Research Article  Open Access
A Digital ModeMatching Control System Based on Feedback Calibration for a MEMS Gyroscope
Abstract
A digital modematching control system based on feedback calibration, where two pilot tones are applied to actuate the sense mode by the robust feedback controller, is presented for a MEMS gyroscope in this paper. A dualmass decoupled MEMS gyroscope with the integrated electrostatic frequency tuning mechanisms, the quadrature correction electrode, and the feedback electrode is adopted to implement modematching control. Compared with the previous modematching method of forward excitation calibration, the proposed modematching scheme based on feedback calibration has better adaptability to the variation in the frequency of calibration pilot tones and the quality factor of the sense mode. The influences of calibration pilot tone frequency and the amplitude ratio on tuning performance are studied in theory and simulation. The simulation results demonstrate that the tuning error due to the amplitude asymmetry of the sense mode increases with a frequency split between pilot tones and the drive mode and is significantly reduced by the amplitude correction technology of pilot tones. In addition, the influence of key parameters on the stability of the modematching system is deduced by using the average analysis method. The MATLAB simulation of the modematching control system illustrates that simulation results have a good consistency with theoretical analysis, which verifies the effectiveness of the closedloop modematching control system. The entire modematching control system based on a FPGA device is implemented combined with a closedloop selfexcitation drive, closedloop force feedback control, and quadrature error correction control. Experimental results demonstrate that the modematching prototype has a bias instability of 0.63°/h and ARW of 0.0056°/h^{1/2}. Compared with the modemismatched MEMS gyroscope, the performances of bias instability and ARW are improved by 3.81 times and 4.20 times, respectively.
1. Introduction
Over recent years, MEMSbased gyroscopes have found widespread application in the automotive and consumer fields, including rollover detection and electronic stability control, gaming consoles, and image stabilization in digital cameras [1–3]. However, MEMS gyroscopes have yet to satisfy the navigationgrade requirements needed for highprecision applications such as an inertial measurement unit for GPSaugmented navigation, unmanned surveillance vehicles, and aircraft and personal positioning systems [4–6]. A variety of means have been proposed to improve the accuracy of MEMS gyroscopes, such as array gyroscopes with multimasses [7], 3D fabrication techniques with highaspect ratio and effective volume utilization [8], digital control method based on ΔΣ technology [9], and modematching technology [1]. Modematching technology is widely concerned and becomes one of the most promising approaches to improve the accuracy of MEMS gyroscopes.
Under modematched conditions, the sense mode is designed to own the same (or nearly the same) resonant frequency with the drive mode. Due to Qfactor amplification, gyroscopes operated under modematched configuration offer higher sensitivity and better resolution [10]. It is hard to completely match the resonant frequencies between two modes through structural design due to the fabrication imperfections. Various approaches have been developed to reduce the frequency mismatch between two resonance modes in the literature. Postfabrication by depositing selective polysilicon [11] or laser trimming [12], localized thermal stress [13], and some other approaches were utilized in the early stage. However, these approaches, which require a manual tuning effort, are not suitable for mass production and temperature variation. Another effective approach based on electrostatic spring softening was adopted to tune the resonance mode frequencies by an adjustable DC potential [14]. Various means have been developed to acquire the appropriate tuning DC voltage. Evolutionary computation [15, 16] and extremumseeking technique [17–19] are introduced to attain the tuning voltage by seeking the peak of the quadrature signal or sense mode. These approaches are either timeconsuming or to be done offline, in which the dynamic requirements are difficult to meet. Another automatic modematching method owned high control precision in closedloop control, which obtained continuous information about the resonance frequency of the sense mode by introducing a square wave dither signal or injecting outofband pilot tones [20–22]. The phase domain approach based on matching the phase relationship between two modes by PLL can be employed to fulfill the modematching control [23, 24]. However, the phase domain will suffer from phase coupling and uncertain parasitic phase shift stemmed from the readout circuits, which could lead to instability of the control system.
This paper presents a digital modematching control loop based on feedback calibration for MEMS gyroscopes. Two pilot tones located on both sides of the drive signal are used to actuate the sense mode and estimate the frequency mismatch by the robust feedback controller, which enhances the adaptability to the variation in the frequency of calibration pilot tones and the quality factor of the sense mode. In Section 2, the structure principle of microgyroscopes is briefly reviewed. Then we describe the implementation of the control system of modematching and analyze the control characteristics of the system. The control system simulations are given in Section 3. We present experimental result demonstration in Section 4. Concluding remarks are finally given in the last section.
