CORE Scholar CORE Scholar: Wright State University Wright State University
CORE Scholar CORE Scholar: Wright State University Wright State University
CORE Scholar CORE Scholar: Wright State University Wright State University
CORE Scholar
2019
Repository Citation
Li, Shuo, "CMOS Receiver Design for 802.11ac Standard Using Offline Calibrated Active Inductor Based
Band Pass Filter in 90 nm Technology" (2019). Browse all Theses and Dissertations. 2266.
https://corescholar.libraries.wright.edu/etd_all/2266
This Dissertation is brought to you for free and open access by the Theses and Dissertations at CORE Scholar. It
has been accepted for inclusion in Browse all Theses and Dissertations by an authorized administrator of CORE
Scholar. For more information, please contact library-corescholar@wright.edu.
CMOS RECEIVER DESIGN FOR 802.11AC STANDARD
USING OFFLINE CALIBRATED ACTIVE INDUCTOR
BASED BAND PASS FILTER IN 90 NM TECHNOLOGY
by
SHUO LI
2019
Wright State University
COPYRIGHT BY
SHUO LI
2019
WRIGHT STATE UNIVERSITY
GRADUATE SCHOOL
Dec 4, 2019
_______________________
_________
Saiyu Ren, Ph.D.
Dissertation Director
_______________________
_________
Fred Garber, Ph.D.
Interim Chair, Electrical
Engineering
_______________________
_________
Barry Milligan, Ph.D.
Interim Dean of the Graduate
School
Committee on Final Examination:
________________________________
Raymond E. Siferd, Ph.D.
________________________________
Henry Chen, Ph.D.
________________________________
Marian K. Kazimierczuk, Ph.D.
________________________________
John M. Emmert, Ph.D.
ABSTRACT
Li, Shuo. Ph.D, Department of Electrical Engineering, Wright State University, 2019.
CMOS Receiver Design for 802.11ac Standard Using Offline Calibrated Active
Inductor Based Band Pass Filter in 90 nm Technology
Wireless local area network is widely used in industry and people daily life.
More and more mobile devices rely on this technology to perform data communication
with 2.4 GHz and 5 GHz frequency band. As the development of CMOS technology is
able to keep shrinking chip size and increasing circuit integration density, traditional
receiver front end system design. In this dissertation, an active inductor-based band pass
filter is studied and implemented with 90 nm technology. This active inductor design
provides very small area consumption and larger quality factor compared to
monitor and compensate the process variation error of band pass filter center frequency
dissertation with active filter and Hartley image rejection architecture embedded into
the system. The receiver can down-convert a 5.25 GHz signal to a 250 MHz IF signal
with input power from -90 dBm to -50 dBm. The area consumption of entire receiver
iv
Table of Contents
I. Introduction .......................................................................................................................1
1.1 History of WLAN ....................................................................................................1
1.2 Compare 2.4 GHz and 5 GHz .................................................................................2
1.3 Benefit of Using 802.11ac Standard .......................................................................4
1.4 Receiver System of Mobile Device Background Study.........................................4
1.5 Motivation ................................................................................................................7
1.6 Objective ..................................................................................................................8
II. Receiver System Architecture ...........................................................................................9
2.1 Typical Receiver Architecture for Wireless Application ......................................9
2.1.1 Image Rejection .............................................................................................10
2.1.2 Heterodyne Receiver ..................................................................................... 11
2.1.3 Hartley Receiver ............................................................................................12
2.1.4 Weaver Receiver ............................................................................................13
2.2 Sub-Circuits of Receiver System with Expected Performance ..................15
III. Low Noise Amplifier .......................................................................................................18
3.1 Introduction ...........................................................................................................18
3.2 Theoretical Analysis ..............................................................................................18
3.3 Circuit Design Techniques ....................................................................................22
3.3.1 Low Threshold Voltage MOS technique ......................................................22
3.3.2 Varactor ..........................................................................................................24
3.4 LNA Implementation and Simulation results .....................................................26
3.5 Conclusion ..............................................................................................................32
IV. Active Inductor Band Pass Filter.....................................................................................33
4.1 Introduction ...........................................................................................................33
4.2 Area Consumption of Passive Filters ...................................................................34
4.3 Active Inductor-based Band Pass Filter Operating Theory ..............................35
4.4 Simulation Result ..................................................................................................39
V. CMOS Amplitude Peak Detector ....................................................................................43
5.1 Introduction ...........................................................................................................43
5.2 Theoretical Analysis ..............................................................................................43
5.2.1 Two-State Amplitude Peak Detector Implementation ................................43
5.2.2 Output Coupling Jitter Analysis ..................................................................45
5.3 Layout Simulation Results ....................................................................................48
5.4 Conclusion ..............................................................................................................52
VI. Analog Buffer ..................................................................................................................53
6.1 Introduction ...........................................................................................................53
6.1.1 Source Follower Based Unity Gain Buffers.................................................54
6.1.2 Source Coupled Differential Pair Based Unity Gain Buffer ......................55
6.2 Two Stage Common Source Active Load Unity Gain Buffer.............................56
6.2.1 Linearized Small Signal Performance Analysis ..........................................57
v
6.2.2 CSAL Buffer Performance Analysis Based on 90 nm CMOS Design .......59
6.3 Double Sided Active Load Analog Buffer with Source Feedback .....................63
6.3.1 Linearized Small Signal Performance Analysis ..........................................64
6.3.2 DSCSAL Buffer with Source Feedback Performance Analysis Based on 90
nm CMOS Design ..........................................................................................................69
6.4 Comparison to Other Published Work ................................................................72
6.5 Conclusion ..............................................................................................................73
VII. On-chip Self-Calibration System for CMOS Active Inductor Band Pass Filter..............75
7.1 Introduction ...........................................................................................................75
7.2 Process Variation Detection and Calibration ......................................................75
7.3 Charge Pump .........................................................................................................78
7.4 Simulation Results .................................................................................................79
7.5 Conclusion ..............................................................................................................83
VIII. Mixer ...............................................................................................................................85
8.1 Introduction ...........................................................................................................85
8.2 Mixer Design ..........................................................................................................86
8.2.1 SIDO Mixer for Weaver Design ...................................................................86
8.2.2 Gilbert Mixer .................................................................................................87
8.2.3 SIDO Mixer for Hartley Design ...................................................................90
8.3 Simulation Results .................................................................................................90
8.3.1 SIDO Mixer of Hartley and Weaver Design Simulation Results ...............91
8.3.2 Gilbert Mixer of Weaver Design Simulation Results..................................98
8.4 Conclusion ............................................................................................................105
IX. Phase Locked Loop .......................................................................................................106
9.1 Introduction .........................................................................................................106
9.2 System Topology and Sub-Circuit Introduction ...............................................107
9.2.1 Phase Frequency Detector ..........................................................................107
9.2.2 VCO with Quadrature Outputs .................................................................109
9.2.3 Charge Pump and Loop Filter ................................................................... 112
9.2.4 N-Divider Block ........................................................................................... 113
9.3 Simulation Result ................................................................................................ 115
9.4 Conclusion ............................................................................................................ 119
X. Analog 90 Degree Phase Shifter ...................................................................................120
10.1 Introduction .........................................................................................................120
10.2 RC-CR Network ..................................................................................................120
10.3 90 Degree Phase Shifter Simulation Results .....................................................123
10.4 Conclusion ............................................................................................................125
XI. System Performance ......................................................................................................127
11.1 Performance Comparison between Hartley and Weaver System ...................127
11.1.1 Hartley System Simulation Results ............................................................127
11.1.2 Weaver System Simulation Results ............................................................129
11.2 Proposed Heterodyne plus Hartley Receiver Simulation Results ...................131
vi
11.3 Conclusion ............................................................................................................134
XII. Conclusion and Future Works .......................................................................................136
12.1 Conclusion ............................................................................................................136
12.2 Major Contributions ...........................................................................................136
12.3 Future work .........................................................................................................137
12.3.1 Calibration System Optimization ..............................................................137
12.3.2 Active Inductor Based Band Pass Filter Gain Calibration ......................137
References .....................................................................................................................................138
vii
List of Figure
Fig 2.1.1 General architecture of receiver front end design ................................................................ 9
Fig 2.1.2 Image issue existing in receiver system [10] ........................................................................ 11
Fig 2.1.3 Heterodyne receiver architecture [21] ................................................................................. 12
Fig 2.1.4 Hartley image rejection architecture [10] ........................................................................... 13
Fig 2.1.5 Weaver image rejection architecture [24] ........................................................................... 14
Fig 2.1.6 Proposed receiver design combined Heterodyne and Hartley architecture. .................... 15
Fig 2.2.1 Receiver system expected gain and noise distribution. ...................................................... 17
Fig 3.1.1 Schematic diagram of LNA .................................................................................................. 19
Fig 3.2.1 Small signal model of LNA input stage ............................................................................... 19
Fig 3.3.1 MOSFET cascoding model ................................................................................................... 23
Fig 3.3.2 Threshold voltage measurements of normal NMOS and low threshold NMOS .............. 24
Fig 3.3.3 Cross section diagram of NCAP .......................................................................................... 25
Fig 3.3.4 Schematic diagram of band pass filter built with inductor and varactor ........................ 25
Fig 3.3.5 Center frequency of band pass filter under different gate voltages of varactor .............. 26
Fig 3.4.1 Schematic diagram of proposed LNA design. ..................................................................... 27
Fig 3.4.2 Simulation results of center frequency, gain, and 3 dB down bandwidth ........................ 27
Fig 3.4.3 LNA simulation results of -0.5 V and 1 V varactor gate voltages...................................... 28
Fig 3.4.4 LNA S11 plot. .......................................................................................................................... 30
Fig 3.4.5 LNA noise figure plot. ........................................................................................................... 30
Fig 3.4.6 1 dB compression point simulation result of LNA ............................................................. 31
Fig 3.4.7 IIP3 simulation result of LNA .............................................................................................. 32
Fig 4.2.1 Schematic and frequency response figure of conventional band pass filter: (a) Schematic
(b) Bode plot ................................................................................................................................. 35
Fig 4.3.1 Gyrator-C network ............................................................................................................... 35
Fig 4.3.2 Schematic of active inductor ................................................................................................ 37
Fig 4.3.3 Schematic of active inductor-based band pass filter .......................................................... 37
Fig 4.3.4 Small signal model of AIBPF ............................................................................................... 38
Fig 4.4.1 Schematic diagram of proposed AIBPF. ............................................................................. 40
Fig 4.4.2 Frequency response plot of AIBPF ...................................................................................... 40
Fig 4.4.3 AIBPF frequency response under different process corner ............................................... 41
Fig 4.4.4 Process variation effect on AIBPF (simulated with Monte Carlo Analysis) ..................... 42
Fig 5.2.1 Schematic of proposed CMOS peak detector. .................................................................... 44
Fig 5.2.2 Small signal equivalent circuit of CMOS peak detector .................................................... 46
Fig 5.3.1 Layout of proposed CMOS amplitude peak detector using 90 nm technology ............... 48
Fig 5.3.2 Jitter results with 195 ohms, 580 ohms, and 700 ohms feedback resistance .................... 49
Fig 5.3.3 Output jitter amplitude data plot of feedback resistance sweeping from 195 ohms to 800
ohms with fine simulation between 500 ohms and 600 ohms. .................................................. 50
Fig 5.3.4 DC detection simulation result of input amplitude sweeping from 0.1 V to 0.5 V at 6.0
GHz frequency. ............................................................................................................................. 50
Fig 6.1.1 Source coupled differential pair unity gain buffer. ............................................................ 55
viii
Fig 6.2.1 Two-stage CSAL analog buffer. ........................................................................................... 56
Fig 6.2.2 Two-stage CSAL linearized AC circuit ................................................................................ 57
Fig 6.2.3 CSAL buffer AC analysis plot. ............................................................................................. 61
Fig 6.2.4 Gain error vs. input/output amplitude. ............................................................................... 61
Fig. 6.3.1 Two stage CSAL buffer with source feedback. .................................................................. 63
Fig. 6.3.2 Small signal model of DSCSAL buffer with source feedback .......................................... 64
Fig. 6.3.3 Layout of proposed DSCSAL buffer .................................................................................. 69
Fig. 6.3.4 AC analysis plots of DSCSAL buffer. ................................................................................. 71
Fig. 6.3.5 Gain Error vs Input/Output Amplitude ............................................................................. 71
Fig 7.2.1 Proposed calibration system block diagram of AIBPF ...................................................... 77
Fig 7.2.2 Process variation effect of AIBPF: (a) Without process variation (b) With process
variation (process parameters towards the slow corner ........................................................... 77
Fig 7.3.1 Charge pump schematic diagram ........................................................................................ 78
Fig 7.4.1 Cadence simulation result of AIBPF output and two bias voltages using proposed
calibration system with slow corner process variation ............................................................. 80
Fig 7.4.2 Yield versus number of Monte Carlo iterations with proposed calibration system ........ 82
Fig 8.1.1 Frequency down-converting path with two mixers in Weaver architecture. ................... 85
Fig 8.2.1 Schematic diagram of SIDO mixer ...................................................................................... 86
Fig 8.2.2 Schematic diagram of Gilbert Mixer .................................................................................. 87
Fig 8.2.3 Conventional R-L-C band pass network............................................................................. 90
Fig 8.3.1 Schematic diagram of SIDO mixer system. ........................................................................ 90
Fig 8.3.2 Quiescent simulation result of SIDO mixer ........................................................................ 91
Fig 8.3.3 Frequency response simulation result of AIBPF modified for SIDO mixer ..................... 92
Fig 8.3.4 Frequency response simulation result of 4 stages AIBPF in series in Weaver design. .... 93
Fig 8.3.5 Time domain transient analysis result of SIDO mixer with AIBPF in Weaver design .... 93
Fig 8.3.6 Noise figure simulation result of first stage frequency conversion system consisted by
SIDO mixer and AIBPFs in Weaver design ............................................................................... 94
Fig 8.3.7 1dB compression simulation result of SIDO mixer in Weaver design .............................. 95
Fig 8.3.8 1dB compression simulation result of SIDO mixer connected with AIBPFs in Weaver
design............................................................................................................................................. 95
Fig 8.3.9 1dB compression point of SIDO mixer in Weaver design .................................................. 96
Fig 8.3.10 IIP3 simulation result of SIDO mixer plus AIBPFs in Weaver design ........................... 97
Fig 8.3.11 Power supply rejection simulation result of SIDO mixer in Weaver design .................. 97
Fig 8.3.12 Schematic diagram of Gilbert mixer. ................................................................................ 98
Fig 8.3.13 Quiescent simulation results of Gilbert mixer. ................................................................. 99
Fig 8.3.14 Frequency response simulation result of AIBPF modified for Gilbert mixer .............. 100
Fig 8.3.15 Frequency response simulation result of 3 stages AIBPF in series. .............................. 100
Fig 8.3.16 Time domain simulation result of Gilbert mixer with AIBPFs ..................................... 101
Fig 8.3.17 Noise Figure simulation result of second stage frequency conversion system (Gilbert
mixer plus AIBPFs) .................................................................................................................... 102
Fig 8.3.18 1 dB compression point simulation result of Gilbert mixer........................................... 102
Fig 8.3.19 1 dB compression point simulation result of Gilbert mixer with AIBPFs .................... 103
ix
Fig 8.3.20 IIP3 simulation result of Gilbert mixer ........................................................................... 103
Fig 8.3.21 IIP3 simulation result of Gilbert mixer plus AIBPFs..................................................... 104
Fig 8.3.22 Power supply rejection simulation result of Gilbert mixer ........................................... 104
Fig 9.2.1 System overview of proposed PLL design ......................................................................... 106
Fig 9.2.2 PFD using DFF circuit ........................................................................................................ 108
Fig 9.2.3 Proposed PFD with zero dead zone ................................................................................... 109
Fig 9.2.4 Schematic of proposed delay cell using in VCO ............................................................... 110
Fig 9.2.5 Schematic of quadrature outputs VCO using propose delay cell ................................... 111
Fig 9.2.6 Loop filter with additional pole ......................................................................................... 113
Fig 9.2.7 Schematic of single phase clock flip-flop ........................................................................... 114
Fig 9.2.8 Operation theory of divide-by-2 circuit ............................................................................ 114
Fig 9.3.1 Simulation result of PFD with 500 ps time difference between two input signals ......... 115
Fig 9.3.2 Schematic of proposed VCO with quadrature phase output and buffers ...................... 116
Fig 9.3.3 Simulation result of proposed VCO with I and Q outputs .............................................. 116
Fig 9.3.4 Phase noise of proposed VCO with 5 GHz operating frequency .................................... 117
Fig 9.3.5 Schematic of PLL with sub-circuit blocks......................................................................... 118
Fig 9.3.6 Frequency response of proposed PLL system. .................................................................. 118
Fig 10.2.1 RC-CR Network with adjustable resistor (active load) ................................................. 121
Fig 10.2.2 Frequency response of high pass and low pass filter with gain mismatch phenomenon
of phase shifter ........................................................................................................................... 122
Fig 11.1.1 Schematic diagram of proposed Hartley system. ........................................................... 127
Fig 11.1.2 Noise figure plot of proposed Hartley system. ................................................................ 128
Fig 11.1.3 Hartley System output waveforms generated by RF and Image signals. ..................... 128
Fig 11.1.4 Schematic diagram of designed Weaver system ............................................................. 129
Fig 11.1.5 Weaver system noise figure. ............................................................................................. 130
Fig 11.1.6 Weaver system output waveforms generated by RF and Image signals. ...................... 130
Fig 11.2.1 Proposed receiver schematic diagram. ............................................................................ 131
Fig 11.2.2 Noise figure of receiver RF section circuits. .................................................................... 133
Fig 11.2.3 Proposed receiver output plots of RF (5.25 GHz) and image (4.75 GHz) input signal
with image rejection of 35.63 dB. .............................................................................................. 133
Fig 11.2.4 Transient simulation results of proposed receiver design: (a) input power of -90 dBm;
(b) input power of -50 dBm. ...................................................................................................... 134
x
List of Table
Table 1.2.1 2.4 GHz WLAN Channel and Center Frequency [4] ....................................................... 3
Table 1.2.2 5 GHz WLAN Non-Overlapping Channel and Center Frequency [4]............................ 3
Table 5.3.1 Simulation Summary and Comparison ........................................................................... 51
Table 6.2.1 Transistor sizes of CSAL buffer optimized for 250fF capacitive load ........................... 60
Table 6.2.2 Summary of single sided 90nm CMOS CSAL analog buffer performance optimized
for 250fF load................................................................................................................................ 62
Table 6.3.1 Transistor sizes of DSCSAL Analog Buffer optimized for 250 fF capacitive load ....... 70
Table 6.3.2 Summary of single sided 90 nm CMOS Double Sided CSAL Analog Buffer with
Source Feedback Performance Optimized for 250 fF load ....................................................... 72
Table 6.4.1 Comparison of the proposed DSCSAL with source feedback to other published unity
gain buffers ................................................................................................................................... 73
Table 7.4.1 AIBPF simulation results .................................................................................................. 81
Table 7.4.2 Power consumption of proposed self-calibration system ............................................... 83
Table 8.3.1 Simulation Result of SIDO Mixer Designed for Hartley System .................................. 98
Table 10.3.1 90 Degree Phase Shifter Output Amplitude Difference ............................................. 125
Table 11.2.1 Proposed Receiver Simulation Results ........................................................................ 132
xi
Acknowledgements
advisor Dr. Saiyu Ren who guide and encourage me during my master and PhD study.
