Fig. 5.2.
The bandwidth of the frequency response is tunable for both slope- and step-
controlled SROs. A larger slope, or a more negative conductance yields a sharper
sensitivity curve and a atter frequency response, and vice versa.
In our UWB-FM receiver, we choose to use the slope-control. Our motivation is
discussed in the following paragraphs.
Since the frequency response of the SRO is used for FM to AM conversion, the
one with the maximum conversion gain is preferred. Direct observation of Fig. 5.2
may not lead to the correct conclusion because the curve shapes are adjustable.
A meaningful comparison should be made when both of them are tuned to their
best performance. So, we perform the bandwidth optimization in the following 3
paragraphs.
We use the slope factor , the changing rate of conductance, in Equ. 3.63 as
the tuning variable for slope-controlled SROs. A set of frequency response curves
are generated as shown in Fig. 5.3, when ranges from 105 to 106 Siemens per
Slopecontrolled Frequency response

1
dG=0.100 mS/ns
dG=0.178 mS/ns
0.9
dG=0.316 mS/ns
0.8 dG=0.562 mS/ns
dG=1.000 mS/ns
0.7
Normalized gain
0.6
0.5
0.4
0.3
0.2
0.1
0
4.25 4.3 4.35 4.4 4.45 4.5 4.55 4.6 4.65 4.7 4.75
Frequency (GHz)
Figure 5.3: The optimization of slope-controlled frequency response.
second (S/s). If we apply the UWB-FM signal (triangular wave modulated) to such
FM-AM conversion curves and change the horizontal axis to time, the curves would
represent exactly 1 period of the waveform in time.
The rst-harmonic amplitudes of these AM waves can be calculated by

1
Aslope = f () cos d (5.1)

where f () and are the AM waves and their phase. A parameter sweep shows
that the best FM-AM conversion curve is obtained when = 0.359 106 S/s, or
0.359 mS/ns, and the normalized output amplitude is 0.485.
We perform the same procedure to step-controlled SROs by using G1 , the
negative tank conductance as the sweeping parameter. A G1 ranging from 100 uS
to 3 mS yield the frequency response curves shown in Fig. 5.4. Applying Equ. 5.1 to
this set of curves, we obtain the best FM-AM conversion curve when G1 = 0.681
mS. The normalized output amplitude is 0.387.
The comparison of best conversion gain of slope- and step-controlled SROs are
listed in Tab. 5.1. The theoretically best conversion gain, 0.5, is also listed as a
reference. It is provided by a cosine shaped curve with 250 MHz bandwidth.
It can be observed from Tab. 5.1 that slope-controlled SROs oer a nearly
optimum frequency response. Therefore, we use a slope-controlled start-up for the
SRO.
Stepcontrolled Frequency response

1
G1=3.162 mS
0.9 G1=1.334 mS
G1=0.562 mS
0.8 G1=0.237 mS
G1=0.100 mS
0.7
Normalized gain
0.6
0.5
0.4
0.3
0.2
0.1
0
4.25 4.3 4.35 4.4 4.45 4.5 4.55 4.6 4.65 4.7 4.75
Frequency (GHz)
Figure 5.4: The optimization of step-controlled frequency response.
SRO Type Freq. Resp. Curve Max. FM-AM Conv. Condition
Slope-cntl Gaussian 0.485 = 0.359 mS/ns

Step-cntl Equ. 3.65 0.387 G1 = 0.681 mS
Ideal Cosine 0.5 BW3dB = 250 MHz
Table 5.1: The comparison of FM-AM conversion gain of slope-, step-control and
the ideal case.
Conductance Waveform
1
Conductance (mS)
0.5
0.5
1
0 10 20 30 40 50 60 70 80 90
Time (ns)
Tailcurrent Waveform
1.5
Current (mA)
0.5
0 10 20 30 40 50 60 70 80 90
Time (ns)
Figure 5.5: The waveform of the total conductance and the tail current for the SRO.
5.2.4.2 Oscillation Envelope

When the conductance transitions from negative to positive is slope-controlled, the
oscillation envelope of an SRO has a Gaussian shape (3.50), as shown in Fig. 3.2.
On the other hand, step-controlled conductance transitions from negative to positive
yield exponential growing and decaying envelope (Equ. 3.51), as shown in Fig. 3.3.
A step-controlled quenching is preferred, because the peak oscillation amplitude
appears when the biasing current is maximum. For a slope-controlled quenching,
the peak amplitude happens when the conductance polarity is changing. At this
time, the biasing current is not maximum, which may lead to gain compression and
reduced dynamic range.
5.2.4.3 Tank Conductance and Biasing Current Waveform

A slope-controlled start-up and a step-controlled quenching yield a saw-tooth-like
waveform. The designed conductance waveform is plotted in Fig. 5.5. In the gure,
the slope of the positive to negative conductance transition is set to 0.359 mS/ns
according to Tab. 5.1.
The conductance waveform can be eciently generated by a saw-tooth current

source. The waveform of this source is shown in Fig. 5.5. The maximum current is
1.2 mA determined by the voltage swing requirement and the passive tank resistance.
The current ramp takes 15 ns, this implies a current slope of 0.08 mA/ns.
At the starting-up phase, MOSFETs of the cross-coupled pair are in weak inver-
sion, the negative transconductance provided by the pair is
1 1 Id 1 Itail
GM = gm = = (5.2)
2 2 nVT 4 nVT
where gm , Id and n are the small-signal transconductance, drain current and slope
factor of one transistor, and VT is the thermal voltage. So before transistors M1 and
M2 shown in Fig. 5.1 are saturated, the negative conductance is proportional to the
biasing current. The IBM 90-nm NMOS we used has a slope factor, n, of 1.37. So
in this linear region, the conductance slope is calculated by
0.08 mA/ns
= = 0.561 mS/ns. (5.3)
4 1.37 26 mV
As the current grows larger, the I-V relation deviates from the exponential law.
The absolute value of the conductance slope is becoming smaller and smaller. The
critical point happens at this phase. When Itail reaches 340 uA, the positive-to-
negative transition of the conductance occurs. Ideally, at this critical point, the
conductance slope should be 0.359 mS/ns.
When current grows large, the I-V behavior of pair transistors is more and more
close to the square law. Then the negative transconductance is more accurately
approximated by
1 1p 1p
GM = gm = 2KId = KItail , (5.4)
2 2 2
where K = Cox W/L. From this equation, we can determine the width of the MOS-
FETs. First of all, the width inuences the gain of the SRO. A larger width yields
a larger -GM and thus a larger negative area below this -GM curve. Secondly, the
width aects the position of the critical point. A smaller width yields a compara-
tively higher current density and thus an early transition and a longer square-law
region.
We select the width based on the amount of gain needed. For a given input RF
signal strength, specic LNA gain and output oscillation amplitude, the gain of the
SRO is a determinable constant. Since the maximum biasing current is set by the
voltage swing, the slope is xed by the frequency response, only width is adjustable
to provide the specic amount of amplication. As a result, we determine the
transistor width to be 12 um.
When the receiver needs the gain control, for example, when a communication
Im
1 2
Re
1 2
Figure 5.6: Pole positions of an SRO.
range is smaller than the maximum designed distance, then a smaller gain is still
attainable after fabrication by slightly tuning the maximum current or the rise-time
of the ramp. These adjustments detune the frequency response and the oscillation
amplitude from their optimum values. But since decreasing the gain implies that a
shorter communication distance and a better input SNR are available, the receiver
performance is not degraded.
The GM compression eect shown in Equ. 5.4 is actually desirable. The reasons
are explained in the following 2 paragraphs.
First of all, the absolute conductance has an upper bound for certain frequency
stability. Equ. 3.15 in Chapter 3 is a high-Q assumption we made to simplify a
parametric oscillator response into a free oscillation. Its graphical representation is
shown in Fig. 5.6. A passive resonant tank has a pair of conjugate poles at position
1, and negative conductance moves poles to position 2 to start an oscillation. The
parameters
1
0 = LC
is the natural oscillation frequency,
G0 G1 (t)
= 2C
is the damping factor, and
p
= 02 2 is the damped oscillation frequency.
GM Nonlinearity
1
0.9
0.8
0.7
0.6
GM / gm
0.5
0.4
0.3
0.2
0.1
0
0 0.05 0.1 0.15 0.2 0.25 0.3
Voltage Swing (V)
Figure 5.7: Negative GM nonlinearity.
When Equ. 3.15 is satised, the damping oscillation frequency is approximately the
natural frequency, i.e.
0 . (5.5)
If the negative GM is too large, however, Equ. 5.5 does not hold any more. Then
frequency stability is jeopardized.
Secondly, the tail current has a lower bound for certain voltage swing, since the
voltage swing across a resonant tank is linearly proportional to the biasing current,
Vsw Rtank Itail . (5.6)
In conclusion, we desire a certain slope-controlled conductance waveform across
the resonant tank to provide an optimum frequency response for FM-AM conver-
sion. Current biasing is suitable to generate this waveform for 2 reasons. Firstly,
when MOSFETs are in weak inversion, the negative conductance is linearly propor-
tional to biasing current. Secondly, when the MOSFETs are saturated, the negative
conductance is weakly dependent on biasing current (they have a square-root re-
lationship, as shown in Equ. 5.4), which relaxes the trade-o between frequency
stability and voltage swing.
5.2.5 Dynamic-range Power Trade-o

In Subsection 5.2.4, we propose the current waveform, which has a maximum of 1.2
mA. This value is a trade-o between the dynamic range and the power consumption.
All previous analysis assumes a linear time-varying negative conductance. However,

the practical negative conductance generated by a cross-coupled pair is oscillation-
amplitude dependent. A curve showing this nonlinearity is plotted in Fig. 5.7.
Oscillators utilize this nonlinearity as the automatic amplitude control mechanism.
Super-regenerative oscillators have 2 operation modes, the linear and the log-
arithmic mode. The latter one is normally undesirable because it performs the
dynamic range compression. Intuitively, in logarithmic mode, large signals have
smaller gain than that of small signals and noise. So the SNR at the output is in-
herently reduced, even if the amplication is noiseless. The gain compression caused
by GM nonlinearity also has this undesirable dynamic-range compression eect.
The solution to prevent the dynamic-range compression is to increase the power
dissipation. A larger biasing current leads to a larger available voltage swing, and
more linear gain when the swing is small.
The maximum biasing current is determined based on this trade-o. Intensive
simulations are performed with the biasing current gradually increased from a small
value, until an acceptable output SNR is obtained. In this way, certain amount of
gain nonlinearity is tolerated and minimum power consumption is guaranteed.
5.2.6 The Inductor

5.2.6.1 Inductor Design Objective
Our design goal for the SRO is to reduce power consumption. For an RLC resonant
tank the dissipated power is expressed as
0 Etank
Ploss = (5.7)
Q
where 0 , Etank and Q are the resonant frequency, energy stored in the tank, and
the quality factor. In order to reduce power dissipation, we could either increase Q
or decrease Etank . However, the latter method reduces the oscillation voltage swing
as well, since Etank can be further expressed as
1 2
Etank = CVsig (5.8)
2
where C is the tank capacitance and Vsig is the voltage across the capacitor. The
minimal oscillation amplitude is determined by the noise
kT
Vn2 = , (5.9)
C
and an SNR requirement

(Vsig/ 2)2 Etank
SN R = = , (5.10)
Vn2 kT
where k and T are the Boltzmann's constant and absolute temperature. A clear
power SNR trade-o is obtained by combining Equ. 5.7 and 5.10
SN R Q
= . (5.11)
Ploss 0 kT
Since 0 kT is practically a constant, if we want high SNR and low power consump-
tion, maximizing the Q factor is the ultimate way.
5.2.6.2 Inductor Selection
In IBM 90-nm RF CMOS design library, a scalable model for symmetrical inductors
within a certain dimension range is available. More specically, within an outer
dimension of 75-to-300 um, a coil width of 7.02-to-15 um and a line spacing of 3-to-
5 um, we have a symmetrical inductor fully modeled without EM simulations. Our
goal is to gure out the parameter combination yielding the highest Q factor and a
self-resonance frequency reasonably above the operating frequency of 4.5 GHz.
Since the trade-os are multidimensional, an enumeration of all parameter com-
binations with a cost function of Q factors is a straightforward solution. However,
this brute-force searching process is computationally intensive. In order to shrink
the design space, several heuristics are considered.
First of all, an M1 metal ground plane is used to prevent capacitive coupling
to the substrate. This measure reduces the induced current owing in the ohmic
substrate, which increases the Q factor. Furthermore, noise coupling between the
inductor and the substrate is also shielded. The penalty of this measure is the
reduced self-resonant frequency caused by the increased capacitance.
Secondly, the minimum line spacing of 3 um is used to maximize the magnetic
coupling among adjacent lines. This measure increases the inductance and the
Q factor. However, the self-resonant frequency is also inuenced because of the
increased capacitance between adjacent conductors.
Intensive simulations are performed to enumerate the value of outer dimension,
conductor width and the number of turns. As a result, the inductor with the highest
Q factor and a reasonably high self-resonant frequency is selected with its parameters
shown in Tab. 5.2

Parameter Value
Outer dimension 300 x 300 um

Line width 15 um
Line spacing 3 um
Number of turns 2
Q factor 23
Self-resonance frequency 6.65 GHz
Table 5.2: Inductor parameters.
5.2.7 Power Supply, Oscillation Amplitude and Oxide In-

tegrity
For the SRO shown in Fig. 5.1, the DC value of the oscillation is the supply voltage.
The maximal voltage swing can reach up to twice as large as this value. While these
properties are desirable for low-power high-performance oscillators, high-voltage op-
eration threatens the integrity of gate oxide. As the device dimensions shrink down,
oxide thickness also decreases. Thin oxide is vulnerable to stressing voltages. The
thickness of gate oxide for IBM 90-nm CMOS is 1.4 nm, which has a breakdown
voltage of 1.32 V. Even below this value, gradual oxide degradation still happens
under a long-duration of voltage stress.
5.2.7.1 Power Supply Consideration

