Applications of Switched-Capacitor

Circuits in Active Filters and

Instrumentation Amplifiers
 Dr. William R. Grisé
[email protected]
Department of IET
Morehead State University

ABSTRACT

 This paper will explain the basic concepts underlying the operation of the switched
capacitor, as well as the use of switched-capacitors to realize compact and versatile
circuits already familiar to the undergraduate student of electronics. One set of example
circuits include easily tunable active filters; specific examples of filter designs that
incorporate switched-capacitors will be developed, and the use of a commercially
available switched-capacitor integrated circuit, the MF10, to implement the designs will
be shown. Another example circuit is an instrumentation amplifier that is more compact
and has a higher CMRR than the conventional realization. Linear Technology's
LTC1043 serves as the vehicle for this circuit. By demonstrating the utility of the
modern switched-capacitor IC in these two important electronic functions, it is hoped
that instructors and students in engineering technology will include the study of the
switched-capacitor in advanced electronics courses.

I. INTRODUCTION

This paper aims to show how the switched-capacitor concept can be used to realize a
wide variety of active filters that have the advantages of compactness and tunability.  In
particular, the explanations and design examples presented here will use mathematical
tools familiar to the electronics technology and engineering undergraduate student. We
will not use the Z-transform, which is the rigorously correct tool for analyzing sampled-
data waveforms.  

The paper will present the following topics. First, the basic ideas behind the use of the
switched-capacitor to replace resistors in active filter circuits will be explained.  Second,
the use of the switched-capacitor to implement lossless, lossy, and differential
integrators, which are the backbone of many switched-capacitor filter circuits,
especially those based on National Semiconductor's MF10 IC [1].  Third, example
designs of active filters using the MF10 will be presented. 

Before detailing the operation of switched-capacitor circuits, it will be useful to

understand the motivation behind, and applications of, these circuits. Basically,
switched-capacitor techniques have been developed in order to allow for the integration
on a single silicon chip of both digital and analog functions. Because very large scale
integrated (VLSI) circuits rely on MOS transistors and pico-farad range MOS
capacitors, any realization of analog circuits on a chip will have to use these elements.
By comparison, conventional analog circuits use the ratio of resistances to set the
transfer functions of amplifiers, and the magnitudes of resistances to determine the
operation of current-to-voltage and voltage-to-current converters. Finally, the values of
RC products are used in active filters and signal generators to determine the frequency
responses of those circuits. When one moves to the silicon chip and strives to achieve
the same functionality in a much reduced area and using the tools of MOS technology,
this is what one discovers. First, switches, small-value capacitors, and decent op-amps
are easy enough to realize in MOS technology. Second, using that same technology, it is
very difficult and wasteful of silicon die area to make resistors and capacitors with the
values and accuracy encountered in audio and instrumentation applications [2,3]. As we
will see in the subsequent sections, designers have overcome these difficulties by
realizing that (1) resistors can be replaced by MOS switches that are rapidly turned on
and off, and MOS capacitors, and that (2) the time constants arising from these
simulated resistances and the MOS capacitors are given in the form of capacitance
ratios. The fact that capacitor ratios control the time constants means that these
constants now can take advantage of the superior matching of capacitances fabricated
on silicon, as well as their ability to track each other with temperature. 

If these are undoubted advantages for the VLSI designer, what can the board-level
circuit design expect to achieve with the use of switched-capacitors? For one thing, as
we will see shortly, not only are the time constants of the switched-capacitor circuit
superior in their control, but these time constants are tunable through the simple
expedient of changing the frequency of the clock pulses that drive the circuit.
Furthermore, the integrated circuit packages that are now available support a number of
filtering functions in one package, thus reducing footprints needed on circuit boards to
realize a given set of analog functions. 

Although switched-capacitors were developed in order to meet the need to incorporate

analog, active filters on silicon along with digital functions, they have since found many
other uses [2]. These include, besides filters, instrumentation amplifiers, voltage-to-
frequency converters, data converters, programmable capacitor arrays, balanced
modulators, peak detectors, and oscillators.