Besides my advisor, I would like to thank Dr. Raymond E. Siferd who shares his
Kazimierczuk, and Dr. John M. Emmert for their great academic suggestions and time.
Special thanks to my defense observer, Dr. Travis E. Doom for his kindness and time.
I pursued both master and doctoral degree in Wright State University, and I had a
wonderful time in these 7 years. I am not only learning knowledge in here, but also
growing up and ready for the future career challenges. Many thanks to the Department
of Electrical Engineering staff and students for their encouragement and selfless help.
Thanks to my parents Wei Li and Zirong Yu for their unparalleled love, unconditional
trust, and endless patient. Thanks to my parents-in-law Qilie Zhang and Xiaohong Xu
for their continued support and encouragement. Thanks to my wife Xiaomeng Zhang
for her love and countless sacrifices to help me complete this dissertation. She is always
my best friend, my best research partner, and my most enthusiastic cheerleader in this
amazing journey.
xii
Finally, I would like to extend my gratitude to my grandparents Yongzheng Li and
Shuwei Liu. Even though they passed away before the completion of my doctoral
degree, I know they would be proud of me and I will forever be grateful for the
xiii
I. Introduction
State of the art, the Wireless Local Area Network (WLAN) is widely used in high
speed communication applications [1]. Unlike traditional wired Local Area Network
(LAN) with fixed device locations, the WLAN provides the benefit of mobility to
connected devices. People can carry their devices to anyplace in WLAN signal covered
area and maintain the connection to the network. Also, WLAN is a good solution for
the situation where a large number of devices potentially may connect to the network
at the same time, like shopping mall or coffee house. Moreover, with the explosive
growth of home intelligence products, WLAN becomes the essential technique to share
information and commands to all devices, and the router can be the only equipment that
needs to wire to the internet. Therefore, the receiver design for WLAN technology has
in 1970, and almost 30 years later, in 1997, the first standard of WLAN: 802.11 was
introduced by IEEE. After then, many additional or modified standards are updated to
improve the WLAN performance. The first two standards: 802.11 and 802.11b only use
unlicensed 2.4 GHz frequency band for communication, and later on the 802.11a
standards employed 5 GHz band to increase data transmitting speed [2]. Recently, the
mobile smart devices benefit from the recent modified standards like 802.11g/n/ac with
wider channel width and advanced communication techniques that allow more
1
1.2 Compare 2.4 GHz and 5 GHz
With IEEE 802.11 standard [3], devices can use two frequency channels: 2.4 GHz
and 5 GHz to transmit data. Since the 2.4 GHz frequency band is an unlicensed band
and it is the first frequency band that used on WLAN technology, nearly all the devices
with Wi-Fi function supports this frequency band. As shown in Table 1.2.1, there are
14 channels assigned for 2.4 GHz WLAN operating, and only 3 of them have the non-
overlapping feature (channel 1, 6 and 11) [4]. Therefore, the large amount of equipment
using limited number of frequency channels at 2.4 GHz make this frequency band
usually very crowded and with the possibility of interference between devices. However,
frequency channels shown in Table 1.2.2 to transmit data. With the less occupation
density of each channel, the 5 GHz band has better connection stability due to the
bandwidth (up to 160 MHz with 802.11ac standard) than 2.4 GHz (maximum 40 MHz
with 802.11n standard), and every channel is non-overlapping with its adjacent
channels, so that a single channel at 5 GHz band can carry more information than 2.4
GHz. However, 2.4 GHz frequency band has its own advantage in signal coverage range
due to the less path loss with lower frequency. As described in Eq 1.2.1, when signal
4𝜋𝑑
Loss = 20 log10 (1.2.1)
𝜆
2
Table 1.2.1 2.4 GHz WLAN Channel and Center Frequency [4]
Channel Center Frequency (GHz)
1 2.412
2 2.417
3 2.422
4 2.427
5 2.432
6 2.437
7 2.442
8 2.447
9 2.452
10 2.457
11 2.462
12 2.467
13 2.472
14 2.484
Table 1.2.2 5 GHz WLAN Non-Overlapping Channel and Center Frequency [4]
Channel Center Frequency (GHz)
36 5.18
40 5.19
44 5.22
48 5.24
52 5.26
56 5.28
60 5.3
64 5.32
100 5.5
104 5.52
108 5.54
112 5.56
116 5.58
120 5.6
124 5.62
128 5.64
132 5.66
136 5.68
140 5.7
144 5.72
149 5.745
153 5.765
157 5.785
161 5.805
165 5.825
3
1.3 Benefit of Using 802.11ac Standard
802.11ac standard was released in 2013, and it revealed channels with 80 and 160
MHz bandwidth (for example: channel 50 with center frequency of 5.25 GHz) to
transmit and receive signal. The largest improvement of 802.11ac standard shows in the
maximum data transmission speed and number of connections. Devices working under
the new standard can allow more data transmitting at the same time than the previous
standards due to the large bandwidth. Similarly, if the bandwidth assigned to single user
or device keeps the same as previous standards, the 802.11ac channel accommodate
more users or devices to operate at the same time. Moreover, 802.11ac channels are
able to support 256-QAM (Quadrature Amplitude Modulation) [5], which also leads a
better performance than old techniques. Also, 802.11ac employs MU-MIMO (Multi
User Multi Input Multi Output) technique that provides a large number of access point
for low-configuration Wi-Fi devices (like smart phone) connections. Thus, 802.11ac
standards is good for indoor network that built for personal Wi-Fi equipment and home
intelligence products.
blooming market of mobile smart devices. The key features of mobile IC wireless
communication system design include high performance, low power consumption, and
area efficiency. To realize such demands, CMOS technology is a good solution as it has
been widely used in Very Large Scale Integrated (VLSI) circuit implementation for
several decades with its characteristics of low power, low cost and high density.
In recently published papers, many works are focusing on the receiver system
design compatible with 5 GHz WLAN standard [6] [7] [8] [9]. Most of the designs also
4
provide the capability of using 802.11ac feature and support maximum 160 MHz
channel bandwidth. Typically, the receiver front-end design for WLAN system consists
of amplifiers, filters, and mixers. Based on design specifications, these circuits are
designed with different techniques and architectures. Authors in [6] propose the
receiver design using direct conversion architecture to minimize the image impact
created by local oscillator and reduce the complexity of system implementation. Also,
the current-reuse and subthreshold techniques are employed to achieve ultra-low power
consumption in amplifiers and phase locked loop. In order to address the process
variation of CMOS receiver design, authors in [6] also present a low noise amplifier
circuit with dynamic bias control node for self-calibration. However, the penalty of this
ultra-low power receiver system design is the large on-chip area consumption taken by
multiple passive inductor employed in circuits. As shown in chip micrograph of [6], the
passive inductors take more than 50% of the on-chip space for RF front-end system,
which results in the increased cost and size of the entire design.
In reference [8] and [9] implementations, both works implement the receiver
system with direct conversion architecture. The difference between these two designs
to the one introduced in [6] is that both [8] and [9] employ a 1-to-N transformer to
achieve the amplification function at the first stage of entire receiver design. The
benefits of this architecture are reduced power consumption and increased linearity
compared to the conventional active Low Noise Amplifier (LNA) design, as the passive
component is famous for low noise, low power and good linearity characteristics. The
authors in [9] also add a variable gain amplifier stage between LNA and mixer to adjust
the gain accordingly with the input signal amplitude. This design scheme further
improve the noise and linearity performance as the system can determine the best gain
value based on input signal strength. Moreover, the variable gain stage allows the
5
system to be more tolerant of larger input power range and higher input saturation value.
Nevertheless, these two works in [8] and [9] suffer the same drawback as the system
proposed in [6], which is the large on-chip area used to place passive inductors. The die
photograph of reference [8] indicates that the inductors take more than 80% area of the
These three referenced works all employ the direct conversion architecture to
eliminate image effect. However, the major issues of direct conversion system are the
flicker noise and complex channel selected filter design. As the RF signal is down-
converted into baseband region, the DC current flowing through circuit creates the
flicker noise in active components [10]. The low pass filter after mixer stage needs to
be built with good noise suppression capability and linearity, which increases the design
difficulty.
Besides, all the above reported designs are completed with multiple on-chip
passive inductors which results in large on-chip area consumption and low quality
factor (Q). Therefore, authors in [11] [12] [13] [14] [15] build filters by using active
inductor instead of passive inductor in the system. Such active inductor can offer better
area efficiency along with higher quality factor and post fabrication tunability compared
Even though, the active inductor has many advantages in on-chip CMOS circuit
design, the major issue of this technique is the process variation effect during chip
fabrication. Many factors affect the accuracy of wafer production, such as temperature,
pressure and doping concentrations [16]. Consequently, the electrical properties like
sheet resistance and threshold voltage will be different between transistors, although
they are designed to have exactly same parameters. Such parameters process variation
happens on every element throughout a whole chip and it becomes more and more
6
critical with CMOS technology scaling down [17]. So, the practical result of the active
post fabrication performance, and in many cases, this variation is the key reason of
testing failure. Thus, to successfully replace the passive inductor by active design, a
post fabrication calibration system is needed to compensate the error caused by process
variation.
1.5 Motivation
latest technique to scale down the communication chip size. Meanwhile, customers
always want their equipment to have longer battery life and more powerful performance.
Therefore, the mobile smart device must be equipped with a receiver design that
operates with the newest communication standard with minimum area consumption, to
save space for high capacity battery to extend device operating time. To realize these
parameters, every single circuit block built on-chip should provide high performance
in minimum space.
In receiver chain system, band pass filter is a key component that is designed to
receive the desired signal and filter out unwanted noise. Traditionally, the band pass
filter is realized by passive on-chip L-C circuit. To achieve high quality factor, low
power consumption within small system area, CMOS active inductor based band pass
filter is implemented to replace the passive design and save chip area [11] [18] [19]
[20].
In order to make the active inductor operating with designed performance after
fabrication, a built-in automatic detecting and calibrating circuit is desired for on-chip
7
1.6 Objective
Implement an active inductor-based band pass filter that can replace the
Design a calibration system for active band pass filter with process variation
Implement all sub-circuits that RF receiver chain needed to meet the standard
Build and simulate the receiver system with active band pass filter in CMOS
90 nm technology.
8
II. Receiver System Architecture
The primary function of the receiver system for wireless applications is to extract
the information embedded in the high frequency carrier. Thus, the receiver chain should
provide sufficient gain for the RF input signal to amplify the wanted signal, while
generating minimum noise. Moreover, the system must have good channel selectivity
and image suppression to ensure the data input into DSP block is pure. As shown in Fig
2.1.1, a typical front end receiver includes a RF band pass filter to select the RF input
signal; an LNA to amplify the weak and noisy signal; a mixer to down-convert signal
to baseband or intermediate band which can be easily handled by following DSP blocks;
a local oscillator source (usually made by phase locked loop) to generate desired
frequency signal for carrier removal. This typical receiver design is developed into
9
2.1.1 Image Rejection
Image signal is one of the vital problems of receiver system, and all the referenced
works in background study section of chapter 1 did not report this parameter. The
rejection of this unwanted signal can be done either in analog or digital design phases.
In this dissertation, the proposed receiver system performs the image rejection feature
with analog circuit design to reduce the design complexity of future digital circuit
blocks.
Fig 2.1.2 demonstrates the issue existing in a receiver system, where fRF is desired
radio frequency input signal frequency, fLO is local oscillator frequency, fIF is called
intermediate frequency and equals to | fRF - fLO|, fIM is the image signal and its frequency
is fIM = | fRF - 2fLO|. Such unwanted image signal can pass through the RF input filter
and overlap the fIF causing degradation of wanted signal quality. To solve this issue,
three different types of analog image rejection receiver architecture are widely used,
10
Fig 2.1.2 Image issue existing in receiver system [10]
For Heterodyne receiver, an image reject filter is placed before mixer to eliminate
the image signal and the system structure is shown in Fig 2.1.3 [21]. The benefit of this
design is low complexity of the entire system. However, the drawbacks are selectivity
and sensitivity of the channel select filter. The choice of fIF is important. Lower fIF needs
high selectivity for filter, while higher fIF increases design complexity of following
circuitry. To solve the problem, more IF steps are needed with the help of off-chip filters
11
Fig 2.1.3 Heterodyne receiver architecture [21]
technique to clear the image signal. Theoretically, when considering both image and
Where VRFin, VRF, and VIM represent the amplitude of total input, wanted input, and
image input, respectively. The ωRF and ωIM describe the angular velocity of wanted and
image signal. After applying quadrature mixing, the down-converted signal is divided
𝑉𝑅𝐹 𝑉𝐼𝑀
𝑉𝑄 (𝑡) = cos(𝜔𝑅𝐹 − 𝜔𝐿𝑂 )𝑡 + cos(𝜔𝐿𝑂 − 𝜔𝐼𝑀 )𝑡 (2.1.3)
2 2
Subtracting VI(t)’ from VQ(t) and the calibrated IF signal equals to:
12
𝑉𝐼𝐹𝑜𝑢𝑡 (𝑡) = 𝑉𝑅𝐹 cos(𝜔𝑅𝐹 − 𝜔𝐿𝑂 )𝑡 (2.1.5)
without the effects of image problem. However, this approach adds the gain mismatch.
If the two paths are mismatched, the image signal cannot be perfectly canceled at IF
presented in Fig 2.1.5 [24]. The first stage mixer outputs VA(t) and VB(t) are the same
13
Fig 2.1.5 Weaver image rejection architecture [24]
𝑉𝑅𝐹 𝑉𝐼𝑀
𝑉𝐵 (𝑡) = 𝑉𝑄 (𝑡) = cos(𝜔𝐼𝐹1 )𝑡 + cos(𝜔𝐼𝑀 )𝑡 (2.1.7)
2 2
After the second stage mixer, the two IF signals VC(t) and VD(t) are shifted to:
𝑉𝑅𝐹 𝑉𝐼𝑀
𝑉𝐶 (𝑡) = 𝑉𝐴 (𝑡) × sin 𝜔𝐿𝑂2 𝑡 = − cos(𝜔𝐼𝐹 )𝑡 + cos(𝜔𝐼𝑀 )𝑡 (2.1.8)
4 4
𝑉𝑅𝐹 𝑉𝐼𝑀
𝑉𝐷 (𝑡) = 𝑉𝐵 (𝑡) × cos 𝜔𝐿𝑂2 𝑡 = cos(𝜔𝐼𝐹 )𝑡 + cos(𝜔𝐼𝑀 )𝑡 (2.1.9)
4 4
And the image signal is eliminated at the output of the subtractor of VD(t) minus
In this dissertation, with the consideration of noise and power consumption, a low
selected for signal processing to meet the specification of 802.11ac standard. The data
path shown in Fig 2.1.6 is realized with LNA, band pass filter (also have image rejection
system is also built and simulated to compare the performance with Hartley structure.
14
Fig 2.1.6 Proposed receiver design combined Heterodyne and Hartley
architecture.
The main specifications of the proposed receiver system include input sensitivity,
image rejection ratio, noise figure, and power consumption. With the requirement of
802.11ac standard, the minimum input sensitivity (input level) is -73 dBm, and the
maximum is -30 dBm [3]. Based on power and voltage converting formula listed in Eq
2.2.1 for 50 Ω system, these number are equivalent to 70.8 uV and 10 mV in amplitude
unit, respectively.
𝑃(𝑑𝐵𝑚)−10
𝑉𝑎𝑚𝑝 = 10 20 (2.2.1)
Where P represent the input power and Vamp is the equivalent amplitude. Such RF
signal needs to be processed and convert into large amplitude and low frequency IF
dissertation, a 9-bit ADC [25] is assumed to connect with proposed receiver chain.
15
Theoretically, the maximum ADC input signal amplitude is calculated as:
𝑉𝑑𝑑 −2∗𝑉𝑇
𝑉𝑎𝑚𝑝𝑚𝑎𝑥 = (2.2.2)
2
estimated as 0.3 V using Eq 2.2.2. The LSB amplitude value of referred ADC is derived
as:
𝑉𝑎𝑚𝑝𝑚𝑎𝑥
𝑉𝐿𝑆𝐵 = (2.2.3)
2𝑁 −1
Where, N is the number of ADC bits which is 9 in the referenced work. The LSB
amplitude is computed as 587 µV. In order to achieve ADC 9 bits input amplitude
requirement with largest power RF input signal, and at least 2 bits with minimum input
node. Moreover, the amplitude of signal injected into mixer RF input node must be
large enough (in mV level) to make circuit work functionally. Therefore, the first two
stages amplifier should offer sufficient gain to amplify the weak RF input signal.
The noise performance is another parameter that is very important for front end
design. Eq 2.2.4 describes the noise effects on the receivers based on cascaded amplifier:
𝐹2 −1 𝐹 −1 𝐹𝑁 −1
𝐹 = 𝐹1 + + 𝐺3∗𝐺 + ⋯ + 𝐺 [26] (2.2.4)
𝐺1 1 2 1 ∗𝐺2 ∗…∗𝐺𝑁−1
Noted that, in Eq 2.2.4, F is the noise factor of entire receiver chain. FN and GN
are the noise factor and gain of Nth stage circuit, respectively. It can be seen that, if F1
is small and G1 is large, the first stage circuit dominates the entire system noise
performance.
