Transcript
8/2/2019 JAIN MOHIT
1/70
UNIVERSITY OF CINCINNATI
Date: ___________________
I, _________________________________________________________,
hereby submit this work as part of the requirements for the degree of:
in:
It is entitled :
This work and its defense approved by:
Chair: _______________________________ _______________________________
_______________________________ _______________________________ _______________________________
August 5, 2005
Mohit Jain
Masters of Science
Electrical Engineering
Design Procedures for Sigma Delta Modulators
Dr. Joseph H. Nevin
Dr. Carla Purdy
Dr. Fred R. Beyette
8/2/2019 JAIN MOHIT
2/70
Design Procedures for Sigma Delta Modulators
A thesis submitted to the
Division of Research and Advanced Studies,University of Cincinnati
in partial fulfillment of the requirements for the degree of
MASTER OF SCIENCE(Electrical Engineering)
in the
Department of Electrical and Computer Engineering and Computer Science,College of Engineering, University of Cincinnati
August 2005
by
Mohit JainB.E (Electronics and Instrumentation),
Institute of Engineering & Technology, Devi Ahilya University, Indore, India,
June 2002
Thesis Advisor and Committee Chair: Dr Joseph H. Nevin
8/2/2019 JAIN MOHIT
3/70
ii
Abstract
With the emergence of complex, non-linear mixed signal systems it becomes important to devisestrategies to model them accurately at the system level. Though advanced CAD tools are
available for digital design, analog design is still a kind of empirical art which mainly involves
selection of architecture, determining specifications for individual analog blocks and
implementation of the system with minimization of non-idealities.
For a long time, most of the building block specifications and non-idealities were explored at the
transistor level which involved a lot of time for each run and multiple runs to achieve the
required specifications. To explore a new method for modeling of a mixed signal circuit is the
main point of investigation of this thesis. In order to explore the new method of modeling, a
Sigma-Delta A/D converter is considered. These converters are gaining popularity as they are
insensitive to circuit imperfections and component mismatch. Design of Switched-Capacitor
(SC) Sigma-Delta modulators requiring optimization of a large set of parameters has been
explored and a top-down method for design of such systems is presented. The research work
deals in detail with the modeling of the converter, choosing the appropriate analog blocks at the
transistor level and various trade offs in implementation of the same.
8/2/2019 JAIN MOHIT
4/70
iii
8/2/2019 JAIN MOHIT
5/70
iv
Dedicated to My Parents
8/2/2019 JAIN MOHIT
6/70
v
Acknowledgements
I would like to extend my heartfelt gratitude to my thesis advisor Dr. Joseph H. Nevin for his
continuous guidance and encouragement in getting the best out of this research work. It would
not have been possible to move ahead without his brilliant insight in analog circuit design always
stayed as my source of motivation to delve deeper in to the subject.
I would also like to thank Dr. Carla Purdy and Dr. Fred R. Beyette Jr. for being on the technical
committee for my thesis defense and for providing critical suggestions to make this document
better. I am very thankful to my lab mates Sanjith, Lakshminarayanan and Ravikanth, for their
constant inputs to problems encountered during this work.
I am very thankful to my friends in Cincinnati especially Aditya, Kalyan, Hetal, Pranay, Pratima,
Rajasundaram, Rashmi, Ratan, Vasant, Vishal and Vishesh for their support all through the
Masters program and making my stay at Cincinnati a memorable one.
Last but not the least; I would like to thank my parents and my sister for their love, constant
support and encouragement. I am very grateful to my parents who have been a source of
inspiration for me and without their sacrifices and efforts this work would have been impossible.
8/2/2019 JAIN MOHIT
7/70
vi
TABLE OF CONTENTS
1. Introduction............................................................................................................................. 11.1. Introduction..................................................................................................................... 11.2. Research Motivation and Objective ................................................................................ 21.3. Thesis Outline ................................................................................................................. 3
2. Fundamentals of Sigma-Delta Data Converters ..................................................................... 42.1. Introduction..................................................................................................................... 42.2. Quantitative analysis of Analog to Digital Conversion.................................................. 42.3. Limitations of Nyquist-rate ADC ................................................................................... 52.4. Oversampling as Compared to Nyquist rate sampling ................................................... 62.5. Sigma Delta Modulators and Noise Shaping.................................................................. 7
2.5.1. First and Second order Sigma Delta Modulator ..................................................... 92.6. Architectures of Sigma Delta Modulators .................................................................... 12
2.6.1. Single loop (Low and High order) Modulators..................................................... 122.6.2. Multi-stage (Cascaded) Modulators...................................................................... 132.6.3. Multi-bit Modulators............................................................................................. 14
2.7. Non idealities in SDM .................................................................................................. 152.7.1. Circuit Noise ......................................................................................................... 162.7.2. Operational Amplifier Non- idealities ................................................................... 17
2.7.2.1. Limited DC Gain........................................................................................... 172.7.2.2. Limited Bandwidth and Slew Rate ............................................................... 182.7.2.3. Saturation...................................................................................................... 18
2.7.3. Comparator Offset and Hysteresis ........................................................................ 183. System Level (Behavioral) Modeling of ? ? Modulator ...................................................... 20
3.1. Introduction................................................................................................................... 203.2. Modeling Approaches................................................................................................... 213.3. Design Methodology..................................................................................................... 23
3.3.1. Topology Selection: .............................................................................................. 233.3.2. Signal Scaling ....................................................................................................... 263.3.3. Model for switching noise .................................................................................... 273.3.4. Model for op-amp noise ........................................................................................ 293.3.5. Model for limited DC gain.................................................................................... 303.3.6. Model for limited BW and SR.............................................................................. 31
4. Circuit Design and Details .................................................................................................... 324.1. Introduction................................................................................................................... 324.2. Topology selection........................................................................................................ 324.3. Integrator design ........................................................................................................... 35
4.3.1. Clock generator..................................................................................................... 384.3.2. Fully Differential Folded-Cascode Op Amp......................................................... 39
8/2/2019 JAIN MOHIT
8/70
vii
4.3.3. Bias circuit ............................................................................................................ 414.3.4. Common-Mode Feedback Circuit......................................................................... 44
4.4. Regenerative Track and Latch Comparator design....................................................... 454.5. Simulation..................................................................................................................... 47
5. Conclusion ............................................................................................................................ 51
APPENDIX A........................................................................................................................... 53APPENDIX B........................................................................................................................... 55BIBLIOGRAPHY......................................................................................................................... 57