II. BASIC SWITCHED-CAPACITOR OPERATION 

The essence of the switched-capacitor is the use of capacitors and analog switches to
perform the same function as a resistor.  This replacement resistor, along with op-amp
based integrators, then forms an active filter.  Before delving too far into actual filter
designs, however, it makes sense to ask why one would want to replace the resistor with
such an apparently complex assembly of parts as switches and capacitors.  It would
seem from the multiplication of parts that the switched-capacitor would be area
intensive.  As a matter of fact, for the resistor values that one seeks in certain filter
designs, this is not the case.  Furthermore, the use of the switched-capacitor will be seen
to give frequency tunability to active filters.  Figure 1[2, 3] shows the basic setup for a
switched-capacitor, including two N-channel Metal-Oxide Semiconductor Field-Effect
Transistors (NMOS) and a capacitor.  There are two clock phases, , which are
non-overlapping. The MOSFET's, either M1 or M2, will be turned ON when the gate
voltage is high, and the equivalent resistance of the channel in that case will be low,
. Conversely, when the gate voltage goes LOW, the channel
resistance will look like .  With such a high ratio of OFF to ON
resistances, each MOSFET can be taken for a switch. Furthermore, when the two
MOSFET's are driven by non-overlapping clock signals, then M1 and M2 will conduct
during alternate half-cycles.   

Figure 1.  Two NMOSFET's, driven by alternating, non-overlapping clock signals,

comprise the basic switched capacitor network. 

This makes the two-MOSFET arrangement equivalent to a single-pole, double-throw

switch (SPDT).  One can now use a symbolic switch picture, as in Figure 2a below, to
represent the circuit.  The operation of this circuit is as follows. When the switch is
thrown to the left, the capacitor will charge up to . When the switch is thrown to the
right, the capacitor will discharge down to/charge up to . As a result of these
consecutive switching events, there will be a net charge transfer of
.  Now, if one flips the switch back and forth at a rate of
 cycles/sec, then the charge transferred in one second is
, which of course has the units of current.  One can
claim that an average current, . If    is much higher
than the frequency of the voltage waveforms, then the switching process can be taken to
be essentially continuous, and the switched-capacitor can then be modeled as an
equivalent resistance, as shown below in Figure 2b.  The value of the equivalent
resistance is given by: 

                       (Eq. 1)

Therefore, this equivalent resistance, in conjunction with other capacitors, and Op-amp
integrators, can be used to synthesize active filters.  It is now clear from Equation (1)
how the use of the switched-capacitor leads to tunability in the active filters, by varying
the clock frequency. 

                        (a)                                                                       (b)             

Figure 2. Equivalent resistor model for switched capacitor circuit in Fig. 1. 

This equivalent resistance has features which make it advantageous when realized in
integrated-circuit form: 

(a)    High-value resistors can be implemented in very little silicon area. For
example, a 1-M resistor can be realized with a 10-pF capacitor switched at
a clock rate of 100 kHz.

(b)   Very accurate time constants can be realized, because the time constant is
proportional to the ratio of capacitances, and inversely proportional to the
clock frequency:

. Capacitor ratios, especially in monolithic form,

are very robust against changes in temperature, and clock frequencies can
also be strictly controlled, so that accurate time constants are now available
in the switched-capacitor technology. 

The principal constraint in using the switched-capacitor is that inherent in all sampled-
data systems: the clock frequency must be much higher than the critical frequency set
by the RC products in the circuit. Furthermore, on either side of the analog switches,
i.e., the MOSFET's, there must be essentially zero-impedance nodes (voltage sources). 
There are a number of other constraints which the unsuspecting designer/user might
overlook [3, p. 725]: 

(a)    The equivalent resistance formed by the action of the switched-capacitor

cannot be used to close the negative-feedback path in an op-amp all by
itself.  One must recall that to ensure stability, the op-amp's feedback path
must be closed continuously, while the switched-capacitor is a sampled-data
construction of a resistor, and thus not continuous.

(b)   Circuit nodes cannot be left floating.  That is, there must always be a
resistive path to ground so that charge does not build up on the capacitor
plates.
 (c)    The bottom plates of the MOS capacitors must be connected to ground or to
a voltage source. There is an intrinsic, parasitic capacitance associated with
the MOS capacitor's bottom plate [4]. This parasitic capacitance can be
between 5% and 20% of the desired value; furthermore, it behaves
nonlinearly with voltage [4].  Therefore, it must be connected to AC ground
or a voltage source so that this nonlinear portion of the capacitance will not
affect the overall response of the switched-capacitor filter. In practical terms,
this means that capacitive voltage dividers with three or more capacitors, and
circuits that switch both ends of a capacitor in sequence to the inputs of an
op-amp, are used.