Minimal supply voltage is desired for low-power operation and maximal oscillation
voltage swing. For a nominal voltage supply of 1 V, a 20% variation of this voltage
is normally acceptable. So the positive voltage swing of this SRO is limited to be 1.2
V. The verication of dielectric integrity to be discussed in a later paragraph shows
that this maximum of 1.2 V is not going to cause oxide breakdown or degradation.
Further increasing the voltage swing and reducing the power consumption requires
the decreasing of supply voltage.
However, the minimum supply voltage is limited by LNA which is likely to use
a cascode structure for reverse isolation. Moreover, the input transistor is likely
to be biased in strong inversion for a decent transit frequency, which needs a gate-
source voltage of around 600 mV. Taking these factors into consideration, we decide
a supply voltage of 900 mV.
5.2.7.2 Oscillation Amplitude

We pursue the maximum oscillation amplitude despite its adverse inuence on power
consumption, because the lack of receiver back-end specication requests the uni-
versality and adaptability of this receiver front-end. In other words, this frond-end
should not impose severe noise requirement to its following stages. This objective
demands a large output amplitude.
Given the supply voltage of 900 mV and a maximum voltage limit of 1.2 V, the
voltage swing amplitude should not exceed 300 mV, i.e. a single-ended peak-to-peak
swing of 600 mV and a dierential peak-to-peak swing of 1200 mV.
5.2.7.3 Dielectric Integrity Verication
The absolute maximum oxide breakdown voltage can be estimated according to
IBM's technology design manual [33]. For a product lifetime of 100 KPOH (thousand
of power-on hours), a maximum temperature of 85 C, an area of gate oxide of
2 200um 100nm and a power-supply voltage of 900 mV, we have a maximum
permitted transient voltage of 1.6 V. So a 300 mV voltage swing amplitude (i.e. 1200
mV dierential peak-to-peak oscillation) is not going to cause oxide breakdown or
degradation.
5.2.8 Simulation Result

A transient simulation is performed to the super-regenerative oscillator. The simu-
lation result is shown in Fig. 5.8.
The stimulus to the SRO is a 4.5 GHz RF carrier being modulated by a 1 MHz
triangular wave, with a modulation bandwidth of 500 MHz, as specied in the
UWB-FM specication. A period of the modulation signal is shown at the top of
Fig. 5.8, from 0 to 1000 ns. At t=0 and 1000 ns, the subcarrier has the minimum
value, yielding an RF signal with an instantaneous frequency of 4.25 GHz. At
t=500 ns, the subcarrier is at its maximum, generating the RF signal with 4.75
GHz instantaneous frequency. At t=250 and 750 ns, the subcarrier is 0, so the RF
frequency at those time is 4.5 GHz.
The single-ended voltage output of the SRO is shown in the middle of Fig. 5.8.
The frequency response of the SRO can be observed. When the input signal has
a frequency of 4.5 GHz, the SRO outputs the maximum oscillation amplitude of
around 0.3 V. The deviation of frequency from its center yields the reduction of
output amplitude. When the RF frequency is at 4.25 or 4.75 GHz, the output
amplitude is the minimum.
The biasing current waveform is shown in the bottom of Fig. 5.8. In every
biasing period of 35 ns, the current ramps from 0 to 1.2 mA in 15 ns. Then it drops
to 0 at once and keeps a 0 value for the rest 20 ns. This SRO draws 0.257 mA in
average from a 0.9 V supply, yielding a power consumption of 231 uW.

Subcarrier
0.5
Voltage (V)
0.5
0 100 200 300 400 500 600 700 800 900 1000
Time (ns)
SRO Oscillation Output
0.3
0.2
Voltage (V)
0.1
0.1
0.2
0 100 200 300 400 500 600 700 800 900 1000
Time (ns)
SRO Biasing Current
1.5
Current (mA)
0.5
0
0 100 200 300 400 500 600 700 800 900 1000
Time (ns)
Figure 5.8: The baseband signal, output oscillations and the SRO biasing current.
5.3. PEAK DETECTOR 53
Vi Vo
Figure 5.9: The simplest peak detector.
5.3 Peak Detector

Peak detectors extract slow-varying oscillation envelope by following the oscillation's
peaks. They are frequently used as simple amplitude demodulators. In this project,
a peak detector is designed to demodulate an AM signal, which is converted from
UWB-FM by the super-regenerative oscillator.
The design goal for this peak detector is the maximum conversion gain and low
power dissipation. The conversion-gain denition of peak detectors are identical
with that of mixers, which is the output envelope magnitude divided by the input
oscillation amplitude.
5.3.2 The Dierential Peak Detector

The simplest peak detector is shown in Fig. 5.9, which consists of a diode, a capacitor
and a resistor. Assuming an ideal diode for simplicity and an RC time constant much
larger than the incoming signal, we can describe its operation as follows. When a
peak comes and Vi > Vo , the diode conducts; Vo follows Vi and the capacitor is
charged. After the peak, Vi < Vo , the diode is tuned o; the capacitor hold the peak
voltage and the resistor slowly discharges the capacitor. The discharge speed is set
in such a way that when the next peak comes, Vo is always slightly smaller than
Vi to close the diode, charge the capacitor and track the pulse while maintain the
smoothness of the output envelope.
The circuit used in our design is shown in Fig. 5.10. Two diode-connected
MOSFETs, M1 and M2, substitute the diode in Fig. 5.9 to track both positive and
negative peaks of a dierential oscillation. A tail current source M3 discharges the
capacitor instead of the resistor. A dummy duplicate of this circuit, M1'M3' copies
the same voltage drop to output a reference voltage, so that Vo will be zero when
oscillation envelope is zero.
The dierential structure is used for 2 reasons. First of all, it tracks the dieren-
tial output peaks of both polarities. So the discharge speed could be twice as much
Vdd Vdd
Vip Vin
M1 M2 M1 M2
M3 Ctail + Vo - M3
Figure 5.10: The dierential NMOS peak detector.
as before without inuencing envelope smoothness. A faster discharge is desirable in
this receiver because the oscillation envelope is exponentially shaped pulses, which
are sharp and require fast tracking. The second reason to use a dierential struc-
ture is that it behaves as a balanced load to the symmetrical oscillator. Maintaining
symmetry is good for common-mode noise rejection.
The design parameters to be determined are the biasing current, the capacitor
value and the physical dimension of M1 and M2.
5.3.3 Biasing and Conversion Gain

In this subsection, we explain that the 2 source followers, M1 and M2 should be
biased in deep weak-inversion to maximize the peak detector's conversion gain.
Traditionally, diodes and bipolar transistors are used as the nonlinear switches.
Both types of them have an exponential V-I relationship. The transfer characteristic
of peak detectors built from these devices can be approximated as
r
2Vi
Vo = Vi VT ln , (5.12)
VT
where Vi , Vo and VT are the input and output amplitude and thermal voltage,
respectively [34].
When MOSFETs are working in weak-inversion, the V-I relationship is still ex-
ponential, expressed as
Vgs
Id = Id0 exp (5.13)
nVT
where n>1 is called the nonideal or slope factor, and VT is the thermal voltage.
The transfer characteristic of peak detectors using weakly inverted MOSFETs is

adapted from Equ. 5.12 as
r
2Vi
Vo = Vi nVT ln . (5.14)
nVT
For MOSFETs working in strong inversion, the square-law applies
1 W
Id = n Cox (Vgs Vth )2 . (5.15)
2 L
A sinusoidal excitation 2Vi cos (t) applied to the input of MOSFET M1 or M2
yields a drain current
1 W 2
Id = n Cox Vgs + Vi cos (t) Vth (5.16)
2 L
1 W 2
= n Cox Vgs Vth + 2 Vgs Vth Vi cos (t) + Vi2 cos2 (t) (5.17)
,
2 L
where
Vgs = VG Vs , (5.18)
VG and Vs are the DC gate voltage and transient source voltage. Since M1 and M2
are alternatively switched on and o for 50% of the time by nearly identical peak
amplitudes, they share the bias current evenly. And the average tail current of M1
and M2 is limited by half of the biasing current, i.e.
T
Itail 1 2 W h 2 i
= Id = n Cox Vov + 2Vov Vi cos (t) + Vi2 cos2 (t) dt (5.19)
2 T2 2 L
Vi2

1 W 2
= Id = n Cox Vov + , (5.20)
2 L 2
where Itail and Id are the tail biasing and average drain current, and
Vov = Vgs Vth = VG Vs Vth (5.21)
is the transient overdrive voltage. For M1' and M2', the tail current is also evenly
shared and the same square-law applies
Itail 1 W 2
= Id0 = n Cox Vgs0 Vth . (5.22)
2 2 L
Equating Equ. 5.20 and 5.22, we have
r
Vi2
Vov = (Vov0 )2 , (5.23)
2
Transfer Characteristics of Peak Detectors

0.5
Bipolar
0.45 MOSFET: weak inversion
MOSFET: strong inversion
0.4
0.35
Output Envelope (V)
0.3
0.25
0.2
0.15
0.1
0.05
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Input Oscillation Amplitude (V)
Figure 5.11: Transfer characteristics of peak detectors.
where
Vov0 = VGS
0
Vth = VG VS0 Vth (5.24)
is the overdrive voltage of MOSFET M1' and M2'. The output voltage of the peak
detector is dened as
Vo = Vs VS0 (5.25)
= Vov0 Vov . (5.26)
Substituting Equ. 5.23 into 5.26 yields
r
Vi2
Vo = Vov0 (Vov0 )2 . (5.27)
2
Equ. 5.27 is the transfer characteristic of a peak detector using square-law nonlinear
devices.
A comparison of conversion gains of peak detectors using MOSFETs working
in weak- and strong-inversion, is now possible by using Equ. 5.14 and 5.27. Fig.
5.11 plots the transfer characteristics of peak detectors utilizing bipolar transistors,
MOSFETs working in weak-inversion with a slope factor of 1.5 and those working in
strong inversion with an overdrive voltage of 0.5 V. As shown in the graph, bipolar
peak detectors have the highest conversion gain of 0.873. Weakly inverted MOSFETs
cause a slightly inferior conversion gain of 0.826. Strongly inverted MOSFETs yields
the worst result, a conversion gain of 0.578, which is signicantly lower than the other
two.
It is worth mentioning here that peak detectors fabricated by standard CMOS
technologies can also perform as the bipolar ones by using dynamic threshold MOS-
FETs (DTMOSs). In essence, DTMOSs are PMOSs or triple-well NMOSs with
connections from their bodies to their gates, which form parasitic bipolar transis-
tors. However, DTMOSs are not used in this design since the use of PMOSs needs
AC coupling and triple-well NMOSs require 3 additional masks. In other words, the
extra gain squeezed out by DTMOSs is too little to justify the trade-os.
5.3.4 Parameter selection

The design parameters to be determined are the biasing current Itail , the source
capacitor value Ctail and the width of M1 and M2. The previous subsection con-
cludes that MOSFETs should be biased in deep subinversion to have the maximum
conversion gain. This implies a large transistor width and a small tail current.
The largest transistor width is limited by the parasitic capacitances between the
two gates. Since the peak detector is in shunt with the resonation tank, all these
capacitances contribute to the tank. The frequency stability can be severely inu-
enced if these parasitics dominate the tank capacitance. Considering the inductor
we selected, a 655 fF total capacitance is required for the tank to resonate in 4.5
GHz. We limit the parasitics to 10% of this total value, i.e. 65 fF, so that a 20%
variation of parasitics causes less than 1% of frequency deviation.
The tail current and source capacitor together determine the discharge time
constant. This time constant is an intrinsic trade-o between amplitude accuracy
and output dynamic range. A slow discharge accurately follows the peak, but only
a small envelope variation can be followed. On the other hand, a fast discharge can
follow fast envelope variation, but it detects not only the peaks, but also the rising
curve before peaks. In our design, the maximum voltage drop between 2 peaks is
around 30 mV, in a period of 0.22 ns. This determines the current-capacitance ratio
to be
Itail Vdrop 30 mV 30 uA
= = = . (5.28)
Ctail Tperiod 0.22 ns 0.22 pF
The source capacitor should be much larger than, or should take into account,
the input capacitance of the following stage. Otherwise the discharge time constant
would be dierent. In our design, a voltage buer for measurement purpose is the
load of the peak detector. We choose a Ctail of 100 fF, then the biasing current is
calculated by Equ. 5.28 to be 15 uA.
It is worth mentioning here that the kT/C noise should not be considered when
choosing the value of Ctail . Sooner or later, after the envelope detection, a band-
pass lter has to be applied to extract the subband signal. The bandwidth of this
Peak Detector Transfer Characteristic

0.2
0.18
0.16
Output Envelope Amplitude (V)

0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
0 0.05 0.1 0.15 0.2 0.25 0.3
Input Oscillation Amplitude (V)
Figure 5.12: The simulated transfer characteristic.
bandpass lter is smaller than the bandwidth determined by Ctail . So a large source
capacitor cannot improve the noise performance in this receiver front-end.
5.3.5 Simulation Results
5.3.5.1 Transfer Characteristic
The transfer characteristic showing the voltage amplitude of the input and the out-
put is plotted in Fig. 5.12. From the graph, it can be observed that a 0-to-300
mV oscillation amplitude yields a 0-to-180 mV output envelope, which indicates a
conversion gain of around 0.6. Furthermore, the nonlinearity of conversion gain can
be observed from Fig. 5.12 (and also Fig. 5.11) when the input oscillation ampli-
tude is small. This reduced conversion gain sets the minimum detectable oscillation
amplitude (about 40 mV), which can be benecial since the small oscillations caused
by noise are suppressed.
The peak detector's response to the practical signal generated by an SRO is
plotted in Fig. 5.13. As shown in the graph, the envelope is extracted from the
oscillation, with a certain amount of attenuation.
The peak detector draws 30 uA from the 0.9 V supply, yielding a power con-
sumption of 27 uW.
SRO Oscillation
300
200
100
Voltage (mV)
100
200
300
0 50 100 150 200 250 300 350 400 450 500
Time (ns)
Peak Detector Output

150
100
Envelope (mV)
50
0
0 50 100 150 200 250 300 350 400 450 500
Time (ns)
Figure 5.13: The oscillation and its envelope.