8/2/2019 JAIN MOHIT
9/70
viii
LIST OF TABLES
Table 3.1: Comparison of several modeling techniques............................................................... 22Table 4.1: Transistor sizes for Folded Cascode Op Amp ............................................................. 40Table 4.2: Transistor W/L ratios for Bias Circuit (R B = 3k) ........................................................ 43
8/2/2019 JAIN MOHIT
10/70
ix
LIST OF FIGURES
Figure 2.1: Quantization noise PSD for Nyquist rate and Oversampled converters ...................... 7Figure 2.2: Block diagram of Sigma Delta Modulator ................................................................... 8Figure 2.3: Linearized model of a Sigma Delta Modulator ............................................................ 8Figure 2.4: First order Sigma Delta Modulator .............................................................................. 9Figure 2.5: Second order Sigma Delta Modulator ........................................................................ 10Figure 2.6: Magnitude vs Frequency plot for first, and higher order ? ? modulators .................. 11Figure 2.7: Multi-stage (Cascaded) ? ? modulator ....................................................................... 14Figure 3.1: Theoretical SNR of a (1 z -1 )n modulator ................................................................. 24
Figure 3.2: Maximum SNR achievable by modulators of order n with coincident zeroes........... 25Figure 3.3: Second order sigma-delta modulator using Simulink ................................................ 25Figure 3.4: Effect of scaling on integrator outputs ....................................................................... 26Figure 3.5: Simulink model of Switching Noise .......................................................................... 28Figure 3.6: Effect of kT/C Noise on PSD of ideal sign wave ....................................................... 29Figure 3.7: Model for op-amp noise ............................................................................................. 29Figure 3.8; Model for limited op-amp gain .................................................................................. 30Figure 3.9: Effect of gain factor on SNR...................................................................................... 30Figure 3.10: Op-Amp flow chart .................................................................................................. 31Figure 4.1: Modified second order modulator architecture .......................................................... 33Figure 4.2: Variation of SNR with a1 and a2 ............................................................................... 34
Figure 4.3: Output density at Integrator 1 and 2 for a1 = 0.5 and a2 varying from 0.1 to 0.9...... 34Figure 4.4: Output density at Integrator 1 and 2 for a1 = 0.4 and a2 varying from 0.1 to 0.9...... 35Figure 4.5: Parasitic insensitive switched-capacitor integrator ................................................. 36Figure 4.6: Two phase non-overlapping clocking scheme ........................................................... 37Figure 4.7: Fully differential implementation of switched capacitor integrator........................... 38Figure 4.8: Schematic for non-overlapping clock generation....................................................... 38Figure 4.9: Fully differential folded-cascode Op Amp................................................................. 40Figure 4.10: Conventional cascode mirror and Wide-swing cascode current mirror ................... 42Figure 4.11: Bias Circuit............................................................................................................... 43Figure 4.12: Switched capacitor common-mode feedback (CMFB) circuit................................. 45Figure 4.13: Regenerative feedback comparator .......................................................................... 47
Figure 4.14: AC analysis of folded cascode Op Amp showing gain and phase margin............... 49Figure 4.15: Comparator Output .................................................................................................. 49Figure 4.16: FFT (2 14 ) magnitude plot for sigma-delta modulator output.................................... 50Figure 4.17: PSD plot for sigma-delta modulator output ............................................................. 50
8/2/2019 JAIN MOHIT
11/70
x
8/2/2019 JAIN MOHIT
12/70
1
1. Introduction
1.1. Introduction
The emergence of powerful digital signal processing for telecommunication and multimedia
applications implemented in CMOS VLSI technology creates the need for high- resolution
analog-to-digital converters that can be integrated in fabrication technologies optimized for
digital circuits and systems. However, the same scaling of VLSI technology that makes possible
the continuing dramatic improvements in digital signal processor performance also severely
constrains the dynamic range available for implementing the interfaces between the digital and
analog representation of signals.
Conventional converters are becoming increasingly difficult to implement as they need precise
analog components in the ir filters and conversion circuits and such conversion circuits can be
very vulnerable to noise and interference. Oversampling converters on the other hand make
extensive use of digital signal processing, with the fact that fine- line VLSI is better suited for
providing fast digital circuits than for providing precise analog circuits.
Oversampled A/D converters based on delta-sigma modulation combine sampling at rates well
above the Nyquist rate with negative feedback and digital filtering in order to exchange
resolution in time for that in amplitude. These converters are especially insensitive to circuit
8/2/2019 JAIN MOHIT
13/70
2
imperfections and component mismatch, and therefore provide a means of exploiting the
enhanced density and speed of scaled digital VLSI circuits so as to avoid the difficulty of
implementing complex analog circuit functions within a limited analog dynamic range.
1.2. Research Motivation and Objective
Choosing the specifications for an analog circuit for given design objectives is potentially a very
complicated and time consuming process. This task is further made difficult by the fact that the
number of specifications to deal with is usually very large, and varies over wide ranges from one
application to another.
Design of an analog system primarily consists of three obstacles: [1]
Selection of architecture.
Determining the specifications of the analog building blocks necessary to implement the
chosen architecture.
Minimization of the effects of circuit non- idealities.
For a long time, most of the building block specifications and non-idealities were explored at the
transistor level. This required simulation over the full range of these variations which took
multiple simulations even for single architectures. With this method, the entire procedure must
be carried out for each of the architectures selected, which accounts for a lot of time too.
8/2/2019 JAIN MOHIT
14/70
3
With the constant changes in technology, it is more amenable to consider a design process that is
independent of technology. In present day System-on Chip design there is an increasing demand
for systematic top-down design methodology. This methodology enables the designer to choose
the right parameters for the individual building blocks constituting the system, starting from the
system specifications.
In the design of high-resolution Switched-Capacitor (SC) Sigma-Delta modulators we typically
have to optimize a large set of parameters, including the performance of the building blocks, in
order to achieve the desired signal-to-noise ratio. This work aims to suggest a designmethodology which the designer can use to tackle these large parameters.
1.3. Thesis Outline
Chapter 2 discusses the concepts and architecture of Sigma-Delta modulators long with a brief
introduction to other conversion methods. This chapter further explores various noise sources
that are encountered in practical implementation of the Sigma-Delta modulators and their effect
on the performance. Chapter 3 deals with various design methodologies that currently exist and
proposes Simulink models for modeling of electronic systems. The chapter then extends to
derive the Simulink models for sigma-delta modulators. Chapter 4 presents circuit design issues
of various building blocks for these modulators and aims at designing a sigma-delta modulator
for prescribed performance using the methodology developed in previous chapters. A conclusion
chapter gives some discussion on the future work.
8/2/2019 JAIN MOHIT
15/70
4
2. Fundamentals of Sigma-Delta Data Converters
2.1. Introduction
Analog-to-digital interfaces are becoming increasingly important as translators between the real
analog world and efficient digital processing systems. A/D converters have existed for along
time and a number of implementation techniques for them have been developed, but in the recent
times sigma-delta modulators have gained popularity. This chapter examines what makes sigma-
delta modulators different from other conversion techniques. Merits and de-merits of the
different topologies and techniques used for high performance sigma-delta modulators are also
discussed. The chapter ends with a discussion of various non-idealities encountered in practical
implementation of the modulator which degrade the modulator performance.