(d)   The noninverting pin of the op-amp should be kept at a constant voltage. If
this pin is connected to the signal in some way, then the virtual short circuit
between op-amp inputs means that the inverting input is no longer a virtual
ground, and so an undesirable alteration of filter response due to the MOS
capacitor's parasitic capacitance will occur (see item (c) above). 

III. SWITCHED CAPACITOR INTEGRATORS 

The op-amp integrator is the most frequently chosen building block for switched-
capacitor filters.  The standard RC integrator is shown in Figure 3a, and its analysis and
description can be found in any electronics text [5, 6]. 

 Figure 3a. Standard op-amp integrator, with transfer curve

   To review, the voltage transfer function of this circuit is given by

                 (Eq. 2)

where . Now, if one replaces the resistor by its switched-capacitor

equivalent, as shown in Figure 3b, and use Eq. 1 as the resistor's value, one finds 

                 (Eq. 3) 

 

Figure 3b. Switched-capacitor integrator.

  Again, one notes the fact that this new integrator has no resistors, which take up
excessive silicon die area. Also, the -3 dB frequency, , depends on a ratio of
capacitances, not on an RC product.  The tolerances for ratios are much easier to control
than the tolerances for products. Finally, this characteristic frequency of the integrator is
inherently settable with a simple change in the clock frequency. 

The typical values of capacitances used in switched-capacitor technology range from

0.1 pF to 100 pF. These are low enough values that the stray capacitances of the MOS
switches, of the interconnects, and of the "plates" of the switched-capacitors themselves
can all have a significant effect on the desired frequency response of the filters designed
with switched-capacitors. The effects of stray capacitance have been reduced greatly by
dual-switch configurations [2, 7]. Figure 4 shows explicitly the clock phasing of the
MOS switches which acts to eliminate the transient charge transfer through the stray
capacitances, Cs1 and Cs2, also indicated in the figure.  In essence, charge transfer only
takes place through the capacitor . Figures 5a and 5b show both the inverting and
noninverting stray-insensitive integrator. The noninverting stray-insensitive integrator is
obtained simply by switching the clock phasing on transistors M2 and M4.

  

Figure 4. Arrangement of extra MOSFET's and clock signals to make switched

capacitor circuit insensitive to stray capacitances.

  
 Figure 5a.  Switch setting used to realize stray-insensitive inverting integrator.

Figure 5b.  Switch setting for stray-insensitive non-inverting integrator.

 Because of the importance of the integrator to switched-capacitor filters, it is necessary

to be familiar with the variants of the integrator.  These include the summing integrator,
the differential integrator, the integrator/summer, and the lossy integrator. All of these
play a role in the synthesis of switched-capacitor filters. The summing integrator, shown
below in Figure 6, has a response given by: 

               (Eq. 4)

  
 Figure 6.  Summing integrator.

  

Figure 7 shows the differential integrator.  The easiest way to understand this circuit is
to look at what happens to the charge accumulation on the capacitor  when the
switches are thrown to the left.  In this case, the capacitor charges up to a value of
. When the switches are thrown to the right, the charge on the capacitor is
poured into the op-amp's summing node. The average current, assuming the switching
rate (= clock frequency) is high enough, is given by

     (Eq. 5)

 This results in a stray-insensitive output voltage of 

                      (Eq. 6)

 where  .
  

 Figure 7.  Differential integrator.

 The lossy integrator provides a simple, first-order lowpass response with gain. This
circuit is realized by placing a switched-capacitor (i.e., a simulated resistor) in parallel
with a feedback capacitor, Figure 8. In general, the easiest way to analyze the response
of more complex switched-capacitor circuits such as this one is to replace all switched-
capacitors with their resistor equivalents. Once the transfer function is found for a
circuit with resistors (and discrete capacitors), then the switched-capacitor equivalents
of the resistors (Eq. 1) can be placed back in the transfer function to obtain the final
result. For the lossy integrator, the analysis proceeds as follows: 

          

where the "0" in  refers to the virtual ground at the op-amp's

inverting input. The transfer function is obtained using these resistor equivalents: 

          (Eq. 7)

Now substituting the switched-capacitor equivalents for the resistors from Eq. 1, one
obtains

 
                         (Eq. 8)

where   is the critical, or -3 dB, frequency of the lowpass filter.  Eq.
8 has the form of a lowpass filter multiplied by a gain proportional to the ratio of the
two switched-capacitors.