In summary, to ensure the data integrity and minimal noise impact generated from
receiver system, the expected gain and noise distribution is presented in Fig 2.2.1. The
16
first two stages circuit should provide 30 to 40 dB gain and generate no more than 3 dB
noise. Besides LNA is expected to have around 20 dB gain and less than 2 dB noise
figure to guarantee the entire chain having good noise performance based on Eq 2.2.4.
Since most mixer circuits are gain-loss circuit, an amplifier stage is needed to be built-
entire receiver, the loss of mixer stage must be within -10 dB to 0 dB.
17
III. Low Noise Amplifier
3.1 Introduction
As discussed in Chapter 2, usually a Low Noise Amplifier (LNA) is the first circuit
block of a receiver chain. It amplifies the signal captured by antennas and minimizes
the noise effects on the receivers. Since LNA design is so critical to the entire front-end
degradation of active circuit. In this dissertation, a classic passive inductor based single
ended tuned LNA shown in Fig 3.1.1 is employed with cascode and source de-
generation techniques.
series to build the pre-select filter which can provide the lowest input power reflection
(S11) at targeting frequency. To amplify the signal and reduce the input capacitance
and M2 transistors. The output node inductor and varactor form a band pass filter whose
resonating frequency can be tuned with bias voltage to improve the circuit gain,
suppress unwanted noise, and calibrate the center frequency if LNA suffers the process
sections. The goal of this design is to maximize the gain of amplifier at desired center
Since LNA is connected to off-chip circuit, the input node equivalent impedance
should equal to general standard 50 Ω. Fig 3.2.1 demonstrates the simplified small
18
signal model of LNA input consisted of CB, LG, source inductor (LS), and parasitic
19
In small signal model, CiM is generated by input Miller effect and it can be
calculated as:
It is noted that Av1 is the gain of first transistor and it approximately equals to -2
when M1 and M2 are the same size [27]. Cgd is the gate to drain capacitance of M1
transistor. Substitute Av1 value into Eq 3.2.1, and the Miller effect capacitance is:
𝐶𝑖 = 3𝐶𝑔𝑑 (3.2.2)
The equivalent input impedance based on the small signal equivalent circuit is
expressed as:
1 1
𝑣𝑖𝑛 = (𝑠𝐿𝐺 + 𝑠𝐶 + 𝑠𝐶 ) ∗ 𝑖𝑖𝑛 + 𝑠𝐿𝑆 (𝑖𝑖𝑛 + 𝑔𝑚1 𝑣𝑔𝑠1 ) (3.2.3)
𝐵 𝑔𝑠𝑒𝑞
Where,
𝑖
𝑣𝑔𝑠1 = 𝑠𝐶 𝑖𝑛 (3.2.5)
𝑔𝑠𝑒𝑞
1 1 𝑔
𝑣𝑖𝑛 = (𝑠𝐿𝐺 + 𝑠𝐶 + 𝑠𝐶 ) ∗ 𝑖𝑖𝑛 + 𝑠𝐿𝑆 (1 + 𝑠𝐶 𝑚1 ) ∗ 𝑖𝑖𝑛 (3.2.6)
𝐵 𝑔𝑠𝑒𝑞 𝑔𝑠𝑒𝑞
Through Eq 3.2.6, the input equivalent impedance of LNA can be calculated as:
𝑣𝑖𝑛 1 1 𝑔
𝑍𝑖𝑛 = = 𝑠𝐿𝐺 + 𝑠𝐶 + 𝑠𝐶 + 𝑠𝐿𝑆 (1 + 𝑠𝐶 𝑚1 ) (3.2.7)
𝑖𝑖𝑛 𝐵 𝑔𝑠𝑒𝑞 𝑔𝑠𝑒𝑞
1 1
𝑠(𝐿𝐺 + 𝐿𝑆 ) + 𝑠𝐶 + 𝑠𝐶 =0 (3.2.9)
𝐵 𝑔𝑠𝑒𝑞
In this chapter, the following parameters are used for capacitance estimation:
20
𝑓𝐹
CGSO = 0.252 𝜇𝑚
𝑓𝐹
CGDO = 0.252 𝜇𝑚
𝑓𝐹
𝐶𝑜𝑥 = 12.33 𝜇𝑚2
And,
2
𝐶𝑔𝑠𝑒𝑞 = 𝐶𝑔𝑠 + 3𝐶𝑔𝑑 = 3 𝐶𝑜𝑥 ∗ 𝑊 ∗ 𝐿 + CGSO ∗ 2 ∗ 𝑊 + 3 ∗ CGDO ∗ 𝑊 (3.2.10)
Substituting all the constant into Eq 3.2.10, the Cgseq is estimated as 265 fF when
width and length of M1 transistor equal to 100 μm and 200 nm respectively. The
transconductance measured from proposed LNA input transistor is 3657.66 μA/V under
DC operating point. Plug Cgseq and gm1 into Eq 3.2.8, source inductor LS is estimated as
For the imaginary part, s is substituted by jω0, and ω0 represents angular frequency
converted from designed center frequency of 5.25 GHz. Eq 3.2.9 is expressed as,
1 1
(𝐿𝐺 + 𝐿𝑆 )𝜔0 = +𝜔 (3.2.11)
𝜔 0 𝐶𝐵 0 𝐶𝑔𝑠𝑒𝑞
1 𝐶𝐵 +𝐶𝑔𝑠𝑒𝑞
𝐿𝐺 + 𝐿𝑆 = 𝜔 2 ( 𝐶 ) (3.2.12)
0 𝐵 𝐶𝑔𝑠𝑒𝑞
3.84 nH based on Eq 3.2.13. The value of LS has already been estimated as 3.62 nH, so
The input impedance matching has a huge impact to LNA noise performance [28].
In this model, the gate and source inductors are placed and calculated to create 50 ohms
21
resistance matching with off-chip load. Matching the off chip source resistance
On the output node, the drain inductance can be estimated by the operating
And,
This RF band pass filter at LNA output establishes a RF input bandwidth and
Assume M1 and M2 transistor in Fig 3.1.1 have the same size. The two parasitic
capacitance Cdb2 and Cgd2 are 334.75 fF and 25.2 fF, respectively. Substitute these two
number and 100 fF load capacitance (include Cload and varactor) into Eq 3.2.15, the
output node total capacitance is 459.2 fF. Plug the Ctotal into Eq 3.2.14, the drain
output stage band pass filter can filter out the noise out of the wanted RF signal band
and also reduce the image signal impact on the entire receiver.
The above theoretical analysis describes all LNA components initial parameters
that suit for 802.11ac specification. In the real circuit design, those numbers are adjusted
power consumption due to the low supply voltage. However, the shrinking supply
voltage range also limits the output signal swing especially for cascoding design shown
22
in Fig 3.3.1. If both transistors have the same threshold voltage VT, the output maximum
amplitude Voutamp is
𝑉𝐷𝐷 −2𝑉𝑇
𝑉𝑜𝑢𝑡𝑎𝑚𝑝 = (3.3.1)
2
Typically, the supply voltage of 90 nm is 1.2 V and threshold voltage for NMOS
is approximately 0.5 V (strong on) as shown in Fig 3.3.2. Put these two numbers into
Eq 3.3.1, the maximum output amplitude is only 0.1 V. In order to increase the voltage
range at output node, low threshold voltage MOSFET is used in this design to mitigate
stacking threshold issue. Based on Fig 3.3.2 simulation result, the low threshold voltage
NMOS (lvtnfet) can be conducted at 0.4 V (strong on). Substitute this threshold voltage
into Eq 3.3.1 along with 1.2V supply voltage, the output signal amplitude can reach 0.2
23
Fig 3.3.2 Threshold voltage measurements of normal NMOS and low threshold
NMOS
3.3.2 Varactor
MOS varactor is widely applied on voltage control system as its capacitance can
be tuned by gate voltage. In 90 nm process, the offered NMOS type varactor (NCAP)
has a bias tuning range from -0.5 V to 1 V. The NCAP device is built with “NFET in n-
well” structure as shown in Fig 3.3.3. It is formed by thin gate-oxide over n-well, and
well region.
To verify the adjustable capacitance properties of NCAP, a testing band pass filter
circuit is built and simulated as demonstrated in Fig 3.3.4. The center frequency of this
filter is decided only by inductor and varactor value. In the simulation setup, the
inductance and varactor geometry size are fixed, and only use Vbais to control the band
pass filter center frequency. Simulation results presented in Fig 3.3.5 express the center
24
frequency changing under different Vbias values. It can be seen that when bias voltage
equals to -0.5 V, the filter has the highest center frequency of 5.87 GHz because of the
smallest capacitance generated at output node by varactor. When bias voltage change
to 1 V, varactor creates the largest capacitance which results in the lowest center
frequency of filter.
Fig 3.3.4 Schematic diagram of band pass filter built with inductor and varactor
25
Fig 3.3.5 Center frequency of band pass filter under different gate voltages of
varactor
The cascode source de-generation LNA used in this dissertation is built and
in Fig 3.4.1, and the load capacitance (Cload + Cvaractor) is set to be 100 fF as previous
analysis. Using the theoretical parameter that calculated in previous sections, the LNA
employed in this dissertation provides center frequency of 5.25 GHz as shown in Fig
3.4.2. The gain at center frequency is measured as 25.24 dB, and the 3 dB down
bandwidth is 717.65 MHz. Since the major task of LNA is suppressing the system noise,
the gain and bandwidth shown on simulation result is sufficient to amplify the wanted
signal within relatively narrow frequency band and attenuate unwanted signal.
26
Fig 3.4.1 Schematic diagram of proposed LNA design.
Fig 3.4.2 Simulation results of center frequency, gain, and 3 dB down bandwidth
27
Quality factor (Q) is calculated as:
𝑓𝐶
𝑄=𝑓 (3.4.1)
−3𝑑𝐵
circuit. Substitute the simulation result into Eq 3.4.1, the Q of designed LNA is 7.32.
This low Q value of implemented circuit is coming from the passive inductor poor
quality factor property. As discussed in section 3.1, the passive components have the
advantages of low operating noise and non-sensitive to PVT variation, which are the
key properties for LNA design. Therefore, to compensate the sacrificed quality factor
and gain, an active inductor-based band pass filter is added in the following stages
Fig 3.4.3 LNA simulation results of -0.5 V and 1 V varactor gate voltages
28
Another introduced feature of designed LNA circuit is the capability of adjusting
by changing the gate voltage of varactor, the total output capacitance is changed. Thus,
the center frequency of LNA can be tuned even after fabrication. In Fig 3.4.3, when
gate voltage of varactor set to -0.5 V, the output node capacitance is minimum, and
LNA has the highest center frequency of 5.32 GHz. When gate voltage equals to -0.5
V, the maximum output capacitance results in the lowest center frequency of 5.13 GHz.
Such frequency difference is large enough for designer to use this LNA on multi-
To evaluate the input impedance matching quality of designed LNA, the Scattering
is widely used to study the two ports network power waves of reflection and incidence.
S11 parameter describe the input port voltage reflection coefficient, and smaller value
of this number means more power can be delivered to output. Moreover, a good
impedance matching circuit is also a pre-filter stage as introduced in section 3.1. Only
the signal within selected frequency band is allowed to enter the system and attenuate
other noise signal strength including image signal. Fig 3.4.4 contains the S11 plot of
designed LNA input matching with 50 ohms resistance as requested by system. At 5.25
GHz center frequency point, the measured S11 equals to -23.13 dB, and it indicates that
a good matching and power transferring performance of this LNA when connected with
The Noise Figure (NF) is used to measure the degradation of Signal-to-Noise Ratio
(SNR) of designed system. A lower value of this parameter indicates better performance
of LNA. In Fig 3.4.5, the designed LNA provides 2.1 dB NF value at center frequency
29
of 5.25 GHz. Since the gain presented in previous simulation result is 25.24 dB, this
number is sufficient to assure a good noise performance of the entire receiver system.
30
1 dB compression point simulation results are used to evaluate the linearity of
testing system. Ideally, for a linear system, the output changing follows the input
varying with a constant ratio. However, in practical circuit design, large input signal
power causes the saturation of system and the output is no longer linearly following
input power increment. Therefore, the point of input power that leads the output
dropping 1dB from its expected gain is the 1 dB compression point. Fig 3.4.6 shows
the designed LNA has 1 dB compression point of -7.12 dBm at selected channel center
Another parameter that widely used to describe the system linearity is Input
Inferred Third order Intercept Point (IIP3). The third order signals are generated by
inputting two fundamental signals into system with frequency close to each other within
targeting band. IIP3 value is the point that the fundamental signals and third order
signals are all at the same power level. The IIP3 value of proposed LNA circuit is shown
in Fig 3.3.11 as -1.54 dBm with the third order signal located at 5.2 GHz.
31
Fig 3.4.7 IIP3 simulation result of LNA
3.5 Conclusion
In this chapter, a cascode source de-generation low noise amplifier is designed and
simulated with 90 nm technology. As the first circuit block of receiver system, this LNA
has a good input matching property with –23.13 dB S11 results at designed center
frequency of 5.25 GHz. The 3 dB down bandwidth is 717.65 MHz, so all wanted signal
reserved and amplified by LNA, and out of band noise is attenuated and filtered out.
The gain and noise figure of designed LNA is 25.24 dB and 2.1 dB at center frequency,
respectively. Such simulation results lay a good foundation of entire front end receiver
noise performance. The power consumption of this stage circuit is 2.44 mW with input
32
IV. Active Inductor Band Pass Filter
(The discussion in the following chapter is substantially drawn from [29], where we first reported the development and
4.1 Introduction
variation, passive inductors are employed with varactor to implement the first stage
circuit of receiver system. The tradeoff of this design is large inductor area and the poor
quality factor of the output band pass filter which must be improved by other circuitry
in the receiver. The solution provided in this dissertation is placing an Active Inductor-
based Band Pass Filter (AIBPF) right after LNA stage to improve the signal quality and
system performance. In conventional design, the passive band pass filter is one of the
most common circuit block in electric circuit design as it has many advantages like low
power consumption, low noise, strong tolerance of large current [30], and high
passive inductor-based band pass filter has some critical weakness, such as passive
inductor filter lacking wide range tunability [31] and complicated high order filter
design being very time-consuming and difficult. The largest drawback is the extremely
large on-chip passive inductor size. Thus, active CMOS inductor becomes more and
more attractive in recent years. An active inductor only takes 1-10% the area of a
passive inductor with same inductance [32]. The AIBPF offers a wider frequency tuning
range by adjusting the bias voltage in the circuit. With only one stage design, the AIBPF
can provide higher gain and quality factor (Q) than traditional passive design, and
potentially can reach even higher value with multiple AIBPFs in cascaded. The
adjustable gain feature of AIBPF is also good for on-chip circuit design, as the active
filter can be easily converted into VGA with little modification [15].
33
4.2 Area Consumption of Passive Filters
The conventional band pass filter schematic and frequency response graph is
presented in Fig 4.2.1, and the transfer function of this 2nd order filter can be developed
1
𝑣𝑜𝑢𝑡 𝑠𝐿∗
𝐻(𝑠) = = 1
𝑠𝐶
1 (4.2.2)
𝑣𝑖𝑛 (𝑠𝐿+ )𝑅+𝑠𝐿∗
𝑠𝐶 𝑠𝐶
capacitive components in circuit getting larger. So, for band pass filter design, the
general function is
Eq 4.2.4 indicates that higher order design can give designer more space to
However, the tradeoff of building high order band pass filter is dramatically
increasing design complexity and area consumption. For System on Chip (SoC)
application, with the technology process scaling down, the unit area cost increasing
rapidly. Passive components especially inductor, take huge on-chip design area
comparing with active components like transistor. In recent years, this issue draws more
attention as the latest wireless systems with most advanced technology are pursuing
34
Fig 4.2.1 Schematic and frequency response figure of conventional band pass
filter: (a) Schematic (b) Bode plot
are proposed, and Gyrator-C network is one of the most popular active circuits of
1948 [34]. It describes a two-port device containing two transconductors and one output
capacitor to achieve the inductor function as shown in Fig 4.3.1 [35]. The equivalent
35
𝐶
𝐿𝑒𝑞 = (𝐺 (4.3.1)
𝑀1 ∗𝐺𝑀2 )
transconductance GM1 and M2 to transconductance GM2. M3, M4, and M5 with bias
inputs together with IB control the quiescent values of GM1, GM2, and GM3. The detailed
small signal analysis for this active inductor is developed in [15] resulting in
𝑍𝑖𝑛 = 𝑅𝐿 + 𝑠𝐿 (4.3.2)
In Eq 4.3.2,
𝐶
𝐿=
𝐵𝐺𝑀1 𝐺𝑀2
𝐺 − 𝐵𝑔𝑑3
𝑅𝐿 ≈
𝐵𝐺𝑀1 𝐺𝑀2
𝐺𝑀3
𝐵≈
𝑔𝑑3 + 𝑔𝑑4
𝐺 ≈ 𝐺𝑀2 + 𝐺𝑀3
And gd3, gd4 are the drain conductance of M3 and M4. The result is a value of L
controllable by GM1, GM2, GM3, and C together with a small series resistor (RL)
controllable by GM1, GM2, GM3. It is noted that M3 provides another degree of freedom
(GM3) together with GM1 and GM2 to control the value of L and Q of the active inductor.
producing transconductance GM0 and channel resistance rds0 into Fig 4.3.3 to control
voltage gain. The small signal equivalent circuit for the AIBPF including a load
36
Fig 4.3.2 Schematic of active inductor
37
Fig 4.3.4 Small signal model of AIBPF
In Fig 4.3.4, Leq and RP are the inductance and parallel loss resistance of active
(𝜔0 𝐿𝑒𝑞 )2
𝑅𝑃 = (4.3.3)
𝑅𝐿
𝑅𝑃
𝑅𝑒𝑓𝑓 = 𝜔 (4.3.4)
0 𝐿𝑒𝑞
So, the center frequency (fC) of this Gm-C network band pass filter is
1 𝐺 𝐺 𝐺
𝑓𝐶 = 2𝜋 = √4𝜋2 (𝑔𝑀1 𝑔𝑀2 𝐶𝑀3 (4.3.5)
√𝐿𝑒𝑞 𝐶𝑙𝑜𝑎𝑑 𝑑3 𝑑4 𝑙𝑜𝑎𝑑 𝐶)
𝑅𝑒𝑓𝑓
𝑄= (4.3.6)
𝜔0 𝐿𝑒𝑞
As seen in Eq 4.3.5, the center frequency is tunable by varying GM1 and GM2, and
GM3 with the transconductance is proportional to the quiescent bias current and
transistor gate voltage; so, post fabrication center frequency tune-ability is possible with
38
4.4 Simulation Result
In this dissertation, all the circuit constructions and simulations are performed by
Cadence using 90 nm technology. There are two methods used to analyze the process
variation: corner analysis and Monte Carlo analysis. Corner analysis is simulating the
different running speed of circuits assuming all transistors are fabricated at corner
process conditions. For both PMOS and NMOS transistors, each of them has three
corners and they are: Slow (S), Typical (T), and Fast (F). Such corners usually indicate
the transistor speed changing based on PVT effect. For example, the slow corner
reflects the impact of low operating voltage and high environment temperature. The
corner analysis is used to estimate the circuit performance under extreme PVT variation.