GM (mS) SNRo (dB) NF Budget (dB)
100 20 9
20 14 3
10 11.3 0.3
1 4.2 -6.8
Table 5.3: Output SNR vs. LNA transconductance and noise gure budget.
5.4 Low-noise Amplier

The LNA performs several functions in this receiver front-end. First of all, it pream-
plies the received weak RF signal to a certain level such that noise in following
stages is comparatively insignicant. Secondly, it behaves as an impedance match-
ing network between the antenna and the following SRO. Thirdly, it shields the
high-power oscillation of SRO from coupling back into the antenna. As a result, the
LNA has to satisfy several requirements in order for the system to operate properly.
5.4.1.1 Input Impedance

The characteristic impedance of the antenna is 50 ohm given as a specication. For
a dierential input, an o-chip balun of 1:2 (50:100) impedance transformation ratio
is assumed. A hybrid ring would be a practical example.
5.4.1.2 Gain and Noise Figure

The requirements of the gain and noise gure for the LNA are derived from a
test-bench simulation with the already designed SRO, peak detector and an ideal
transconductance block. In essence, we plug in the worst-case RF excitation and
nd out the transconductance that yields the minimum acceptable output SNR.
More specically, from Section 4.2, the worst-case RF signal is -80 dBm (Equ. 4.4)
when the receiver is 10 meters away from the transmitter. And the required output
SNR is 11 dB (Equ. 4.12) for a data-rate of 100 kbps and a bit-error-rate of 106 .
For super-regenerative receivers, we argue the necessity of using the transcon-
ductance instead of voltage or power gain to characterize an LNA. When an SRO
is following an LNA, it modies the load resistance of the LNA constantly. The
negative conductance behaves as a gain-boosting amplier when it is in shunt with
the load impedance of an LNA (since it increases the load resistance). The volt-
age and power gain is also changing, so they are time-varying. On the other hand,
transconductances can be load-resistance independent.
Intensive simulations, to be discussed in Section 7.2, yield the relationship be-

5.4. LOW-NOISE AMPLIFIER 61
tween LNA transconductance and output SNR listed in Tab. 5.3. As shown in the
table, there is a trade-o between the gain and noise gure requirements. A high
gain suppresses the noise of following stages so the LNA NF dominates the receiver
NF. On the other hand, a low gain brings the noise of following stages into play and
the NF requirement is tightened. A sensible choice for LNA specication from Tab.
5.3 is a transconductance of 20 mS and a noise gure of 3 dB.
In the LNA design process, it is convenient to use a constant resistance, the
passive tank resistance of 1046 ohm, as the load. Then, the voltage gain of the LNA
is calculated as
Av = GM RL = 20.9, (5.29)
and the power gain is
Vo2
RL RS
S21 = Vi2
= A2v = 41.8 = 16.2 dB. (5.30)
RL
RS
5.4.1.3 Reverse Isolation

As discussed in Section 4.2, ETSI (European Telecommunications Standards In-
stitute) imposes a narrow-band spurious emission limitation of less than -47 dBm
EIRP (Eective Isotropic Radiated Power) [32]. In our design the SRO produces
high-power oscillation at 4.5 GHz, which needs to be attenuated to satisfy the spu-
rious emission specication.
A maximum oscillation amplitude of 0.3 V generated by the SRO has a power of
2
Vrms 2 0.32
Posc = = = 0.17 mW = 7.6 dBm, (5.31)
RL 1046
when applied on the 1046 ohm load resistance. The required LNA isolation can then
be calculated as
S12 = 47 dBm Posc = 39.4 dB. (5.32)
5.4.1.4 Specication Summary

The LNA specications are summarized in Tab. 5.4. Besides these requirements
that must be satised, minimum power consumption is the goal to pursue.
5.4.2 The Schematic

The LNA schematic is shown in Fig. 5.14. Essentially, it is a dierential degenerated
common-source stage with cascode and neutralization. More specically, MOSFETs
M1 and M2 are the common-source input stage with their sources degenerated by
Ls, to provide a real part of the input impedance. MOSFETs M3 and M4 are the
Parameter Value
Input Impedance 100 ohm (dierential)

GM > 20 mS
S21 > 16.2 dB
NF < 3 dB
S12 < -39.4 dB
Bandwidth 500 MHz
Table 5.4: LNA specication summary.
Vdd Vdd
Vdd Vdd
M8 M3 Vfreq M4
M7
M5 M6
Vdd
Vdd Vdd
Vi+ Pad Pad Vi-

Lg
M1 M2
Ls
Figure 5.14: The schematic of the LNA.

Matching
Network
Rs
M1
Vs
Cgs
Ls
Figure 5.15: Simplied LNA input.
cascode which provide reverse isolation and increase the output impedance to avoid
loading the oscillator tank. Extra reverse isolation is provided by the neutralization
MOSFETs M5 and M6. The matching network at each side is formed by the degen-
eration inductor Ls , biasing inductor Lg , a shunt and a series capacitor, the ESD
diodes, a C4 pad and a bond-wire. MOSFETs M7 provides the biasing voltage to
the gates of M1 and M2 through Lg .
We adopt the dierential structure because of its common-mode noise rejection
and even-order harmonics suppression. Common-source input stage is chosen for its
high power gain compared with common-gate or common-drain stages. Inductive

1
degeneration is capable of noise matching and GM boosting besides its impedance
matching functionality. We use the cascode stage not only to provide reverse iso-
lation, but also to prevent the output impedance from loading the oscillation tank.
Neutralization is applied for the insucient isolation of the cascode stage.
Despite the popularity of this topology, we perform a power-driven design pro-
cedure which is quite dierent from the classical methodology [25], which is, from
our point of view, noise-driven.
5.4.3 Power-driven Design Procedure
The design procedure discussed in this subsection derives the minimum power for a
certain gain and bandwidth, given the device characteristics of a certain fabrication
process.
5.4.3.1 Minimum Ft
We derive the relationship between the LNA transconductance and the transition
frequency of the input stage.
The input part of the LNA is redrawn in Fig. 5.15. We neglect the cascode
and neutralization MOSFETs since they do not contribute or inuence the LNA
transconductance. The gate to source capacitance, Cgs , is drawn explicitly. If the
impedance is matched, the current owing into the dashed box is
vs
is = . (5.33)
2Rs
Furthermore, if the matching network is lossless, maximum power transfer is achieved
and all the current is owing into the degenerated resistor. Then the voltage across
Cgs is expressed as
is vs
vgs = = = jQvs (5.34)
jCgs jCgs 2Rs
where Q is the quality factor of the input network
1
Q= . (5.35)
2Cgs Rs
Then the single-ended transconductance of LNA is expressed as
id vgs gm T
GMsingle = 2 GM = = jQ gm = = , (5.36)
vgs vs j2Cgs Rs 2jRs
where gm and T is the small-signal transconductance and angular transit frequency
of the input MOSFET. From Equ. 5.36, we observe that for an operating frequency
of 4.5 GHz, a source impedance of 50 ohm, and GM > 20 mS the transit frequency
is
2 GM 2jRs
ft = > 18 GHz (5.37)
2
5.4.3.2 Minimum Current Density

The DC operating points of an IBM 90-nm RF NMOS is extracted from a gate
voltage sweep. The plots of transit frequency and current density vs. gm/Id are
given in Fig. 5.16 and 5.17. We use the parameter gm/Id to indicate the degree of
inversion. From Fig. 5.16, we observe that a gm/Id of no less than 14 is required to
achieve the transit frequency of 18 GHz. In Fig. 5.17, we pick a current density of
20 uA/um, which yields a gm/Id of about 12.
1 The
gate to source voltages, Vgs , of M1 and M2 are increased to Q times the input voltage,
Vi+ or Vi . So the equivalent transconductance of the LNA is also Q times larger than the
transconductance of the dierential pair, as will be shown in Equ. 5.36.
Transit Frequency vs. gm/Id

100
90
80
70
Transit Frequency (GHz)
60
50
40
30
20
10
0
0 5 10 15 20 25
gm/Id
Figure 5.16: Transit Frequency vs. gm/Id.
Current density vs. gm/Id

3
10
2
10
Current Density (uA / um)
1
10
0
10
1
10
2
10
3
10
0 5 10 15 20 25
gm/Id
Figure 5.17: Current density vs. gm/Id.

Gate Capacitance vs. gm/Id

6
5.5
5
Gate Capacitance (fF)
4.5
3.5
3
0 5 10 15 20 25
gm/Id
Figure 5.18: Gate capacitance vs. gm/Id.
Until now, the power dissipation is still unclear but the drain current to MOSFET
width ratio is determined.
5.4.3.3 Minimum Power
According to our bandwidth requirement of 500 MHz and the center frequency of
4.5 GHz, the quality factor of the matching network is
f0
Q= = 9. (5.38)
BW
Substituting this Q into Equ. 5.35, we can derive the Cgs requirement of
1
Cgs = = 39 fF. (5.39)
2QRs
The gate capacitance as a function of gm/Id is shown in Fig. 5.18 for a transistor
of 3.6 um. When gm/Id=12, we have a gate capacitance of around 5 fF. We need
to scale it 8 times larger to satisfy the Cgs requirement, which yields a MOSFET
width of 30 um.
Finally, the minimum biasing current is calculated as
Imin = 20 uA/um 30 um = 600 uA. (5.40)
And the dierential structure consumes, therefore, 1.2 mA.

Vdd Vdd Vdd
RL RL Vo
Vbias Vo V
i
Vx Vbias
Vi
(a) CS-CG (b) CD-CG
Figure 5.19: Unilateralization techniques.
5.4.4 Reverse Isolation

The signal at the output of an amplier can be coupled back to the amplier's
input by parasitics. For super-regenerative receivers, the ampliers output has large
voltage swing caused by periodic oscillations. This characteristic of an SRO imposes
stringent requirement of reverse isolation to the LNA in front of it.
The reverse coupling is mainly caused by the drain to gate capacitance, Cgd , of a
common-source stage. Commonly used techniques to mitigate the reverse coupling
of Cgd are the unilateralization and neutralization [35]. In this LNA design, the
reverse isolation specication of -39.4 dB is achieved by applying both techniques.
Simulations show that a 10% mismatch of the neutralization MOSFETs still leads
to acceptable results.
5.4.4.1 Unilateralization
Unilateralization techniques plug in devices in series with the signal path to atten-
uate signals in one direction. The two commonly used unilateralization method are
the CS-CG and the CD-CG combinations, as shown in Fig. 5.19. Both schemes
are actually 2-stage designs exploiting the shielding eect of the CG stage. The
penalties paid are the increased voltage or current consumption.
In this design, we apply the CS-CG topology, or a cascode. The motivation is
that the output resistance of a cascode stage is higher than that of a CD-CG stage.
This prevents the output resistance of the LNA from loading the oscillation tank of
the following stage.
The shielding eect of a cascode stage can be demonstrated by its reverse transfer
function
vx 1
= ro , (5.41)
vo gm ro + Ro1
+ sCgs ro + 1
where gm , ro and Cgs are the transconductance, drain-to-source resistance and gate-
to-source capacitance, and Ro1 are the output resistance of the CS stage. In this
expression, Cgd of the cascode is neglected for simplicity. In a reasonable design, the
second and the third term of the denominator in Equ. 5.41 cannot be much larger
2
than 1. So the main attenuation is given by the intrinsic gain of the cascode stage.
Since 90-nm short-channel devices have their intrinsic gain no more than 10, we
increase the length of the cascode MOSFETs to 200 nm. Further increase of length
has a diminished return of intrinsic gain but a more serious ft trade-o. Smaller ft
of the cascode transistors yields a smaller gm that is incapable of suppressing Miller
eect, and a large Cgs that lowers the 2nd pole. The 200 nm transistors provide an
intrinsic gain of 20, which implies a reverse isolation of -26 dB. Accounting for the
-8 dB attenuation caused by Cgd coupling, we still need -6 dB isolation.
The preceding analysis shows that barely cascoding cannot provide sucient
reverse isolation for our receiver.
5.4.4.2 Neutralization
Neutralization techniques plug in devices in parallel with the signal path to produce a
180-degree phase shift. Because the LNA has a dierential structure, it is convenient
to cross couple the output back using the equal amount of coupling capacitance. As
shown in Fig. 5.14, transistor M5 and M6 are identical with M1 and M2. The signal
coupled by Cgd of M5 has the same amplitude as that coupled by Cgd of M1, but is
180-degree out of phase. So the two signals cancel each other. The same eect is
produced between M2 and M6.
The eect of cancellation highly depends on the matching of Cgd , which is not
only process- but also voltage-dependent. When devices are perfectly matched, the
simulation result shows an extra -18 dB attenuation is obtained, which yields a
total reverse isolation of -52 dB. A 10% mismatch analysis shows that all the LNA
specications are still satised.
5.4.5 Input Impedance Matching

Compared with the classical inductive degenerated impedance matching, several
unique problems are tackled in this design. Firstly, the super-regenerative oscillator
2 The
Miller eect is mitigated by letting 1/gm Ro1 , i.e. gm Ro1 1 which implies that Ro1
and ro are in the same order of magnitude. Furthermore, the 2nd pole at the node vx should not
be dominant, i.e. gm/Cgs = gm ro/Cgs ro BW , so signals within the bandwidth yield sCgs ro not
much larger than 1.
Id
Zin
Cgs gm ro RL
Ls
Figure 5.20: Simplied model of inductive degenerated common-source.
following the LNA is a time-varying load. The large load variation leads to input
mismatch for bilateral ampliers. Secondly, special issues arise when neutralization
capacitors are used together with inductive degeneration. Last but not least, low-
power input stage causes trouble when low-Q matching is desired.
5.4.5.1 Load Dependency of Input Impedance

The simplied model to calculate the input impedance of inductive degenerated
common-source stage is shown in Fig. 5.20. A commonly used formula to express
the input impedance is derived by the gyration as
1 1
Zin = + jLs (1 + ) = + jLs + T Ls (5.42)
jCgs jCgs
where = T/j is the forward current gain. This expression, however, implicitly
assumes RL ro . For most of LNAs, this condition is true or can be made true by
cascoding. In a super-regenerative receiver, where an LNA is loaded by an SRO, we
can never make this assumption.
An SRO varies the tank conductance periodically between positive and negative
values. The maximum sensitivity of the SRO is obtained at the critical point, when
the total tank conductance is zero, as shown in Equ. 3.42. At this time point,
however, the load resistance to the LNA is innite. So we need a more accurate
input impedance expression.
The exact input impedance of the circuit shown in Fig. 5.20 is, after elaboration,