2.2. Quantitative analysis of Analog to Digital Conversion
Analog to Digital conversion of a signal requires two operations, namely Sampling and
Quantization. In the sampling process a continuous time signal is sampled at uniformly spaced
times intervals (T s). The signal can be reconstructed back to the continuous time provided the
sampling frequency (f s) is at least twice the bandwidth of the signal (f b) as per Nyquist sampling
theorem.
8/2/2019 JAIN MOHIT
16/70
5
Quantization refers to the process of mapping an infinite number of input amplitude values to a
finite number of output amplitude values. Thus Quantization error refers to the difference
between the original analog amplitude and the quantized digital amplitude. The quantizer in any
ADC is a non- linear system. To make its analysis tractable we can linearize it by modeling it as a
noise source e[n] which is added to sampled signal x[n], to produce the quantized output signal
y[n], i.e.
y[n] = x[n] + e[n] Equation 2-1
If x[n] is active, e[n] can be approximated as an independent random number uniformly
distributed between ?/2[2]. The quantization noise may be treated as white noise with power:
121 22 /
2 /
2 == +
deePe Equation 2-2
This is independent of frequency, f s. Also the spectral density of e is white (i.e. a constant over
frequency) and all of its power folds into the frequency band f s / 2. Then the spectral density of
sampled noise is given by
ss
e
f f P
f E 1
12)(
== Equation 2-3
2.3. Limitations of Nyquist-rate ADC
Performance of Nyquist rate converters is limited by the technology in which they are fabricated
as the resolutions of these converters rely on the matching between the components. For an N-bit
ADC, the matching should be at least 1/2 N. However, matching of components to greater than 10
8/2/2019 JAIN MOHIT
17/70
6
bits is difficult in any normal CMOS process technology. Hence, high resolution data converters
are extremely difficult to attain without the use of techniques such as trimming of components or
calibration.
The case where f s=2f b is known as Nyquist rate sampling and the anti-aliasing filter before the
ADC should have a very sharp cut-off. This is a huge drawback especially if these ADCs are
used in receiver chains as the design and integration of such analog filters is no n-trivial.
2.4. Oversampling as Compared to Nyquist rate sampling
Oversampled conversion is a technique that improves the resolution obtained from
straightforward Nyquist rate conversion. Oversampling refers to acquiring signals from the
analog waveform at a rate significantly higher than the Nyquist rate. Each of these samples is
quantized by an N bit ADC. Since quantization is described by equation 2-1, the total amount of
noise power injected into the sampled signal, x[n] is given by equation 2-2. This is the same
noise power produced by a Nyquist rate converter but its frequency distribution is different
because of the higher sampling rate. In the oversampled case the noise power is uniformly
distributed between f s /2 to f s /2 .
As shown in Figure 2.1 the noise power in the oversampled case has been spread over a
bandwidth equal to sampling frequency f s which is much greater than the signal bandwidth. Only
a relatively small fraction of noise power falls in the band [ -f b f b], and the noise power outside
signal band can be greatly attenuated with a digital low-pass filter following the ADC. After the
8/2/2019 JAIN MOHIT
18/70
7
low-pass filtering has been performed, the signal can be downsampled to the Nyquist rate
without affecting the signal to noise ratio. The collective operation of low-pass filtering and
down-sampling is known as decimation.
IEIE IVIV
1 \ TXLVW5 DWH&RQYHUVLRQ
2 YHUVDPSOHG&RQYHUVLRQ
3 H I
Figure 2.1: Quantization noise PSD for Nyquist rate and Oversampled converters
We define the oversampling ratio, OSR, as
b
s
f
f OSR
2= . Then the quantization noise power in-
band is reduced to [3, Chapter 14]
=OSR
Pe1
12
2
Thus every doubling of OSR decreases the quantization noise power by one-half, or equivalently
3 dB.
2.5. Sigma Delta Modulators and Noise Shaping
As seen in the section above, the in-band noise can be substantially reduced by oversampling.
Sigma-delta converters go a step ahead and use feedback in addition to oversampling to further
8/2/2019 JAIN MOHIT
19/70
8
suppress in-band quantization noise. This can be accomplished by what is termed as Noise
Shaping, wherein the noise transfer function is made to have a high-pass filter characteristic.
This decreases the in-band quantization noise but amplifies the noise outside the signal
bandwidth. The noise power outside the bandwidth can then be removed by the digital
decimation filter.
Figure 2.2 shows a block diagram of sigma-delta modulator and the linearized model of the same
are shown in Figure 2.3. The input to the circuit feeds to the quantizer via an integrator, and the
quantized output feeds back to subtract from the input signal. The feedback forces the averagevalue of the quantized signal to track the average input. Any persistent difference between them
accumulates in the integrator and eventually corrects itself. The integrator is modeled as a
function H(z) and quantizer as a noise source e(n).
Figure 2.2: Block diagram of Sigma Delta Modulator
Figure 2.3: Linearized model of a Sigma Delta Modulator
8/2/2019 JAIN MOHIT
20/70
9
From the linear model in the figure above we can derive output
)(.)(1
1)(.
)(1)(
)( z E z H
z X z H
z H zY
++
+=
Thus the signal experiences a different transfer function than the error.
2.5.1. First and Second order Sigma Delta Modulator
To realize a first order noise shaping, we make1
1)(
=
z z H , Giving us a signal transfer function
STF (z) = z-1 and Noise transfer function N TF (z) = 1- z -1 . Thus the input appears without any
change at the output (but with a delay) and noise goes through a high-pass filter. A block
diagram of such a choice is shown in Figure 2.4.
Figure 2.4: First order Sigma Delta Modulator
For this configuration, the in-band noise power can be calculated as [3, chapter 14]
32' 1
3
OSR
PP ee
And corresponding SNR as
SNR max = 6.02N + 1.76 5.17 + 30log (OSR)
8/2/2019 JAIN MOHIT
21/70
10
Where N refers to the number of bit the quantizer has. Here we see that doubling the OSR gives
an SNR improvement of 9dB as compared to 3dB for oversampling without noise shaping.
Though easy to implement, the quantization noise from a first-order modulator is highly
correlated [4] and the oversampling ratio needed to achieve resolution greater than 12 bits is
prohibitively large.
The second order sigma-delta modulator is a widely used one. This modulator has a noise
transfer function which is a second-order high pass function. For this modulator the signaltransfer function S TF (z) = z -1 and the noise transfer function is given by NTF(z) = (1 z -1 )2. The
block diagram implementing these functions is shown in Figure 2.5.
Figure 2.5: Second order Sigma Delta Modulator
In a way similar to above, it can be shown that in-band noise power and SNR are,
54' 1
5
OSR
PP ee
SNR max = 6.02N + 1.76 12.9 + 50log (OSR)
Thus, we see here that doubling the OSR improves the SNR for a second-order modulator by
15dB.