Thus, the corner analysis is always used as coarse evaluation of system PVT tolerance.
component mismatch. It uses more complicate mathematic model to simulate the PVT
variation and lets designer do detailed analysis of circuit performance under different
variation. This method is good for the design that has already passed the corner analysis
Fig 4.4.1 presents the schematic diagram of proposed AIBPF, and the frequency
response plot of this circuit is shown in Fig 4.4.2 with center frequency adjusted to 5.25
GHz. The gain of this system at center frequency is 28.8 dB and 3 dB down frequency
vulnerable to process variations and component mismatch during fabrication. Fig 4.4.3
shows the frequency response of AIBPF with the same design parameter but different
process corner. AIBPF produces a lower center frequency under slow-slow corner and
40
Fig 4.4.3 AIBPF frequency response under different process corner
out of 3 dB down bandwidth caused by process variation is typically more than 70%
based on simulation results in 90 nm CMOS technology when AIBPF is built for high
in Fig 4.4.4 is based on the results of 50 Monte Carlo simulations. The Monte Carlo
results show that only 8 of the 50 Monte Carlo results (16%) have the center frequency
within the desired 3 dB bandwidth. This corresponds to 84% of the post fabrications
filters needing calibration to reset the center frequency within the desired 3 dB
bandwidth.
post-fab calibration system is needed to correct center frequency changes and increase
the yield of integrated circuits incorporating AIBPFs. The calibration system should be
able to capture the center frequency error, analyze variation type, and compensate
41
process variation automatically. In next 2 chapters, two key components: amplitude
detector and analog buffer are introduced to be part of the self-calibration system. The
CMOS amplitude detector is used to realize the error acquisition function. The analog
buffer is placed inside the system to help increase the driving ability of on-chip circuit
and isolate adjacent circuit blocks. The completed AIBPF self-calibration system is
Fig 4.4.4 Process variation effect on AIBPF (simulated with Monte Carlo
Analysis)
42
V. CMOS Amplitude Peak Detector
(The discussion in the following chapter is substantially drawn from [37], where we first reported the development and
5.1 Introduction
technology, active inductor has been applied in some SoC designs especially for band
pass filter application to save area [15]. Nevertheless, the fatal drawback of this solution
is process variation which results in center frequency and gain changing. To fix such
circuit block [38]. A wide output range CMOS amplitude peak detector becomes an
important integral part of calibration stage to realize the real-time monitoring function.
In this dissertation, a unique simple peak detector reported in [39] is used and
The schematic diagram of the proposed CMOS peak detector is depicted in Fig
3.2.1, which consists of two stages. First stage is referred from [39] to detect input
As shown in Fig 5.2.1, Ibias is the current source generated by transistor M2. I1 is
the current flowing through Rfeedback, essentially M1 transistor gate leakage current,
which is very small. Thus, the voltage drop across Rfeedback is quite small. First stage
input amplitude increases, to have M1 current IM1 balanced with Ibias, the gate/drain
voltage of M1 must be reduced to adjust the current. Therefore, the drain voltage of M1
43
(V1) can successfully express the input amplitude changing, and V1 DC voltage
However, V1 is also the DC voltage for M1 gate, and theoretically the value of V1
must be greater than threshold voltage of M1 to keep M1 on. To satisfy such voltage
limitation, the first stage output V1 varying range becomes smaller. Refer to the data of
[39], by using only first stage circuit, the output DC voltage range is 400 mV to 50 mV
supply voltage. It appears M1 is operating in cutoff region for some cases unless the
blocking capacitor C1 is small and the input DC voltage couples to M1 gate directly.
From the data, it can also be seen that the detection output reflects the input amplitude
inversely, and the ratio between maximum input peak to peak value and supply voltage
is 0.56, which is not suitable for designs require large bias voltage adjustment range.
44
The objective of this paper is to enlarge the peak detector output range and have the
To flip the detection results generated from first stage, an active load amplifier
(M3 and M4 transistors), is added as second stage circuit shown in Fig 5.2.1. The output
Where,
1 1
𝑟𝑜𝑢𝑡2 = (𝑔 ≈𝑔 (5.2.2)
𝑚3 +𝑔𝑑𝑠3 +𝑔𝑑𝑠4 ) 𝑚3
Since gds3 and gds4 are much smaller than gm3, rout2 is inversely proportional to
gm3. In order to have large rout2 at vout node, M3 must have a small width, as the trans-
conductance of transistor is proportional to its width [26]. At the same time, enlarging
the width of M4 transistor can increase the trans-conductance value, gm4 in Eq 5.2.1.
With the transistors width setup based on above analysis, the second stage active load
amplifier enhances the detection result range, also inverts the DC output trend from
Output coupling jitter (also called AC gain) is very important for peak detector
design. As the expected detection result is a DC voltage, and the jitter coupled on output
must be as small as possible. Ideally, the jitter gain of first stage circuit is zero, so that
the output signal of peak detector is a perfect DC signal, and next stage circuit result
can be more accurate. To analyze jitter performance of this peak detector, a small signal
equivalent circuit is employed in Fig 5.2.2 by assuming C2 (1 pF) is much larger than
gate capacitance of M4 and parasitic capacitance of M1, M2. The goal is to find the
45
lowest jitter gain with respect to Rfeedback to reduce the jitter amplitude coupling on
i2 is the AC current flowing through rout1, which equals to rds1 in parallel with rds2.
(𝑣𝑖𝑛 −𝑣1 )
𝑖1 = 𝑅 (5.2.5)
𝑓𝑒𝑒𝑑𝑏𝑎𝑐𝑘
(𝑣𝑖𝑛 −𝑣1 )
𝑖2 = 𝑔𝑚1 𝑣𝑔𝑠1 − 𝑅 (5.2.6)
𝑓𝑒𝑒𝑑𝑏𝑎𝑐𝑘
46
In order to minimize the jitter amplitude at Vout node, the jitter gain of first stage
according to Eq 3.2.9. Based on first stage simulation results, the maximum value for
gm1 is 5133.3 μA/V2, and the minimum value is around 1426.8 μA/V2 on the basis of
transistor width of M1, 18 μm and M2, 1.5 μm. Thus, the resistance range of feedback
resistor is within 195 Ω and 700 Ω. Rfeedback is also used to block the AC current flowing
The second stage active load amplifier circuit is also designed to have low
frequency equation:
1
𝑓−3𝑑𝐵 = 2𝜋∗𝑟 (5.2.10)
𝑜𝑢𝑡 ∗𝐶𝐿
Therefore, when rout2 getting larger, the circuit has smaller 3 dB down frequency.
Referring to the analysis results from section 5.2.1, M3 transistor size need to be small
to have large equivalent resistance. So that, with the reduced M3 transistor size, the 3
dB down frequency value of the second stage circuit decreased, and the active load
amplifier circuit can further weaken the coupling jitter amplitude at Vout node. In
conclusion, in order to enlarge the voltage difference between every two adjacent DC
detection results and minimize the coupling jitter amplitude on DC output, the M3
47
5.3 Layout Simulation Results
Based on section 5.2.1 and 5.2.2 analysis, the proposed CMOS peak detector is
implemented in CMOS 90 nm technology with layout shown in Fig 5.3.1. To verify the
variable that sweeping from 195 Ω to 800 Ω. Input signal has 100 mV amplitude and
6.0 GHz frequency. Fig 5.3.2 contains the output jitter results with Rfeedback of 195 Ω,
580 Ω, and 700 Ω, and it shows the proposed peak detector having the minimum output
jitter amplitude when feedback resistance equals to 580 Ω. Fig 5.3.3 summarizes the
48
jitter amplitude changing with Rfeedback changing step size of 10 Ω between 195 Ω and
800 Ω. Through the graph, 580 Ω feedback resistor gives the minimum jitter amplitude,
72.35 µV. However, when Rfeedback is within 550 Ω and 650 Ω range, the jitter amplitude
is near the same. Based on output jitter results, in this paper, Rfeedback is set to be 580 Ω,
Fig 5.3.2 Jitter results with 195 ohms, 580 ohms, and 700 ohms feedback
resistance
49
Output Jitter Magnitude (uV)
700
600
500
400
300
200
100
0
195 200 300 400 500 510 520 530 540 550 560 570 580 590 600 650 700 750 800
Feedback Resistance (Ohoms)
Fig 5.3.3 Output jitter amplitude data plot of feedback resistance sweeping from
195 ohms to 800 ohms with fine simulation between 500 ohms and 600 ohms.
Fig 5.3.4 DC detection simulation result of input amplitude sweeping from 0.1 V
to 0.5 V at 6.0 GHz frequency.
50
Layout simulation results in Fig 5.3.4 show the proposed peak detector can well
distinguish different input amplitude with 1.0 pF capacitive load and 6.0 GHz input
frequency. As seen in Fig 5.3.4, with 500 mV input amplitude, the peak detector output
is 680 mV. The ratio between maximum input peak to peak value and supply voltage is
0.83, and that means the capability of input amplitude acceptance for proposed design
is better than the design in [39], whose ratio equals to 0.56. The largest output difference
between two adjacent 100 mV input amplitude (200 mV to 300 mV) is about 186 mV,
and the smallest is 106 mV while the input amplitude changing from 400 mV to 500
mV. Such large output range makes next stage circuit recognize the active circuit
working status easily. If PVT variation happens, the peak detector’s output DC value
changes to reflect the error. Then, the calibration circuit can be activated by this error,
Table 5.3.1 summarizes the performance comparison of this proposed design with
previous reported designs. It can be seen that the proposed design is very competitive
compared with other peak detector designs. The operating frequency is the highest and
51
Low power benefits from more advanced technology, and small size of current
source transistor in first stage that limits the current. Cadence layout simulation results
show this proposed peak detector can detect input amplitude of 100 mV to 500 mV
5.4 Conclusion
The CMOS peak detector introduced in this chapter can be used for on-chip self-
calibration application. The proposed peak detector can successfully detect different
range is 1.0 GHz to 10.0 GHz with 1 pF capacitive load. The power consumption is 0.4
mW with input amplitude of 500 mV and frequency of 6.0 GHz and power supply of
1.2 V. The low power cost, high input frequency range and wide output voltage range
make this proposed CMOS peak detector is suitable for most on-chip applications.
52
VI. Analog Buffer
(The discussion in the following chapter is substantially drawn from [41], where we first reported the development and
6.1 Introduction
The unity gain analog buffer is a key component for RF mixed signal integrated
circuit design. The task of the unity gain analog buffer is to drive a relatively large on
chip capacitive load (CL) over a wide frequency range with near unity gain while
presenting a small input capacitance (Cin) to the previous stage. Also, the buffer must
be able to isolate two stages circuit and prevent the interference from one to another.
For high frequency operation, one objective is to maximize the 3dB bandwidth for
a specified ratio CL/Cin. The required value of Cin and the ratio of CL/Cin varies
application may typically require the on-chip buffer to present a Cin value of
should be noted that the effective output load Cout includes the load capacitance CL and
the self-loading capacitance on the output node of the buffer (CSL). Linearity is another
important requirement for an on chip buffer since it must be capable of operating with
include maximum input amplitude, variation of gain over the range of input amplitudes
(gain accuracy), total harmonic distortion (THD), low frequency gain, and offset error
buffer design based on a range of input magnitudes where the variation in gain is less
than 2%. The tolerance for gain variation varies for a particular application, so
53
6.1.1 Source Follower Based Unity Gain Buffers
The source follower has been the classical circuit for implementing a unity gain
buffer [27]. Since the source is the output node, the device threshold varies due to the
body effect as the input increases resulting in relatively poor linearity. The body effect
can be alleviated by cascading NMOS and PMOS source follower circuits and/or
connecting the source to bulk in an isolated well; however, gain accuracy (linearity),
offset, and loss of gain are usually not sufficient for many applications. Several
modifications have been made to the basic source follower circuit to reduce offset and
increase linearity. Negative feedback has been introduced in various ways to increase
gain accuracy [42] [43] [44]. These designs, which include a combination of a feedback
amplifier and source follower improve gain accuracy but suffer from reduced
with reduced value of Vdd. Also, the introduction of a high gain amplifier requires
assessment of stability with the addition of a compensation capacitor in most cases and
with the source follower buffer is the tradeoff between source follower transistor size
and 3dB bandwidth. The output resistance of the source follower is proportional to
proportional to transistor size for a given value of overdrive (VGS–VT). Reducing the
output resistance by increasing the source follower transistor size has the detrimental
effect of increasing the value of Cin and CSL of the buffer resulting in a relatively low
54
6.1.2 Source Coupled Differential Pair Based Unity Gain Buffer
differential pair as shown in Fig 6.1.1 [45] [46]. The operation of the differential pair is
based on the current generated by the common source current sink (M5) being split by
matched differential pair M1 and M2 and the matched pair M3 and M4. Ignoring
dynamic drain resistance, near unity gain will be obtained if gm1 = gm2 = gm3 = gm4.
The thresholds of M1 and M2 vary with input/output voltages due to the body effect.
Also, the variation of the P-channel trans-conductance with large signal input and
output signals do not match the variation of N-channel trans-conductance. Both of these
effects will limit the buffer input range and linearity (gain accuracy). Reducing output
increases the difficulty to maintain all transistors operating in saturation for submicron
implementations with low voltage Vdd technologies, thereby reducing the maximum
A two-stage common source active load (CSAL) unity gain buffer will be
developed and analyzed to provide a basis for an improved buffer that is proposed in
this section. The two-stage circuit is shown in Fig 6.2.1. The circuit is similar to the
source coupled differential pair unity gain buffer discussed in section 6.1.2 except for
the absence of the tail current source (M5). The circuit does not operate as a matched
source coupled differential pair, but rather as a cascade of two common source
amplifiers with active loads. Input biasing (offset) is provided by the driving circuit or
realized with a capacitive coupled input with a high impedance voltage bias on the gate
of M1. Output offset is set by adjusting the sizes of M3 and M4 relative to M1 and M2.
56
Fig 6.2.2 Two-stage CSAL linearized AC circuit
The linearized AC equivalent circuit for the two-stage buffer is shown in Fig 6.2.2.
The dynamic drain resistance has been ignored in the small signal equivalent circuit
(rds>>1/gm) to simplify the analysis and provide insight into the primary performance
drivers.
The product of the first and second stage gains (gm1/gm2) * (gm3/gm4) should be near
0 dB for unity gain. Note that there is no requirement for matching transistors for unity
gain which provides flexibility in choosing transistor sizes based on buffer input
−𝑔𝑚4
𝑃2 = 𝐶 (6.2.3)
𝑆𝐿1 +𝐶𝑆𝐿
It is noted that stability is not an issue with this buffer since the gain of both stages
is low.
gate widths [42]. Representative values in this dissertation are used below to facilitate
𝑓𝐹
𝐶𝑔𝑠 ≅ 1.0 𝜇𝑚 [𝑊(𝜇𝑚)] (6.2.7)
𝑓𝐹
𝐶𝑔𝑑 ≅ 0.35 𝜇𝑚 [𝑊(𝜇𝑚)] (6.2.8)
𝑓𝐹
𝐶𝑠𝑏 = 𝐶𝑑𝑏 ≅ 0.6 𝜇𝑚 [𝑊(𝜇𝑚)] (6.2.9)
Eq 6.2.11 where it is seen that the values are not only a function of transistor widths,
but also will vary with the overdrive (VGSQ-VTN) for M1 and M4 and (VSGQ-|VTP|) for M2
and M3, where VTN and VTP represent the threshold voltage of nmos and pmos transistor.
𝑊
𝑔𝑚𝑁 ≅ 𝐾𝑝𝑁 × (𝑉𝑔𝑠 − 𝑉𝑇𝑁 ) (6.2.10)
𝐿
𝑊
𝑔𝑚𝑃 ≅ 𝐾𝑝𝑃 × (𝑉𝑠𝑔 − |𝑉𝑇𝑃 |) (6.2.11)
𝐿
The problem becomes one of finding the combinations for W1, W2, W3, and W4
to maximize the bandwidth of the two-stage response under the unity gain constraint:
required value of Cin. Under the parasitic capacitance model, Cin ≈ 1.7 fF/μm * W1 (μm);
capacitance model above. As stated in Eq 6.2.2 and Eq 6.2.3, the two poles setting the
58
bandwidth are P1 and P2. CSL1 is typically much smaller than CSL2 + CL, which implies
choosing a relatively small gm2 and relatively large gm4. A design procedure considering
of gm2.
Set the input offset to the desired value (typically 0.6 V for 90 nm CMOS
maximize the two stage 3 dB bandwidth with the two-stage low frequency gain near
obtain desired output offset and more accurately determine the low frequency gain, 3
Perform large signal transient analysis to determine input dynamic range that
CSAL buffer design with Cin ≈ 5 fF (W1 = 3 um) and CL = 250 fF. The circuit is
59
designed for an input offset voltage of 0.6 V and output offset is approximate 0.6 V.