1 gm ro
Zin = + jLs || (ro + RL ) + Ls , (5.43)
jCgs Cgs ro + RL + sLs
where || denotes in parallel with. We can observe that when RL is small, Equ. 5.43
simplies to 5.42. On the other hand, when RL becomes large, the real part of the
input impedance decreases. At the critical point, when all the positive resistance
given by the tank and the LNA are canceled, the input impedance has no real part.
Intuitively, when the drain to ground resistance (RL in this case) is innite, no
current is owing into the drain and no current is owing out from the source. Source
degeneration, which utilizes the source current as a feedback, cannot be functional.
We may expand this conclusion to all source current involved impedance matching
techniques, such as a common-gate input stage.
The LNA load and input resistance changing with the quenching wave is plotted
in Fig. 5.21. Our LNA designed so far has a dierential output resistance of around
15.8 kohm. When the absolute value of tank resistance is much (10 times) smaller
than 15.8 kohm, as denoted between the dashed lines in Fig. 5.21, the LNA is
matched to the input.
Improving the performance further to achieve input impedance matching at crit-
ical point is fundamentally dicult. At the output side, to drive an innite resis-
tance, a voltage output should be used. But low output impedance loads the tank
and brings down the Q factor, which involves several dimensional trade-os. At
the input side, we need a low impedance to load the output of the LNA in order
not to inuence the input matching. This implies that a current buer should be
inserted in, between the LNA and the SRO. This implication justies the cascoding
applied in the rst place. Further cascoding does make the LNA more unilateral,
but increased power consumption outweighs the benet.
5.4.5.2 Neutralization
A simple inductive degeneration generates a real part of input impedance being pro-
portional to the source inductor, Ls . However, the neutralization capacitors change
the input impedance and destroy the proportionality. As a result, a maximum of
input resistance exist.
A simple way to show the eect is to observe the current ow through gate-drain
and neutralization capacitances, as shown in Fig. 5.22. Assuming
vg1 = vg2 = vg (5.44)
vd1 = vd2 = vd (5.45)
and
Cgd = Cn = C , (5.46)
1
Conductance (mS)
0.5
0.5
1
0 10 20 30 40 50 60 70 80 90
Time (ns)
Load Resistance of LNA
10
Resistance (Kohm)
10
0 10 20 30 40 50 60 70 80 90
Time (ns)
Input Resistance of LNA
150
Resistance (ohm)
100
50
0 10 20 30 40 50 60 70 80 90
Time (ns)
Figure 5.21: The variation of load and input resistance of LNA.

in1 in2
Cn Cn
Vg1 Cgd i1 Vd1 Vd2 i2 Cgd Vg2
Cgs gm ro RL RL ro gm Cgs
Ls Ls
Figure 5.22: Neutralization Currents.
we have
i1 = (vd vg ) sC (5.47)
in1 = (vd vg ) sC (5.48)
i2 = (vd vg ) sC (5.49)
in2 = (vd + vg ) sC . (5.50)
Therefore, the current owing out of node vg1 via capacitances Cgd and Cn is
ig1 = i1 in1 = 2vg sC . (5.51)
This extra current ow, ig1 , caused by Cgd and Cn , changes the input impedance
of the LNA from Equ. 5.43 to

1 gm ro 1
Zin = + jLs || (ro + RL ) + Ls || . (5.52)
jCgs Cgs ro + RL + sLs 2jC
The input resistance of the LNA with and without the inuence of Cgd and Cn
(the real part of Equ. 5.42 and 5.52), as a function of the degenerated inductance,
Ls , are plotted in Fig. 5.23. Two phenomena can be observed from the graph.
Firstly, to achieve a certain matching resistance, neutralization requires a much
larger Ls value. Secondly, Cgd and Cn cause a maximum of input resistance looking
into the gate. Before this maximum happens, an increase of the source inductance,
Ls , yields a proportional increase of gyrated gate resistance, as is predicted by Equ.
5.42 and 5.43. However, increasing Ls further after the peak of gate resistance can
only reduce this resistance.

Input Resistance vs. Source Inductance

200
With Neutralization
180 Without Neutralization
160
140
120
Rin (ohm)
100
80
60
40
20
0
0 1 2 3 4 5 6 7 8 9 10
Ls (nH)
Figure 5.23: Input resistance vs. source inductance.
5.4.5.3 Low power, Low-Q Matching

Passive matching techniques can be divided into 2 categories, the narrow-band and
wide-band matching. The narrow-band matching turns the center frequencies of all
the cascading lters to the operating frequency. For low-loss networks, the insertion
loss could be close to 0 dB at the center frequency. The drawback is that the
bandwidth is determined by the Q factor of the network and the operating frequency.
The wide-band matching, on the other hand, tunes the cascading lters away from
the center in such a way that the stacked frequency response is at in the passband.
This type of matching techniques decouples the bandwidth dependencies on the Q
factor and the operating frequency. The penalty paid is the insertion loss, which
cannot be zero even if loss-less components are used. This LNA uses a narrow-band
matching technique, which yields the smallest insertion loss for a given number of
stages and the given quality of components.
In order to achieve the 500 MHz bandwidth, we need the Q factor of the matching
network to be
f0 4.5 GHz
Q= = = 9. (5.53)
BW 500 MHz
Following analysis shows that this Q requirement is dicult to achieve for low-power
LNAs.
The matching network and its equivalent simplied representation are shown in
Fig. 5.24. In the schematic, Lb is caused by the bond-wire, which is specied as
2 nH. The bond-pad and ESD (Electrostatic Discharge) diodes are all capacitance
paths to the ground, which constitutes parts of Cp0 . Cp , another capacitor is added
0
to Cp so they resonate with Lb at the operating frequency. The Q factor of this L
Vdd
Vdd
Rs Lb Pad Cs
M1
Vs Cp Lg
Cgs
Ls
R1 R2
Rs Lb Cs Cgs
Vs Cp Lg1 Lg2 3Cov Rdeg
Ls
Figure 5.24: Matching network and its equivalent representation.
match is
0 L 2 4.5 109 2 109
Q1 = = = 1.13. (5.54)
Rs 50
The resistance looking to the source after this transformation, denoted as R1 , is
R1 = Rs 1 + Q21 = 50 1 + 1.132 = 114 .

(5.55)
At the input transistor's side, we split the shunt inductor Lg into Lg1 and Lg2 for
interpretation purpose only. In Subsection 5.4.3, we have select the transistor's
width to obtain a Q2 of around 10 when Rdeg = 50 . This can be transformed into
an equivalent parallel resistance of
R2 = Rdeg Q22 = 5 k. (5.56)
The rest of the network transforms this 5 kohm resistance down to 114 ohm, then
the matching is done. This transformation network, Cs and Lg1 should have a Q
factor of r
R2
Q3 = = 6.62 , (5.57)
R1
which requires the inductor to be
R2
Lg1 = = 26.7 nH. (5.58)
0 Q2
Lg2 is resonating with the input capacitance of the transistor at 4.5 GHz, which is
1
Lg2 = = 14.2 nH, (5.59)
02 Cin
where
Cin 3Cov ||Cgs = 88 fF (5.60)
is the input capacitance looking into the gate of M1. The total inductance of Lg is
then
Lg = Lg1 || Lg2 = 9.3 nH. (5.61)
The preceding analysis argues that using loss-less components and fairly unlim-
ited values, the minimum-power LNA satisfying all the specications are feasible.
However, two practical problems destroy this feasibility. First of all, the limited
Q factor of Lg jeopardizes the gain and the noise gure of the LNA. For example,
an inductor of Q=20 yields an equivalent shunt resistance of 5.25 kohm, which is
nearly equal to the input resistance, R2 , calculated in Equ. 5.56. This leads to a
3 dB decrease of signal power and a 1.76 dB increase of noise power. This implies
a 3 dB gain reduction and 4.76 dB noise gure degradation. Secondly, one induc-
tor's value, Lg = 9.3 nH, is prohibitively large. It is worth mentioning that using
a shunt-inductor topology has yielded the minimum inductance possible, since the
2 inductors, Lg1 and Lg2 are in parallel. Even if the large physical dimension is
tolerable, a symmetrical inductor of 18.6 nH has too low a self-resonant frequency
to be of any use.
The fundamental cause of these problems is the small power consumption and
thus the smallCin . If power is doubled, i.e. Cin is doubled, then Lg2 is halved, R2 is
quartered and Lg1 is thus halved, according to Equ. 5.59, 5.56 and 5.58, respectively.
Therefore, we are dealing with the ultimate power-gain-noise trade-o, but from a
power-aware perspective. We increase the width of the input MOSFETs and biasing
current to increase the gain and decrease the noise gure until they satisfy the LNA
specications. Adequate performance is achieved when the biasing is twice as much
as the minimum derived in Subsection 5.4.3.
5.4.6 ESD Protection

Electrostatic discharge is considered at the beginning of the design of the matching
network. The total capacitance between the chip-input to the ground, shown as Cp0
in Fig. 5.24, is resonating with Lb at 4.5 GHz. So it has a value of
1
Cp0 = = 625 fF. (5.62)
02 Lb
ESD Current
1.5
Current (A)
1
0.5
0
0.5 0 0.5 1 1.5 2 2.5 3
Time (us)
Pad Voltage
3
Voltage (V)
0
0.5 0 0.5 1 1.5 2 2.5 3
Time (us)
Figure 5.25: The transient voltage and current of an ESD event.
The smallest C4 bond-pad, having a dimension of 85 85 um, accounts for 50 fF.
The parasitics of Cs also take several fF. The rest capacitance budget is given to
ESD devices.
We utilize the double-diode ESD protection scheme with minimum dimension for
its minimum additional capacitance. A 2 kV HBM (Human Body Model) ESD event
is applied to the input and ground. The voltage at the input node and the current
owing through the diodes are plotted in Fig. 5.25. We can observe a maximum
current of 1.3 A and a maximum voltage of 2.2 V.
The gate-dielectric integrity is veried according to IBM's technology design
manual [33]. For a product lifetime of 100 KPOH (thousand of power-on hours), a
maximum temperature of 85 C, an area of gate oxide of 2 60um 100nm and a
gate-biasing voltage of 594 mV, this protection scheme can withhold a total amount
of 18 minutes of this type of ESD events.
The same ESD event applied between the input and the Vdd is also veried with
no gate-oxide breakdown.

The s-parameter simulation is performed to verify the gain, matching and reverse
isolation of the LNA.
The S11 of the LNA from 4 to 5 GHz is shown in Fig. 5.26. It can be observed
from the plot that the LNA is matched well at 4.5 GHz, with an S11 of -37.5 dB. At
4.25 and 4.75 GHz, the S11 is 5.3 and 6.5 dB, respectively. S11 is not matched to
-10 dB in the whole 500 MHz bandwidth since low power low-Q matching involves a
S11
0
10
15
S11 (dB)
20
25
30
35
40
4 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Frequency (GHz)
Figure 5.26: Input port reection coecient, S11.
trade-o, discussed in Subsection 5.4.5. Instead, signal is allowed to be attenuated
to 50% of its power at the edge of the bandwidth. The -3 dB S11 happens at 4.13
GHz and 4.99 GHz frequency. Hence, for a low-loss input matching network, the -3
dB bandwidth is around 860 MHz.
The forward power gain, S21, is plotted in Fig. 5.27. As shown in the graph,
a maximum gain of 16 dB is obtained at the center frequency, which indicates an
LNA transconductance of 20 mS. The -3 dB bandwidth is around 220 MHz, limited
by the load of the LNA. This bandwidth barely means that the load resonant tank
has a Q factor of
f0 4.5 GHz
Q= = = 20.5 . (5.63)
BW 220 MHz
The reverse power gain of the LNA is shown in Fig. 5.28. As shown in the
graph, a minimum of 52 dB reverse isolation is obtained, satisfying the S12 < -39.4
dB specication. This performance guarantees that the ETSI spurious emission
limitation is satised.
The noise gure of the LNA is shown in Fig. 5.29. As shown in the graph, the
minimum noise gure of 2.5 dB is achieved at 4.5 GHz. At 4.25 and 4.75 GHz, the
noise gure is 2.9 and 2.8 dB, respectively.
Two current paths each draws 1.22 mA from the 0.9 V supply. Therefore, the
LNA consumes 2.2 mW power. This continuous power consumption can be reduced
by duty cycling using an auxiliary circuit, which biases the LNA dynamically during
the SRO sampling period and turns it o during the quenching period.
S21
16
14
12
10
S21 (dB)
2
4 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Frequency (GHz)
Figure 5.27: Forward power gain, S21.
S12
50
55
60
S12 (dB)
65
70
75
4 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
Frequency (GHz)
Figure 5.28: Reverse power gain, S12.

Noise Figure
4.4
4.2
3.8
Noise Figure (dB)
3.6
3.4
3.2
2.8
2.6
4 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5

Frequency (GHz)
Figure 5.29: The noise gure of the LNA.

