Beruflich Dokumente
Kultur Dokumente
1639
61
Index TermsAngular velocity, gyroscopes, microelectromechanical systems (MEMS), offset, quadrature error, sensors,
stability, zero-rate output (ZRO).
I. INTRODUCTION
NGULAR VELOCITY sensors have applications in areas
such as inertial navigation, automotive safety, and stability control systems, platform stabilization, including picture
stabilization in camcorders and cameras, and robotics, to name
but a few. In particular, automotive chassis control systems such
as Electronic Stability Program/Electronic Stability Control
(ESP/ESC) have created a rapidly growing market for lowto medium-priced, medium-accuracy angular velocity sensors
which could be realized with micromachining [1], [2].
Traditionally, angular velocity has been measured with a
rotating wheel gyroscope, which is based on the conservation
of angular momentum. This has been replaced by fiber-optic
and ring laser gyroscopes in precision applications. (Strictly
speaking, the term gyroscope refers to the traditional rotating
wheel device, whereas the term angular velocity sensor refers
to angular rate measurement devices in general. However, they
are used interchangeably in the literature, and the convention
will be followed here as well.) Optical gyroscopes are the most
accurate angular velocity sensors available at the moment, and
Manuscript received June 29, 2007; revised August 21, 2007; accepted August 25, 2007. This work was supported in part by VTI Technologies, Vantaa,
Finland, and in part by the Finnish Funding Agency for Technology and Innovation (TEKES). The associate editor coordinating the review of this paper and
approving it for publication was Prof. William C. Tang.
The authors are with SMARAD-2/Electronic Circuit Design Laboratory, Helsinki University of Technology, FI-02150 Espoo, Finland (e-mail:
mhs@ecdl.tkk.fi; ljaalton@ecdl.tkk.fi; karih@ecdl.tkk.fi).
Digital Object Identifier 10.1109/JSEN.2007.908921
1640
quired phase accuracy in the demodulation can also be impractically strict. However, it is widely known that an effective way
to cancel this signal is to apply a dc voltage to properly placed
electrodes in the sensor element [5][7]. There are also several
other ways to compensate or cancel the mechanical quadrature
signal, either electrostatically or purely electronically. These
methods will be discussed briefly in Section IV-A1. Mechanical
adjustment or trimming of the sensor element is also possible,
but is highly dependent on the actual design and manufacturing
process and will not be addressed in this paper.
This paper will focus on these two nonidealities and the relationships between them. Different sources of the ZRO that have
been reported in the literature will be discussed, and they are
given a uniform representation. Ways to minimize the effect of
these sources on the final ZRO will be considered. How the electrostatic quadrature compensation performed with a dc voltage
affects the ZRO will be analyzed. Different ways to implement
the control loop for the compensation will be presented, concentrating on a case where the compensation voltage is generated
with a digital-to-analog converter (DAC), and the controller is
digital [4], [8].
This paper is organized in an introductory part formed by this
section together with Section II, and three other parts formed by
Sections IIIV. First, in the other half of the introductory part, in
Section II, the general properties of a vibratory microgyroscope
will be presented.
The first of the three other parts is formed by Section III. In
this section, various implementations of the quadrature compensation loop will be studied. The focus will be on a case where
the compensation voltage is generated with a DAC, and the controller is digital. In particular, extending the resolution with
techniques will be studied.
The second part is formed by Section IV. Here, different
sources of the ZRO will be discussed, and how they manifest
themselves in the final ZRO after the synchronous demodulation
will be shown. The effect of the electrostatic quadrature compensation performed with a dc voltage will also be analyzed.
The third part is formed by Section V. Here, experimental results will be given to support the theory that has been presented.
This section is divided into two parts, such that the first part
examines different implementations of the quadrature compensation loop and the second part studies the ZRO components of
one particular microgyroscope.
Finally, in Section VI, the paper ends with some concluding
remarks.
II. VIBRATORY MICROGYROSCOPES
A vibratory microgyroscope is composed of two orthogonal
mechanical resonators. A schematic drawing of such a system
without any nonidealities is shown in Fig. 1. In the figure, the
four rollers are used to express the fact that when the mass
is moving along the -direction, the -directional spring and
and
have no effect on the behavior, and vice
damper
versa. Thus, the system comprises two 1-degree-of-freedom res, the damper
, and the
onators, one formed by the spring
, the damper
, and
mass , and the second by the spring
the mass . Boths springs are assumed to be massless.
