A New Electronic Feedback Compensation Method

for Rate Integrating Gyroscopes

Parsa Taheri-Tehrani1, A. Dorian Challoner2, Oleg Izyumin3, Bernhard Boser3, and David Horsley1
Email: ptaheri@ucdavis.edu dahorsley@ucdavis.edu
1
University of California, Davis, CA, USA, 2Inertialwave Inc., USA and 3University of California, Berkeley, CA, USA

Abstract— Towards the objective of a rate integrating (a) (b)

gyroscope (RIG) without a minimum rate threshold and
performance limited only by electrical and mechanical thermal
noise, we present our progress on a new, generalized electronic
feedback method for the compensation of resonator damping
asymmetry (anisodamping) and stiffness asymmetry
(anisoelasticity) with a new method of RIG operation using
self-precession. This enables overcoming the precession
angle-dependent bias error and minimum rate threshold, two
issues identified by Lynch for a MEMS RIG [1]. To correct
angle-dependent bias, we augment the electronic feedback force of
the amplitude regulator with a non-unity gain output distribution
matrix selected to correct for anisodamping. Using this method,
we have decreased the angle dependent bias error by a factor of
30, resulting a minimum rate threshold of 3.5 dps. To further
improve RIG performance, an electronically-induced self-
precession rate is incorporated and successfully demonstrated to
lower the rate threshold. The RIG’s output noise is also evaluated,
demonstrating an ARW of 11 mdps/√Hz, similar to rate gyro
operation at same amplitude.
Keywords — Rate integrating gyroscope; anisodamping
compensation; disk resonator gyroscope; whole-angle mode;
quadrature cancellation; self-precession
I. INTRODUCTION
MEMS vibratory gyroscopes operates based on Coriolis
force coupling between two distinct resonance modes of a single
resonator and can be divided to two operational modes: 1- rate
mode (RG) in which angular rate is measured and 2- rate
integrating mode (RIG) in which the angle is measured. In RG,
one of the resonance modes is driven as the drive mode and the
Coriolis force caused by rotation induces vibration in the other
resonance mode whose amplitude gives a measure of rotation
rate. On the other hand, in RIG, both modes are driven, so an
angle dependent energy controller, or more generally viewed as
common mode velocity feedback [1], can be implemented to
allow free precession of vibration angle between two resonance
modes [2]. In RG, the scale factor is dependent on Q factor of
the sense mode and vibration amplitude of the drive mode, and
increasing either of them will increase scale factor and as result
decrease noise (ARW) of the gyroscope [3]. Unlike RG, in RIG,
the scale factor is independent of quality factor and the
amplitude of vibration, and is only dependent on geometry of
the resonator through Coriolis coupling between two resonance
modes known as angular gain [2].
Major sources of error in RIG comes from anisoelasticity
and anisodamping. The former can be traditionally solved using
a quadrature cancellation loop [4], which is also adopted here.
Axis A
Axis B
Fig. 1: (a) SEM picture of the DRG (b) resonance mode
shapes of the 2θ wineglass modes.
On the other hand, anisodamping error has the same phase as
Coriolis force, its direct cancellation still has limited success [8].
In this paper, we present a disk resonator gyroscope (DRG)
(Fig. 1) with a ring-down time 10,000 times shorter (0.1s),
frequency 50 times higher (250 kHz) and resonator volume
1,875,000 times smaller than the classical 30 mm quartz HRG
(600 μm in diameter) in whole-angle mode operation first
presented in [2], with an electronic feedback method for
anisodamping compensation. The control algorithms are
implemented at based-band using a modification of our earlier
controller using the method of averaging [4]. In addition, a
method of RIG operation with a continual self-precession
rate [5] is investigated to facilitate identification and
compensation of general anisotropy, improve gyroscope
performance and eliminate the threshold rate. Furthermore, the
noise characterization of RIG is studied and compared with the
RG performance of the same device.
II. CONTROL APPROACH
The FPGA-based stand-alone platform presented in [2] is
used to carry on the experiments on the RIG. The block diagram
of this platform is illustrated in Fig. 2: the digital hardware
including phase-lock loop, modulation, and demodulation are
implemented in an FPGA using Verilog programming language
with reference clock of 200 MHz. The control algorithms are
implemented in a softprocessor embedded in the FPGA with a
sampling rate of 1600 Hz. The 16 bit DACs and ADCs have a
sampling rate of 2 MHz.
A. Energy and quadrature control
Methods for open-loop anisoelasticity compensation has
been comprehensively studied in [6,7]. However; further closed-
loop feedback control [1-4] is needed to compensate for the
residual anisoelasticity. The anisoelasticity error also known as
quadrature can be quantified using equation 1:
Q  2xc ys  yc xs  (1)