Chapter 6
Auxiliary Circuits
6.1 Introduction
Super-regenerative receivers need auxiliary circuits to work properly. First of all,
every SRO needs a time-varying biasing. The complexity of biasing circuits depends
on the waveform to be applied to SROs. To obtain the desired FM-AM conversion,
a large voltage swing and low-power dissipation, this design uses a special waveform
which complicates the biasing circuit considerably. Secondly, power reduction is
possible if the LNA is shut down during the quenching period. Super-regenerative
receivers are inherently sampling circuits which deal with only discrete time signals.
LNA is only useful in certain periods to provide gain and matching. At the rest of
the time, it can be put to rest for power saving consideration as long as its reverse
isolation is not compromised. Thirdly, a voltage buer is required to output the
demodulated subcarrier. This stage is mainly designed for testing and measurement
purpose. Oscilloscopes are assumed to be the load. Lastly, an on-chip current
reference circuit is designed for the biasing of all the functional blocks.
The detailed design of the biasing waveform generator for SRO, the dynamic
biasing circuit for LNA, the output buer and the current reference are presented
in Section 6.2, 6.3, 6.4 and 6.5, respectively.
6.2 Biasing Waveform Generator for SRO

The biasing waveform generator produces the saw-tooth like current waveform spec-
ied in Subsection 5.2.4 to bias the SRO. We assume that an external clock is
available. The duty cycle of the clock should be tunable in order for the SRO to
have a variable gain. The peak of biasing waveform is also designed to be exter-
nally tunable, so that the external control of the oscillation amplitude, gain, and
frequency response of the SRO is possible after the fabrication.
80
6.2. BIASING WAVEFORM GENERATOR FOR SRO 81
Saw-tooth Voltage Generator Linear Transconductance
Vdd Vdd
SRO
Vclk M1 C
Vamp Rf
Ibias
Ri M2
+
M3 M4 OTA
M5 M6
Figure 6.1: The schematic of the biasing waveform generator.
6.2.1 The Schematic

The circuit of this biasing waveform generator is shown in Fig. 6.1. In essence, the
circuit rst converts the external rectangular clock signal into a saw-tooth voltage
waveform, which is then converted into a current waveform by a linear transconduc-
tance block.
6.2.2 Saw-tooth Voltage Generator

The saw-tooth voltage is derived from an external rectangular clock using the left
half of the circuit shown in Fig. 6.1. The circuit consists of a capacitor, 2 switches
M1 and M2, a resistor Ri and a current mirror formed by transistor M3 and M4.
An external DC voltage Vamp is converted to a DC current by Ri . The current
is then duplicated to the switched-capacitor circuit by the current mirror. On a
positive period of the clock, M1 is open and M2 is closed. The capacitor, C, is
charged by the DC current, and the voltage across the capacitor is growing linearly
with time. When a positive to negative transition of the clock occurs, M2 becomes
open and M1 is closed, which short-circuits the capacitor and discharges it sharply.
This short is continued until the next positive cycle begins. In this way, the shape
of the quenching waveform described in Fig. 5.5 is generated in a voltage form, as
shown in the second row of Fig. 6.4.
The amplitude of the saw-tooth waveform can be controlled by the external
voltage Vamp . A large voltage yields a large current charging the capacitor, and for
a given period, a large ramp amplitude, and vice versa.
6.2.3 Linear Transconductance

The linear transconductance converts a voltage into a current with the shape of
the waveform preserved. The circuit is shown in the right part of Fig. 6.1, which
82 CHAPTER 6. AUXILIARY CIRCUITS
SingleSided Amplitude Spectrum: Biasing Waveform

1
0.9
0.8
0.7
0.6
|Y(f)|
0.5
0.4
0.3
0.2
0.1
0
0 100 200 300 400 500 600 700 800 900 1000
Frequency (MHz)
Figure 6.2: Amplitude spectrum of the biasing waveform.
consists of a resistance Rf , an operational transconductance amplier (OTA) and a
current mirror formed by MOSFETs M5 and M6.
The OTA, transistor M5 and Rf form a negative feedback loop. For a given gate
voltage of M5, a current is generated. Resistor Rf converts this current into a voltage
and feeds it back to the positive input of the OTA. If this voltage is higher than the
voltage across the capacitor, the OTA applies a positive incremental voltage to the
gate of M5, which yields an increased current and a reduced voltage at the positive
input of the OTA. In this way, the negative feedback linearize the transconductance
of M5, and a transconductance of
Io 1
GM = (6.1)
Vin Rf
is obtained, when the loop-gain is large. Transistor M6 accurately duplicates this
current to bias the SRO.
6.2.4 OTA Design

In the biasing waveform generator shown in Fig. 6.1, the main power consumer is
the OTA. Therefore, the design goal for this OTA is to minimize the power while
achieving the performance requirements.
6.2.4.1 Specication
The power of an OTA is determined by its gain-bandwidth product. For a cascode
structure, the DC gain is a fairly constant value while the bandwidth is not. So
we need to specify the minimum bandwidth requirement for the OTA. A discrete
Fourier transform yields the amplitude spectrum of the saw-tooth biasing waveform.
Its normalized version is plotted in Fig. 6.2. As can be observed from the graph,
the main power of the biasing waveform is concentrated in frequencies below 500
MHz. Therefore, we use a 500 MHz bandwidth, i.e.
BW = 500 MHz, (6.2)
as our design specication for the closed-loop linear transconductance. A smaller
bandwidth leads to waveform distortion and less controllability of the SRO.
Assuming a single-pole behavior, the gain of the OTA is expressed as
ADC
Av = , (6.3)
1 + j ffc
where ADC and fc are the DC-gain and -3 dB bandwidth, respectively. Then the
close-loop transfer function is
Av gm ADC gm
GMcl = = , (6.4)
1 + Av gm Rf 1 + ADC gm Rf + j ffc
where gm is the small-signal transconductance of M5. The bandwidth of the closed-
loop transconductance is
BW3dB = (1 + ADC gm Rf ) fc = 500 MHz. (6.5)
Assuming a cascode topology is going to be used and an intrinsic gain of 10 is
available for this technology, ADC is around 100 or 40 dB. Since the output current
of M5 is a biasing waveform, gm is signal dependent and time-varying. To simplify
the calculation, we further assume that the loop-gain is mainly supplied by the OTA,
i.e.
gm Rf 1. (6.6)
Then the bandwidth specication for the OTA can be calculated as
BW3dB
fc = 5 MHz. (6.7)
1 + ADC
Equ. 6.6 determines a trade-o between the power consumption and tracking
errors at the start of ramps. If we assume gm Rf 1, we need a fast OTA with large
fc , consuming quite some power. On the other hand, if we assume gm Rf 1, it is
invalid at the starting up of a ramp, since the drain current of M5 is small, so is the
transconductance, gm . In this period, the loop-gain is severely overestimated and a

Vdd
Vin Vip
Figure 6.3: The schematic of the OTA.
current slope deviating from its voltage reference can be observed. The simulation
result showing this eect is plotted in Fig. 6.4.
6.2.4.2 The schematic

The schematic of the OTA is shown in Fig. 6.3. In essence, the folded cascode
structure is used for its high gain compared with those using uncascoded current
sources as loads. Since the input voltage ranges from 500 mV to 900 mV, a pair of
NMOSs is used for the input.
The lengths of all the transistors are scaled to 200nm for the improved intrinsic
gain and reduced threshold voltage. All transistors are biased in moderate inversion
to balance the speed and power consumption. More specically, a gm/Id of 15 is
selected, which yields a current density of 10 uA/um and a transit frequency of 1
GHz. The NMOS transistors have a basic width of 10 um except the input ones,
which are 40 um wide. PMOSs are 3 times larger in width. As a result, 100 uA is
consumed in each current path. A total current consumption is 400 uA.

A transient simulation is performed and the result of the external clock, saw-tooth
voltage, op-amp virtual ground, the gate voltage of the tail current source and the
biasing voltage for the SRO are shown in Fig. 6.4. The amplitude control voltage,
Vamp , is set to 700 mV.
From the saw-tooth waveform at the second row, we can observe the charge-
injection eect caused by the small capacitor and comparatively large switches.
The small capacitor is desirable since a small current is needed to charge it. Large
switches are used so that the conducting resistance is small. Charge injection is not
External Clock
1000
Voltage (mV)
500
0
0 20 40 60 80 100 120 140
Time (ns)
Sawtooth Voltage Waveform
1000
Voltage (mV)
800
600
0 20 40 60 80 100 120 140
Time (ns)
Virtual Ground
1000
Voltage (mV)
800
600
0 20 40 60 80 100 120 140
Time (ns)
OpAmp Output Voltage
600
Voltage (mV)
400
200
0 20 40 60 80 100 120 140
Time (ns)
SRO Biasing Current
1.5
Current (mA)
0.5
0
0 20 40 60 80 100 120 140
Time (ns)
Figure 6.4: Transient simulation result of the biasing waveform generator.

a serious issue in this application because the continuous charging and discharging
current conducts the charges quickly and the limited bandwidth of the feedback loop
lters the sharp pulses.
By carefully inspecting the biasing current waveform shown in the bottom of Fig.
6.4, a slope deviation can be observed at the starting part of a ramp. This eect
is caused by the small transconductance of M5 and M6 when the biasing voltage at
their gates is low. At this phase, Equ. 6.6 is not satised and the loop-gain of the
transconductance loop is small, yielding this tracking error. Careful simulations are
performed to formulate Equ. 6.6 in such a way that the tracking error only appears
before the SRO is reaching its critical point. So the biasing slope at the critical
point is not inuenced by this tracking error.
The op-amp constantly drains 408 uA from the 0.9 V power supply. The current
sensor path (Rf and M5), op-amp biasing and the switched capacitor consume 26
uA, 10 uA and 4 uA, respectively. As a result, a total power of 403 uW is consumed.
Two measures can be taken to trade performance o for less power. Firstly,
the 500 MHz bandwidth specied in Equ. 6.2 can be compromised, which leads to
signicant waveform distortion. Secondly, we can assume that the gm of M5 has a
larger contribution to the loop gain by violating Equ. 6.6, which causes larger slope
deviation at the starting part of a ramp. Both measures degrade the controllability
of the SRO.
The power dissipation of this biasing circuit is 1.6 times as large as that consumed
by the SRO. This extra power consumption, caused by the complex saw-tooth cur-
rent generator brings us back to the selection of the biasing waveform for the SRO.
Although we intend to save power by using a slope-controlled SRO, its complex
biasing causes power overhead that outweighs the benet. An SRO working in
step-controlled state is, under the circumstances, worth a try (see Section 7.3).
6.3 Dynamic Biasing for LNA

LNAs used in super-regenerative receivers do not need to work all the time. SROs
are sampling ampliers, which amplify only a portion of the signal in a certain
period. For the power-saving purpose, LNAs can be turned o when SROs are in
the quench period or when oscillations have already been built up.
6.3.1 LNA Minimum Start-up Time Analysis

An LNA cannot share the same clock with an SRO in a receiver. As analyzed in
Chapter 3, the maximum sensitivity of an SRO happens at the clock transition
point. Signals right before a transition are at least as important as they are after
6.3. DYNAMIC BIASING FOR LNA 87
Dynamic Ibias
Biasing
Network Lg
Cs
M1 M2
Cp
Ls
Figure 6.5: The equivalent biasing network for LNA.
the transition. So an LNA should start working before an SRO is turned on.
How long an LNA should start before an SRO depends on the characteristics
of both of them. Intensive analysis in Chapter 3 deals with the sensitivity curve of
SROs. This subsection analyzes the warm-up time of the LNA introduced in Section
5.4.
The schematic involving the dynamic biasing of the LNA is shown in Fig. 6.5.
Transistor M2 is the input stage of the LNA. Inductors Ls and Lg and capacitors
Cs and Cp are used for input power matching. Transistor M1 is a part of a current
mirror, which feeds the biasing voltage through Lg to the gate of M2. The biasing
current source is controlled by switching events and thus it can be treated as a step
stimulus.
When a steep power-up current is generated, it faces 2 paths. The rst path
is the resistance of the diode-connected MOSFET M1, which is around 1/gm . The
second path is formed by RF components Lg , Cs , Cp , Ls and parasitics of M2.
At high frequencies, this second path has a lower impedance than that of M1. So
the initial part of the power-up current is bypassed, without generating a voltage
through M1.
A slow warm-up is undesirable. However, to accelerate this process is dicult
since the bypassing components are also parts of the power matching network, which
should not be modied. Simulation result (plotted in the bottom of Fig. 6.11) shows
that the warm-up time of the LNA is around 5 ns.
6.3.2 Dynamic Biasing Circuit

The circuit, which periodically turns o the LNA, is plotted in Fig. 6.6. Because the
prestarting of the LNA is required, a delay-line is created to provide a time constant
internally without referencing to another clock.
The whole circuit consists of a ramp voltage generator, a comparator, a NOT
and a NAND gate to deal with logics, and a switch, M6, to quench the current
Ramp Voltage Generator Phase Delay
Vdd Vdd
Vclk Vswt
M1 C M6
M4 M5
M2 Vrmp +
CMP Vcmp Iref Ibias
Vref
M3
LNA
Figure 6.6: The schematic of the dynamic biasing network for LNA.
mirror.
The ramp voltage generator converts the negative portion of the clock into a
ramp. The detailed mechanism is identical with the other one discussed in Subsec-
tion 6.2.2.
The comparator constantly compares the voltage integrated by the capacitor to
an external reference voltage, Vref . By doing this, Vref is able to control the amount
of phase delay of the positive to negative transition edge.
The logics enforce the LNA to work at both the warm-up and the amplication
period. The waveform of the node voltages are shown in Fig. 6.7. In the gure, Vclk
is inverted at rst. Then its quenching parts are converted to ramps, denoted as
Vr . After that, the comparison between Vref and Vr generates positive to negative
transitions of Vcmp . It is these transitions that trigger the start-up of the LNA.
Finally, the NAND combines the warm-up period and the working period of the
LNA into one signal, Vswt , which controls the switch of the current mirror M6.
When Vswt is positive, M6 is open, and the LNA works.
The physical dimension of the switch, M6, involves the trade-o between charge
injection and current leakage. A large M6 has a small conducting resistance and
the LNA is quenched completely. However, charge injection can be clearly observed.
On the other hand, a small M6 has a relatively large on-resistance. When Iref is
owing through this resistance, the voltage drop leads to a leaking current owing
in the LNA. A compromise between the two yields the dimension of M6 as L=100
nm and W=10 um.
6.3.2.1 Comparator
The schematic of the comparator is shown in Fig. 6.8. The input stage is an N-input
dierential pair with a current mirror as the load. Its main purpose is to provide
gain. The second stage is an NMOS driving a PMOS current source. Extra gain
6.3. DYNAMIC BIASING FOR LNA 89
Vclk
/Vclk
Vrmp Vref
Vcmp
Vswt
Ibias
Amplification Quenching Warmup Amplification

Period Period Period Period
Figure 6.7: Dynamic biasing waveforms.
and a rail-to-rail voltage swing are provided by this stage. Since the load of this
comparator is an NAND gate, which has a small dimension, the comparator also
use small transistors (L=200 nm, W=1 um). Biasing current of 10 uA is consumed
in each path. Thus, a total current consumption is 30 uA.
6.3.2.2 Logic Gates

The transistor level NOT and NAND gates are shown in Fig. 6.9. Since the inverter
drives M1, M2 and the NAND gate, it has a width of 10 um for the NMOS and
20 um for the PMOS. The NAND gate only drives the switch M6, so it has the
minimum dimension. Because NMOSs in the NAND gate is in series, they have the
same width as the PMOSs.