In the configuration depicted in Fig. 1, the masses of both
- and -directional resonators are identical and equal to .
(2)
Here, the terms
, and
are defined in Fig. 1.
and
were already defined above. The terms
The masses
and
represent the nonproportional damping, and
and
the anisoelasticity [9]. These terms have been added in
order to take the nonidealities of the mechanical system into acis the angular velocity of the system about
count. Finally,
and
are external forces used to excite the
the axis, and
resonators.
Now, on the basis of (2), the operation of a vibratory
gyroscope can be understood as follows. The -directional
resonator, often called the primary resonator or the drive
resonator, is driven to oscillation at its resonance frequency
which is also known as the operating frequency, using external excitation . The oscillation amplitude
is controlled and hence the -directional displacement can be
written as
(3)
is the vibration amplitude, determined by an external
where
controller.
By solving the -directional EoM with the assumption that
, the movement of the -directional resonator, often
SAUKOSKI et al.: ZERO-RATE OUTPUT AND QUADRATURE COMPENSATION IN VIBRATORY MEMS GYROSCOPES
1641
(4)
is the magnitude of the transfer
Here,
function of the secondary resonator from force to displacement
, defined as
at the operating frequency
(5)
and
(6)
where
is the ratio between the resonance frequencies of the primary and secondary resonators, the latter defined
, and
is the quality factor of the secas
. Here, it is
ondary resonator, defined as
assumed that the angular velocity signal is narrowband, i.e., the
and do not vary significantly over the signal
values of
bandwidth.
Let us first consider the ideal case, where the off-diagonal
. Now, the secondary resonator remains
terms
is applied
at rest unless a -directional angular velocity
to the device. In that case, the Coriolis effect couples the movement from the primary resonator to the secondary resonator. The
movement of the secondary resonator can be written as
(7)
From (7), it can be seen that the secondary resonator movement contains the amplitude-modulated angular velocity infor.
mation around the carrier frequency
Next, let us consider the effect of the off-diagonal terms in the
causes an output signal cor2-D EoM. The anisoelastic term
. This output
responding to an angular velocity
signal has a 90 phase shift when compared with the actual Coriolis signal, and hence is referred to as the mechanical quadrature signal. (The term mechanical quadrature signal is used
to distinguish it from other signals which will be introduced
later and are also in quadrature with respect to the Coriolis
signal.) On the other hand, the nonproportional damping term
causes an output signal corresponding to an angular ve. This signal is in-phase with the Coriolis
locity
signal. Whereas the Coriolis signal can be distinguished from
the mechanical quadrature signal by phase-coherent demodulation, it is impossible to differentiate the Coriolis signal from the
output inflicted by nonproportional damping.
The analysis is also applicable to vibratory gyroscopes with
torsional resonators by replacing the masses with moments of
inertia, displacements with angles, and forces with torques.
Depending on the value of , the gyroscope can be said to be
,
operated either in low-pass mode or mode-matched. If
the element operates in the low-pass mode, indicating that the
operating frequency is below the resonance frequency of the
secondary resonator. The sensitivity of the gyroscope element
,
can be increased by reducing . At the extreme, when
.
the gyroscope is said to be mode-matched, i.e.,
, increasing both
Now, the Coriolis signal is amplified by
sensitivity and resolution when compared with low-pass mode.
However, this would lead to an impaired open-loop bandwidth,
together with possible issues with gain stability and linearity and
the phase instability of the output signals.
For the analysis in the following sections, it is assumed that
the element is operated in the low-pass mode, in open-loop con. Further, it will be assumed that
and the
figuration
. If, for example,
mode separation are high enough for
and
, then
, a value that is small
for certain
enough to be considered negligible. The required
falls rapidly when is increased.
III. QUADRATURE SIGNAL COMPENSATION
As was shown in Section II, the off-diagonal terms in the
spring matrix of the 2-D EoM (2) cause a mechanical quadrature signal to the secondary resonator. This section will present
different ways to implement the control loop for electrostatic
quadrature compensation performed with a dc voltage [5][7].