The Bode plots (based on Cadence simulations) for stage 1, 2, and the output of
90nm CMOS CSAL buffer are shown in Fig 6.2.3. Referring to Fig 6.2.1, Stage 1 output
noted that the first stage transfer function has a low frequency gain of 4.28 dB and a 3
dB down bandwidth of 4.28 GHz which are relatively high because the loading is only
the self-loading capacitance delineated in Eq 6.2.5. The second stage gain and 3 dB
down bandwidth are -6.32 dB and 3.17 GHz with the 250 fF output load. The transistor
sizes are adjusted as described in the previous procedure to obtain maximum output 3
dB bandwidth with a flat low frequency gain near 0 dB. The output result which is the
product of stage 1 and 2 indicates a 3 dB bandwidth of 2.27 GHz and a low frequency
bandwidth.
Table 6.2.1 Transistor sizes of CSAL buffer optimized for 250fF capacitive load
Transistor Width (μm) Length (nm)
M1 3 100
M2 1 100
M3 12 100
M4 8 100
60
Fig 6.2.3 CSAL buffer AC analysis plot.
61
Table 6.2.2 Summary of single sided 90nm CMOS CSAL analog buffer
performance optimized for 250fF load
Input
Low
Input Load Range/ Input/Output
Freq 3dB BW THD Power
Cap Cap Gain Offset
Gain
Accuracy
140mV/<
≈5fF 50fF -2.79dB 3.79GHz 2% -54.3dB 0.6V/0.7V 730μW
variation
140mV/<
≈5fF 250fF -3.23dB 2.27GHz 2% -54.3dB 0.6V/0.7V 731μW
variation
Transient analysis was performed to determine bandwidth, gain, output offset, gain
accuracy THD, and power dissipation. A plot of variation in gain versus input amplitude
is shown in Fig 6.2.4 indicating a maximum input amplitude with less than 2% gain
variation of 140 mV when driving a 250 fF load, and the THD is -54.3 dB when assessed
Performance of the CSAL analog buffer is summarized in Table 6.2.2 for the
design optimized to drive 250 fF load when driving 250 fF and also when driving 50
fF. The results above indicate that the CSAL buffer has good 3 dB bandwidth for a
relatively large ratio of output capacitive load to input capacitive load (CL/Cin).
(1) Large signal input magnitude range for a 2% gain accuracy is relatively small.
The two stage CSAL buffer discussed above uses a first stage with an active load
P-channel pull up and an active load N-channel pull down on the second stage. It is
difficult to obtain a symmetric result for rising and falling inputs and outputs causing
limitations with regard to offset error, gain accuracy, and THD. These performance
issues are improved by a proposed double sided CSAL analog buffer with source
feedback.
62
6.3 Double Sided Active Load Analog Buffer with Source Feedback
A proposed two stage Double Sided Common Source Active Load (DSCSAL)
analog buffer with source feedback is shown in Fig 6.3.1. Input biasing (offset) is
provided by the driving circuit or realized with capacitive coupled input with a high
For this double-sided circuit, M9 and M10 are operating in the linear region with
(Vgs-VT) rather than a tail current source. Also, the parasitic self-loading capacitance at
at the source of M1 and M4 and the source of M5 and M8. The circuit can be
implemented with M9 and M10 replaced by resistor/capacitor parallel pairs, but the
63
effective area is increased. It is important to note that M9 and M10 must operate in the
linear region (not as current sources) to realize the performance described below. This
The left side of the DSCSAL buffer is identical to the single sided CSAL buffer
previously discussed except for the addition of M9. The right side mirrors the left side
with p-channel transistors replaced by n-channel and vice versa. When the p-channel
pulling up. As will be seen the complementary actions of the two sides improves
The linearized AC equivalent circuit is shown in Fig 6.3.2. The dynamic drain
resistances have been ignored in the small signal equivalent circuit since rdsx>>1/gmx.
To obtain insight into the operation of the DSCSAL buffer, a small signal transfer
function will be developed for the case where the right side small signal parameters are
equal to the left side parameters; that is: gm1=gm5; gm2=gm6; gm3=gm7; gm4=gm8; Cgs1=Cgs5;
Fig. 6.3.2 Small signal model of DSCSAL buffer with source feedback
64
Then the first stage transfer function is:
𝑣𝑜1𝐿 𝑣𝑜1𝑅
= (6.3.1)
𝑣𝑖𝑛 𝑣𝑖𝑛
Then the left side source voltage for M1 and M4 can be evaluated:
Where gmx=gm1+gm2+gm4.
𝑔𝑚1 +𝑠𝐶𝑖𝑛𝐿 𝑔
𝑚2
−𝑖𝑜1𝐿 = [𝑔𝑚1 − (𝑔𝑚1 + 𝑔𝑚2 ) ( 𝐴+𝑠𝐵
)] 𝑣𝑖𝑛 + [𝑔𝑚2 − (𝑔𝑚1 + 𝑔𝑚2 ) (𝐴+𝑠𝐵 )] 𝑣𝑜1𝐿 −
𝑔
𝑚2
[(𝑔𝑚1 + 𝑔𝑚2 ) (𝐴+𝑠𝐵 )] 𝑣𝑜𝑢𝑡 (6.3.8)
Where
65
𝑔
𝐺𝑜1 = − 𝑔𝑚1
𝑚2
𝑔𝑚4
(𝑔 +𝑔𝑚2 )
𝑔𝑚9 𝑚1
𝐺1 = 𝑔
𝑔𝑚2 (1+ 𝑚4 )
𝑔𝑚9
𝑔𝑚4 +𝑔𝑚9
𝑍𝑜1 = 𝑔
𝐶𝑠 − 𝑚2 𝐶𝑖𝑛𝐿
𝑔𝑚1
𝑚2 𝑔 (𝑔𝑚4 +𝑔𝑚5 )
𝜔𝑜1 = √ (𝐶
𝑖𝑛𝐿 +𝐶𝑆 )𝐶𝑆𝐿1
𝑣𝑜1𝐿 𝑣𝑜𝑢𝑡
(𝑠 = 0) = 𝐺𝑜1 + 𝐺1 [ (𝑠 = 0)] = 𝐺𝑜1 + 𝐺1 (6.3.11)
𝑣𝑖𝑛 𝑣𝑖𝑛
For gm9→∞ and Cs→0 (no tail transistors), Eq 6.3.10 yields the expected result for
It is seen from Eq 6.3.10 that vo1L/vin is dependent on vout/vin due to the feedback
The output voltage is determined by output current from both the left and right
sides which results in a factor of two for the current driving the total output capacitance
CT.
66
1
𝑣𝑜𝑢𝑡 = 2𝑖𝑜2𝐿 (𝑠𝐶 ) (6.3.15)
𝑇
Where CT=CSL2+CL+CSR2.
Combine Eq 6.3.14 and Eq 6.3.15 yields the second stage small signal response
𝑠 𝑠 𝑣 (𝑠)
𝑣𝑜𝑢𝑡 (𝑠) 𝐺𝑜2 (1+ )+𝐺2 (1+ ) 𝑖𝑛
𝑍𝑜2 𝑍2 𝑣𝑜1𝐿 (𝑠)
= 𝑠 𝑠2
(6.3.16)
𝑣𝑜1𝐿 (𝑠) 1+ +
𝜔𝑜2𝑄𝑜2 𝜔𝑜2 2
Where
−𝑔𝑚3 +𝐾𝑔𝑚2
𝐺𝑜2 = − 𝑔𝑚4 (1−𝐾)
𝐾𝑔𝑚1
𝐺2 = − 𝑔
𝑚4 (1−𝐾)
𝑔𝑚3 −𝐾𝑔𝑚2
𝑍𝑜2 = − 𝑔𝑚3 𝐷
𝑔
𝑍2 = 𝐶 𝑚1
𝑖𝑛𝐿
2𝑔𝑚4 (1−𝐾)
𝜔𝑜2 = √ 𝐷𝐶𝑇
2𝑔𝑚4 (1−𝐾)
𝜔𝑜2 𝑄𝑜2 = 2𝑔𝑚4 𝐷+𝐶𝑇
𝐶𝑖𝑛𝐿 +𝐶𝑆
𝐷=𝑔
𝑚1 +𝑔𝑚2 +𝑔𝑚4 +𝑔𝑚9
𝐶𝑇 = 𝐶𝐿 + 𝐶𝑆𝐿2+𝐶𝑆𝑅2
𝑣𝑜𝑢𝑡 𝑣𝑖𝑛 𝐺2
(𝑠 = 0) = 𝐺𝑜2 + 𝐺2 [ (𝑠 = 0)] = 𝐺𝑜1 + (6.3.17)
𝑣𝑜1𝐿 𝑣𝑜1𝐿 𝐺𝑜1 +𝐺1
For 𝑔𝑚5 →∞ and 𝐶𝑆 →0 (no tail transistors), Eq 6.3.16 collapses to the expected
It is seen from Eq 6.3.17 that vout/vo1L is dependent on vo1L/vin due to the feed ward
67
from the input (via source of M1) to the drain of M4 (output).
The first stage transfer function in Eq 6.3.10 shows that the addition of the tail
transistors in the DSCSAL buffer results in second order pole response in the
denominator and the addition of a zero in the numerator. The values of ωo1, Qo1, and
Zo1 can be adjusted to not only increase the first stage frequency response, but also to
interact with the second stage response to obtain a unity gain buffer with a significantly
increased 3dB bandwidth. Specifically, the Qo1 is adjusted so that the first stage is
underdamped with a modest peak frequency response. The second stage response given
by Eq 6.3.16 also has a second order response in the denominator; however, the
relatively large load capacitance (CL) results in a small Qo2 with over damping. The
proper selection of parameters results in the combination of the first stage underdamped
response and the second stage over damped response producing a flat output (vout/vin)
From Fig 6.3.1, it is seen that the addition of the transistors operating in the
resistive mode results in a reduction in the magnitudes of vgs1 and vsg5 for the input
transistors creating degenerative source feedback for the first stage of both sides.
Degenerative source feedback has the effect of mitigating the non-linear effects of the
large signal variation at the input [47]. It is also noted that the source of the second
stage output active load transistors M4 and M8 are also connected to the feedback node
(drains of M9 and M10) resulting in degenerative source feedback from the second stage
output nodes. The source feedback on the output active loads results in a more linear
output active resistor magnitude (1/gm) as the output signal magnitude increases. The
net effect is to improve the large signal linearity (input range/gain accuracy) as a result
of the degenerative source feedback. This result will be demonstrated in the next section
where the proposed DSCSAL analog buffer with source feedback is designed in 90nm
68
CMOS technology.
nm CMOS Design
technology to assess performance with Cadence tools. The buffer is initially designed
using schematic for assessing performance and optimizing transistor sizes followed by
The resulting layout for the DSCSAL buffer with source feedback is shown in Fig 6.3.3.
69
Table 6.3.1 is the transistor sizes for the proposed DSCSAL buffer with feedback
The Bode frequency plots for stage 1, 2, and the outputs are shown in Fig 6.3.4,
which indicate a 3 dB bandwidth of 4.4 GHz and a low frequency gain of 0.16 dB output
response.
In Fig 6.3.4 the right and left side first stage plots are the frequency response of
vo1R(s)/vin(s) and vo1L(s)/vin(s). Note the second order result as previously discussed with
the under damped response showing a peak magnitude in the vicinity of 4 GHz.
The vout(s)/vin(s) frequency result combines the two stage transfer functions that
have been shaped by selecting the proper sizes of the transistors to obtain maximize 3
dB bandwidth with a flat unity gain. The right and left side second stage plots are the
vout(s)/ vo1R(s) and vout(s)/ vo1R(s) results with the overdamped second order response.
Table 6.3.1 Transistor sizes of DSCSAL Analog Buffer optimized for 250 fF
capacitive load
Transistor Width (μm) Length (nm)
M1 3 100
M2 0.5 100
M3 5 100
M4 16 100
M5 7.5 100
M6 0.6 100
M7 7 100
M8 24 100
M9 50 100
M10 80 100
70
Fig. 6.3.4 AC analysis plots of DSCSAL buffer.
Gain (%)
3
-1
-2
-3
-4
-5
-6
10
20
30
40
50
60
70
80
90
100
110
120
130
140
150
160
170
180
190
200
210
220
230
240
250
260
270
280
290
300
input(mV)
71
Table 6.3.2 Summary of single sided 90 nm CMOS Double Sided CSAL Analog
Buffer with Source Feedback Performance Optimized for 250 fF load
Input
Low
Input Load Range/ Input/Output
Freq 3dB BW THD Power
Cap Cap Gain Offset
Gain
Accuracy
270mV/
≈10fF 50fF 0.16dB 6.43GHz <2% -63.3dB 0.6V/0.57V 1025uW
variation
270mV/
≈10fF 250fF 0.16dB 4.4GHz <2% -63.3dB 0.6V/0.57V 1058uW
variation
Large signal transient analysis was performed to determine low frequency gain,
output offset, gain accuracy, THD, and power dissipation. A plot of variation in gain
verses input amplitude is shown in Fig 6.3.5 indicating a maximum input amplitude
with less than 2% gain variation of 260 mV when driving a 250 fF load. The total
Performance of the DSCSAL analog buffer is summarized in Table 6.3.2 for 250
fF and 50 fF loads. It is noted that the transistor sizes are selected to optimize
Comparing the single sided CSAL buffer performance in Table 6.2.2 to the
DSCSAL with source feedback in Table 6.3.2, it is seen that all parameters of
72
Table 6.4.1 Comparison of the proposed DSCSAL with source feedback to other
published unity gain buffers
Low Input
Load Offset
Design Tech Freq 3dB BW Range/Gain THD Power
Cap Error
Gain Accuracy
---/ <1.8%
[42] 0.35um -0.13dB 87MHz -59dB 29mV 4.8mW
variation
13pF
This 270mV/<2%
90nm 0.17dB 153MHz -63.7dB 26mV 1.06mW
Work variation
---/<1.6%
[43] 0.35um -0.14dB 100MHz -60dB ---- ----
variation
1pF
This 270mV/<2%
90nm 0.09dB 1.97GHz -63.3dB 26mV 1.06mW
Work variation
Previous work that was found was based on CMOS technologies with larger
feature sizes driving larger capacitive loads. To provide a comparison, the DSCSAL
buffer performance was assessed when driving the same capacitive load reported in
each of the previous published work. As seen in Table 6.4.1, the DSCSAL buffer has a
significantly increased 3 dB bandwidth for each comparison even though the proposed
buffer was designed to optimize performance with a 250 fF load. Other performance
parameters are comparable or improved for the proposed buffer (where data permitted
comparisons). A key parameter missing for other published work was the input
capacitance.
6.5 Conclusion
A proposed double sided common source active load unity gain buffer with source
feedback implemented in 90 nm CMOS technology. This buffer has a very wide unity
73
gain 3dB bandwidth of 4.4 GHz with an output to input capacitive load ratio of 250
fF/10 fF. This design (optimized for the 250 fF load) has a 3dB bandwidth of 6.43 GHz
for a 50 fF load and 1.97 GHz for a 1 pF load. The double sided design together with
maintaining a less than 2% variation in gain for input amplitudes up to 270 mV. The
total harmonic distortion is -63.3 dB. The double sided design provides flexibility in
choosing transistor sizes to obtain unity gain with minimal offset error over a very wide
bandwidth. The active load, low gain design eliminates stability issues with no
reduced bandwidth. The Vdd to Vss path includes an active load diode connected
resistor with very low drain to source voltage drop. This topology facilitates keeping
the common source transistors in the saturation mode thereby making the buffer
architecture suitable for submicron CMOS technologies with low rail voltages. The
proposed new buffer architecture is an attractive option for on-chip submicron CMOS
applications requiring a very wide unity gain bandwidth, large CL/Cin ratio and excellent
74
VII. On-chip Self-Calibration System for CMOS Active
7.1 Introduction
From the content of chapter 4, the process variation has a huge impact on AIBPF
circuit performance. In order to maintain the active filter operating with designed
to compensate the process variation. Previous work have addressed either off chip post
fabrication calibration techniques for correcting AIBPF center frequency and Q values
[15] [32]or self-compensation solution for low frequency base band active filter design
functionality. The proposed system can capture the center frequency error, analyze
Determine post fabrication center frequency that has been altered by process
value; 3) Maintain the corrected center frequency after calibration process. The
proposed design is based on the principle that the amplitude of the AIBPF output should
be the largest at the desired center frequency as shown in Fig 7.2.1. Three sine wave
signals are generated with the same amplitude: signal 1 at a frequency fL = fC - f6dB,
75
signal 2 at a frequency fC and signal 3 at a frequency fH = fC + f6dB. The three signals are
sequentially applied in short bursts to the input of the AIBPF with the sequence repeated
14 times. A peak detector converts the AIBPF output amplitude for each of the three
signals to three different DC signals which are fed to the following comparator
sequentially. The peak amplitude DC value converted from the fC input signal is stored
by a track and hold circuit as reference data that is then compared with other two DC
signals. For each sequence of input signals, on-chip logic determines which of the three
peak detector signals is the largest to facilitate incrementally adjusting the bias control
in the proper direction to bring the desired frequency (fC) to be located between fL and
fH. The feedback mechanism shown in Fig 7.2.1 operates to adjust VBias1 and VBias2 in
desired center frequency. Changing both bias voltages in this manner resulted in a faster
convergence compared to fixing one and changing the other. This technique also had
the effect of minimizing the change in gain and Q of calibrated filter compared to the
pumps that receive up or down pulses from the logic circuitry during each sequence of
If process variation effects are minimal, the AIBPF has maximum amplitude at the
center frequency as shown in Fig 7.2.2(a). Then the logic circuitry produces 0 on both
up and down inputs to the charge pumps and bias controls are unchanged. This is
repeated for the 14 repetitions of the three input signals with no change at the end of
the sequence. If the post fab process parameters have shifted towards the slow corner,
relative positions of fL, fC, and fH are as presented in Fig 7.2.2(b). In this case, the
comparator generates a pulse signal associated with the slow corner indicator while
keeping a 0 at the fast corner indicator. The pulse signal causes a 1 to be sent to the up
76
input of charge pump 1 and to the down input of charge pump 2 in Fig 7.2.1. The output
value of both charge pumps is changing, and the VBias1 and VBias2 of AIBPF are
compensated to move the AIBPF center frequency toward the desired value. After a
sequence of 14 input signals, the resulting bias voltage will have placed fc in the middle
and the bias value is held constant until next calibration process.