2. ModeMatching System Design and Analysis
2.1. MEMS Gyroscope Design
Structural schematic of the dualmass decoupled MEMS gyroscope is shown in Figure 1 [25]. The MEMS gyroscope structure consists of two identical substructures. Due to the decoupling effect of decoupled beams (including the drive and sense decoupled beams), the drive and sense frames can vibrate along only one direction (axis or axis), whereas the proof masses vibrate along two orthogonal directions to achieve Coriolis coupling between the drive and sense modes. The kinematic coupling of two substructures in the drive direction is implemented by drivecoupling suspension beams. Similarly, a complex lever system, which consists of the coupling crossbeams, sensecoupling suspension beams, and leversupporting beams, is used to achieve the kinematic coupling of two substructures in the sense direction. The kinematic couplings in the drive or sense direction ensure that two substructures vibrate with the same frequencies in the inphase and antiphase modes. Although the drivedecoupled beam is optimized to suppress the mechanical coupling between the drive and sense modes, fabrication imperfections can lead to an undesired mechanical crosstalk (quadrature error) between two modes. Quadrature correction electrodes, with the help of dedicated closedloop quadrature cancellation electronics, are designed to cancel the quadrature error by applying the automatically adjusted DC potentials. In order to obtain the maximum mechanical sensitivity, the frequency of the driving mode is basically tuned to be identical with that of the sense mode by finite element simulation. However, unavoidable residual stress, fabrication tolerances, and temperature variation lead to mismatched mode frequencies. Therefore, the frequency tuning mechanisms are designed to automatically tune the sense mode frequency to approximate the drive modal frequency by using the electrostatic spring softening capability of the tune combs.
When the tuning voltage is applied on the frequency tuning electrodes, the subsequent electrostatic negative stiffness is presented as [9–11] where is the number of tuning combs, is permittivity, is comb thickness, is the length, and is the gap of the comb. The ideal mode matching can be achieved by changing the stiffness of the proof mass in the sense mode. When the tuning voltage is exerted on tuning combs, the resonant frequency of the sense mode can be described as where , , and are the original resonant frequency, stiffness, and proof mass of the sense structure, respectively. And . The equation can be simplified further as where is the mechanical parameter and . The structure parameters of the dualmass decoupled MEMS gyroscope are shown in Table 1. In the dualmass decoupled MEMS gyroscope with frequency tuning mechanism, a 35 Hz mode mismatch error between the drive mode and sense mode is designed in the structure.

2.2. Proposed ModeMatching Method
The proposed modematching control loop in the paper is shown in Figure 2. refers to the transfer function of the sense mode, represents the feedback gain, refers to the interface gain, represents the demodulation compensation term, LPF (lowpass filter) is used to filter the highfrequency signal after demodulation, and then the filtered signal is fed into a PI (proportional integral) controller to generate DC voltage for frequency tuning. Different from previous frequency calibration methods which inject directly pilot tones into the sense mode in the forward path and check the amplitude difference between the tones [21, 22], two pilot tones () in the proposed calibration means, whose frequencies are located at both sides of the drive mode symmetrically, are introduced after the frontend preamplifier and then exerted on the sense mode through force feedback path.
Disconnecting the feedback control loop of the tuning voltage and neglecting the influence of Coriolis signal, the transfer function of the modematching system in the openloop conditions can be expressed as
Assuming that the openloop gain is much greater than one, equation (4) can be simplified as where is resonant frequency of sense mode and is the Qfactor of the sense mode.
Figure 3 compares two tuning approaches based on pilot tone excitation. Obviously, there is a significant difference between the two openloop systems. The openloop system in the previous calibration method is similar to a bandpass system, while the openloop system in the proposed calibration method is similar to a band elimination system. Two selected pilot tones ( and ), whose frequencies are referenced to the resonant frequency of the drive mode () and located symmetrically at the both sides of the drive mode, are applied to excite the sense mode. The difference of amplitudefrequency responses between two tones points out the direction of frequency tuning and is used to control the frequency tuning mechanism to implement the mode matching. In the proposed method shown in Figure 3(b), once the resonant frequency of the sense mode () is higher than that of the drive mode (), the tuning control circuit will be stimulated to increase the tuning voltage because the amplitude response of sideband is much greater than that of sideband . On the contrary, when the amplitude response of sideband is much less than that of sideband , which demonstrates that the resonant frequency of the sense mode is already less than that of the driving mode, the tuning control circuit will be promoted to decrease the tuning voltage. Finally, the frequencies of the drive mode and sense mode are matched when the amplitude responses in two sidebands of and are equal. The previous approach shown in Figure 3(a) has a similar control process.