8/2/2019 JAIN MOHIT
22/70
11
In order to achieve high resolution by pushing more quantization noise outside the signal band,
higher order NTFs have to be implemented. shows a magnitude vs. frequency plot for first-order
to fourth-order NTF. As can be observed the in-band error decreases but the out-of-band error is
amplified. The next section deals with various architectures that can be used to implement higher
order NTFs.
Figure 2.6: Magnitude vs Frequency plot for first, and higher order ? ? modulators
8/2/2019 JAIN MOHIT
23/70
12
2.6. Architectures of Sigma Delta Modulators
With the foregoing discussion we can conclude that sigma-delta modulators give us 3 degrees of
freedom, namely, the quantization level (N), the OSR and the order of the modulator. Based on
these degrees of freedom, different modulator topologies can be obtained. Sigma-delta
modulators can be classified into two broad groups: single-loop and multi-loop cascaded
modulators.
2.6.1.
Single loop (Low and High order) Modulators
Low order sigma-delta modulators, the first and second order discussed above, offer advantage
of having guaranteed stability, have a simple loop filter design and simple circuit. But these
modulators have low SNR and very high oversampling ratios are required for achieving high
SNR, at the same time they are more prone to idling tones. Idle tones are stable sequences output
by a nonlinear network in the prolonged absence of input; they are manifested as multiple high-
amplitude single-frequency components in the output of the converter.
In principle, arbitrary higher-order loop filters can be configured by connecting more integrators.
The high-order modulator is expected to have a high SNR even for modest oversampling ratios,
provide better immunity to idling tones and, as they involve simply expanding the configuration,
they have a simple circuit design. However the stability of these loops becomes precarious for
loop filters of order greater than 2. Linearized analysis, which results in N TF(z) = (1 z -1 )L, is not
a reliable predictor of the stability, since the 1-bit quantizer is a grossly nonlinear element whose
equivalent gain varies abruptly with the value of its input. It is shown [5, chapter 3] that for
8/2/2019 JAIN MOHIT
24/70
13
guaranteed stability the equivalent quantizer gain must be high. Thus, with the fixed output
amplitude, this will be achieved only if the quantizer input is small. To ensure this, the maximum
amplitude of input must be restricted to fairly low values, so as to inversely decrease the
dynamic range. In practice, the loop filter need to be carefully designed and the stability may be
signal dependent.
2.6.2. Multi-stage (Cascaded) Modulators
The multi-stage sigma-delta modulator consist of first and/or second order modulators stages in
cascade as shown in Figure 2.7. The input to the first stage is the signal to be modulated. The
consecutive stages modulate the quantization error of the previous ones. The outputs of the
stages are suitably combined so that at the output of the modulator only the input signal and
signal-independent noise shaped by (1-z -1 )L are present. This design gives very high SNR and is
free of stability problems associated with the higher-order single stage converters as multi-stage
modulators inherit the excellent stability properties of their low-order stages.
As there are no stability constraints in their design, any increase of their order by cascading more
stages results to an increase of the SNR. The limit to this increase is the matching of the digital
and analog parts which requires complex switched-capacitor circuits. The component of the
output noise due to leakage of the lower-order shaped noise becomes relatively larger compared
to the increasingly higher-order shaped noise term of the final stage.
8/2/2019 JAIN MOHIT
25/70
14
1
1
1
z z
1
1
1
z z
1
1
1
z
z
Figure 2.7: Multi-stage (Cascaded) ? ? modulator
Y1 = z-1X + (1 - z -1 ) E1
Y2 = z-1 (-E1) + (1 - z -1 ) E2
Y3 = z-1 (-E2) + (1 - z -1 ) E3
Y = z -2Y1 + z -1 (1- z -1 ) Y2 + (1- z -1 )2 Y3
Y = z -3X + (1 - z -1 )3E3
2.6.3. Multi-bit Modulators
The above mentioned configurations can be implemented using multi-bit quantizers; employing
a multi-bit quantizer can increase SNR by 6 dB for every doubling in the number of quantizer
levels. Thus, a multi-bit modulator can achieve resolution comparable to that of a single-bit
modulator at a lower sample rate, which is a significant advantage in applications requiring high
8/2/2019 JAIN MOHIT
26/70
15
bandwidth such as digital video. Furthermore, if multi-bit quantizers are used in single-loop
structures, the system is more stable as the hard non-linearity of the quantizer is less pronounced
in this case. The extra quantization levels allow for large dither signals at the quantizer input,
hence eliminating idling tones.
There are some limitations to the applicability of multi-level quantization within sigma-delta
modulators. The primary limitation is that a multi-bit DAC needs to be incorporated along with
an ADC. While a single-bit feedback DAC is linear, the multi-bit DAC is limited by component
matching of its individual elements. Further, the error generated due to the DAC non- linearity isonly shaped by STF and hence will not be suppressed by the modulator noise-shaping.
Consequently the DAC linearity should be as high as the desired modulator resolution. This
becomes difficult to achieve without element trimming. Another limitation stems from the fact
that the digital decimation filter hardware following the multi-bit quantizer must allow for multi-
bit inputs. Different techniques have been presented that alleviate the linearity problem
encountered in multi-bit quantizer. These include like digital correlation [6], dynamic element
matching [7] and mismatch shaping [8] to suppress the nonlinear DAC noise.
2.7. Non idealities in SDM
The implementation of various sigma-delta modulators above assumed the use of ideal
components, which is not true in practice. There are several non-idealities which generate
additional noise in the practical sigma-delta modulator, on top of the quantization noise. This
section discusses various non idealities and how they limit the performance of a sigma-delta
8/2/2019 JAIN MOHIT
27/70
16
modulator. The proper modeling of these non-idealities can help predict the performance of the
modulator at the early stages of design. Here we only discuss the non-idealities and their
modeling is discussed in next chapter.
2.7.1. Circuit Noise
Intrinsic noise refers to noise that is generated in the device itself, as opposed to noise that
couples in from an external interfering source, Intrinsic noise cannot be eliminated by shielding,
filtering, or circuit layout, since it is a property of the device, but its value can be altered by
choice of circuit topology and component size. The two major sources of noise in sigma-delta
modulators are the thermal noise of the switches and the op-amp noise.
Thermal noise is caused by random fluctuation of carriers due to thermal energy and is present
even at equilibrium. Because of this, it needs to be taken into account for both the switches and
op-amps in a switched-capacitor circuit. Thermal noise has a white spectrum and a wide band,
limited only by the time constants of the switched capacitors or the bandwidths of the op-amps.
As shown in [5, chapter 11] for a sampled circuit with a resistor R and capacitor C, the noise can
be found by modeling the resistor as having a noise source in series with a power source equa l to
the Johnson noise 4kTR? f. The total noise power can be found by evaluating the integral
( ) C
kT df
fRC
kTRe T
=+
=0
22
21
4
Where k is the Boltzmanns constant and T the absolute temperature. Thus we see that the
thermal noise is generated in the resistor, the total noise power depends only on the capacitor.