The focus will be on a case where the compensation voltage is
generated with a DAC, and the controller is digital [4], [8].
A. Continuous-Time Compensation
A straightforward continuous-time quadrature compensation
loop is presented in Fig. 2. If the feedback part is ignored for
is inflicted
a while, the force
by the primary resonator movement, as described in Section II.
The force excites the secondary resonator with the transfer
, causing the uncompensated quadrature
function
movement
. The resulting signal goes through the
and is synchrodisplacement-to-voltage conversion
nously down converted to dc. Then, it is filtered to the desired
bandwidth with a low-pass filter, yielding the output signal
.
In the feedback part, the signal is first brought to a controller
, which outputs the voltage for the quadrature compensation electrodes. In the micromechanical element, the compensation voltage is converted to an electrostatic force and modin a proper phase. Finally,
ulated to the operating frequency
. The
the result is summed to the uncompensated force
includes the dynamics from the
transfer function
compensation voltage to the resulting force. In the real device,
1642
Fig. 3. A feedback loop with A/D and D/A conversion for quadrature signal
compensation.
the conversion from voltage to force and the further up converoccur simultaneously. Depending on the sensor elsion to
ement, the relationship between the quadrature compensation
voltage and the amplitude of the force generated can be linear,
or it can have a higher order dependence, such as a quadratic
one.
Typically, the quadrature signal needs to be zeroed at dc, requiring the controller to have a pole in the origin. The simplest solution would be a single integrator with
. Because of the low-pass mode operation, the compensation bandwidth is most likely limited by the low-pass filter
after the demodulation. If the bandwidth needs to be increased
beyond this limitation, a more complex controller with zeroes
that provide phase lead at higher frequencies might be required.
In this case, a complete PID (proportional-integrator-differentiator) controller can be used, for instance.
B. Compensation Using DAC With Digital Controller
When the quadrature signal is converted to the digital domain and the controller is implemented with digital signal processing (DSP), the loop takes the form shown in Fig. 3. In the
figure, the signal is A/D converted before demodulation, and
down converted and filtered in the digital domain. The A/D converter (ADC) could also be located after the low-pass filter, in
which case the demodulation is performed in the analog domain. The demodulation can also be performed using subsampling techniques.
A fundamental issue in the digital controller is that because
of the limited resolution of the feedback DAC, the output signal
can exhibit oscillations between two values even when
the actual linear compensation loop is stable. This can happen if
there is an integrator in the loop and the signal at the controller
input changes by more than one least-significant bit (LSB) when
the DAC output changes by one LSB. Now, it might occur that
the controller input cannot be brought to zero with any DAC
inteoutput level. The integrator in the controller
grates the residual signal until the DAC output changes by one
LSB. Then, the integrator output starts moving in the opposite
direction, until it returns to the original value, after which the
oscillation cycle starts over again. Because of the phase shifts
and in
and
, respectively, the oscillation cou, and causes
ples to the Coriolis output, attenuated by
SAUKOSKI et al.: ZERO-RATE OUTPUT AND QUADRATURE COMPENSATION IN VIBRATORY MEMS GYROSCOPES
1643
and
are the dc and ac components of the exciwhere
tation signal, respectively. As the electrostatic force is proportional to the square of the voltage, the resulting primary resonator motion is then
(9)
where
is the primary resonator transfer function from
excitation voltage to displacement. Further, the gain and phase
shift of the secondary resonator detection circuit (displaceand ,
ment-to-voltage converter) are assumed to be
respectively.
1) Mechanical Quadrature Signal: The mechanical quadrature signal [5], [9], [11], [12] is generated when the primary resonator movement couples directly to the secondary resonator as
a result of the nondiagonal terms in the spring matrix of the
2-D EoM. The magnitude of the force generated is directly proportional to the primary resonator displacement. Hence, the mechanical quadrature signal at the output of the detection circuit
can be written as
(10)
where is a parameter that determines the magnitude of the
coupling. With the parameters defined in Section II, is equal
to
.
Four different ways to handle the mechanical quadrature
signal can be identified. First, it can be removed or brought to
a desired level at the sensor element by either careful design
[14], [15] or by tight process control and by postmanufacturing
trimming and screening.