(a)
(b)
Compared with the previous tuning approach, the amplitude responses of two sidebands in the proposed approach, especially under high quality factor conditions, can obtain better signaltonoise ratio due to the band elimination characteristic in the openloop system. At the same time, the frequency difference between pilot tones and the drive mode can be increased to further improve the signaltonoise ratio. The above characteristics are conducive to improving the control performance of the system. In addition, the proposed approach enhances the adaptability to the variation in the frequency of calibration pilot tones and quality factor of the sense mode, which will be discussed in detail in the next section.
In summary, compared with the previous openloop tuning, offline tuning, and closedloop tuning shown in the Figure 3(a) method, the proposed tuning approach, which is an inrun calibration technique and conducive to more precise realtime tuning control, is expected to implement a better SNR and get better bias instability and has better adaptability to the frequency of calibration pilot tones and quality factor of the sense mode.
2.3. System Characteristics
2.3.1. Frequency Limit of Calibration Pilot Tones
In order to analyze the pilot tone frequency limit of the proposed modematching technique based on feedback calibration, the openloop amplitude phase response of the frequency tuning system is analyzed. Figure 4 illustrates the block diagram of openloop frequency tuning. The PI controller is removed compared with the closedloop modematching system shown in Figure 2. We focus on the openloop output of LPF with varying tuning voltages which will be fed into the PI controller to generate the tuning voltage in the closedloop modematching system. The openloop tuning parameters are shown in Table 2.

Assuming that the openloop gain is much greater than one, the transfer function of the openloop tuning system can be expressed as
The responses in two pilot tones can be written as
The LPF module is applied to filter invalid highfrequency signals generated by multiplying and . Then the output of LPF can be derived. where
Figure 5 illustrates the openloop LPF output with varying tuning voltages in different frequencies of calibration pilot tones according to the simulation parameters in Table 2.
(a)
(b)
The resonant frequency of the sense mode is equal to that of the drive mode when the tuning voltage reaches 5.16 V under the given parameters. Five different frequency splits between calibration pilot tones and the drive mode vary from 10 Hz to 200 Hz as shown in Figure 5. The frequencies of five calibration pilot tones are representative and located in different locations, including those closing to the driving mode, those closing to the sense mode, those between the driving mode and the sense mode, or those away from the sense mode. The LPF outputs with different frequency splits shown in Figure 5(a) indicate good consistency during openloop tuning procedure and are basically the same with the adjustment process in the closedloop modematching control. And the tuning voltages are basically negatively correlated with the outputs of the LPF in different frequency splits. The monotonic characteristic facilitates the realization of closedloop control, which proves that the proposed modematching method has good adaptability to the frequency change of calibration pilots.
However, simulation results shown in Figure 5(b) illustrate that the LPF outputs have a great correlation with the frequency location of the calibration pilot tones in the forward modematching method. The LPF outputs do not increase or decrease monotonically with the increase of tuning voltage in certain calibration pilot tones, especially in 10 Hz or 25 Hz frequency split between calibration pilot tones and the drive mode, which means that the ideal mode matching can hardly be achieved by the traditional closedloop control theory. Simultaneously, the LPF outputs decrease monotonically with the increase of tuning voltage in 100 Hz or 200 Hz frequency split between calibration pilot tones and the drive mode; nevertheless, the output signal of LPF is extremely weak and is likely to be submerged in noise due to the large quality factor of the sense mode, which results in the subsequent signal not being processed. The above analysis shows that the forward modematching method is only suitable for lowquality factors.
Simulation results demonstrate that the proposed tuning method has good redundancy for the variation in calibration pilot tone frequencies and the quality factor of the sense mode, which is significantly better than the forward modematching method where the frequency of calibration pilot tones must be chosen to be within a certain frequency band [26], that is,
In addition, five zero crossing points shown in Figure 5(a) locate at different tuning voltages, which will result in frequency tuning errors shown in Table 3. The frequency tuning errors are mainly due to the amplitude asymmetry between two calibration pilot tones in the sense mode. This will be highlighted in the next section.