978-1-4673-6939-8/16/$31.00 ©2016 IEEE

Authorized licensed use limited to: Johns Hopkins University. Downloaded on March 03,2023 at 18:00:23 UTC from IEEE Xplore. Restrictions apply.
 (a)
(b) (c)
Fig. 2: (a) Block diagram of the RIG controller. (b) Measured
controller performance. Energy is controlled at an amplitude
of 70 nm, and the Q/E ratio is 0.004. (c) Measured bandwidths
of the energy controller (140 Hz) and the quadrature
controller (105 Hz).
where, xc and x s are in-phase and quadrature portions of the X
axis resonance mode signal after demodulation and yc and y s
are in-phase and quadrature portions of the Y axis. The
anisodamping error and its compensation will be discussed later
in the paper.
In addition to closed-loop controller used to null
anisoelasticity, a closed-loop energy controller is needed to
control the oscillation amplitude [2,4]. The energy is calculated
by:
E  xc2  xs2  yc2  ys2
The implementation of closed-loop controllers on quadrature
and energy, result in measured stabilized quadrature and energy
for all angles illustrated in Fig. 1 (b). The measured bandwidth
of the controllers are illustrated in Fig. 1 (c) respectively
showing a BW of 140 Hz and 105 Hz.
B. Anisodamping compensation
To compensate for anisodamping error a modified angle
dependent controller from [4] is used. The modified equation is:
 FEx  cos 
 Ey 