8/2/2019 JAIN MOHIT
28/70
17
The integrators of a switched capacitor modulator may include more than one input branch
(typically two, input and feedback), each contributing to the total noise power.
2.7.2. Operational Amplifier Non-idealities
A major component of a switched capacitor sigma-delta modulator is the integrator made using
an Op-Amp. An ideal integrator has infinite dc gain, infinite bandwidth, no slew rate limitation
and no saturation limits.
2.7.2.1. Limited DC Gain
The ideal integrator with infinite dc gain has the transfer function
1
1
1)(
=
z
z z H
But the dc is limited by the circuit constraints and hence causes integrator leak. The
consequence of this integrator leak is that only a fraction of P 0 of the previous output of the
integrator is added to each new input sample. The integrator transfer function in this case
becomes
10
10
1)(
= zP zg
z H
And the dc gain is H 0 = g0 /(1-P 0). The limited gain at low frequency reduces the attenuation of
the quantization noise in the baseband and consequently results in an increase of the in-band
quantization noise.
8/2/2019 JAIN MOHIT
29/70
18
2.7.2.2. Limited Bandwidth and Slew Rate
The integrators constructed from amplifiers with a single dominant pole are observed to have an
exponential impulse response. The time constant of the response, t, can be nearly as large as the
sampling period T. This constraint is considerably less stringent than requiring the integrator to
settle to within the accuracy of the A/D converter. Simulation results indicate that for values of t
larger than the sampling period, the modulator becomes unstable.
In typical sampled-data analog filters, the UGB of the operational amplifier must be at least an
order of magnitude greater than the sampling rate. However, simulations show that for sigma-
delta modulators implementation of the integrator using operational amplifiers with UGB
considerably lower than this, and with correspondingly inaccurate settling, will not impair the
sigma-delta modulator performance, provided that the settling process is linear.
2.7.2.3. Saturation
The limited output swing of Op-Amps is a major concern with decreasing bias voltages. This
requires use of modified architectures to make sure that the signal swing at the output of the
integrators is within the limits. This is further discussed in the next chapter.
2.7.3. Comparator Offset and Hysteresis
The impact of comparator non- idealities is much lower than those of the integrator due to the
position of the comparator in the modulator loop. The imperfection in the comparator can be
considered as another noise source which adds to e[n], the quantization error. However, this
8/2/2019 JAIN MOHIT
30/70
19
noise from the extra source is subjected to noise shaping by the modulator and so its affect on
SNR degradation is not significant.
8/2/2019 JAIN MOHIT
31/70
20
3. System Level (Behavioral) Modeling of ? ? Modulator
3.1. Introduction
Choosing specifications for an analog circuit is potentially a very complicated and time
consuming process. Thus, in a design it is better to perform analysis at the system architectural
level before starting transistor level design; this reduces the number of design iterations and
helps exploring the design options better. The ultimate goal being the low level circuit
parameters dictated by the selected architecture and desired performance.
Sigma-Delta Modulators being mixed-signal nonlinear circuits pose significant problem in
estimation of their performance. Hence in order to achieve the performance objectives, it is
required that various parameters describing the circuit performance be optimized carefully prior
to and during modulator design. For a systematic design of Sigma-Delta Modulator we can
enumerate the following tasks
Topology and accompanying topology parameter selection
Analysis of the impact of circuit non-idealities on the system performance and specificationof the resulting building block performances.
Design, optimization and simulation of the building blocks and the system followed by
performance evaluation.
8/2/2019 JAIN MOHIT
32/70
21
3.2. Modeling Approaches
Various approaches have been used for accurate modeling of Sigma-Delta Modulation which
includes device models, circuit macro-models, time-domain macromodels, finite-difference
equations, table-lookup models, harmonic balance methods and behavioral models [5, chapter
14].
Device models are small and large signal models of active devices used in a circuit simulation
such as SPICE, though accurate, they take extremely long simulation time. Circuit macromodels
as discussed in [9] are models of circuits made up of several active and passive components; it is
less complex that the original circuit and uses circuit specifications as model parameters. Though
this approach provides a good accuracy, the speed improvement with respect to device models is
poor. Time domain macromodels [10, 11] are based on a set of time-domain equations derived
for a specific circuit and use circuit specifications as control parameters but are designed strictly
for transient analysis and are generally not used with a circuit simulator.
Finite-difference equations are based on the z-transform function of sampled-data circuits. Using
difference equations results in small and efficient simulation programs, such as MIDAS [12] for
oversampled converters and SWITCHCAP2 for switched-capacitor circuits. They provide
quickest simulation of all methods but are poor in non- idealities modeling capabilities. Table-
lookup models use a two-step approach to modeling. The first step is to extract a table of input
and output points for the original circuit with the use of a high accuracy simulator followed by
8/2/2019 JAIN MOHIT
33/70
22
using the stored table of points instead of the original circuit for further transient simulations
[13]. This method seems limit the simulated performance of oversampled converters to 80 dB [5,
chapter 14].
Envelope following or harmonic distortion balance method [14], simulates clocked circuits in
which the waveform is similar from cycle to cycle. However, the states of an oversampled
converter can change significantly from cycle to cycle and the states are not periodic. Hence, the
simulation of oversampled converters is an inappropriate target for the harmonic balance
method. Table 13.1 gives comparison of various models [5, chapter 14]
Model Advantages Disadvantages
Device Models Based on device physics Too slow for discrete-time
systems
Circuit-based macromodels Can add model features that
approach device model accuracy
Only a factor of 10 speed
improvement over device
models
Time-domain macromodels Produces quick simulations; can
model dynamic errors
Not reusable because
equations depend on load and
feedback configuration
Difference equations Simulates the quickest of all Custom implementations not
directly reusable; cannot
model dynamic errors
Table-lookup models Acceptable speedup over device
models; can achieve accurate
modeling of static errors
Unclear how to optimize table
building; only model static
errors; size of table increases
with the square of the circuit
states; tables are not reusable
Table 3.1: Comparison of several modeling techniques
8/2/2019 JAIN MOHIT
34/70
23
Thus a simulation strategy that provides good speed of simulation, modeling capabilities and
reusability is needed. A way to do this is behavioral modeling where the circuit is split with
multiple nonlinearities into a set of linear circuits that are solved explicitly before simulations [5,
chapter 14]. During simulation, the program determines the correct linear circuit to use and
applies the pre-solved equation to calculate the next piece of the transient response.
This chapter proposes a top down design methodology for design of Sigma-Delta Modulators
which involves choosing the initial topology followed by behavioral modeling to selectappropriate circuit parameters. The program Simulink has emerged as a powerful tool for
mathematical analysis in recent times, but its use to simulate electronic circuits has been very
limited. The method used here uses Simulink for behavioral modeling of electronic components
and overcomes the disadvantages of the other methods by providing an accurate, fast and flexible
method with good user interface for simulating non- linear mixed signal like Sigma-Delta
Modulators.