Simulations show that a Vref setting from 900 mV to 580 mV yields an LNA turn-o
period ranging from 0 to 20 ns, i.e. 16 mV/ns. After the LNA and the SRO are
plugged in, simulations further show that a power-o period ranging from 0 to 17.5
ns produces almost identical SRO output amplitude compared with one another.
For a 17.5 ns power-o, the control voltage, Vref should be 620 mV.
Vdd Vdd
Vbiasp
Vcmp
Vrmp Vref
Vbiasn
Figure 6.8: The schematic of the comparator.
Vdd Vdd
A B
X Y
Y
Figure 6.9: The schematic of the NOT and NAND gate.

6.4. OUTPUT BUFFER 91
The transient results of node voltages and the biasing current are plotted in Fig.
6.10 and 6.11. The simulation is performed at a control voltage of 620 mV.
Fig. 6.10 is the simulated version of Fig. 6.7, showing the external clock, the
inverted clock, the voltage ramp, the comparator output and the switching signal.
Fig. 6.11 plots the switching signal again with the gate voltage of the mirror,
the gate voltage of the input pair and the LNA current. The slow start-up behavior
of the LNA, as analyzed in Subsection 6.3.1, can be clearly observed.
The power consumption of this dynamic biasing circuit consists of the static
power, consumed by the rst stage of the comparator and the current mirror, and
the dynamic power, consumed by the logic gates, the switched-capacitor and the
second stage of the comparator. All the dynamic power caused by gate switching
is neglected in this analysis since the power supply voltage is low and the gates are
small. The static current consumed is 30 uA, consisting of 20 uA by the rst stage of
the comparator and 10 uA by the current mirror of the LNA (M4 in Fig. 6.6). The
average of the dynamic current consumed is between 5.7 uA to 11.4 uA, depending
on the duty-cycle of the comparator output. In detail, the switched capacitor draws
5.7 uA in average and the second stage of the comparator draws 0 to 5.7 uA. In
conclusion, this dynamic biasing circuit draws around 36-to-42 uA current from the
0.9 V power supply, which indicates a power consumption of 32-to-38 uW.
At a control voltage of 620 mV (the best power saving situation without perfor-
mance degradation), the average biasing current of the LNA is 601 uA single-ended
and 1.20 mA in total, compared with the 2.44 mA current consumption if the LNA
is not quenched. Thus, 50.7% power reduction is obtained.
6.4 Output Buer

The output buer duplicates the oscillation envelope generated by the peak detector
and drives external instruments.
6.4.1 Device Testability

This design takes the testability into consideration. Four control inputs, namely Vclk ,
Vamp , Vf rq and Vref are implemented for performance ne-tuning after fabrication.
Vclk and Vamp x the slope and amplitude of the quenching wave for SRO. Their
combination controls the gain, sampling speed, output pulse shapes and frequency
response of the super-regenerative receiver. Vf rq is used to adjust the receiver's
operating frequency. Vref controls the warm-up time of the LNA. By tuning these
inputs, the receiver can be congured to the optimum after fabrication.
Many of these postfabrication congurations need feedback. The ideal output

External Clock (Vclk)

1000
Voltage (mV)
500
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Inverted Clock (/Vclk)
1000
Voltage (mV)
500
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Voltage Ramp (Vrmp)
1000
Voltage (mV)
800
600
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Comparator Output (Vcmp)
1000
Voltage (mV)
500
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Switching Voltage (Vswt)
1000
Voltage (mV)
500
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Figure 6.10: The transient simulation result of the dynamic biasing network.
Switching Voltage (Vsw)

1000
800
Voltage (mV)
600
400
200
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Gate Voltage of the Mirror (M4 and M5)
1000
800
Voltage (mV)
600
400
200
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
LNA Biasing Voltage
600
Voltage (mV)
500
400
300
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
LNA Drain Current
1.5
Current (mA)
0.5
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Figure 6.11: The transient simulation result of the LNA under dynamic biasing.

Envelope (mV) 150
100
50
0
0 50 100 150 200 250 300 350 400 450 500
Time (ns)
Amplitude Spectrum of the Ouput Pulses
Normalized Amplitude Spectrum
0.8
0.6
0.4
0.2
0
0 50 100 150 200 250 300 350 400
Frequency (MHz)
Figure 6.12: Output pulses and their power spectral density.
signal which contains abundant information is the pulses at the output of the peak
detector. By observing this signal, the operating frequency, gain, pulse shapes,
sampling speed and frequency response can all be deduced.

The output buer is designed to amplify the weak signal from the peak detector to
drive external instruments. Furthermore, it should be able to shield the instruments
from inuencing the operation of the receiver.
The output signal to be buered is plotted at the top of Fig. 6.12. It can be
observed that a maximum voltage swing of 150 mV is required for the output buer.
By performing a discrete Fourier transform to the output signal, its normalized
amplitude spectrum is plotted at the bottom of Fig. 6.12. As shown in the graph,
a bandwidth of around 300 MHz is needed to preserve most of the output signal
power.
The signal source, i.e. the peak detector, has a specic output capacitance of 100
fF, which should not be inuenced. Otherwise, the discharge time constant of the
peak detector would be modied and a reduced conversion gain and slow tracking
behavior could be observed.

At the load side, the input impedance of measurement instruments is specied
as 10 kohm resistance in parallel with 20 pF capacitance.
6.4.3 The schematic

The schematic of the output buer is shown in Fig. 6.13. It is a 2-stage Op-amp
with unity-gain feedback.
The rst stage is a folded cascode OTA with NMOS input dierential pair. We
choose a folded cascode input for its large gain compared with single dierential
pair and small voltage headroom requirement compared with a telescopic structure.
The NMOS input pair is chosen since the input biasing is relatively high, around
570 mV.
The second stage is an NMOS common source with a PMOS current source. The
main purpose of this stage is to drive the 20 pF capacitive load. A source follower
is not used in this design since the limited voltage swing of the previous stage and
the voltage shift prevents the output-to-input connection.
The lengths of all the transistors are scaled to 200nm for the improved intrinsic
gain and reduced threshold voltage. All transistors are biased in moderate inversion
to balance the speed and power consumption. More specically, the overdrive voltage
is selected to obtain a gm/Id of 15, which yields a current density of 10 uA/um and
a transit frequency of 1 GHz. The second stage has a width of 600 um (1800 um for
the PMOS), in order to have a biasing current of 6 mA and a slew rate of 0.3 V/ns.
The rst stage is a 3 times scaled version of the OTA designed in Subsection 6.2.4
to drive the second stage with reasonable speed.
Miller compensation is performed to achieve a phase-margin of 60 degrees. In
detail, a 1.81 pF capacitor is plugged in, shown as Cc in Fig. 6.13.
The input capacitance of this buer stage is around 128 fF, slightly larger than
the 100 fF capacitor used in the peak detector. Simulations show that the sub-
stitution of the 100 fF capacitor used in the peak detector by this 128 fF input
capacitance of the buer yield indistinguishable results.
The buer amplier has its own power supply so that the power consumption of
this buer and the receiver can be measured separately.

An AC small-signal analysis yields the open-loop amplitude and phase response after
the frequency compensation, plotted in Fig. 6.14. As shown in the graph, a phase
margin of 60 degree is obtained.
A transient simulation is performed, and the voltage output of the peak detector
and the buered version are plotted in Fig. 6.15. From the graph, nearly identical
Ibias Ibias Vout
+ Vdd Vin M1 M2

Cc Load
Ibias Ibias
Figure 6.13: The schematic of the output voltage buer.
waveforms are observed.
This output buer draws 9.86 mA from a 0.9 V power supply, i.e. a power
consumption of 8.87 mW.
6.5 Current Reference

The biasing currents of all the circuit blocks are provided by an on-chip current
reference.
6.5.1 The schematic

A peaking current source is used in this design, as shown in Fig. 6.16. Essentially,
M1 and its K times scaled up version, M2, have a square-law I-V relationship, so the
voltage between their gates are nonlinearly depending on their drain currents, which
are forced to be equal by a current mirror (M3 and M4). On the other hand, the
resistor, R, has a linear I-V relationship. In order for the voltage between the gates
of M1 and M2 and the voltage across the resistor to be equal, the drain current of
M1, or M2, has to be a certain value (besides zero), dened by the cross point of
the nonlinear curve and the linear one. M5 is a start-up device which guarantees a
current ow at zero condition. The capacitor, C, bypasses voltage variation caused
by noise and lters ringing at the start-up phase.
The design parameters for this circuit are:
R the resistance value
K the ratio of M2's W/L to M1's W/L
W/L the width-to-length ratio of M1.

6.5. CURRENT REFERENCE 97
Amplitude Response
60
50
40
Amplitude (dB)
30
20
10
10
20
0 1 2 3 4 5 6 7 8 9
10 10 10 10 10 10 10 10 10 10
Frequency (Hz)
Phase Response
0
50
Phase (degree)
100
150
200
0 1 2 3 4 5 6 7 8 9
10 10 10 10 10 10 10 10 10 10
Frequency (Hz)
Figure 6.14: Bode plot after Miller compensation.
In order to minimize the power-supply sensitivity, R should be maximized [36].
Considering chip area and process variation, R is chosen to be 1 kohm. For a peaking
current source, the current is not a monotonic growing function of the power supply.
A maximum current exist for a voltage supply variation. Therefore, to minimize
the power supply sensitivity, K is chosen in such a way that the current is locally
independent on the voltage supply around the nominal current value. This condition
happens for a Vdd of 0.9 V when K = 7. At last, we adjust W/L to 30 um / 1 um
for a nominal current of 100 uA.
6.5.2 Power Supply and Temperature Sensitivity

DC-sweep simulations are performed to obtain the power-supply and temperature
sensitivity of the output current.
The current vs. the supply voltage is plotted at the left of Fig. 6.17. The peak
of the current is tuned to the nominal power supply voltage of 0.9 V. The derivative
of output current to supply voltage at Vdd =0.9 V is 0. Furthermore, a 10% voltage
variation (from 0.81V to 0.99V) causes a maximum output change from 95.64 uA
to 95.35 uA, i.e. a 0.3% current variation.
A temperature variation from 0 to 85 degrees Celsius yields a biasing current
from 88 to 115 uA as plotted at the right of Fig. 6.17. Thus, a positive temperature
coecient of 0.318 uA/K is obtained.


740
720
700
680
Voltage (mV)
660
640
620
600
580
560
0 50 100 150 200 250 300 350 400 450 500
Time (ns)
Buffer Output
740
720
700
680
Voltage (mV)
660
640
620
600
580
560
0 50 100 150 200 250 300 350 400 450 500
Time (ns)
Figure 6.15: Voltage output of the peak detector and the buered version.
6.5. CURRENT REFERENCE 99
Vdd
M3 M4
M5
R
M2
M1 C
Figure 6.16: The schematic of the peaking current reference.
Power Supply Sensitivity Temperature Sensitivity

96 120
95.5 115
95 110
94.5 105
Ibias (uA)
Ibias (uA)
94 100
93.5 95
93 90
92.5 85
0.7 0.8 0.9 1 1.1 0 20 40 60 80
Vdd (V) Temperature (Celsius)
Figure 6.17: Power-supply and temperature sensitivity of the output current.

Power Supply
1
0.8
Voltage (V)
0.6
0.4
0.2
0
0.2
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Output Current (Without Startup Device)
120
100
Current (uA)
80
60
40
20
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Output Current (With Startup Device)
120
100
Current (uA)
80
60
40
20
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Figure 6.18: Power-up behavior.
6.5.3 Power-up Behavior

Transient simulations are also performed to examine the power-up behavior. The
supply voltage of 0.9 V is applied as a 1 ns step, shown at the top of Fig. 6.18. The
responses of the circuits with and without the start-up device (transistor M5) are
shown at the bottom and middle of Fig. 6.18, respectively. From the gure, it is
observed that transistor M5 accelerates the start-up of the current reference circuit
signicantly.
Chapter 7
Receiver Performance
7.1 Introduction
In Chapter 5 and 6, circuit blocks of the UWB-FM super-regenerative receiver are
presented. This chapter evaluates the performance of the receiver as a whole.
Section 7.2 introduces the simulation setup and presents the results. A test-
bench is designed to simulation the system consisting of all the building blocks in
the circuit level. Simulation results are compared with the receiver specication
derived in Chapter 4 to show the validity of the design.
The global optimization of power consumption is discussed in Section 7.3. This
procedure leads to further power reduction, since the individual power minimization
for each block does not necessarily yield the minimum power in total.
7.2 Test-bench Simulation

7.2.1 Transient Noise Analysis
Super-regenerative receivers are nonlinear time-varying dynamical systems. Fur-
thermore, super-regeneration is the transient response of such a dynamical system.
Therefore, fast circuit simulation techniques such as periodic steady-state in time
domain and harmonic balance in frequency domain neglect super-regeneration eect
completely. Only the transient analysis, using the method of time domain numerical
integration, can simulate the behavior of super-regenerative receivers.
Noise is a stochastic process. In time domain, noise is simulated by a random
noise generator. Depending on the power of a noise source, the random noise gen-
erator produces a time series of random numbers with a certain variance.
The transient noise analysis is a time domain circuit simulation technique. It
uses the time-domain integration method to calculate the transient response of a
dynamical system with internal noise. At each integration step, random numbers
101
102 CHAPTER 7. RECEIVER PERFORMANCE
Modulator Receiver Front-end

R1
Vosc
Vsc VCO R2 LNA PD BUF Vbpf
+
Venv BPF
Vfrq SRO
DB WG Iref
Vref Vclk Vamp
Figure 7.1: Block diagram of the receiver test-bench.
are generated at rst, representing the value of noise sources at that time. Then,
these noise signals, together with other deterministic signals are used as stimuli to
the system. After that, nodal analysis is performed to calculate the instantaneous
node voltages and loop currents. At last, the voltages across capacitors and the
currents owing through inductors are calculated by numerical integration for the
next time step.
We use the transient noise analysis to characterize the noise performance of the
super-regenerative receiver. The simulation procedure consists of 2 steps. First, a
transient analysis is performed to the receiver under excitation of the UWB-FM RF
signal. The RMS value of the output is the signal power. Then, several passes of
transient noise analysis are performed to the receiver with no RF excitation. The
mean of the RMS value of the outputs is the noise power. In this way, the output
SNR is obtained and the noise gure of the receiver can be calculated.
There are three issues need to be justied before applying the aforementioned
simulation procedure. Firstly, superposition property is implicitly assumed. This
assumption is valid as long as the SRO is working in the linear region, as indicated
in Equ. 3.31. Secondly, the output noise is implicitly assumed to be stationary and
ergodic. Otherwise, many long noise sample paths are required to calculate the noise
power. Thirdly, the noise samples to calculate the statistical variance is assumed
to be enough. Two passes of 5000 ns transient noise simulation results are used to
calculate the RMS noise power. We do not have enough computation power to do
a more intensied simulation.