  

Figure 8.  Lossy integrator, or first-order lowpass filter with gain.

  

IV. SWITCHED CAPACITOR BIQUADRATIC FILTERS

 The biquad configuration [8] normally features a lossy inverting integrator, a lossless
inverting integrator, and a unity-gain-inverting amplifier.  In the standard active RC
configuration, this requires three op-amps.  However, the switched-capacitor realization
of the biquad needs only two op-amps to perform the same function. One op-amp
performs the lossy inverting integration function, while the second op-amp performs
lossless, noninverting integration. Although one can design an adequate switched-
capacitor version of the biquad by making a resistor-by-resistor replacement in the
standard RC biquad filter, such an implementation has been found to have unacceptably
wide capacitance spreads, especially when higher filter Q's are sought [2]. Instead,
Figure 9 shows the biquad filter with improved capacitance ratios. This circuit provides
the highpass and bandpass responses. Just as with the analysis of the lossy integrator, a
fairly complete analysis of this circuit will be made here. 
  

Figure 9.  Biquadratic bandpass and highpass filter provided by switched capacitor
network, with improvements needed to reduce spread in capacitor values.

 The form of the bandpass filter function is at least clear: 

                     (Eq. 9) 

The use of a stray-insensitive switched-capacitor ( ) with alternating clock phases

makes possible the noninverting form above. In order to complete the analysis, one has
to calculate the highpass filter response.  The output node of the first op-amp, which
gives the highpass filter response, can be seen to be the superposition of two signals at
the summing node of the op-amp:                                                            

  (Eq. 10a)
  With   the expression in curly brackets
simplifies, giving

(Eq. 10b)

 In order to obtain finally the bandpass transfer function, Eq. 10b is inserted into Eq. 9.
The result of this manipulation is

                 (Eq. 11) 

Gathering like terms together, and multiplying both sides by , one obtains

              (Eq. 12)

 With , and isolating , the transfer function for the bandpass filter is
finally found:
                           (Eq. 13)

 The corresponding expression for the highpass filter is simply obtained by inverting

Eq. 9. Recalling that , one finds readily 

                 (Eq. 14)

 Although Eq. 13 does in fact display the standard form for the frequency response of a
bandpass filter, there is nothing in the expression that gives the gain of the filter circuit. 
In fact, from the form of the transfer function, and from simulation, it can be seen that
the circuit in Figure 9 has no resonant gain. The same applies to the highpass filter
expression in Eq. 14.  It is clear that a useful active filter circuit formed from switched-
capacitors must possess some voltage gain in the passband. Of course, it is possible to
simply "tack on" an amplifier to the output of the circuit. However, an elegant solution
is to turn the second op-amp in Figure 9 into a summing integrator, in which the input
signal is injected into the second op-amp via an equivalent resistance derived from
another switched-capacitor. Figure 10 shows the implementation of this idea. In Figure
10, the SPDT switches, labeled S1, are shown closed in the first half of the clock cycle
.  

The analysis of this circuit in order to derive the bandpass filter transfer function
follows a path similar to the one followed in Eq. 13, although more tedious. We will
simply present the main results here and proceed to the simulation of the circuit. 
Superposition of input signals to each of the op-amp's input pins gives 

               (Eq. 15)

 After much manipulation, one finds that the transfer function for the bandpass filter is 
         (Eq. 16)