3.3. Design Methodology
3.3.1. Topology Selection:
The following formula gives a relationship among SNR, OSR and Modulator order for a SDM
with a one-bit quantizer, assuming an input amplitude equal to the modulator output amplitude
and a perfect differential noise transfer function of (1 z -1 )2 [3].
SNR =
+ +12
2 .12
.23
10log10 L L M L
8/2/2019 JAIN MOHIT
35/70
24
It can be used to as a first step in order to determine the OSR and Modulator order that will be
required to obtain a particular SNR. Figure 3.1 shows the theoretical maximum SNR that could
be achieved using higher order modulators for various OSR; following it is Figure 3.2 which
shows the maximum SNR that can be achieved practically. The SNR can be improved a little by
spreading the NTF zeros across the band of interest in a manner that minimizes the in-band
noise. [5, chapter 4]
This formula needs to be used with a caution though as it gives a good indication for low OSR
values but with increase in OSR the validity of the formula decreases. Figure 3.1 suggests that aSNR of 160 dB would result from a 5 th order NTF at an oversampling ratio of 64, but Figure 3.2
shows that due to the limitation imposed by stability, the achievable SNR is 60 dB lower.
Figure 3.1: Theoretical SNR of a (1 z -1 )n modulator
8/2/2019 JAIN MOHIT
36/70
25
Figure 3.2: Maximum SNR achievable by modulators of order n with coincident zeroes
The next step is estimation of the maximum SNR for a given order and topology using
behavioral simulations. As mentioned earlier, Simulink has been used here to perform behavioral
simulations (Figure 3.3). This diagram clearly shows the advantage of this approach over others.
As could be seen the implementation is easy and all basic buildings are provided in the library.
The integrator has been implemented using a discrete filter block and comparator using the block
sign (output =1 for input > 0 else output = -1).
Figure 3.3: Second order sigma -delta modulator using Simulink
8/2/2019 JAIN MOHIT
37/70
26
3.3.2. Signal Scaling
Signal scaling refers to selection of gain factors in front of each integrator. It is necessary to
place an appropriate attenuation factor to avoid clipping at the integrator outputs for large input
signals. The goal of signal scaling is to maximize the dynamic range by using all of the available
swing without clipping [15]. Figure 3.4 shows the simulated output waveforms for a second
order modulator with and without scaling. As could be seen the outputs at integrators can grow
very large (which is impossible to achieve with the implementation) if scaling is not done.
Figure 3.4: Effect of scaling on integrator outputs
A one-bit quantizer has the property that the signal gain that comes before it is unimportant, as it
only detects the polarity. Improper scaling will have a tremendous impact on the dynamic range
of the modulator. It is shown [16] that if the input to the quantizer is distributed between Vqi
8/2/2019 JAIN MOHIT
38/70
27
and the output is 1, then the power of the quantization error is 1-Vqi+V 2qi and the distribution is
not uniform for Vqi ?1. This shows that pre-quantization gain affects the quantization error.
Different approaches have been proposed to solve this problem [17, 18, 19]. One pragmatic
approach is to consider a linearized quantizer model and adjust the quantizer gain such that the
total gain around the outermost feedback loop is unity. This empirical technique has little
rigorous justification but simply produces analytical results that agree with simulation.
Following are the criteria that should be kept in mind while deciding upon the gain coefficients:
All of the output swing of the integrators should be used without signal clipping.
The number of unitary capacitors used should be minimal.
The digital scaling should be easily implementable (applies for multistage and multibit
modulator).
Once the order and topology is determined, various non- idealities (both intrinsic and extrinsic)
that would be encountered in the implementation of the system can be incorporated to analyze
their effect on the performance.
3.3.3. Model for switching noise
The switch thermal noise is superimposed to the input voltage x(t) leading to
)(.4
)()( t nCsKT
t xt y +=
8/2/2019 JAIN MOHIT
39/70
28
The factor of 4 accounts for the two path through which noise is sampled (during F1 and F2)
and the fully differential structure of the circuit. The n(t) denotes a Gaussian random process
with unity standard deviation. If there are two input branches (to incorporate two different
coefficients, one for the signal and one for the feedback), each branch has to be modeled with a
separate kT/C noise.
The switching noise is modeled as shown in Figure 3.5 and is incorporated as an addition to the
input to the modulator.
Figure 3.5: Simulink model of Switching Noise
The effect of kT/C noise on the output PSD is shown in Figure 3.6. As could be seen the
introduction of this noise raises the output noise floor and as predicted by the equation the
increase of capacitance reduces the noise.
8/2/2019 JAIN MOHIT
40/70
29
Figure 3.6: Effect of kT/C Noise on PSD of ideal sign wave
3.3.4. Model for op-amp noise
The model shown in Figure 3.7 is used to simulate the effect of the operational amplifier noise.
Then V n represents the total RMS noise voltage of the operational amplifier referred to the input.
The effect of this noise on the output PSD is similar to the above shown kT/C noise.
Figure 3.7: Model for op-amp noise
8/2/2019 JAIN MOHIT
41/70
30
3.3.5. Model for limited DC gain
The following model in Figure 3.8 shows the implementation of practical a Op-Amp with limited
DC gain a. Figure 3.9 shows the effect of limited gain on the SNR for a second order modulator.
1
1
1)(
=
z
z z H
Figure 3.8; Model for limited op-amp gain
Figure 3.9: Effect of gain factor on SNR
8/2/2019 JAIN MOHIT
42/70
31
3.3.6. Model for limited BW and SR
The behavioral model of limited bandwidth and slew rate is implemented using the MATLAB
function which calculates output y depending on the input x and value of UGB and SR. The
following flow chart shows the algorithm for the same.
=1
2 UGB
t x
SR1 =
Ts t
( ) y x x x SR t eTs t
=
sgn( ). . 11
( ) y x e Ts= 1 / y x SR Ts= sgn( ). .
Figure 3.10: Op-Amp flow char t
8/2/2019 JAIN MOHIT
43/70
32
4. Circuit Design and Details
4.1. Introduction
The previous chapter discussed the new design methodology using Simulink and how the
required circuit performance matrices could be determined using the same. This chapter focuses
on the circuit design of the building blocks needed to implement the sigma-delta modulator.
These include switched-capacitor integrators, operational amplifiers, comparators and the clock
generation circuitry.
An experimental prototype sigma-delta modulator is designed for audio frequency range of 20
kHz to achieve a SNR of 80 dB [15] in a TSMC035 process with a single power supply of 3.3 V.