7.2. TEST-BENCH SIMULATION 103
C2 Frequency Response of the Bandpass Filter

L2 5
R1
Vin Vbpf
50
28.1 uH 225 pF 10
Gain (dB)
L1 C1 R2
563 nH 11.3 nF 50
15
20
1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4
Frequency (MHz)
Figure 7.2: Bandpass lter and its frequency response.
7.2.2 The Test-bench

The schematic used for test-bench simulation is plotted in Fig. 7.1. The receiver
front-end to be evaluated consists of all the circuit blocks introduced in the previous
chapters, namely the LNA, SRO, peak detector (PD), waveform generator (WG) for
SRO, dynamic biasing (DB) for LNA, output buer (BUF) and the current reference
(Iref ).
The UWB-FM RF signal is generated by an ideal voltage-controlled oscillator
(VCO) being controlled by a 1 MHz triangular subcarrier (Vsc ). The VCO oscillates
at 4.5 GHz when Vsc is zero, and it has a gain of 500 MHz/V. Vsc has a peak-to-peak
value of 1 V and a DC value of zero. In this way, the output of the VCO is an
FM modulated signal having a center frequency of 4.5 GHz and a bandwidth of 500
MHz.
According to Equ. 4.4, the received power is -80 dBm for a communication range
of 10 meters. This signal power should generate a 31.6 uV voltage across the input
impedance of LNA of 100 ohm. To take the LNA matching into consideration, we
plug in two 50 ohm resistors, R1 and R2, and double the voltage output of VCO to
63.2 uV. As a result, the input power to the LNA is -80 dBm if it is power matched.
The output of the receiver is passed through a bandpass lter (shown as BPF in
the gure). This lter has two functions. Firstly, it recovers the subcarrier from the
amplitude modulated pulses, so the mission of the receiver front-end is accomplished.
Secondly, the processing gain is obtained by this lter, so that the noise gure or
the sensitivity of the receiver can be measured.
A second-order Butterworth lter, implemented in the LC-ladder form, is used
in this test-bench. The schematic and the frequency response of this lter are shown
in Fig. 7.2. As shown in the graph, the lter has a 2 MHz center frequency and a
400 kHz bandwidth. As a result, it passes through the FSK modulated subcarrier
while eliminates the out-of-band noise as much as possible.
The reason that this lter is not integrated on-chip is because the back-end of the
receiver, i.e. the FSK demodulator, may not require this lter at all. For example,
if the noncoherent FSK detection is performed, two detuned bandpass lters can
demodulate the FSK into binary signal without preltering.


A transient simulation is performed to the test-bench. The voltage waveforms of
the input subcarrier, SRO oscillation, its envelope and the output after bandpass
lter are plotted in Fig. 7.3. As shown in the graph, the receiver is capable of
recovering the subcarrier. Since the FM signal is converted to a 100% modulated
AM signal, and the noncoherent detection of this DSB-AM signal neglects the phase
information completely, the 1 MHz subcarrier is recovered as a 2 MHz signal.
The transient noise simulation is performed to the test-bench with zero RF input.
The voltage outputs of the SRO, the peak detector and the bandpass lter are plotted
in Fig. 7.4. The SRO oscillation in this simulation is caused by the noise only.
Comparing the amplitudes of the pulses from Fig. 7.4 to that shown in Fig.
7.3, we observe that the noise generates larger pulses than the signal does. In
other words, noise power is larger than the signal power. The only property that
distinguishes signal from noise is that the signal power is concentrated in a specic
frequency band while the noise power is evenly distributed. The bandpass lter in
the test-bench lters out all the out-of-band noise, so that noise power is reduced
while the signal power is left intact. The comparison of the Vbpf signals shown in
Fig. 7.4 and Fig. 7.3 shows this SNR improvement.
The output signal-to-noise ratio is calculated by the RMS value of the signal
divided by that of the noise. A conservative estimation using the peak-to-peak
voltage of the signal divided by the largest peak-to-peak voltage of the noise yields
an output SNR of 14.3 dB.
The whole receiver excluding the output voltage buer draws 2.273 mA in average
from a 0.9V voltage supply. Thus, a power consumption of 2.046 mW is obtained.
7.3 Global Power Reduction: A Step-controlled SRO

7.3.1 Motivation
Compared with a slope-controlled SRO, a step-controlled SRO yields at least 20%
reduction of dynamic range, and consumes more power for the same voltage swing,
as discussed in Section 5.2. Because of these reasons, the slope-control mechanism
is chosen in the rst place.
However, the slope-control requires a complex saw-tooth waveform generator to
bias the SRO. This generator consumes extra power, which is even more than that
of the SRO, as discussed in Section 6.2. It is possible to reduce this extra power but
the penalty paid is a performance degradation.
On the other hand, the biasing circuit for a step-controlled SRO is much simpler.
It can be implemented by a current mirror with its gates periodically pulled down
7.3. GLOBAL POWER REDUCTION: A STEP-CONTROLLED SRO 105
Subcarrier (Vsc)
0.5
Voltage (V)
0.5
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)
SRO Oscillation (Vosc)
200
Voltage (mV)
200
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)
Peak Detector Output (Venv)
150
Envelope (mV)
100
50
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)
Bandpass Filter Output (Vbpf)
4
Envelope (mV)
4
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)
Figure 7.3: Receiver outputs under signal excitation.


300
200
100
Voltage (mV)
100
200
300
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)

150
Envelope (mV)
100
50
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)

1
0.5
Envelope (mV)
0.5
1
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)
Figure 7.4: Receiver outputs without input signal.

Conductance (mS)
0.5
0.5
1
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Tailcurrent Waveform
1
Current (mA)
0.5
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Figure 7.5: Conductance and Tail current waveform for step-control.
by a clocked switch.
Therefore, applying step-control may lead to global power reduction, despite the
increased power and the inferior performance of the SRO.
7.3.2 The Adapted Design for Step-control

The slope-controlled super-regenerative receiver front-end already designed consists
of an LNA, an SRO, a peak detector, an output buer, a current reference, the
dynamic biasing circuit for the SRO and that for the LNA. In the adapted version
applying the step-control, the SRO is tuned slightly, the biasing waveform generator
is substituted by a switch, and the logic is inverted for the dynamic biasing of the
LNA. The rest of the circuit blocks, namely the LNA, the peak detector, the output
buer and the current reference, are utilized in the new design without modication.
7.3.2.1 The Step-controlled SRO

Biasing Waveform In Subsection 5.2.4, it is discussed that a step-controlled
start-up and quenching needs a rectangular waveform. The ideal conductance wave-
form yielding the optimum frequency response is plotted in the top of Fig. 7.5.
In the plot, the negative conductance is set to -0.681 mS according to Tab. 5.1.
And the duration of the negative conductance is 16.5 ns determined by the certain
amount of gain amplifying the RF signal to an oscillation amplitude of 300 mV.
The conductance waveform is generated by a rectangular-pulse current source.
The required waveform of this source is shown in the bottom of Fig. 7.5.
Biasing Current and Transistor Widths For a certain negative conductance,
there is still one degree of freedom to select the biasing current or transistor widths.
This selection involves the dynamic range - power trade-o discussed in Subsection
5.2.5. Basically, a small biasing current and large transistors are power saving but
the dynamic range is limited. A large biasing current requires small transistors to
generate a certain negative conductance. This large current density yields large
overdrive voltages and the oscillation amplitude is voltage limited. So an optimum
current density exists for a certain negative conductance and voltage supply.
Extensive simulations show that this optimum is achieved when the biasing tail
current is 1 mA and the width of each transistor of the cross coupled pair is 5 um.
7.3.2.2 The Dynamic Biasing Circuit

The dynamic biasing circuit for the step-controlled SRO and the LNA is shown
in Fig. 7.6. Both the SRO and LNA are biased by current mirrors with switches
attached to the gates. When either one is quenched, the corresponding switch is
closed to pull down the gate voltage and shut down the current ow.
The switch quenching the SRO is directly controlled by an external clock. The
other one quenching the LNA is controlled by a delayed version of the external clock.
As is shown in the top of Fig. 7.6, the delay-line is a logically inverted version of
the one discussed in Section 6.3. The simulated node voltages and biasing currents
are shown in Fig. 7.7.
7.3.3 Test-bench Simulation Results

Test-bench simulations similar to those introduced in Section 7.2 are performed to
evaluate the step-controlled super-regenerative receiver. The simulation results are
shown in Fig. 7.8 and Fig. 7.9, for signal and noise excitation, respectively. The
same conservative estimation using the peak-to-peak voltage of the signal divided
by the largest peak-to-peak voltage of the noise yields an output SNR of 13.2 dB.
The step-controlled super-regenerative receiver excluding the output voltage
buer draws 1.996 mA in average from a 0.9V voltage supply. Thus, a power con-
sumption of 1.796 mW is obtained.
7.3.4 The Power Reduction of the Step-controlled Receiver

In the previous sections, two super-regenerative receiver front-ends, one slope- and
the other step-controlled, are designed for UWB-FM reception. Test-bench sim-
ulations indicate that both designs satisfy the system specication. Under this
condition, the minimum power consumed by the 2 receivers is compared.

Ramp Voltage Generator Phase Delay
Vdd
Vref
Vclk M1 C CMP Vcmp
+
M2
Vswt
M3
Vdd
Vamp
SRO LNA
Vclk M4 Vswt M5
Figure 7.6: The dynamic biasing circuits for the SRO and the LNA.
Slope-controlled Step-controlled
SRO 257 uA 428 uA

Dynamic Biasing for SRO 448 uA 100 uA
Total 705 uA 528 uA
Table 7.1: The comparison of current consumption between the slope- and step-
controlled receivers.
External Clock
1000
Voltage (mV)
500
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Voltage Ramp
1000
Voltage (mV)
800
600
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Comparator Output
1000
Voltage (mV)
500
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
LNA Switching Voltage
1000
Voltage (mV)
500
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
LNA Biasing Current
Current (mA)
0.5
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
SRO Biasing Current
Current (mA)
0.5
0
0 10 20 30 40 50 60 70 80 90 100
Time (ns)
Figure 7.7: Delay-line voltages and biasing currents of the SRO and the LNA.
Subcarrier (Vsc)
0.5
Voltage (V)
0.5
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)
300
200
Voltage (mV)
100
0
100
200
300
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)
150
Envelope (mV)
100
50
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)
4
Envelope (mV)
4
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)
Figure 7.8: Receiver outputs under signal excitation.


300
200
100
Voltage (mV)
100
200
300
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)

150
Envelope (mV)
100
50
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)

1
0.5
Envelope (mV)
0.5
1
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (ns)
Figure 7.9: Receiver outputs without input signal.

Besides the power consumed by the LNA, peak detector and the biasing, which
are nearly equal for the two receivers, the power dissipation of the SRO and its
dynamic biasing circuit is listed in Tab. 7.1. We can observe from the table that
although the slope-control yields a smaller SRO power consumption than the step-
control does, as predicted theoretically in Subsection 5.2.4, the extra power used to
generate the complex waveform outweighs its benet. In contrast, the step-control
is power consuming in principle, but the simple power management circuit leads to
overall power saving.

Chapter 8
Conclusions
8.1 Summary of Results

UWB-FM systems, targeting on short-range low-data-rate wireless applications, de-
mand receivers with low complexity and low power consumption. In this research, a
fully integrated super-regenerative receiver in IBM 90-nm RF CMOS technology is
designed to detect 500 MHz bandwidth UWB-FM signals at 4.5 GHz. Circuit sim-
ulations show that a receiver sensitivity of -82.2 dBm is attainable for a 100 kbps
baseband data-rate and 106 bit-error-rate. The whole circuit draws an average of
2 mA from a 0.9 V supply.
Tab. 8.1 compares the performance of this work with another two recently
published UWB-FM receivers. As shown in the table, this receiver demonstrates
higher sensitivity while consumes remarkably less power.
The contribution of this work also includes an innovative optimization of the
quenching waveform for WBFM detection and a novel low-power driven design pro-
cedure for LNAs.
In conclusion, super-regenerative receivers, due to their simplicity and low power-
consumption, are promising for short-range low-data-rate UWB-FM applications.
Tong [12] Gerrits [37] This work
RF frequency 4 GHz 4 GHz 4.5 GHz

RF bandwidth 2 GHz 500 MHz 500 MHz
Sensitivity N/A -74 dBm -82.2 dBm
Data rate N/A 62.5 kbps 100 kbps
4 6
Bit error rate N/A 10 10
Power consumption 19.4 mW 9.6 mW (w/o LNA) 1.8 mW
Technology 0.18 um CMOS 0.18 um BiCMOS 90 nm CMOS
Table 8.1: Performance comparison of UWB-FM receivers.
114
8.2. FUTURE WORK 115
8.2 Future Work

This research demonstrates the feasibility and advantage of performing UWB-FM
detection by super-regenerative receivers. Layout, fabrication and physical mea-
surement of this design is necessary to make the point more concrete. Industrial
application of this design requires further reliability and yield analysis.
UWB-FM is designed to have multiuser capability and robustness to narrow-
band interference [28, 38]. The multiuser capability is briey discussed in Section
4.2. The feasibility of using this receiver for subcarrier FDMA requires further study
of the LNA linearity. Qualitatively, the susceptibility to narrow-band interference is
the most at the center frequency, 4.5 GHz, and the least at 4.25 and 4.75 GHz. The
quantitative relation of tolerable interference strength versus frequency is a subject
for future study.
The current design of this receiver assumes an external clock to quench the os-
cillation periodically. However, it leaves the freedom of generating this clock to the
user. Self-quenching is possible by the addition of a comparator and a monostable
multivibrator. In this conguration, oscillations are quenched as long as the output
envelope reaches a certain level. Self-quenching leads to very simple implementation
and presumably low power consumption. The penalty paid is the reduced dynamic
range. Whether the dynamic range reduction is acceptable for UWB-FM applica-
tions is a subject for future research.
RF-MEMS technology may reduce the power consumption considerably. Ac-
cording to Section 5.2, the lower bound of power consumption is ultimately limited
by the quality factor of the resonator. In the literature, bulk acoustic wave (BAW)
resonators are increasingly used to obtain super-regenerative receivers working at
micro-watt power levels. The extra benet of using high-Q devices includes the
selectivity and frequency stability improvements.