Fig 7.2.2 Process variation effect of AIBPF: (a) Without process variation (b)
With process variation (process parameters towards the slow corner
77
The current implementation employs an off chip Arbitrary Wave Generator (AWG)
to generate the desired three input signals (50 ns CW burst at fL, 100 ns CW burst at fC,
signal value that will place the fc signal at the peak output amplitude.
The on-chip calibration circuitry should draw as little power as possible when the
AIBPF is operating on line (not in calibration mode). To realize this goal, a virtual
power structural [49] is employed in most circuit blocks of the calibration system. In
calibration mode, the virtual power transistor is turned on, and delivers supply power
to the calibration system. If calibration control switch is off, all calibration system
circuits with virtual power structural are shut down by turning off the virtual power
transistors, and the power consumption caused by leakage current of all additional
78
A standard charge pump [50] shown in Fig 7.3.1 is used to provide and change the
voltage of VBias1 and VBias2 by SC (slow corner) and FC (fast corner) signals as discussed
in Section 3. Since the AIBPF is very sensitive with bias voltage changing, the current
flowing through M1 and M2 must be very small to have fine output changing step size.
As shown in Table 7.3.1, the width of M1 to M4 is already close to the minimum value.
To further decrease the output voltage changing step size, the transistor lengths are
increased to 5 µm.
the one designed in chapter 4. Process variation is simulated by Monte Carlo analysis
with 20 sampling points. Based on the operation presented in the previous section, the
sequence of 3 signals with same amplitude but different frequencies (5 GHz, 5.25 GHz,
and 5.5 GHz) are the inputs to the AIBPF during calibration. If process variation has a
minimal effect, the 5.25 GHz frequency output of filter has the largest amplitude, and
the bias input remains unchanged during calibration process. Otherwise, one of the
other two test signals (5 GHz and 5.5 GHz) will generate the largest output amplitude
of the AIBPF. The top waveform at the first iteration of Fig 7.4.1 shows 5 GHz (fL)
signal amplitude is the largest among three test signals where process variation has
79
shifted the filter operation towards slow corner. The calibration system collects this
information and generates pulse signals to the up node and down node of charge pump
1 and 2, respectively. These signals result in the changing of VBias1 and VBias2, so that the
center frequency is adjusted in an iterative manner during calibration period. At the end
of the test sequence, the 5.25 GHz output regains the largest amplitude, and the final
compensated bias value is stored by charge pump and the calibration process is turned
off. The AIBPF uses this new bias voltage value to obtain the 5.25 GHz targeted center
Fig 7.4.1 Cadence simulation result of AIBPF output and two bias voltages using
proposed calibration system with slow corner process variation
80
Table 7.4.1 AIBPF simulation results
Post-fab before Post-fab after
Parameters Pre-fabrication
calibration calibration
Center Frequency 5.25 GHz 5.86 GHz 5.26 GHz
Gain 28.8 dB 41.6 dB 37.9 dB
Bandwidth 140 MHz 40 MHz 50 MHz
Quality Factor (Q) 37.5 146 105
Linearity
≤ 1.5 mV ≤ 200 µV ≤ 400 µV
(input amplitude)
Dynamic Range 34.17 dB 29.8 dB 29 dB
The results for a typical filter design sequence is summarized in Table 7.4.1, where
the key filter parameters are shown for 1) prefabrication filter designed to specifications
with typical PDK model parameters, 2) post fabrication performance prior to calibration
performance after calibration. The linearity parameter is the maximum sine voltage
amplitude based on the 1 dB compression point and dynamic range is the maximum
RMS sine wave amplitude divided by the RMS in band noise. As seen in Table 7.4.1,
the desired center frequency has been obtained as a result of the calibration sequence.
However other key parameters for the post fabrication filter have changed due to
process variations and calibration adjustments to VBias1 and VBias2. The gain and Q
prefabricated design. For some applications the increased gain and Q and reduced
bandwidth may be a plus; however, the designer may have to incorporate additional
flexibility in the design to adjust gain, Q, and bandwidth after calibration. Previous
work has addressed incorporating mechanisms in the AIPBF design for adjusting post
81
Fig 7.4.2 Yield versus number of Monte Carlo iterations with proposed
calibration system
Among the 20 Monte Carlo analysis sampling results, there are 15 cases where the
center frequency needs adjusting due to the process variations. With this number, the
yield (ratio of good circuits to total fabricated circuits) for uncalibrated circuits is only
5/20 = 25%. Within these incorrect function results, the proposed calibration system
successfully corrects 11 of them by adjusting the biases automatically, and the rate of
finished product yield is increased to 16/20 = 80%. The 4 cases where calibration is not
successful is due to a significant reduction in filter gain due to process variation. The
filter response for all three calibration frequencies is so low that the calibration system
cannot detect the difference in amplitude resulting in no change for a calibration run.
This significant improvement in yield can lower the cost of fabrication and allows users
to have the ability to restore the AIBPF center frequency when needed after chip is
fabricated.
82
Table 7.4.2 Power consumption of proposed self-calibration system
Average power at slow corner Calibration On Calibration Off
AIBPF 3.7 mW 3.7 mW
Charge Pump 0.181 mW 0.183 mW
Other Self-Calibration Circuits 0.523mW 0.003mW
of yield, a plot of yield versus number of Monte Carlo iterations is shown in Fig 7.4.2
with number of iterations simulated up to 100. From the figure, the yield is between 80
and 84% for iterations more than 10. Therefore, 20 Monte Carlo seeds was considered
in Table 7.4.2 when operating at 5.25 GHz desired center frequency. During calibration
process, the calibration system consumes 0.704 mW extra power (Charge Pump plus
consumption. The charge pumps continue to operate when calibration is off, since
charge pumps are used to supply and maintain the bias voltage. The other self-
calibration circuitries require only 0.003 mW power when virtual power transistors are
turned off during normal operation. The calibration system consumes 0.186 mW power
7.5 Conclusion
post fabrication process variation effects on 5.25 GHz active inductor band pass filters
is presented in this chapter. The system is designed to correct the center frequency error
by detecting the filter output amplitude and adjust bias voltages of AIBPF. The results
83
of 20 Monte Carlo analysis simulation run indicates the calibration system successfully
relocates the center frequency at the desired value for most of the expected post
boosted from 25% to 80% by incorporating the proposed calibration system with
AIBPF. Since all the circuits are constructed without passive inductors, the potential for
higher Q and gain while operating at RF frequency are still preserved, and the total area
consumption of AIBPF plus proposed self-calibration system is still much smaller than
single stage passive inductor band pass filter with the same center frequency. Also, the
virtual power technique guarantees the proposed calibration system adds relatively
84
VIII. Mixer
8.1 Introduction
In CMOS receiver chain system, mixer is an essential circuit block due to its
frequency translating function. In most cases, the information sending out from
capability. This carrier frequency is created by Local Oscillator (LO), and it is also sent
to receiver for frequency conversion. However, at receiver side, due to the circuit
complexity and power consumption, the signal sending into ADC must have carrier
signal removed, and this task is completed by mixer circuit in front end design.
As stated in chapter 2, both Hartley and Weaver frequency conversion systems can
reject the image and improve the signal quality. The Hartley system only needs one
path and one 90-degree shifter built in one path to perform the image rejection function.
The Weaver architecture employs two mixers in each frequency shifting path to
demodulate the wanted signal and remove the image part as demonstrated in Fig 8.1.1.
A SIDO mixer is placed in the first stage to be compatible with the output from amplifier
in Fig 2.1.6. The Gilbert mixer is used as the second stage down-converting component
as its differential inputs can provide high gain to improve the output signal amplitude.
85
Since the receiver designed in this dissertation chooses the low IF architecture, the
mixer implemented in this chapter is integrated with active inductor-based band pass
filter to filter out high frequency carrier and reserve the IF signal. Meanwhile, the active
circuits also provide the benefit of gain improvement and area efficiency.
Weaver image rejection architecture uses two stages of mixer to demodulate the
RF signal into IF band, and the first stage circuit uses SIDO design shown in Fig 8.2.1
86
At this stage, signal with 5.25 GHz of targeting channel center frequency is down-
converted to 750 MHz signal by SIDO mixer when local oscillator generates 4.5 GHz
LO signals. Furthermore, since next stage Gilbert mixer is designed with two
differential inputs, the single-input-dual-output feature of SIDO mixer can convert the
input signal from previous RF stage single pin mode into differential format. The output
750 MHz signal feeds into next stage Gilbert mixer for further processing.
Fig 8.2.2 shows the schematic of Gilbert cell mixer. In order to design the mixer
with reasonable power consumption, the sum of current flowing through both M1 and
M8 are set to be no more than 5 mA. So, in the following calculation, both transistors
passed current and gate voltage are set to 2.5 mA and 0.5 V respectively. Thus, the
87
Since the trans-conductance value of transistor is proportional to circuit gain, all
the transistors need to work at saturation region to have the largest trans-conductance
and sufficient gain of mixer design. Apply the long channel NMOS saturation current
In Eq 8.2.2, W and L represents the transistor width and length. Based on single
threshold voltage (VT0) is 0.386 V, process gain (Kn) is 1.03 mA/V2, and channel length
modulation (λn) is 0.43 V-1. For M8 transistor, substitute current limitation and gate
voltage assumption to Eq 8.2.2, the total width is 123 μm by solving the equation below:
1 𝑊
𝐼𝑑𝑠8 = 2.5𝑚𝐴 = 2 ∗ 1.03𝑚𝐴/𝑉 2 ∗ 400𝑛𝑚
8
∗ (0.5 − 0.386)2 ∗ (1 − 0.43 ∗ 0.5) (8.2.3)
To make M1 transistor working in saturation region, VDS must be greater than (VGS
– VT0), so that v1 in Fig 8.2.2 is set to be 0.2 V. And current flowing through M1
As current passing M1 is divided in two, if M2 and M3 have the same size, and
M4, M5, M6, and M7 are identical, the current flowing through R4 is 1.25 mA.
Assuming the output waveform has 1.0 V offset voltage, then R4 and R5 should equal
using Eq 8.2.5, and the result is 10.8 mA/V. Apply trans-conductance equation to M2
transistor:
88
2∗1.03𝑚𝐴/𝑉 2 ∗𝑊2 ∗1.25𝑚𝐴
10.8𝑚𝐴/𝑉 = √ (8.2.6)
400𝑛𝑚
1 1.03𝑚𝐴 𝑊
0.625𝑚𝐴 = 2 ∗ 4
∗ 400𝑛𝑚 (𝑉𝑙𝑜 − 𝑉2 − 0.386)2 (1 − 0.43(𝑉𝑜𝑢𝑡𝐷𝐶 − 𝑉2 ))(8.2.8)
𝑉2
Assume V2 = 0.4 V, Vlo = 1.2 V, and VoutDC = 1 V, then the Vbias and width of
Therefore, all the components parameters are solved to high conversion gain with
different from real value, due to the unconsidered parasitic capacitance and resistance.
However, this IF output signal is noisy and still coupled with high frequency
carrier signal, thus a band pass filter is needed to reserve only the certain frequency
range mixer output signal that contains the wanted channel data and pass them to next
If the filter is built with traditional RLC parallel structure as shown in Fig 8.2.3,
targeting channel (channel 50 of 802.11ac). If output capacitance is set to be 100 fF, the
inductance usually means large consumption of on-chip area. Therefore, this band pass
filter is implemented with active inductor in chapter 4 to minimize the circuit area usage.
89
Fig 8.2.3 Conventional R-L-C band pass network
The SIDO mixer used in Hartley architecture is the same as the SIDO mixer
employed in Weaver. By feeding the 5 GHz LO signal into the circuit, mixer is able to
down-convert the wanted 5.25 GHz RF signal to 250 MHz low frequency IF signal.
Attach the AIBPF designed with 250 MHz center frequency to mixer, the Hartley
system perform the frequency conversion with only one stage mixer.
Both SIDO and Gilbert mixers are built and simulated in Cadence software with
90
8.3.1 SIDO Mixer of Hartley and Weaver Design Simulation Results
The schematic diagram of designed SIDO mixer system is shown in Fig. 8.3.1,
and the simulation results of mixer and filter are measured with system internal node to
get the circuit properties with actual load. As discussed in previous section, both
Weaver and Hartley employ the same SIDO mixer to perform the frequency conversion.
To make the SIDO mixer output have offset of 0.6 V, the quiescent operating point
simulation is performed with Vbias set to be a variable as shown in Fig 8.3.2. The result
offset of the circuit. Besides, under this bias voltage, the quiescent current flowing
though mixer is 292.7 µA, which results in the total power consumption of 351.24 µW.
91
The frequency response waveform of AIBPF modified in this stage is
demonstrated in Fig 8.3.3 with center frequency of 750 MHz and bandwidth of 588
MHz. In order to further improve the gain and bandwidth performance, there are 4
AIBPFs connected in series to form the filter stage. The result in Fig 8.3.4 indicates that
this multi-stages AIBPF provides gain of 18.292 dB and bandwidth of 255 MHz.
The time domain transient analysis simulation result is shown in Fig 8.3.5 with
5.25 GHz frequency and 10 mV amplitude input sine wave. The 750 MHz output
waveform has 119.3 mV amplitude and the linear gain calculated by transient analysis
result is 11.93.
Fig 8.3.3 Frequency response simulation result of AIBPF modified for SIDO
mixer
92
Fig 8.3.4 Frequency response simulation result of 4 stages AIBPF in series in
Weaver design.
Fig 8.3.5 Time domain transient analysis result of SIDO mixer with AIBPF in
Weaver design
93
The Fig 8.3.6 contains the noise performance of the first stage frequency down-
conversion system (SIDO mixer plus AIBPFs). As shown in simulation result, the noise
figure of this stage is 23.7 dB at converted signal center frequency of 750 MHz. Even
though, this number is a little higher than some of the mixer design in literature,
considering multiple active circuits is built in this stage and provide large gain, this NF
value is acceptable. Besides, with the low noise and large gain design of previous
amplifier stages, the noise performance of mixer stage does not show large impact on
entire receiver system noise parameter. Furthermore, the large gain feature of the SIDO
mixer based first frequency conversion stage can mitigate the noise effect of second
Fig 8.3.6 Noise figure simulation result of first stage frequency conversion system
consisted by SIDO mixer and AIBPFs in Weaver design
94
Fig 8.3.7 1dB compression simulation result of SIDO mixer in Weaver design
Fig 8.3.8 1dB compression simulation result of SIDO mixer connected with
AIBPFs in Weaver design
95
Fig 8.3.7 and 8.3.8 demonstrate 1 dB compression point of SIDO mixer and entire
frequency conversion system including SIDO mixer and AIBPFs, respectively. For
SIDO mixer, the compression point is -12.35 dBm. If the design is integrated with
AIBPF, then the compression point drops to -27.9 dBm. It can be seen from these two
graphs that the tradeoff of adding extra gain to the system is the degradation of linearity.
The IIP3 simulation result of SIDO mixer is -4 dBm which can be found in Fig
8.3.9 with down-converted third order signal of 749 MHz. Under the same testing setup,
the IIP3 result of SIDO mixer plus AIBPFs is -24.82 dBm shown in Fig 8.3.10.
To evaluate the capability of designed SIDO mixer suppressing the supply power
variation to its output, the Power Supply Rejection (PSR) simulation is performed. The
smaller value of this result indicates the system is less sensitive to the variation. As
shown in Fig 8.3.11, the SIDO mixer has PSR value of -125 dB at frequency of 750
MHz.
96
Fig 8.3.10 IIP3 simulation result of SIDO mixer plus AIBPFs in Weaver design
Fig 8.3.11 Power supply rejection simulation result of SIDO mixer in Weaver
design
97
The SIDO mixer designed for Hartley system has the same architecture with
transistor size and control voltages properly adjusted. The simulation results are listed
in Table 8.3.1.
Table 8.3.1 Simulation Result of SIDO Mixer Designed for Hartley System
Parameters Value
Center Frequency 250 MHz
1 Stage AIBPF Gain 11.7 dB
4 Stages AIBPF Gain 48.16 dB
1 Stage AIBPF 3 dB Bandwidth 385.27 MHz
4 Stages AIBPF 3 dB Bandwidth 156.91 MHz
SIDO + AIBPF Noise Figure 21.85 dB
SIDO + AIBPF 1 dB Compression Point -39.17 dBm
SIDO + AIBPF IIP3 -34.07 dBm
PSR -137.82 dB
8.3.12, and the simulation results of mixer and filter are measured with system internal
The quiescent operating point simulation result of Gilbert mixer is shown in Fig
8.3.13 with bias voltage of 596.9 mV and current of 800 µA when output DC voltage
98
Fig 8.3.13 Quiescent simulation results of Gilbert mixer.