 F   [k p ( E0  E )  k I ( E0  E )dt ]G  
 sin  
 cos sin  
G  I   
 sin  cos 
where equation (3) is the PI controller introduced in [4] if G is
an identity matrix. In this equation E0 is energy set point, kp and
kI are the controller gains, and FEx and FEy are in-phase force
amplitudes to control the energy of the resonator in X and Y
Authorized licensed use limited to: Johns Hopkins University. Downloaded on March 03,2023 at 18:00:23 UTC from IEEE Xplore. Restrictions apply.
(a) (c)
(b)
Fig. 3: (a) Gyro output with 200 dps rate applied with rate
table. (b) Residual error before and after compensation. (c)
Fourier components of error before and after compensation.
The dominant error terms are 2θ and 4θ harmonics.
modes, and θ is the angle (sensor’s output). Equation (4) shows
the modified matrix G in which δτ (anisodamping error) and θτ
(primary axis of damping) is used to calculate its value.
The anisodamping error can be characterized through
several test presented in [2]. In this paper, we use a method by
applying a rate to calculate the anisodamping value and primary
axis of damping. Due to existence of anisodamping, the angle
precession measured from gyroscope includes 2θ angle
dependent error. By subtracting a linear fit (Fig. 3 (a)) from the
rate output and performing Fourier analysis on the residual (Fig.
3 (b)), the nθ components of the residual can be calculated. Also
frequency sweeps in both X and Y axis confirms a 13.3%
difference in quality factor and 12% different in gain of each
axis due to transducer electrode gap difference and amplifiers
gains.
After implementation of compensation for anisodamping
(2) and gain difference due to transducer electrode gap difference
and amplifiers gains, the RIG is tested under rate from a rate
table for three different conditions presented in Fig. 4. The figure
shows the angle dependency of the residual error calculated
using the method presented in Fig. 3, and the measurements are
compared with simulation models made using a modified
(3)
(4)
Fig. 4: Measured and simulated rate dependence of the
residual error: as-fabricated gyro, gain compensated, and
anisodamping compensated.
version of original base-band model presented in [4], showing a
significant reduction in residual error.
C. Phase-Locked loop (PLL)
The reference signal for phase-locked loop (PLL) used in this
work unlike [2] is a combination of signals from axis X and axis
Y. The signals of the feedback loop to the PI controller of the
PLL are shown in equation (5) and (6), Ls being the quadrature
and Lc being in phase portions of L signal:
Ls  2xc xs  yc ys 
Lc  xc2  xs2  yc2  ys2
With this reference signal the PLL will be always locking to
a signal allowing 360 degree rotation. The PLL block diagram
is illustrated in Fig. 5 (a). Having such a PLL will create an angle
dependent frequency of oscillation which will exchange
between resonance of X and Y (Fig. 5(b)). By looking at the
frequency of the NCO as the gyroscope rotates the frequency of
the modes can be extracted. Fig. 5 (c,d) illustrates the frequency
offset from PLL center frequency, showing a 0.63 Hz difference
between the two resonance modes of the gyroscope. The 2θ
components the amplitude and also direction of the maximum
and minimum resonances aligned with X and Y axes.
D. Quadrature error
By measuring the effort needed to cancel the quadrature one
can measure the primary axis of stiffness. According to [4], the
quadrature cancellation effort should follow a 2θ dependency
(Fig. 6(a)). Using Fourier series fit (Fig. 6(b)), the primary axis
of stiffness is measured to be 81.1° (Fig. 6(c)). This method of
characterization of quadrature could be potentially used to
compensate for all anisoelasticity using full position matrix
feedback control in the same manner presented here to
compensate for anisodamping.
III. PERFORMANCE LIMITS
A. Self-Precession
For further characterization of the gyroscope performance,
a method of self-precession rate is used, in which the gyro’s
vibration pattern is precessed at a constant rate by applying in-
(d)
(c)
Fig. 5: (a) PLL logic introduced by Lynch and implemented
here. (b) PLL Frequency changes between ω1 and ω2
depending on the angle, (c,d) Δf = 0.63 Hz at the operation.
Authorized licensed use limited to: Johns Hopkins University. Downloaded on March 03,2023 at 18:00:23 UTC from IEEE Xplore. Restrictions apply.
(a) (b)
(c)
(5)
(6)
Fig. 6: (a) 2θ dependence of the quadrature force measured
and Fourier fit. (b, c) Fourier analysis of the quadrature force
to find the primary axis of stiffness.
phase forces to the X and Y axes. After re-modulation, these
forces amount to antisymmetric velocity feedback at the
resonator’s natural frequency in our generalized view of RIG
operation and control [5]. The modified energy control equation
selected to apply this approach with our basedband controller is
presented in Equation (7):
( 2)
 FEx   FEx  0  1 cos 
F     ASP    (7)
 Ey   FEy  1 0   sin  
where ASP is the amplitude of the force applied to create the
self-precession rate. A benefit of using this method is that the
noise sources coming from rate table (e.g. vibration and
electromagnetic interference) are eliminated and the applied rate
may exceed the limits of the rate table. Most important, the
minimum rate threshold or rate bias can be greatly reduced or
eliminated [5]. Fig. 7(a) illustrates the calibration done with the
DRG driven at an amplitude of 70 nm and varying levels of ASP,
resulting in a force sensitivity of 0.000915 nN/dps. This self-
precession test is done for compensated and uncompensated
RIG, showing a nonlinearity in uncompensated operation
resulting from anisodamping error. Also the minimum rate in
which the RIG could self-precess is characterized using this
method, resulting in 100 dps for uncompensated and 3.5 dps for
compensated versions showing comparable results to the recent