Second, it can be compensated electronically by injecting a
signal with the same amplitude but opposite phase to the input
of the secondary resonator detection circuit, effectively cancelling the mechanical quadrature signal. This might be difficult
to achieve, as both the amplitude and the phase of the feedback
signal need to be carefully controlled. Although the amplitude
1644
(11)
where is a parameter that determines the magnitude of the
nonproportional damping. Again, with the parameters defined
.
in Section II, is equal to
The only way to minimize this source is to minimize the parameter by mechanical design. To keep the source stable together with the gain
, needs to be kept stable.
3) Electrical Cross-Coupling in Sensor Element: In electrical cross-coupling [12], the primary resonator excitation
signal couples to the secondary resonator detection circuit
through stray capacitances. Because only the time-varying
can couple through capacitances,
component
the cross-coupled signal at the output of the detection circuit
can be written as
(12)
where is a parameter that determines the magnitude of the
cross-coupling.
There are several methods to avoid electrical cross-coupling.
First, and most obviously, the stray capacitances should be made
as small as possible. Next, in the case of differential detection,
SAUKOSKI et al.: ZERO-RATE OUTPUT AND QUADRATURE COMPENSATION IN VIBRATORY MEMS GYROSCOPES
(13)
where is a parameter that determines the magnitude of direct
excitation.
The only ways to eliminate the effect of direct excitation are
either to ensure the orthogonality of the primary resonator excitation force and the secondary resonator motion by process
control or by postprocess trimming, or by exciting the primary
resonator in such a way that the secondary resonator cannot
respond to it directly. It should be observed that if the secondary resonator movement rises from the primary resonator
movement through anisoelasticity, the resulting signal will be
in quadrature, whereas the signal resulting from direct excitation is in-phase with the Coriolis signal.
To keep the direct excitation force and, hence, the output
signal constant, the term should be constant, together with
. As the latter term varies in order to conthe product
trol the vibration amplitude of the primary resonator, the term
needs to be minimized until the required level of ZRO stability
also needs to be kept constant
is achieved. The gain
in order to keep the output signal caused by direct excitation
stable.
The direct excitation plagues not only vibratory gyroscopes
with capacitive excitation, but also those with piezoelectric excitation. In [18], the sources of the parasitic secondary signal in a
tuning fork gyroscope with piezoelectric excitation and piezoresistive detection are analyzed. In the reference, direct excitation
of secondary movement is referred to as actuation unbalance.
5) Cross-Coupling in Electronics: Due to parasitic stray capacitances or inductances in the electronics, various signals can
couple to the secondary resonator detection circuit and inflict
ZRO [4]. The source of coupling can be either the signal from
the primary resonator or a clock signal that is synchronized to
the primary resonator. The contribution of the clock signals, referred to the output of the secondary resonator detection circuit,
can be written as
(14)
and
describe the magnitudes of the cross-coupled
where
signals. Parameter refers to a signal that is in-phase with the
Coriolis signal, and to a signal that is in quadrature with the
Coriolis signal.
The cross-coupling primary resonator output signal, again referred to the output of the secondary resonator detection circuit,
can be written as
(15)
1645
and
describe the magnitudes of cross-coupling.
where
Again, refers to the cross-coupled signal that is in-phase with
the Coriolis signal, and to the signal that is in quadrature with
the Coriolis signal. If identical detection circuits are used for
both the primary and secondary resonators, the primary signal
.
is 90 phase shifted relative to the Coriolis signal, and
This source of coupling can be significant if the primary and
secondary resonator detection circuits are located close to each
other, which needs to be the case if good matching of the components in the circuits is required.
B. Effects of Synchronous Demodulation and Quadrature
Compensation
In the previous section, the different sources of the ZRO were
expressed at the output of the secondary resonator detection circuit in (10)(15). However, because of the synchronous demodulation, these equations do not directly represent the magnitude of the final resulting ZRO. Furthermore, the electrostatic
quadrature compensation performed with a dc voltage affects
the ZRO signal of the gyroscope. In this section, these two effects will be analyzed.