Fig 8.3.14 contains the frequency response waveform of AIBPF modified for this
stage mixer with center frequency of 250 MHz with gain of 11.514 dB and bandwidth
of 160 MHz. There are 3 AIBPFs connected in series building the band pass filter block
for each mixer output node, and the overall performance is shown in Fig 8.3.15 with
center frequency still locating at 250 MHz but higher gain of 36.56 dB and narrow
bandwidth of 77.45 MHz. With this filter stage, the 250 MHz mixer output signal is
The time domain waveform in Fig 8.3.16 indicates the mixer and AIBPFs provide
99
Fig 8.3.14 Frequency response simulation result of AIBPF modified for Gilbert
mixer
100
Fig 8.3.16 Time domain simulation result of Gilbert mixer with AIBPFs
The drawback of Gilbert mixer circuit is the poor noise figure value. Combining
both mixer and AIBPFs, the simulation results in Fig 8.3.17 indicates noise figure of
conversion stage with AIBPFs attached on mixer are shown in Fig 8.3.18 and 8.3.19,
respectively. By adding the AIBPFs to filter out noise and increase gain, the 1 dB
Fig 8.3.20 and 8.3.21 contains the IIP3 simulation results of Gilbert mixer and
shown on graphs, the Gilbert mixer has IIP3 value of -10.27 dBm and then drops to -
101
Fig 8.3.17 Noise Figure simulation result of second stage frequency conversion
system (Gilbert mixer plus AIBPFs)
102
Fig 8.3.19 1 dB compression point simulation result of Gilbert mixer with
AIBPFs
The supply power rejection simulation result of Gilbert mixer is shown in Fig
103
Fig 8.3.21 IIP3 simulation result of Gilbert mixer plus AIBPFs
104
8.4 Conclusion
In this section, the SIDO and Gilbert mixers are designed for building Hartley and
Weaver frequency down-converting system. The AIBPFs are optimized to pass the
targeting low frequency IF signal and filter out the high frequency noise. Both mixers
provide large gain and good quality factor by taking the advantage of active filter. The
SIDO mixer has better noise performance and less power consumption, while the
Gilbert design offers better supply power rejection. The system performance of Hartley
and Weaver design are introduced along with entire receiver chain simulation results in
105
IX. Phase Locked Loop
9.1 Introduction
Phase Locked Loop (PLL) circuit is widely used in modern transceiver design to
provide many of system signals, such as carrier, clock, etc. In WLAN communication
frequency carrier signal before transmitting. Such carrier signal is removed at receiver
section by mixer circuitry to recover the transmitting information for following data
processing. The PLL circuit built in WLAN system is employed to generates high
frequency carrier signal which is also called Local Oscillator (LO) signal in both
transmitter and receiver paths. In order to satisfy system specifications, the PLL must
deliver high stability, short settle time, and low phase noise with reasonable power
consumption [28].
with feedback control structure PLL system to meet the receiver specifications.
106
9.2 System Topology and Sub-Circuit Introduction
Frequency Detector (PFD), charge pump, VCO and N divider as shown in Fig 9.2.1.
Signal from VCO output feeds back to PFD with 1/N original frequency through N
divider to compare the phase and frequency with XO reference signal. If any frequency
variation of VCO is detected, the PFD generates a signal to indicate the variation and
control charge pump adjusting VCO bias voltage until the frequency variation is
eliminated.
and feedback VCO output, the PFD compares the time of each signal rising edge and
two D Flip Flop (DFF) plus a reset AND gate can form a PFD system as shown in Fig
9.2.2. This design provides wide phase difference detection range from -2π to 2π which
ensure the entire PLL system having short response time and good stability [28]. The
operation of this PFD can be described as: 1) At initial state, Ref and VCO are equal to
‘0’, and two DFF output (Q1 and Q2) are also equal to ‘0’. Therefore, the output of
AND gate is 0 and two DFFs are not reset; 2) If the rising edge of Ref arriving at DFF
is earlier than VCO, then Q1 rises to ‘1’ and Q2 remains at ‘0’. 3) When VCO rising
edge arrives DFF2, Q2 becomes ‘1’, which makes the output of AND gate rise to ‘1’
and reset both DFF outputs to ‘0’; 4) System returns to initial state and waits the rising
However, the drawback of this implementation is the dead zone issue which
caused by small phase error between two input signals. In this case, the PFD fail to
107
output the correct signal as the time delay created by signal phase difference is less than
the AND gate delay. So, to avoid the dead zone effect, the feedback portion of PFD
design needs to be removed, and such circuit structure is used in this dissertation as
108
Fig 9.2.3 Proposed PFD with zero dead zone
The proposed PFD circuitry is a two-stage architecture, and the first stage output
is the selection signal of second stage 2 multiplexers. At initial state, both Ref and VCO
are ‘0’, and Select is equal to 1. The second stage multiplexers output the value at input
node ‘1’, and both are ‘0’. If rising edge of Ref signal reaches PFD earlier than VCO,
Select remain at ‘1’, and only Up rises to ‘1’. When the rising edge of VCO arrives at
PFD after certain delay, the Select signal is pulled down to 0, and both multiplexers
output the value at input node ‘0’ which is GND (‘0’). Thus, the PFD finishes one
Generally, there are two types of VCO design that are popular in PLL system
implementation, one is using LC tank, and another one is using delay cell to build ring
109
oscillator. The LC tank based architecture oscillates at passive component resonating
frequency and the tuning feature is usually achieved by controlling the varactor value.
Even though, the LC tank oscillator provides less phase noise, this circuitry suffers
lacking wide tuning range and relatively large on-chip area due to the inductor.
Therefore, in this dissertation, the proposed VCO uses ring oscillator structure to be
area efficiency. Such VCO needs to have low phase noise and wide tuning range along
To reach those specifications, the delay cell introduced in [50] is employed in this
paper, and the schematic is shown in Fig 9.2.4. Such delay cell is consisted of
differential oscillator and CMOS latch to improve the driving ability and phase noise
performance.
110
The tunability of output frequency is achieved by varying the Vcon in Fig 9.2.4.
When M3 and M4 working in linear region, the current flowing through transistor can
be expressed as:
2
𝑊 𝑉𝑑𝑠
𝑖𝑑𝑠2 = 𝜇𝑛 ∗ 𝐶𝑜𝑥 ∗ ∗ [(𝑉𝑐𝑜𝑛 − 𝑉𝑜𝑢𝑡 − 𝑉𝑇𝑛 )𝑉𝑑𝑠 − ] (9.2.1)
𝐿 2
Where μn is the electron mobility in the channel and Cox represents gate
capacitance in unit area. W and L are the width and length of transistor, respectively.
VTn is the threshold voltage of NMOS transistor and Vds is the drain to source voltage
across transistor. Thus, the equivalent resistance R of M3 and M4 are equal to:
𝜕𝑉𝑑𝑠 1
𝑅= = 𝑊 𝑉 (9.2.2)
𝜕𝑖𝑑𝑠 𝜇𝑛 ∗𝐶𝑜𝑥 ∗ ∗[(𝑉𝑐𝑜𝑛 −𝑉𝑜𝑢𝑡 −𝑉𝑇𝑛 )− 𝑑𝑠 ]
𝐿 2
From Eq 9.2.2, with the increment of Vcon, the R decreases. If the parasitic
capacitance is unchanged within the path from M6 and M7 to out+ and out-, the smaller
value of R increases the time constant of changing M6 and M7 gate voltage and
decreases the output frequency due to the stronger latch impact [50].
Fig 9.2.5 contains the VCO architecture using proposed delay cell, and it provide
quadrature phase outputs with buffer connected to enhance the driving and isolation
ability.
Fig 9.2.5 Schematic of quadrature outputs VCO using propose delay cell
111
9.2.3 Charge Pump and Loop Filter
Charge pump converts the output signal of PFD to analog voltage changing and
tune the frequency of VCO. In the proposed PLL system, charge pump design uses the
same circuitry as introduced in section 7.3 of chapter VII, which can pull up or pull
down the control voltage of VCO based on the Up of Down pulse generated from PFD.
component to smooth the control voltage of VCO [28], and it should be placed between
charge pump and VCO. A widely used design of loop filter is employing passive resistor
and capacitors to provide the low pass feature and additional pole as shown in Fig 9.2.6.
To calculate the value of each component in Fig 9.2.6, and gain the flattest
frequency response curve, the damping factor ξ of loop filter needs to be √2/2 as shown
below [50]:
𝜔0
𝜉= ∗ 𝑅𝑍 ∗ 𝐶𝑍 (9.2.3)
2
𝐼
𝜔0 = √2𝜋𝐶𝑐ℎ 𝑁 𝐾𝑉𝐶𝑂 (9.2.4)
𝑍
In Eq 9.2.4, Ich is the charging current of charge pump and the value measured
from the design in section 7.3 is 5.5 μA. CZ is set to be 10 pF and N is the value of
divider stage which is 128 in proposed PLL. KVCO represent the gain of VCO (Hz/V)
which can be found with simulation result, and the recommended resistance of loop
112
Fig 9.2.6 Loop filter with additional pole
To reduce the system consumption and decrease system complexity, the feedback
signal from output of VCO need to pass a frequency division circuit block to lower its
frequency before entering PFD. In this dissertation, the N value is set to be 128 and this
divider features single phase clock flip-flop structure which is compatible with low
power (no static power consumption), area efficiency and high frequency application.
As illustrated in Fig 9.2.8, when circuit is in initial state, both CLK and out node
are equal to ‘0’. Internal node X and Y are equal to ‘1’, and out is ‘0’. In next step,
feedback gets ‘0’ from previous state out node and CLK rises to ‘1’, which maintains
the ‘1’ at X node and pulls down Y to ‘0’. At this state, the out is equals to ‘1’. Then
CLK changes to ‘0’, and feedback is equal to ‘1’. X switches to ‘0’ and pull up Y to ‘1’.
The out keeps ‘0’ as previous state. In the last state, both CLK and D are equal to ‘1’.
X and Y keeps the value as previous state. The out is pulled down to ‘0’ and system
returns to the initial state. Therefore, the system output frequency is half of the input
frequency.
113
Fig 9.2.7 Schematic of single phase clock flip-flop
114
9.3 Simulation Result
The proposed PLL system and its sub-circuits are designed and simulated in
From Fig 9.3.1, the minimal time difference between two input signals that PFD
can differentiate is 500 ps. The average power consumption of this stage is 3.5 μW with
50 MHz input frequency. Fig 9.3.2 is the schematic of proposed quadrature phase output
Fig 9.3.1 Simulation result of PFD with 500 ps time difference between two input
signals
115
Fig 9.3.2 Schematic of proposed VCO with quadrature phase output and buffers
Fig. 9.3.3 demonstrate the waveform of each output node with quadrature phases. All
the plots maintain 50% duty cycle with 0 V to 1.2 V full scale swing, and the average
116
Fig 9.3.4 Phase noise of proposed VCO with 5 GHz operating frequency
Fig 9.3.4 shows the phase noise performance evaluated by Cadence RF with 5
GHz operating frequency. At 1 MHz point, the phase noise is -90.61dBc/Hz, which is
The gain of VCO (KVCO) is equal to 3.8 GHz/V with simulation result. Substitute
this numbers with all the known value introduced in section 9.2.3 into Eq 9.2.4, the
bandwidth is calculated as 4.04*106 rad/s. Plug the bandwidth result along with Cz into
Eq 9.2.3, the Rz value is equal to 35 KΩ when the entire loop has the quickest
stabilization time. CP value is much less than CZ, and it is set to 500 fF in the final loop
filter design.
The completed PLL schematic is shown in Fig 9.3.5, and the reference signal is
set to be 39.0625 MHz, as this value is available for commercial crystal oscillator and
117
Fig 9.3.5 Schematic of PLL with sub-circuit blocks
From the simulation result in Fig 9.3.6, the designed PLL outputs 5 GHz signal
with stabilization time less than 1.2 μs. The average power consumption of this PLL
118
9.4 Conclusion
LO source. Such PLL uses conventional architecture and consisted of PFD, charge
pump, VCO, and dividers. To accommodate the area efficiency purpose of this
dissertation, a ring oscillator structure based VCO is employed to avoid the usage of
on-chip inductor. Meanwhile, such VCO provides quadrature phase outputs for Hartly
The proposed PLL features fast settling time and good stability for maintaining 5 GHz
signal with 50% duty cycle. The average power consumption of entire system is 8.67
119
X. Analog 90 Degree Phase Shifter
10.1 Introduction
As discussed in the first chapter, the key component of building Hartley system is
the 90 degree phase shifter [52]. By having this stage placed in sin(ωLOt) path, the
signals generated from two paths have differential IF component and identical image
part. Thus, the image signal can be easily removed by subtracting two signals using a
architecture, the 90 degree phase shifter must provide good phase shifting accuracy and
minimum gain mismatch error. Moreover, this phase shifter needs to be designed with
the consideration of power and area efficiency to fit entire receiver design objectives
To satisfy the specification listed above, a 90 degree phase shifter using RC-CR
by employing active resistor that can alternate frequency response at output node. This
feature guarantees two signals with 90 degree phase difference having the same
The RC-CR network shown in Fig 10.2.1 is built by two signal paths with active
resistor and capacitor connected in series. The output signals can be expressed as:
𝑅 𝑗𝜔𝑅𝐶
𝑉𝑜𝑢𝑡1 = 1 = 1+𝑗𝜔𝐶 (10.2.1)
1+
𝑗𝜔𝐶
1
𝑗𝜔𝐶 1
𝑉𝑜𝑢𝑡2 = 1 = 1+𝑗𝜔𝐶 (10.2.2)
1+
𝑗𝜔𝐶
120
𝜋
∠𝑉𝑜𝑢𝑡1 = 2 − tan−1 𝜔𝑅𝐶 (10.2.3)
Comparing Eq 10.2.3 and 10.2.4, the phase difference between two output nodes
is 90 degree over all frequency. Moreover, these two paths of the circuitry are equivalent
to low-pass and high-pass filter, respectively. Since the 3dB down cutoff frequency (fC)
of RC circuit has phase of 45 degree, the two output signals can have the same
amplitude at fC point with 90 degree phase difference as shown in Fig 10.2.2. The 3dB
down cutoff frequency (fC) of Vout1 and Vout2 are equal to:
1
𝑓𝐶 = 2𝜋𝑅𝐶 (10.2.5)
If capacitor is set to be 2 pF, then the resistance of each path needs to be 318 Ω to
121
Fig 10.2.2 Frequency response of high pass and low pass filter with gain
mismatch phenomenon of phase shifter
However, if input signal frequency changes, the amplitude of two output becomes
disparate, and this gain mismatch degrade the quality of system image rejection. To
address this issue, either resistor or capacitor in circuitry must be capable of varying its
value without changing physical geometry. In this dissertation, this feature is achieved
with active resistor built by CMOS transistor and controlled by gate voltage.
For transistor working in linear region, the drain to source resistance for N-type
transistor. The W and L describe the width and length of transistor and VT is the
threshold voltage that can turn on the transistor. Noted that, in Eq 10.2.6, the gate to
source voltage VGS must be greater than VT to allow current ID flowing through
transistor. Besides, in order to operate transistor in linear region, the drain to source
122
voltage VDS needs to be smaller than VGS - VT. Thus, the equivalent resistance between
changing the gate voltage. Therefore, the gain calibration can be achieved by adjusting
gate voltage to compensate the gain mismatch error at different input frequency.
The 90 degree phase shifter with self-calibration system is built and simulated in
90nm technology. The schematic diagram is demonstrated in Fig 10.3.1 with built-in
input blocking capacitor to set the signal offset value. The testing frequency band is the
down-converted IF frequency band from 170 MHz to 330MHz. In the initial setup, the
250 MHz input signal sending into phase shifter creates two output signals with same
frequency and amplitude but 90 degree phase difference as shown in Fig 10.3.2.
Fig 10.3.1 Schematic diagram of 90 degree phase shifter using RC-CR network.
123
Fig 10.3.2 90 degree phase shifter two outputs waveforms at 250 MHz
Fig 9.4.3 and Fig 9.4.4 demonstrate the waveforms before and after calibration
when input frequency is 170 MHz and 330 MHz, respectively. In these two simulation
results, the input amplitude is 100 mV. However, due to the frequency change, the
output signals amplitude difference is 20.6 mV for 170 MHz case and 15.5 mV for 330
mV case before calibration. After adjusting the gate voltage of active load transistor,
the amplitude difference between two output signals are 0.5 mV for both 170 MHz and
Fig 10.3.3 90 degree phase shifter two outputs waveforms at 170 MHz: (a) Before
calibration (b) After calibration
124
Fig 10.3.4 90 degree phase shifter two outputs waveforms at 330 MHz: (a) Before
calibration (b) After calibration
590 mV 20.6 mV
170 MHz
500 mV 0.5 mV
590 mV 15.5 mV
330 mV
670mV 0.5 mV
From Table 10.3.1, the two output signals of phase shifter can have minimum gain
10.4 Conclusion
designed and simulated with 90 nm technology. To have the same amplitude of 2 output
signals among demanding frequency band, the CMOS transistor is employed as active
resistor in the circuit and its resistance can be tuned by changing gate voltage. Therefore,
the gain mismatch error between two output nodes created by different input frequency
125
is compensated by adjusting the active resistance. The simulation results indicate the
designed phase shifter can maintain 90 degree difference and same amplitude between
two output signals within the frequency band from 170 MHz to 330 MHz. The average
power consumption of the phase shifter system is 10.19 μW when input signal has 100
126
XI. System Performance
In this section, the Weaver and Hartley systems are implemented and simulated
with Cadence using 90 nm technology. Both architectures are performed the noise
figure, image rejection ratio, and power consumption simulations to evaluate the system
performance and decide which design fits the entire front end receiver better. Detailed
and it contains the IQ paths to achieve image rejection function at the final output node
with the help of 90 degree phase shifter. The noise figure simulation plot is shown in
Fig 11.1.2 with 40.14 dB at selected band center frequency of 5.25 GHz. The power
127
Fig 11.1.2 Noise figure plot of proposed Hartley system.
Fig 11.1.3 Hartley System output waveforms generated by RF and Image signals.
128
The Image Rejection Ratio (IRR) is defined as:
𝑉𝑜𝑢𝑡𝑅𝐹
𝐼𝑅𝑅 (𝑑𝐵) = 20 log10 𝑉 (11.1.1)
𝑜𝑢𝑡𝐼𝑚𝑎𝑔𝑒
and VoutImage is the amplitude of output signal up-converted by image input. Using the
transient analysis results shown in Fig 11.1.3, this IRR value can be calculated as 25.76
dB.
Fig 11.1.4 shows the schematic diagram of designed Weaver system. As stated in
chapter 8, this dual frequency conversion architecture is transforming the 5.25 GHz RF
signal into 750 MHz temporary IF signal by 4.5 GHz LO signal first, then uses second
LO source with 500 MHz frequency to further down-converted the first stage output to
The noise figure is 52.24 dB for the Weaver system and simulation plot is shown
in Fig 11.1.5. The average power consumption is measured as 16.15 mW when input
signal is set to 0.1 mV amplitude. These two numbers are quite large comparing to the
results of Hartley design due to the extra mixers and active filters.
129
Fig 11.1.5 Weaver system noise figure.
Fig 11.1.6 Weaver system output waveforms generated by RF and Image signals.
130
The input frequency of measuring image rejection ratio are 5.25 GHz and 4.75
GHz. Both transient analysis results are presented in Fig 11.1.6, and the rejection ratio
Through the simulation results, Weaver system provides better image rejection
ratio than Hartley design due to its two-stage frequency conversion architecture.