 In order to simulate the performance of switched-capacitor networks, it is necessary to

realize that these circuits are a hybrid of both continuous and sampled-data signals.  In
fact, switched-capacitor circuits are continuous in amplitude and discrete in time.
Because of this combination, simulation with conventional circuit simulators such as
PSPICE presents a problem. The presence of switched voltages means that a transient
analysis must be performed. At the same time, the desire to determine the frequency
response across a wide range of frequencies means that a transient analysis must be
performed for each desired frequency. This is a very time-consuming process because
one has to wait until the steady state is reached. One way to overcome this problem
within a SPICE-type simulator is to implement all the design equations in z-transforms.
These z-domain models will allow one to perform frequency-domain simulation of
complex switched-capacitor circuits. Although the z-transform is the rigorously correct
mathematical tool for the analysis of sampled-data systems, it is not really available to
the majority of engineering technology students. The model element used as the
simulation kernel in SPICE for z-domain analysis is called the storistor [10]. It consists
of conductances, a lossless transmission line for delay effects, and controlled sources. In
order to simulate the frequency response of something as simple as an integrator, one is
required to model as many as seven storistors, four capacitors, and an op-amp
subcircuit. Given the long experience of many students and instructors with the SPICE
program, this might be an acceptable alternative. However, the size of the input files for
even simple switched-capacitor circuits (beyond the integrator), together with the
mathematics required to understand the z-transform, will probably deter many from this
approach. 

This paper will use a perhaps less well-known simulation package called APLAC
[11-12] (originally Analysis Program for Linear Active Circuits). The APLAC program
has been under continuous development since 1972. Since 1988, the Nokia Corporation,
developers of wireless communication products, has sponsored continued improvement
in the system design and electromagnetic capabilities of the APLAC program. The
particular strengths of the APLAC program are its use of object oriented programming
techniques, which permit easy adaptation of models to the circuit environment in which
a component finds itself. Also, the program has a very extensive library of system level
blocks, and the ability to model electromagnetic behavior of components. 

For our present purposes, one of the attractions of the APLAC program is its use of the
convolution integral to model circuits with both frequency-dependent and transient
behaviors. This feature is important for switched-capacitor circuits. The frequency-
dependent portions of the circuit can be analyzed straightforwardly. However, the
frequency response of the time-dependent portions of the circuit (such as switches and
sources) is calculated by creating a frequency-domain equivalent circuit by means of the
convolution integral. 

In the circuit of Figure 10, the following values of components and parameters are
used: 

These values give a resonant (center frequency) voltage gain of A 0 = 10, a Q = 50, and a
center frequency of  f0 = 20 kHz. The results of the simulation are shown below, in
Figure 11.

 Figure 10. Biquadratic switched-capacitor filter with gain setting from summing
integrator. The switches S1 are closed on the first half-cycle of the clock waveform. 

 
Figure 11. APLAC simulation of bandpass filter with gain, from Figure 10. From Probe
tool of simulator, peak gain is 3.32 dB, and center frequency is at f0 = 17.8 kHz.

The results show a simulated gain of only 3.3 dB, and the center frequency is off by ~2
kHz. This inaccuracy is due to the use of ideal MOS switches in the hand calculations
leading up to Eq.  16. The APLAC model for the switches assumes an Ron = 100 , and
an Roff  = 100 k.  APLAC has the capability to optimize the circuit's component values
to achieve desired circuit behaviors.

  

V. APPLICATION OF MF-10 

The MF-10 is a universal switched-capacitor filter supplied by National Semiconductor

[1]. The MF-10 uses the two-integrator loop structure to realize lowpass, highpass,
bandpass, notch, and allpass functions through externally chosen, discrete, resistors. The
actual switched-capacitor integrators are internal to the chip, while the external resistors
give the user flexibility in configuring his/her own response. However, to take
advantage of component tracking with temperature, etc., all responses are designed to
be functions of resistor ratios only.  Figure 12 shows the summing amplifier and two-
integrator cascade internal to each section of an MF-10. The tunability of a particular
filter's critical frequency, , is determined by a logic level applied to a 50/100/CL
frequency ratio programming pin. In other words, the critical frequency will be 

; if the programming pin is tied to ground, the divisor is

100, otherwise, if tied to a HIGH (positive power supply), the divisor will be 50. Figure
12 shows how the notch, bandpass, and lowpass filter functions are realized by the MF-
10. Because the summing amplifier is outside the two-integrator loop, this configuration
will be faster and allow a greater range of operating frequencies. 

 
 Figure 12. Two-integrator loop, with external resistors, used in
MF-10 to realize notch, lowpass, and bandpass filters. 