report of 2.5 dps on a much lower frequency device [8].
However, in terms of compensation effectiveness we note that
our result represents a significant 30X reduction in
anisodamping which appears an order of magnitude more
effective. This performance is also confirmed using rate table
tests showing a 3.5 dps minimum precession rate. To overcome
this compensation residual, an operational self-precession rate
of 725 dps was implemented on the gyroscope and the
gyroscope was tested on a rate table. After applying rates from
50 dps to 500 dps in both directions for 10 full rotations, the
offset of 725 dps self-precession rate was subtracted from the
gyroscope output resulting in a direct angle measurement. The
angular gain, AG, measured in this operation is plotted in Fig. 7
(b), the same as measured for operation without any self-
precession rate. To show the benefit of a RIG operating with
self-precession, an inertial rate of 1 dps (previously undetectable
 (a)
(b) (c)
Fig. 7: (a) Self-precession calibration, showing a linear
behavior and sensitivity of 0.000915 nN/dps for 70 nm
amplitude oscillation. (b) Measured angular gain of a RIG
with fixed 725 dps self-precession rate for different rate table
speeds. (c) Net precession, θN=AG∙θI=-θ–AG∙θS due to inertial
rate input of only N =1 dps after calculated self-precession,
θS is subtracted from measured (total) precession, θ.
by the gyroscope) in both directions, each for 90 seconds, was
measured for the first time (Fig. 7(c)). We demonstrate that a
RIG operated with self-precession can detect input rates below
its compensation residual potentially eliminating rate bias due to
anisotropy.
B. Noise Limits
In this paper, we present, for the first time a method to
measure the noise performance of a RIG, similar to RG, in
anticipation of its graduation to zero rate threshold operation.
The test is done with rate and without rate. In the test with rate,
data is collected for one hour under rotation of 360dps. To
calculate the noise, the residual of angle change measured is
fitted using Fourier series, and the residuals are subtracted (Fig.
8 (a)). Using numerical derivation, the rate is calculated (Fig. 8
(b)), and Allan deviation is performed on the result (Fig. 8 (c)).
Also the Allan deviation of the output of the gyroscope after
integration is also calculated resulting in an ARW of 11
mdps/√Hz the same as the ARW of the gyroscope in rate mode
at the same drive amplitude. The comparison of the data shows
higher noise when rate is applied due to the noise from rate table
and also some singularities in fixed-point math calculations;
inside the softprocessor some low frequency (<10Hz) variation
is introduced in the output which results in higher ARW.
SUMMARY
In this paper, we have presented an improved MEMS disk
resonator gyroscope in rate integrating mode operation. MEMS
resonator gyroscopes suffer from anisoelasticity and
anisodamping errors. It is demonstrated that using a non-unity
matrix in the energy control feedback loop, anisodamping is
compensated to effectively decrease the rate threshold by 30X
to 3.5 dps. To eliminate any rate threshold due to residual
anisotropy, a method of RIG operation with self-precession has
been successfully demonstrated. Using this method, a consistent
Authorized licensed use limited to: Johns Hopkins University. Downloaded on March 03,2023 at 18:00:23 UTC from IEEE Xplore. Restrictions apply.
(a) (b)
(c)
Fig. 8: (a) Residual and residual after subtracting the first 8
harmonics. (b) Residual in dps. (c) Allan deviation plot of RIG
for operation under rate and operation without rate.
scale factor and a rate measurement for 1 dps inertial rate which
is below its anisotropic rate threshold with conventional RIG has
been demonstrated for the first time. In addition, noise limits of
the RIG have been evaluated to be approaching the same thermal
noise limits as in RG mode using the same MEMS and
electronics.
ACKNOWLEDGMENT
This work was supported by DARPA under grant W31P4Q-
11-1-0003. The authors would like to thank Professor Thomas
Kenny and his research group at Stanford University for device
fabrication.
REFERENCES
[1] D. D. Lynch, “MRIG Frequency Mismatch and Quadrature Control,”
International Symposium on Inertial Sensors and Systems (ISISS) 2014,
Laguna Beach, CA, 25-26 Feb. 2014.
[2] P. Taheri-Tehrani, O. Izyumin, I. Izyumin, C. H. Ahn, E. J. Ng, V. Hong,
Y. Yang, T. W. Kenny, B. E. Boser, D. A. Horsley, ”Disk resonator
gyroscope with whole-angle mode operation,” IEEE International
Symposium on Inertial Sensors and Systems (ISISS), 2015, Hapuna Beach,
HI, 23-26 March 2015..
[3] P. Taheri-Tehrani, M. Kline, I. Izyumin, B. Eminoglu, Y.-C. Yeh, Y.
Yang, Y. Chen, I. Flader, E. J. Ng, T.s W. Kenny, B. E. Boser, and D. A.
Horsley, “Epitaxially-Encapsulated Quad Mass Gyroscope with
Nonlinearity Compensation,” IEEE MEMS Conference, Shanghai, China,
January 24 - 28, 2016.
[4] D. D. Lynch, “Vibratory Gyro Analysis by the Method of Averaging”
Proc. 2nd Saint Petersburg Int. Conf. on Gyroscopic Technology and
Navigation, pp. 26-34, May 1995.
[5] D. Challoner, P. Taheri-Tehrani, D. Horsley, Inertial Wave Angle Gyro
[Patent Pending]
[6] P. Taheri-Tehrani, T. Su, A. Heidari, G. Jaramillo, C. Yang, S. Akhbari,
H. Najar, S. Nitzan, D. Saito, L. Lin, and D. A. Horsley, “Microscale
diamond hemispherical resonator gyroscope,” Hilton Head Workshop,
pp. 289–292, June 2014.
[7] T.-H. Su, S. Nitzan, P. Taheri-Tehrani, M.H. Kline, B.E. Boser, D.A.
Horsley, "Silicon MEMS Disk Resonator Gyroscope With an Integrated
CMOS Analog Front-End," Sensors Journal, IEEE , vol.14, no.10,
pp.3426-32, Oct. 2014.
[8] Z. Hu, B. Gallacher “Control and Damping Imperfection Compensation
for a Rate Integrating MEMS Gyroscope,” Inertial Sensors and Systems
Symposium (ISS), 2015, Karlsruhe, Germany, Sept. 22-23, 2015.