However, with less active circuits involved in the design, the Hartley system offers
better noise performance and low power consumption which are the key parameters of
wireless front end implementation. In the following section, the proposed receiver
is presented with circuit blocks in Fig 11.2.1 and the detailed implementation of each
131
Table 11.2.1 Proposed Receiver Simulation Results
Parameter Simulation Results
Gain (dB) 45.8
Bandwidth (MHz) 200
Input Sensitivity (dBm) -90
Noise Figure (dB) (without PLL) 10.72
Image Rejection (dB) 35.63
Power Consumption (mW) 28.42
Maximum Input Power (dBm) -50
The minimum input power that the receiver can detect, and transfer is -90 dBm,
which satisfied the input sensitivity requirement in 802.11ac standard (-73 dBm). Noise
figure value listed in the table is simulated at system highest gain mode. Since the RF
stage circuits (LNA+AIBPF) are implemented with low noise and high gain feature,
the entire front-end noise is suppressed even the IF stage employs two mixers and noisy.
This parameter is calculated based on Eq 2.2.4 with the Noise Figure of RF section and
Hartley system, and the simulation results is contained in Fig 11.2.2 and 11.1.2.
However, even such high gain design in RF phase can pick weak signal (-90 dBm) but
it also limits the input power range, and the maximum unsaturated input power of
Image rejection ratio is simulated with both RF (5.25 GHz) and Image (4.75 GHz)
signals sending into receiver with same amplitude. The output of these two frequency
signal is shown in Fig 11.2.3, and proposed receiver image rejection is calculated as
35.63 dB with Eq 11.1.1. Even though, this result is not comparable with some latest
publications, considering the proposed receiver using IF band and rejecting image
implemented based on I-Q signal format, so these two signals can also be used to
132
From Fig 11.2.4, the minimum and maximum power input result in output
amplitude of 11.5 mV and 195 mV, respectively. These two values can be converted as
4 bits and 8 bits using Eq 2.2.3 under the assumption of ADC in [25] connected to the
Fig 11.2.3 Proposed receiver output plots of RF (5.25 GHz) and image (4.75
GHz) input signal with image rejection of 35.63 dB.
133
Fig 11.2.4 Transient simulation results of proposed receiver design: (a) input
power of -90 dBm; (b) input power of -50 dBm.
11.3 Conclusion
In this chapter, two popular analog image rejection system: Hartley and Weaver
have been designed and simulated. Simulation results indicate that even though the
Weaver design provides better image rejection, the Hartley architecture is more suitable
for mobile WLAN applications, as it has low noise and low power consumption benefits.
A CMOS Hartley receiver is built with all circuits mentioned in previous chapters and
134
tested to verify the entire front-end chain performance. The proposed receiver is
compatible with 802.11ac standard, and it features low noise, high input power
135
XII. Conclusion and Future Works
12.1 Conclusion
standard is presented. This system uses active inductor to replace conventional on-chip
passive inductor for area saving and performance improving purpose. Hartley
Designed active inductor and active inductor-based band pass filter with better
quality factor and on-chip area efficiency than conventional passive design.
Designed a Phase Locked Loop system that outputs quadrature phase 5 GHz
Designed a receiver chain that suitable for mobile WLAN application, and
embedded active inductor-based band pass filter in the system to improve area
136
12.3 Future work
The calibration system introduced in Chapter VII still has room to be improved.
First, the circuitry needs off-chip instrument to deliver reference signals for system
operating. This requirement can increase the complexity and cost of calibration process,
and the chip needs to provide two extra pins for only compensation purpose. Second,
the calibration system cannot correct the center frequency shifting error of filter when
process corner. For future work, a built-in PLL can be used to generate the reference
signal on-chip and construct a fully-automatic calibration system for active inductor
based band pass filter. A better comparator can be designed that maintains fine
overcome process variation for active inductor based band pass filter. However, the gain
deviation due to process variation is another drawbacks that prevent the wide-scale use
of active filter. For the future work, a hybrid system that contains both frequency and
gain calibration can make the active band pass filter more attractive for on-chip circuit
design.
137
References
[1] J. K. K. L. S. P. a. E. K. D. Han, "A fully integrated dual band transceiver for IEEE
802.11a/b/g/j/n WLAN applications using hybrid up/down conversion architecture," in
Proceedings of 2010 IEEE International Symposium on Circuits and Systems, Paris, 2010.
[2] M. Sauter, From GSM to LTE-Advanced Pro and 5G: An Introduction to Mobile Networks
and Mobile Broadband, 3rd Edition, WILEY, 2017.
[3] I. Standard, "IEEE Standard for Information Technology - Telecommunications and
Information Exchange Between Systems - Local and Metropolitan Area Networks - Specific
Requirements - Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer
(PHY) Specificatio". Patent IEEE Std 802.11-2007 (Revision of IEEE Std 802.11-1999), 12
June 2017.
[4] D. M. Dobkin, RF Engineering for Wireless Networks, Newnes, 2011.
[5] M. S. Gast, 802.11ac: A Survival Guide, O'Reilly Media, Inc., 2013.
[6] H. M. L. M. S. Y. R. S. M. a. M. T. A. H. M. Shirazi, "A 980μW 5.2dB-NF current-reused
direct-conversion bluetooth-low-energy receiver in 40nm CMOS," in IEEE Custom
Integrated Circuits Conference (CICC), Austin, 2017.
[7] R. W. e. al., "A 2×2 WLAN and Bluetooth combo SoC in 28nm CMOS with on-chip WLAN
digital power amplifier, integrated 2G/BT SP3T switch and BT pulling cancelation," in 2016
IEEE International Solid-State Circuits Conference (ISSCC), San Francisco, CA, 2016.
[8] A. H. a. B. Razavi, "A Low-Power CMOS Receiver for 5 GHz WLAN," IEEE Journal of
Solid-State Circuits, vol. 50, no. 3, pp. 630-643, 2015.
[9] S. T. Y. e. al, "An 802.11a/b/g/n/ac WLAN Transceiver for 2×2 MIMO and simultaneous dual-
band operation with +29 dBm Psat integrated power amplifiers," in ESSCIRC Conference
2016: 42nd European Solid-State Circuits Conference, Lausanne, 2016.
[10] L. B. O. Miguel D. Fernandes, Wideband CMOS Receivers, Switzerland: Springer, Cham,
2015.
[11] A. D. F. M. a. C. D. K. Ture, "Area and Power Efficient Ultra-Wideband Transmitter Based
on Active Inductor," IEEE Transactions on Circuits and Systems II: Express Briefs, vol. 65,
no. 10, pp. 1325-1329, 2018.
[12] Z. W. J. X. a. N. M. A. L. Ma, "A High-Linearity Wideband Common-Gate LNA With a
Differential Active Inductor," IEEE Transactions on Circuits and Systems II: Express Briefs,
vol. 64, no. 4, pp. 402-406, 2017.
[13] S. H. H. L. a. C. K. J. Song, "A 1-V 10-Gb/s/pin Single-Ended Transceiver With Controllable
Active-Inductor-Based Driver and Adaptively Calibrated Cascaded-Equalizer for Post-
LPDDR4 Interfaces," IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 65,
no. 1, pp. 331-342, 2018.
[14] J. Y. a. Y. Kim, "Global clock distribution on standing wave with CMOS active inductor
loading," in IEEE 60th International Midwest Symposium on Circuits and Systems
(MWSCAS), Boston, MA, 2017.
138
[15] S. Ren and B. Chris, "RF CMOS active inductor band pass filter with post fabrication
calibration," International Journal of Electronics and Communications, pp. 1058-1067, 2013.
[16] S. yadala and N. J. Sadrusham, "Process-Voltage-Temperature (PVT) Variations and Static
Timing Analysis," [Online]. Available: http://asic-soc.blogspot.com/2008/03/process-
variations-and-static-timing.html.
[17] M. Onabajo and J. Silva-Martinez, Analog Circuit Design for Process Variation-Resilient
Systems-on-a-Chip, Springer-Verlag New York, 2012.
[18] M. Y. R. K. A. N. S. H. G. Momen, "A New High Performance CMOS Active Inductor," in
2016 39th International Conference on Telecommunications and Signal Processing (TSP),
Vienna, 2016.
[19] A. K. M. S. D. Ben Issa, "An UWB pulse generator using switching CMOS active inductor
oscillator," in 2016 13th International Multi-Conference on Systems, Signals & Devices
(SSD), Leipzig, 2016.
[20] M. M. M. M. K. B. Sehmi Saad, "A New Low-power, High- Q , Wide Tunable CMOS Active
Inductor for RF Applications," IETE Journal of Research, vol. 62, no. 2, pp. 265-273, 2016.
[21] A. Zolfaghari, Low-Power CMOS Design for Wireless Transceivers, Boston, MA: Springer,
2003.
[22] M. A. I. E. a. S. H. K. Embab?, "Automatic Mismatches Calibration in Hartley Image-Reject
Receiver," pp. 63-66, 2003.
[23] F. Ellinger, "Radio Frequency Integrated Circuits and Technologies".
[24] H. O. F.E. Idachaba, "Analysis of a Weaver, Hartley and Saw- Filter Based, Image Reject
Architectures for Radio Receiver Design," 2012.
[25] S. H. a. B. Razavi, "A 7.1-mW 1-GS/s ADC with 48-dB SNDR at Nyquist rate," in
Proceedings of the IEEE 2013 Custom Integrated Circuits Conference, San Jose, CA, 2013.
[26] S. Voinigescu, High-Frequency Integrated Circuits, Cambridge University Press, 2013.
[27] P. E. A. a. D. R. Holberg, CMOS Analog Circuit Design, Oxford University Press, 2011.
[28] R. H. Caverly, CMOS RFIC Design Principles, Artech House, 2007.
[29] X. Z. H. X. S. R. Shuo Li, "On-chip self-calibration system for CMOS active inductor band
pass filter," AEU - International Journal of Electronics and Communications, vol. 92, pp. 64-
68, 2018.
[30] K. Lacanette, "A Basic Introduction to Filters—Active, Passive, and Switched-Capacitor,"
National Semiconductor, 1991.
[31] A. Sunca, O. Cicekoglu and G. Dundar, "A Wide Tunable Bandpass Filter Design Based on
CMOS Active Inductor," in Ph.D. Research in Microelectronics and Electronics (PRIME),
2012 8th Conference, Aachen, Germany, 2012.
[32] H. Xiao and R. Schaumann, "A 5.4-GHz high-Q tunable active-inductor band pass filter in
standard digital CMOS technology," Analog Integrated Circuit Signal Processing, pp. 51:1-
9, 2007.
[33] H. Zumbahlen, Linear Circuit Design Handbook, Newnes, 2008.
[34] B. D. H. Tellegen, "The gyrator, a new electric network element," Philips Res. Rept, no. 3, pp.
139
81-101, 10 November 1948.
[35] F. Yuan, CMOS Active Inductors and Transformers: Principle, Implementation, and
Applications. 1st Edition, Springer Publishing Company, 2010.
[36] A. S. C. Psychalinos, "Current amplifier based grounded and floating inductance simulators,"
AEU - International Journal of Electronics and Communications, vol. 60, pp. 168-171, 2006.
[37] H. X. X. Z. a. S. R. S. Li, "A low power CMOS amplitude peak detector for on-chip self-
calibration applications," in IEEE National Aerospace and Electronics Conference
(NAECON), Dayton, 2017.
[38] R. V. J. S.-M. a. E. S.-S. A. Valdes-Garcia, "A Broadband CMOS Amplitude Detector for On-
Chip RF Measurements," IEEE Transactions on Instrumentation and Measurement, vol. 57,
no. 7, pp. 1470-1477, 2008.
[39] F. J. a. H. Olsson, "RF detector for on-chip amplitude measurements," Electronics Letters,
vol. 40, no. 20, pp. 1239-1240, 2004.
[40] H.-H. H. a. L.-H. Lu, "Integrated CMOS power sensors for RF BIST applications," in 24th
IEEE VLSI Test Symposium, Berkeley, CA, 2006.
[41] X. Z. S. R. Shuo Li, "High Frequency Unity Gain Buffer in 90-nm CMOS Technology,"
Journal of Circuits, Systems and Computers, vol. 25, no. 07, 2016.
[42] S. H. L. a. T. R. V. G. Xing, "Self-Biased Unity-Gain Buffers With Low Gain Error," IEEE
Transactions on Circuits and Systems II: Express Briefs, vol. 56, pp. 36-40, 2009.
[43] A. L. F. a. C. A. d. R. Filho, "A -60dB THD/100MHz true unity-gain voltage buffer CMOS
circuit," in SBCCI '10 Proceedings of the 23rd symposium on Integrated circuits and system
design, 2010.
[44] H. R. a. J. Kostamovaara, "On the Performance and Use of an Improved Source-Follower
Buffer," IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 54, pp. 504-517,
2007.
[45] G. P. a. S. P. G. Palmisano, "High-performance and simple CMOS unity-gain amplifier," IEEE
Transactions on Circuits and Systems I: Fundamental Theory and Applications, no. 47, pp.
406-410, 2000.
[46] A. L. a. H. Y. Y. Kong, "A highly linear low-voltage source-degeneration transconductor based
on unity-gain buffer," Tsinghua Science and Technology, vol. 14, pp. 698-702, 2009.
[47] K.-C. Kuo and A. Leuciuc, "A linear MOS transconductor using source degeneration and
adaptive biasing,," Circuits and Systems II: Analog and Digital Signal Processing, IEEE
Transactions on, vol. 48, pp. 937-943, 2001.
[48] C. S. E. S.-S. J. H. J. Wang, "Built-in self optimization for variation resilience of analog
filters," in 2015 IEEE Computer Society Annual Symposium on VLSI, 2015.
[49] Z. L. a. V. Kursun, "Charge Recycling Between Virtual Power and Ground Lines for Low
Energy MTCMOS," in 8th International Symposium on Quality Electronic Design
(ISQED'07), San Jose, CA, 2007.
[50] J. E. R. S. Saiyu Ren, "Design and performance of a robust 180 nm CMOS standalone VCO
and the integrated PLL," Analog Integrated Circuits and Signal Processing, vol. 68, no. 3, pp.
285-298, 2011.
140
[51] M. A. E. N. A. J. F. Tayebeh Azadmousavi, "A novel zero dead zone PFD and efficient CP for
PLL applications," Analog Integr Circ Sig Process , vol. 95, no. 1, pp. 83-91, 2018.
[52] W. P. X. Z. B. L. L. Y. a. B. L. P. Li, "Image-Free Microwave Photonic Down-Conversion
Approach for Fiber-Optic Antenna Remoting," IEEE Journal of Quantum Electronics, vol.
53, no. 4, pp. 1-8, 2017.
[53] W. Gao, Energy and Bandwidth Efficient Wireless Transmission, Springer International
Publishing, 2017.
[54] F. E. Idachaba and H. Orovwode, "Analysis of a Weaver, Hartley and Saw-Filter Based, Image
Reject Architectures for Radio Receiver Design," Advanced Materials Research, vol. 367, pp.
199-204, 2012.
[55] H. Uhrmann, R. Kolm and H. Zimmermann, Analog Filters in Nanometer CMOS, Springer-
Verlag Berlin Heidelberg, 2014.
[56] N. I. Margaris, Theory of the Non-linear Analog Phase Locked Loop, Springer-Verlag Berlin
Heidelberg, 2004.
[57] M. J. Pelgrom, Analog-to-Digital Conversion, Springer-Verlag New York, 2013.
[58] S. Majzoub, R. Saleh and R. Ward, "PVT variation impact on voltage island formation in
MPSoC design," in Quality of Electronic Design, 2009. ISQED 2009. Quality Electronic
Design, San Jose, CA, 2009.
[59] S. Birla, N.Shukla, K. Rathi, R. Singh and M. Pattanaik, "Analysis of 8T SRAM Cell at
Various Process Corners at 65 nm Process Technology," Circuits and Systems, pp. 326-329,
2011.
[60] M. Wirnshofer, Variation-Aware Adaptive Voltage Scaling for Digital CMOS Circuits,
Springer, 2013.
[61] M. Onabajo and J. Silva-Martinez, Analog Circuit Design for Process Variation-Resilient
Systems-on-a-Chip, Springer-Verlag New York, 2012.
[62] G. Xing, S. Lewis and T. Viswanathan, "Self-Biased Unity-Gain Buffers With Low Gain
Error," Circuits and Systems II: Express Briefs, IEEE Transactions on, vol. 56, pp. 36-40,
2009.
[63] A. L. Fortunato and C. A. d. R. Filho, "A -60dB THD/100MHz true unity-gain voltage buffer
CMOS circuit," in SBCCI '10 Proceedings of the 23rd symposium on Integrated circuits and
system design, 2010.
[64] G. Palmisano, G. Palumbo and S. Pennisi, "High-performance and simple CMOS unity-gain
amplifier," Circuits and Systems I: Fundamental Theory and Applications, IEEE Transactions
on, vol. 47, pp. 406-410, 2000.
[65] F. Yuan, CMOS Active Inductors and Transformers: Principle, Implementation, and
Applications, Springer, 2008.
[66] A. D. F. M. C. D. Kerim Ture, "Area and Power Efficient Ultra-Wideband Transmitter Based
on Active Inductor," IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS
BRIEFS, vol. 65, no. 10, pp. 1325-1329, 2018.
[67] R. S. H. Xiao, "A 5.4-GHz high-Q tunable active-inductor bandpass filter in standard digital
CMOS technology," Analog Integrated Circuits and Signal Processing, vol. 51, no. 1, pp. 1-
141
9, 2007.
[68] F. K. S. M. Abdullah Yesil, "Electronically controllable bandpass filters with high quality
factor and reduced capacitor value: An additional approach," AEU - International Journal of
Electronics and Communications, vol. 70, pp. 936-943, 2016.
[69] L. G. C. Andriesei, "On the tuning possibilities of an RF bandpass filter with simulated
inductor," in 2007 International Semiconductor Conference, 2007.
[70] R. S. Haiqiao Xiao, "A 5.4-GHz high-Q tunable active-inductor bandpass filter in standard
digital CMOS technology," Analog Integrated Circuits and Signal Processing, vol. 51, no. 1,
pp. 1-9, 2007.
142
`