4.2. Topology selection
Using MATLAB simulation, we can conclude that a second order modulator with an OSR of 128
should be sufficient for obtaining the result. As mentioned above, for a conventional sigma-delta
modulator, the signal range required at the outputs of the two integrators is several times the full
scale analog input range; thus a modified modulator architecture which reduces the signal range
at the outputs of the integrator considerably is used. The modified architecture which is a slight
8/2/2019 JAIN MOHIT
44/70
33
variation of the classical model is shown in Figure 4.1. It is advantageous from the circuit
simplicity, noise and power dissipation to make the two gains of the first integrator equal so that
sampling network can be shared between the input signal and the feedback signal. Thus keeping
a1 = b1, we can derive other parameters from the MATLAB simulation as described below.
Figure 4.1: Modified second order modulator architecture
The variation of SNR with a1 and a2 is shown below in Figure 4.2. As can be seen a1 = 0.5
gives us the maximum SNR, along with Figure 4.3 showing the output density at integrator 1 and
as could be seen output goes farther than 1.1 which is difficult to achieve in a low voltage design.
Hence a1 = 0.4 was chosen which compromises a little of the SNR. Also there is no effect of a2
on the value of SNR (which is predictable as it is only the gain before the quantizer). The output
density at Integrator 2 is used to determine the value of a2. Figure 4.4 shows integrator 2 output
density as function of various integrator gains for a1 = 0.4. As could be seen for a2 = 0.4 the
output goes a little farther than 1.1 which is acceptable [4]. In the similar way a3 is varied and
there is not a huge advantage obtained by making a3 different than a2, hence a3 is made same as
a2 which simplifies the circuit.
8/2/2019 JAIN MOHIT
45/70
34
Figure 4.2: Variation of SNR with a1 and a2
Figure 4.3: Output density at Integrator 1 and 2 for a1 = 0.5 and a2 varying from 0.1 to 0.9
8/2/2019 JAIN MOHIT
46/70
35
Figure 4.4: Output density at Integrator 1 and 2 for a1 = 0.4 and a2 varying from 0.1 to 0.9
A fully differential architecture has been used in this design as it has superior power supply
rejection ratio along with providing twice the output swing for a given supply voltage. In
addition, the symmetry of a fully differential circuit provides for cancellation of even order
distortion components, regardless of their cause.
4.3. Integrator design
The most important component of a sigma-delta modulator is the integrator, the design of which
is discussed below. Figure 4.5 shows a simple realization of a parasitic- insensitive single ended
integrator [3, Chapter 10] which consists of an op-amp, a sampling capacitor C s, a feedback
8/2/2019 JAIN MOHIT
47/70
36
capacitor C f and four switches S1 to S4. In the sampling mode S1 and S4 are on and S2 and S4
are off, allowing the voltage across C s to track V in while the op amp and C f holds the previous
value. In the transition to the integration mode, S4 turns off first, S1 turn off next and
subsequently S2 and S3 are turned on. The charge stored on C s is therefore transferred to C f
through the virtual node.
Figure 4.5: Parasitic insensitive switched-capacitor integrator
To reduce charge- injection effects in SC circuits, a two phase clocking scheme as shown in
Figure 4.6 is used, the clocks are arranged so as to turn off S2 and S4, which are near the virtual
ground node of the op-amp, slightly ahead of their counterparts. Another factor which needs to
be considered during the layout of the integrator is the connection of the bottom plate of the
capacitor across the op-amp. Integrated capacitors are not generally symmetrical and there is a
larger parasitic capacitance to the substrate from the bottom plate than from the top plate. The
capacitors should be connected such that the common plate is driven either directly or through a
switch by a voltage source or the output of the op-amp. This arrangement causes the parasitic
capacitance to have the least effect on the operation of the circuit. Also, substrate noise coupling
is reduced by this arrangement.
8/2/2019 JAIN MOHIT
48/70
37
W
W
W
W
6
6
6
6
Figure 4.6: Two phase non-overlapping clocking scheme
All the switches S1 S4 are realized using minimum sized NMOS so as to minimize parasitics,
while maintaining good conduction for low enough analog common mode voltage. However
transmission gates are used for the switches coupled to the input signal, since they must conduct
over a wide range of voltage.
Figure 4.7 shows a fully differential implementation of the integrator which is similar to the
above mentioned integrator except the difference of the feedback voltage V ref The operation of
the integrator is similar to the one discussed above except that the voltage transferred to the
8/2/2019 JAIN MOHIT
49/70
38
output is the difference between input voltage Vi and the feedback voltage V ref when the
switches S2 and S3 are turned on.
Figure 4.7: Fully differential implementation of switched capacitor integrator
4.3.1. Clock generator
A simple schematic [20] is used to generate such a two-phase non-overlapping clock as shown in
Figure 4.8. The outputs C1 and C2 are a pair of non-overlapping clocks, while the outputs C1A
and C2A refer to slightly advanced version of C1 and C2 respectively. Similarly, the outputs
C1N and C2N are complementary parts of C1 and C2.
Figure 4.8: Schematic for non -overlapping clock generation
8/2/2019 JAIN MOHIT
50/70
39
4.3.2. Fully Differential Folded-Cascode Op Amp
The operational amplifier used in the integrators is the most critical element of the modulator.
The design of proper op-amp involves a great amount of theoretical calculations and empirical
simulations to achieve fast speed and sufficient gain.
Different op-amp topologies were considered and the feasibility to this design was analyzed. A
single stage telescopic cascode op-amp could have offered a high gain and a good bandwidth but
it also has lowest output signal swing (important criteria for low-voltage designs). A two-stage
amplifier can meet the gain and output swing requirements but is slower as compared to the one
stage amplifiers. Thus folded-cascode op-amp as shown in Figure 4.9 is a good choice providing
a good combination of swing and bandwidth with a disadvantage of being somewhat noisier than
other topologies. The basic idea of folded-cascode Op Amp is to apply cascode transistors to the
input differential pair but using transistors opposite in type from those used in input stage.
The following design equations can be used to choose various parameters for the device size in
the Op Amp [21].
Unity Gain Frequency L
mt
C
g 1=
Phase Margin ( )1
6tan Lt
m
C
gPM
=
Slew Rate L
SS
C
I SR
.2=
Cascode Bias Current += o Lt casc V PM C I )tan(1
8/2/2019 JAIN MOHIT
51/70
40
Figure 4.9: Fully differential folded-cascode Op Amp
Transistor W/L
M1, M2 160 / 1.5
M3 114 / 1.0
M4, M5 210 / 1.5
M6, M7 62.5/1.2
M8, M9 37.0 / 1.2
M10, M11 37 .0/ 1. 0
Table 4.1: Transistor sizes for Folded Cascode Op Amp
8/2/2019 JAIN MOHIT
52/70
41
From MATLAB behavioral simulation we obtain that the Op Amp should have gain of at least
60 dB, a UGB of 60 MHz and SR of 150V/s .