116 CHAPTER 8. CONCLUSIONS
Bibliography
[1] J. F. Gerrits and J. R. Farserotu., Ultra wideband fm: A straightforward
frequency domain approach, European Microwave Conference, 2003. 33rd,
pp. 853856, Oct. 2003.
[2] E. H. Armstrong, Signaling systems, US Patent, July 1922.
[3] J. Whitehead, Super-Regenerative Receivers. Cambridge Press, 1950.
[4] P. Favre, N. Joehl, A. Vouilloz, P. Deval, C. Dehollain, and M. Declercq, A 2-v
600-a 1-ghz bicmos super-regenerative receiver for ism applications, Solid-
State Circuits, IEEE Journal of, vol. 33, pp. 21862196, Dec 1998.
[5] A. Vouilloz, M. Declercq, and C. Dehollain, A low-power cmos super-
regenerative receiver at 1 ghz, Solid-State Circuits, IEEE Journal of, vol. 36,
pp. 440451, Mar 2001.
[6] N. Joehl, C. Dehollain, P. Favre, P. Deval, and M. Declerq, A low-power 1-
ghz super-regenerative transceiver with time-shared pll control, Solid-State
Circuits, IEEE Journal of, vol. 36, pp. 10251031, Jul 2001.
[7] J.-Y. Chen, M. Flynn, and J. Hayes, A fully integrated auto-calibrated super-
regenerative receiver in 0.13-m cmos, Solid-State Circuits, IEEE Journal
of, vol. 42, pp. 19761985, Sept. 2007.
[8] I. McGregor, E. Wasige, and I. Thayne, Sub-50w, 2.4 ghz super-regenerative
transceiver with ultra low duty cycle and a 675w high impedance super-
regenerative receiver, Microwave Conference, 2007. European, pp. 13221325,
Oct. 2007.
[9] B. Otis, Y. Chee, and J. Rabaey, A 400 /spl mu/w-rx, 1.6mw-tx super-
regenerative transceiver for wireless sensor networks, Solid-State Circuits Con-
ference, 2005. Digest of Technical Papers. ISSCC. 2005 IEEE International,
pp. 396606 Vol. 1, Feb. 2005.
117
118 BIBLIOGRAPHY
[10] J. L. Bohorquez, J. L. Dawson, and A. P. Chandrakasan, A 350w cmos msk
transmitter and 400w ook super-regenerative receiver for medical implant
communications, VLSI Circuits, 2008 IEEE Symposium on, pp. 3233, June
2008.
[11] J. Gerrits, J. Farserotu, and J. Long, A wideband fm demodulator for uwb ap-
plications, Research in Microelectronics and Electronics 2006, Ph. D., pp. 461
464, 0-0 2006.
[12] T. Tong, J. Mikkehen, and T. Larsen, A 0.18 mum cmos implementation of
a low power, fully dierential rf front-end for fm-uwb based p-pan receivers,
Communication systems, 2006. ICCS 2006. 10th IEEE Singapore International
Conference on, pp. 15, 30 2006-Nov. 1 2006.
[13] M. H. L. Kouwenhoven, High Performance Frequency Demodulation Systems.
Delft University Press, 1998.
[14] F. Moncunill-Geniz, P. Pala-Schonwalder, C. Dehollain, N. Joehl, and M. De-
clercq, An 11-mb/s 2.1-mw synchronous superregenerative receiver at 2.4 ghz,
Microwave Theory and Techniques, IEEE Transactions on, vol. 55, pp. 1355
1362, June 2007.
[15] D. Ash, A low cost superregenerative saw stabilized receiver, Consumer Elec-
tronics, IEEE Transactions on, vol. CE-33, pp. 395404, Aug. 1987.
[16] F. Moncunill, O. Mas, and P. Pala, A direct-sequence spread-spectrum super-
regenerative receiver, Circuits and Systems, 2000. Proceedings. ISCAS 2000
Geneva. The 2000 IEEE International Symposium on, vol. 1, pp. 6871 vol.1,
2000.
[17] F. Moncunill-Geniz, P. Pala-Schonwalder, and F. del Aguila-Lopez, New su-
perregenerative architectures for direct-sequence spread-spectrum communica-
tions, Circuits and Systems II: Express Briefs, IEEE Transactions on, vol. 52,
pp. 415419, July 2005.
[18] M. Pelissier, D. Morche, and P. Vincent, Rf front end of uwb receiver based on
super-regeneration, Ultra-Wideband, 2007. ICUWB 2007. IEEE International
Conference on, pp. 180183, Sept. 2007.
[19] F. Xavier Moncunill-Geniz, P. Pala-Schonwalder, F. del Aguila-Lopez, and
R. Giralt-Mas, Application of the superregenerative principle to uwb pulse
generation and reception, Electronics, Circuits and Systems, 2007. ICECS
2007. 14th IEEE International Conference on, pp. 935938, Dec. 2007.
BIBLIOGRAPHY 119
[20] M. Anis, R. Tielert, and N. Wehn, Fully integrated self-quenched super-
regenerative uwb impulse detector, Wireless Pervasive Computing, 2008.
ISWPC 2008. 3rd International Symposium on, pp. 773775, May 2008.
[21] H. Kalmus, Some notes on superregeneration with particular emphasis on its
possibilities for frequency modulation, Proceedings of the IRE, vol. 32, pp. 591
600, Oct. 1944.
[22] L. Hernandez and S. Paton, A superregenerative receiver for phase and fre-
quency modulated carriers, Circuits and Systems, 2002. ISCAS 2002. IEEE
International Symposium on, vol. 3, pp. III81III84 vol.3, 2002.
[23] J. Ayers, K. Mayaram, and T. Fiez, A low power bfsk super-regenerative
transceiver, Circuits and Systems, 2007. ISCAS 2007. IEEE International
Symposium on, pp. 30993102, May 2007.
[24] F. Moncunill-Geniz, P. Pala-Schonwalder, and O. Mas-Casals, A generic ap-
proach to the theory of superregenerative reception, Circuits and Systems I:
Regular Papers, IEEE Transactions on, vol. 52, pp. 5470, Jan. 2005.
[25] T. H. Lee, The Design of CMOS Radio-Frequency Integrated Circuits. Cam-
bridge University Press, 2nd. ed., 2004.
[26] J. Whelehan, Low-noise ampliers-then and now, Microwave Theory and
Techniques, IEEE Transactions on, vol. 50, pp. 806813, Mar. 2002.
[27] Y. Xue, J. Gerrits, H. Choi, Nikhil, and S. K., Specication for prototyping
from phy and layer 2 (update d3.3.4a), Tech. Rep. IST-507102 D3.3.4b, My
Personal Adaptive Global NET, Dec. 2005.
[28] J. Gerrits, J. Farserotu, and J. Long, Multi-user capabilities of uwbfm commu-
nications systems, Ultra-Wideband, 2005. ICU 2005. 2005 IEEE International
Conference on, pp. 6 pp., Sept. 2005.
[29] F. C. Commission, First report and order, Revision of Part 15 of the Commis-
sion's Rules Regarding Ultra-Wideband Transmission Systems, Apr. 22 2002.
[30] J. Keignart, J.-B. Pierrot, N. Daniele, A. Alvarez, M. Lobeira, J. L. Garcia,
G. Valera, and R. P. Torres, U.c.a.n. report on uwb basic transmission loss,
Tech. Rep. IST-2001-32710, U.C.A.N., Mar. 2003.
[31] C. Mutti, F. Bauer, K. Schoo, M. Monti, F. Platbrood, A. Pollini, J. Gerrits,
J. Ayadi, and I. Siaud, Requirement specication for phy-layer, Tech. Rep.
IST-507102 D3.2.1, My Personal Adaptive Global NET, Jun. 2004.

120 BIBLIOGRAPHY
[32] ETSI, Electromagnetic compatibility and Radio spectrum Matters (ERM);
Wideband Transmission systems; Data transmission equipment operating in the
2,4 GHz ISM band and using spread spectrum modulation techniques; Part 1:
Technical characteristics and test conditions. No. ETSI EN 300 328-1 in Euro-
pean Standard (Telecommunications series), ETSI, v1.3.1 ed., Dec. 2001.
[33] IBM, CMOS 9SF Technology Design Manual, vol. ES 88H1175. IBM, Setp.
2007.
[34] R. Meyer, Low-power monolithic rf peak detector analysis, Solid-State Cir-
cuits, IEEE Journal of, vol. 30, pp. 6567, Jan 1995.
[35] D. Cassan and J. Long, A 1-v transformer-feedback low-noise amplier for 5-
ghz wireless lan in 0.18-/spl mu/m cmos, Solid-State Circuits, IEEE Journal
of, vol. 38, pp. 427435, Mar 2003.
[36] B. Razavi, Design of Analog CMOS Integrated Circuits. McGraw-Hill, 2001.
[37] J. Gerrits, J. Farserotu, and J. Long, A wideband fm demodulator for a low-
complexity fm-uwb receiver, Wireless Technology, 2006. The 9th European
Conference on, pp. 99102, Sept. 2006.
[38] J. Gerrits, J. Farserotu, and J. Long, Low-complexity ultra wideband com-
munications, Circuits and Systems, 2007. ISCAS 2007. IEEE International
Symposium on, pp. 757760, May 2007.

Super-Regenerative Receiver For UWB-FM Rui - Hou - 2008 PDF

Uploaded by

Copyright:

Available Formats

Super-Regenerative Receiver For UWB-FM Rui - Hou - 2008 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Super-Regenerative Receiver For UWB-FM Rui - Hou - 2008 PDF

Uploaded by

Copyright:

Available Formats

Super-regenerative Receiver for UWB-FM

Delft University of Technology

September 20, 2008

A thesis submitted to the Department of Microelectronics, Faculty of Electrical

Engineering, Mathematics and Computer Science, Delft University of Technology,

in partial fulllment of the requirements for the degree of

Prof. John R. Long

Prof. John R. Long

Dr. ing. Leo C. N. de Vreede

Dr. Ko A. A. Makinwa

September 20, 2008

UWB-FM is a low-complexity ultra-wideband (UWB) communication system de-

receivers with low power consumption.

Most of previous work utilized delay-line demodulators for UWB-FM detection.

necessarily power-ecient. Reported power consumption was around 20 mW for a

tion of a UWB-FM receiver by exploiting the super-regeneration principle.

As a result, a fully integrated super-regenerative receiver in IBM 90-nm RF

CMOS technology is designed to detect 500 MHz bandwidth UWB-FM signals at

an average of 2 mA from a 0.9 V supply.

In addition, the contribution of this work includes an innovative optimization of

procedure for LNAs.

In conclusion, super-regenerative receivers are promising for short-range, low-

data-rate UWB-FM applications, due to their simplicity and low power-consumption.

inuence of many individuals.

Firstly, I would like to give my gratitude to my supervisor, Prof. John R. Long,

inspiration, criticism and encouragement.

been a pleasure since then.

full of adventure. Before my graduation, I would like to express my gratitude to all

college students for their kindness and help.

My family is always supporting me, especially during my study in the Nether-

lands. I am very grateful for that.

of my brilliant wife, Cheng Jiang.

Delft, the Netherlands

1.2 Purpose and Scope of the Research . . . . . . . . . . . . . . . . . . . 1

1.3 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2.2 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.3 Current Approach: Super-regeneration . . . . . . . . . . . . . . . . . 5

3.2 Driven Parametric Oscillator Model . . . . . . . . . . . . . . . . . . 8

3.3 Solution of the ODE . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.3.1 General Solution . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.3.2 Particular Solution . . . . . . . . . . . . . . . . . . . . . . . . 12

3.3.3 Complete Solution . . . . . . . . . . . . . . . . . . . . . . . . 15

3.4 Characteristics of an SRO . . . . . . . . . . . . . . . . . . . . . . . . 16

3.4.1 Sensitivity Curve . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.4.2 Oscillation Envelope . . . . . . . . . . . . . . . . . . . . . . . 17

3.4.4 Frequency Response . . . . . . . . . . . . . . . . . . . . . . . 18

4.2 System Specication . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4.2.1 RF Signal Characteristics . . . . . . . . . . . . . . . . . . . . 22

4.2.1.1 Operating Frequency . . . . . . . . . . . . . . . . . . 22

4.2.1.2 Transmitting Power and Propagation . . . . . . . . . 22

4.2.1.3 Noise Power and SNR . . . . . . . . . . . . . . . . . 23

4.2.2 Sub-band Signal Characteristics . . . . . . . . . . . . . . . . . 23

4.2.2.1 Sub-band Signal Bandwidth . . . . . . . . . . . . . . 24

4.2.2.2 SNR Requirement . . . . . . . . . . . . . . . . . . . 24

4.2.3 Receiver Front-end Specication . . . . . . . . . . . . . . . . . 25

4.2.3.1 Noise Figure . . . . . . . . . . . . . . . . . . . . . . 25

4.2.3.3 Spurious Emission . . . . . . . . . . . . . . . . . . . 25

4.2.4 Specication Summary . . . . . . . . . . . . . . . . . . . . . 26

4.3 The FM Demodulation Scheme . . . . . . . . . . . . . . . . . . . . . 26

in partial fulllment of the requirements for the degree of

Dr. Ko A. A. Makinwa

necessarily power-ecient. Reported power consumption was around 20 mW for a

inuence of many individuals.

4.2 System Specication . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4.2.3 Receiver Front-end Specication . . . . . . . . . . . . . . . . . 25

4.2.4 Specication Summary . . . . . . . . . . . . . . . . . . . . . 26

5.2.5 Dynamic-range Power Trade-o . . . . . . . . . . . . . . . . . 47

5.2.7.3 Dielectric Integrity Verication . . . . . . . . . . . . 51

5.3.2 The Dierential Peak Detector . . . . . . . . . . . . . . . . . . 53

5.4 Low-noise Amplier . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

5.4.1.4 Specication Summary . . . . . . . . . . . . . . . . . 61

6.4 Output Buer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91