1) Synchronous Demodulation: After the signals of
(10)(15) are summed at the output of the detection circuit, they enter the demodulator. The signal at the input of the
demodulator is then
(16)
Next, the signal is down converted into two components, the
in-phase component, which also carries the Coriolis information, and the quadrature component. One should be careful in
order not to confuse the quadrature component in the demodulator output with the mechanical quadrature signal, which is
only one part of the quadrature component, albeit a significant
one.
The down conversion is performed by multiplying with
for the in-phase component and with
for
the quadrature component, and by filtering the signal with a
low-pass filter. Assuming that there is a further gain of two after
the filtering, the resulting output signals from the demodulator
are
(17)
1646
and
(19)
has also been applied. From (19),
where the assumption
it can be seen which parameters affect the input-referred ZRO
and how it can be minimized.
In a practical implementation, the mechanical quadrature
can be so high [4], [5] that the degree to
signal
) needs to be controlled to yield
which the phase shift (or
good ZRO stability is impractically strict. If the mechanical
quadrature signal is, for example, 10 times larger than the
full-scale Coriolis signal, then a phase shift of 0.06 causes
the ZRO inflicted by the mechanical quadrature signal to rise
to 1% of the full-scale Coriolis signal. This also means that
a tiny variation in the phase shift, or a variation in the gain
, causes significant variations in the ZRO. Additionally, a mechanical quadrature signal ten times higher than the
full-scale Coriolis signal could already limit the dynamic range
of the secondary resonator detection circuit.
2) Quadrature Compensation: Next, the effect of the electrostatic quadrature compensation performed with a dc voltage
is analyzed. The compensation is performed in a closed-loop
configuration, as described in Section III. The detailed implementation of the feedback loop can be ignored for the purpose
of this analysis. A steady-state operation will be assumed, i.e.,
it will be assumed that there are no transients. This is justified,
as the ZRO is a dc signal by definition. It will also be assumed
that the quadrature compensation at dc is perfect, i.e., there is an
ideal integrator in the feedback loop. As the dynamic operation
will not be analyzed, the detailed controller parameters are not
important as such, as long as they ensure a stable operation of
the feedback loop.
When applied, the quadrature compensation drives the signal
given in (18) to zero. When there are no other components apart
(21)
From (21), it can be seen that while the term
has been removed, the effect of the quadrature compensation is
present through the sine times tangent and cosine times tangent
terms that have appeared in the equation compared with (17).
The error terms are not proportional to the magnitude of the
;
uncompensated mechanical quadrature signal
they depend only on the phase shift terms and .
Now, as it is assumed that the element is operated in low-pass
, then with reasonable values of the phase shift
mode with
, the terms involving the sine times the tangent in (21) are rendered negligible compared with the terms involving the cosine.
Furthermore, the last and the third last term are completely cancelled. The ZRO can then be written as
(22)
The equation shows the components that affect the final ZRO
after quadrature compensation. They are the signal caused by
nonproportional damping, the signal electrically cross-coupled
at the sensor element, the direct excitation, and any cross-coupled clock signals or a component of the signal from the primary
resonator that are in-phase with the Coriolis signal. Further, the
phase shift determines the magnitude of the ZRO. If is small,
then as a result of the cosine function, the ZRO is not very sensitive to variations in . If, for example, varies between 1 and
SAUKOSKI et al.: ZERO-RATE OUTPUT AND QUADRATURE COMPENSATION IN VIBRATORY MEMS GYROSCOPES
1647
Fig. 6. A block diagram of the angular velocity sensor including the bulk micromachined sensor element, the analog part implemented with a custom IC, and the
digital signal processing part implemented with an FPGA chip.
2 , the variation in the ZRO is less than 500 ppm. The ZRO stability is also dependent on the stability of the gain
.
After reducing (22) to input angular velocity, it can be written
as
(23)
and have been
Compared with (19), the terms involving
removed. Otherwise, identical methods can be applied to minimize the input-referred ZRO.
V. EXPERIMENTAL RESULTS
The theory presented in Sections III and IV was studied experimentally by using a developed version of the MEMS angular
velocity sensor originally presented in [4] and [8]. The system
comprises a bulk micromachined sensor element, analog interface electronics implemented with a custom IC, and DSP implemented with a field-programmable gate array (FPGA) chip. A
block diagram of the system is shown in Fig. 6. The actual micromechanical sensor element is shown in Fig. 7. The operating
is
kHz, and the mode separation
frequency
.