The analysis of the transfer functions for the three transfer functions mentioned above
follows the pattern in Eq. 13-14. By inspection, one sees that 

                                     (Eq. 17a)

                                    (Eq. 17b)

                                                  (Eq. 17c)

Eliminating  to obtain , the transfer function for the notch filter is
obtained: 

          (Eq. 18)

Similar manipulations give the bandpass and lowpass responses. The bandpass response
is:

              (Eq. 19) 

The lowpass response is:

              (Eq. 20)

 
VI.   SWITCHED-CAPACITOR IMPLEMENTATION APPLIFIER  LTC1043  
Although the initial impetus for the development of the switched-capacitor was the
opportunity and need to synthesize active filters that would be compatible with
MOSFET technology, the early 1980's found many other uses for the switched-
capacitor. Linear Technology has developed the LTC1043 [9], which contains dual
switched capacitor networks, along with an on-chip non-overlapping clock generator,
oscillator, and charge balancing circuitry. The clock generator controls both of the
switch networks, while the charge balancing circuitry is designed to cancel any effects
due to stray capacitance. The on-chip oscillator has a fixed frequency of 185 kHz. An
external capacitor can be connected across pins 16 and 17 (for the instrumentation
amplifier) to yield any desired clock rate. The desired clock rate can be found from 

; the 24-picofarad capacitance is the internal capacitance

responsible for the oscillator’s fixed frequency.  

Among the circuits developed from the LTC1043 are instrumentation amplifiers, lock-
in amplifiers for detecting extremely small parameter shifts in sensor applications, and
signal conditioners for platinum resistance temperature detectors (RTD), relative
humidity sensors, and LVDT’s. The instrumentation amplifier is a standard op-amp
circuit presented in many electronics texts [5-6], and is designed to amplify small
difference signals such as might be found in measurement or transducer applications. At
the same time, common-mode or noise signals picked up by the lines feeding the
amplifier must be suppressed, especially as these signal levels are often larger in
amplitude than the sought-for difference signals. Figure 13 shows the LTC1043
combined with a standard non-inverting op-amp to give an instrumentation amplifier
with a common-mode rejection ratio (CMRR) of  >120 dB. Figure 14 shows the same
circuit with the ½ LTC1043 as a black box.

Figure 13. Instrumentation amplifier using ½ of LTC 1043 switched-capacitor, along

with LF356/353 op-amp in non-inverting configuration.
  
Figure 14. Switched-capacitor-based instrumentation amplifier, with ½ LTC1043
shown as a black box. The pin numbers in Figure 13 are the pins in the black box in this
figure.

 The operation of this circuit is as follows. First, the dual switch, when flipped to the
left, charges the capacitor C1 up to the difference V1 – V2. Second, on the next clock
pulse, the switches will then dump the charge represented by that voltage difference
onto C2. Third, the continuous clocking from the oscillator will force C2 to eventually
develop a voltage equal to the difference voltage. Finally, the difference voltage, with
the common-mode signal stripped off by the LTC1043 is amplified by the op-amp. It is
interesting to observe several features of this circuit and compare them to the standard
instrumentation amplifier. By using the capacitor C1 (the so-called “flying capacitor”),
the common-mode voltage present at the inputs is looking into a capacitive voltage
divider, between the C1 and the LTC1043’s parasitic capacitance. This parasitic
capacitance is typically less than 1 picofarad, so the AC value of the CMRR is > 120
dB. By comparison, Analog Device’s AD624 instrumentation amplifier can go as high
as 130 dB for high gains, up to 60 Hz. Because of the capacitive voltage divider from
the LTC1043, this instrumentation amplifier shows higher CMRR, over a wider range
of voltage gains, and to a higher frequency.  

VII.  CONCLUSIONS 

This paper has presented the essentials of operation  of switched-capacitor networks,

with a special emphasis on its use in designing active filters. Unlike active filters based
on the conventional op-amp, switched-capacitor filters have critical frequencies that are
easily pin-settable. Furthermore, they require less power than the conventional op-amp
based network because of their reliance on CMOS technology. Finally, for the
functionality provided on a single chip, they take up less room on circuit boards.
Alternate use of the switched-capacitor network in an instrumentation amplifier has also
been presented. The operation of this device is a little easier to digest for some students
than discussion of active filters; it is hoped that instructors and students can use the
information herein to extend their acquaintance with modern integrated circuits. 

The presentation of results here is in a form which instructors and upper-level students
in electronics technology can adapt to the curriculum in technology programs.