Assuming a load capacitance of 5pF, the slew rate requirement gives us the bias current equal to
1.5 mA. Overdrive voltages of 500 mV, 400 mV, 300 mV were chosen for transistor pairs M4-5,
M6-7, and M8-11 respectively giving us the W/L ratios as given in the table. Using these ratios
as a preliminary step and after a few iterations, the Op Amp was made to give the required gain,
bandwidth and slew rate.
4.3.3. Bias circuit
This work utilizes a constant-transconductance bias circuit using wide swing cascode current
mirrors [3, Chapter 6] for generation of reference voltages. A conventional cascode current
mirror is shown in Figure 4.10 (a).Although its output impedance is increased to rds1(rds2gm1),
a cascode current mirror reduces the maximum output-signal swing so that the minimum allowed
voltage for V out is V tn greater than 2V eff , where V eff is the minimum drain-source voltage needed
to keep a transistor working in the saturation region. This loss of signal swing is a serious
disadvantage for modern VLSI technologies.
An alternative circuit shown in Figure 4.10 (b) that does not reduce the signal swing so much
while keeping high output impendence, called a wide-swing cascode current mirror, can be used.
The basic idea of this is to bias the transistors at the edge of the triode region. Thus the minimum
allowable voltage just needs to be greater that (n+1)V eff . If n is unity (keeping I bias = Iin), this
8/2/2019 JAIN MOHIT
53/70
42
mirror can guarantee that all of the transistors are in the saturation region even when V out drops
to small values.
2
/ n
LW 2 /
n LW
LW / LW /
( )21 /
+n
LW
Figure 4.10: Conventional cascode mirror and Wide -swing cascode current mirror
The advantages of the wide-swing current mirror is coupled with a constant-transconductance
bias circuit wherein the transistor transconductances are stabilized using resistor R B and setting
(W/L) 2 = 4(W/L) 3. This configuration causes the transconductance of M3 to be stabilized to
Bm
Rg
13 =
Since all transistor currents are derived from the same biasing network and the ratios of current
are mainly dependent on the geometry, the transconductance of all n-channel transistors is
stabilized to 333) / (
) / (m
D
Diimi g I LW
I LW g =
and for all p-channel transistors 333) / () / (
m D
Dii
n
p
mi g I LW
I LW
g =
The transistors M15 M18 have been added to act as start-up circuitry by injecting currents into
the bias loop in case all the currents in the loop are zero. Once the currents are set up, the start-up
circuit is disabled by M15 and M16 turning off.
8/2/2019 JAIN MOHIT
54/70
43
Figure 4.11: Bias Circui t
Transistor W/L (m/ m)
M1, M 4 15.0 /3. 2
M2 60.0 / 2.0
M3, M12 15.0 / 2.0
M5 3.5/ 3.2
M6, M9, M10 45.0 /3. 2
M7, M8, M11 45.0 / 3.2
M14 15.0 /3. 2
M13, M15, M16, M17 20.0/2.0
M18 4 / 60
Table 4.2: Transistor W/L ratios for Bias Circuit (R B = 3k)
8/2/2019 JAIN MOHIT
55/70
44
4.3.4. Common-Mode Feedback Circuit
One requirement of using fully differential Op Amps is that a common-mode feedback (CMFB)
circuit must be added to establish the common-mode output voltage. This is due to the fact that
while using fully differential Op Amps in feedback applications, the feedback determines the
differential signal voltages but the common mode voltage is not affected. Thus it is necessary to
have a circuit to determine the output common mode voltage and to control it to some specified
voltage. The CMFB ideally will keep this common-mode voltage immovable, preferably close to
halfway between the power-supply voltages, even when large differential signals are present.
Without it, the common-mode voltage is left to drift, since the common-mode loop gain is not
typically large enough to control its value. The performance requirements on the CMFB circuitry
are not nearly as stringent as for the main op amp, because the signal of interest is the difference
between the main op amp outputs.
There are two typical approaches to design of CMFB circuits namely, continuous time and
switched-capacitor approach. The later is popular with the switched-capacitor circuits since it
introduces clock-feedthrough glitches in continuous-time applications. Figure 4.12 shows a
CMFB circuit used in my work. It may be seen that capacitors C C form a voltage divider to
generate the average of the op amp output voltages, which is used to generate control voltage for
the Op Amp current sources. CMOS transmission gates are used to realize the switches
connected to the outputs of the op amp, in order to accommodate a wider signal swing. The
switched capacitors C S (20 fF) are set to about one quarter to one tenth the sizes of the non-
switched capacitors C C (100 fF) so as to avoid common-mode offset voltages and op amp
overload [3, chapter 6]. All the switches are implemented using minimum sized n-channel
8/2/2019 JAIN MOHIT
56/70
45
transistors except for the switches connected to the outputs which are implemented using
transmission gates.
Figure 4.12: Switched capacitor common-mode feedback (CMFB) circuit
4.4. Regenerative Track and Latch Comparator design
The second major building block in a sigma-delta modulator is the comparator which quantizes a
signal in the loop and provides the output of the modulator. The principle design parameters of
this comparator are speed, which must be adequate to achieve the desired sampling rate, input
offset, input-referred noise and hysteresis. Since the comparator appears after the loop gain block
and before the output terminal, non-idealities associated with it are shaped by the loop in the
same way that the quant ization noise it produces is shaped. Therefore, the performance of the
modulator is relatively insensitive to offset and hysteresis (i.e. the tendency that a comparator
might have to stay in the previous direction when it should toggle to another direction) in the
first-stage comparator.
8/2/2019 JAIN MOHIT
57/70
46
A fast regenerative latch [22] without pre-amplification and offset cancellation, as shown in
Figure 4.13, has been used to implement the comparator. It consists of discharge transistors, an
n-channel flip flop with a pair of n-channel transfer gates for strobing, p-channel flip flop, and p-
channel pre-charge transistors.
In this latch, the cross-coupled devices M2A M2B and M3A M3B are strobed at their drains,
rather than sources. Advantages for a flip-flop strobed at drain node over a flip-flop strobed at
source node exist in regard to regeneration speed and offset. Since carrier mobility is nearly
twice faster at zero substrate bias than at a few volt substrate bias conditions, regeneration speed
for the flip-flop is faster for the drain strobing scheme. In addition, since strobing transistors
isolate the flip-flop, load capacitance becomes only the gate capacitance of the flip-flop itself.
Offset voltage caused by a channel length fluctuation, which is estimated as the main source of
total offset voltage, is much lower at zero volt substrate bias. Therefore, transistor channel
lengths can be decreased and the flip-flop speed can be made faster as a result. An inverting
buffer is connected to both the output nodes of the p-channel flip-flop to give the same loading
effect to the flip-flop. Since the analysis is in discrete time, a latch is needed for holding the
output during the feedback. Hence, we can use a SR latch with NOR gates.
8/2/2019 JAIN MOHIT
58/70
47
9' '
96 6
9,1 9,1
&
5
6
top related