In the system, the gyroscope output signals are converted into
ADCs ([10], Ch.
the digital domain with bandpass (BP)
5) after analog signal processing. The analog signal processing
comprises capacitance-to-voltage (C/V) conversion of the signals using continuous-time charge-sensitive amplifiers (CSAs),
Fig. 7. Left: Schematic drawing of the structure of the sensor element. Right:
Scanning electron microscope image of the middle (structural) wafer. (Pictures
courtesy of VTI Technologies, Vantaa, Finland.)
1648
Fig. 9. Spectrum of the resulting Coriolis output when the quadrature signal
varies between two levels as a result of limited compensation resolution. (7281, sampling frequency 125 Hz,
point FFT, Kaiser window with
/s).
= 13
100
FS =
61
FS = 1 92
= 13
SAUKOSKI et al.: ZERO-RATE OUTPUT AND QUADRATURE COMPENSATION IN VIBRATORY MEMS GYROSCOPES
61
1649
Fig. 13. Spectrum of the resulting Coriolis output when the quadrature signal is
compensated with a dc voltage. (7281-point FFT, Kaiser window with
,
/s).
sampling frequency 125 Hz,
FS = 100
= 13
FS = 100
61 DAC is used
= 13, sampling
(24)
The frequency
can be freely chosen, as long as it lies within
the final output bandwidth of the angular velocity sensing
system and the phase shift of the system at that frequency is
known. From (24), it can be seen that there are no electri,
cally cross-coupling components apart from the signal at
whereas, because the electrostatic force is proportional to the
square of the exciting voltage, there are force components at
and
. Because of the high-quality factor of
the primary resonator, the primary movement excited by these
components is negligible, whereas the secondary movement
found.
can be measured and the parameter
and
Then, the terms
are still left. The sum of the last two terms can
be found by disconnecting the input of the secondary resonator
detection circuit from the sensor element, after which only the
signals cross-coupled in the electronics appear at the output. The
can be distinguished from clock signals by
term
letting it vary with time with a known frequency and measuring
the cross-coupling of the time-dependent component. Then,
can be solved by a simple subtraction operation. Finally,
the first term can be found by subtracting all the other components from the total ZRO.
1650
TABLE I
MEASURED ZRO COMPONENTS, REDUCED TO INPUT
of
VI. CONCLUSION
In this paper, issues related to the ZRO of a vibratory microgyroscope were studied. Different sources of the ZRO that have
been reported in the literature were discussed, and they were
given a uniform representation. The discussed sources were: a
mechanical quadrature signal; nonproportional damping; electrical cross-coupling in the sensor element; direct excitation of
the secondary resonator, and cross-coupling in the electronics.
Attention was paid to the stability of each source. Methods
to minimize the effect of different sources were described,
whenever possible. The effects of synchronous demodulation
and electrostatic quadrature compensation performed with a dc
voltage on the final ZRO were analyzed.
Various ways to implement a control loop for the electrostatic quadrature compensation were presented, concentrating
on a case where the compensation is performed with a dc voltage
generated with a DAC and the controller is digital. In particular,
modulation
how to extend the resolution of the DAC by
and, hence, avoid the spurious components in the output signal
resulting from limited resolution was studied.
In the experimental work, a digitally controlled quadrature
compensation loop was implemented, first with a 7-bit HV
DAC with the same HV D/A conDAC, then with a 15-bit
verter as the output DAC. How the spurious components can be
removed without compromising the final SNR was shown.
Next, the ZRO sources for one particular MEMS gyroscope
implementation were studied experimentally. It was found that
cross-coupling in the electronics is the most important ZRO
source in this implementation. This is because the primary and
SAUKOSKI et al.: ZERO-RATE OUTPUT AND QUADRATURE COMPENSATION IN VIBRATORY MEMS GYROSCOPES
TABLE II
DEFINITIONS OF THE MOST SIGNIFICANT SYMBOLS
USED IN THE ZRO EQUATIONS
1651
1652