Theoretically, the change in the frequency of the calibration pilot tones does not affect the bias stability or Allan deviation measurement in the proposed modematching method. However, the actual circuit implementation may affect the final bias stability or Allan deviation measurement. If the frequencies of the calibration pilot tones are too close to that of the drive mode, in order to avoid the Coriolis signal and the quadrature signal coupling into the tuning loop shown in Figure 2, the cutoff frequency of the LPF in the tuning loop shown in Figure 2 must be much smaller than the frequency difference between the calibration pilot tones and the drive mode. Therefore, too low cutoff frequency of LPF will lead to difficulty in parameter designing of LPF and will also affect the dynamic performance and adjustment speed of the tuning control loop. On the contrary, if the frequencies of the calibration pilot tones are too far from that of the drive mode, in order to maintain a large output in the tuning loop, a larger signal needs to be fed back to the input through and the feedback signal may be much larger than the voltage limit of the power supply, which will lead to difficulties in circuit implementation. If the frequencies of the calibration tones are too close to or away from that of the drive mode, it may degrade the performance of the tuning loop system, eventually leading to the deterioration of the bias stability or Allan deviation measurement. Therefore, the frequency difference between the calibration pilot tones and the drive mode needs to be compromised and compatible with the above two situations.
2.3.2. Amplitude Ratio of Calibration Pilot Tones
Different frequency splits between calibration pilot tones and drive mode result in different zero crossing points, which means how much the tuning voltage can reach in the closedloop control system.
According to equation (8), the zero crossing points refer to
Resolving equation (11) under the perfect tuning voltage can figure out the amplitude ratio of two pilot tones.
Because is much smaller than , equation (13) can be simplified as where represents the frequency split between calibration pilot tones and the drive mode. Obviously, the amplitude ratio raises with the increase of frequency split.
Figure 6 verifies the function of equation (13). We set the frequency split . Obviously, when two calibration tones adopt the same amplitude, the amplitude difference between two calibration pilot tones in the sense mode results in error of the zero crossing point, which will eventually cause significant frequency tuning errors shown in Figure 7.
However, when the amplitudes of two calibration tones are corrected according to equation (14), a perfect zero crossing point shown in Figure 6 can be implemented. Figure 7 illustrates the change of frequency tuning error with a frequency split between calibration pilot tones and the drive mode. From Figure 7, the frequency split of 60 Hz between calibration pilot tones and the drive mode results in an error of 0.019% or 0.61 Hz frequency offset, while the frequency split of 200 Hz results in an error of 0.19% or 6.17 Hz offset. The frequency tuning errors mainly stem from imperfect symmetry of amplitude responses of calibration pilot tones in the sense mode. Simultaneously, the frequency tuning errors shown in Figure 7 can be significantly reduced by correcting the amplitude of two calibration tones.
2.3.3. The Impact of the Quality Factor in the Sense Mode
Figure 8 shows the openloop LPF output with different quality factors of the sense mode in a 60 Hz frequency split between calibration pilot tones and the drive mode. Simulation results demonstrate that the openloop LPF outputs in the proposed method are basically not affected by the quality factor change throughout the openloop tuning process. However, the traditional forward modematching method is obviously affected by the quality factor due to the presence of resonance peaks in the tuning process. When the frequency deviation between calibration pilot tones and the drive mode is small and does not satisfy the formula shown in equation (10), the increase of the quality factor will lead to a sharper resonance peak shown in Figure 5(b), which will potentially cause significant amplitude fluctuations, and even lead to unsuccessful closedloop control during the frequency tuning. When the frequency deviation between calibration tones and the drive mode is large and satisfies the formula shown in equation (10), the openloop LPF output signals in the forward modematching method are weak under a large quality factor, which results in a poor signaltonoise ratio. The above analysis indicates that the traditional forward modematching method is more suitable for lowquality factors. Compared with the traditional modematching method, the proposed modematching technique illustrates a good adaptability to the change of quality factor and can significantly improve the signaltonoise ratio of the modematching loop under high quality factors and is especially suitable for large quality factor environments.
(a)
(b)
2.4. Stability Analysis
The proposed modematching technique based on feedback calibration guarantees good compatibility with the variation in the quality factor and frequency split. However, the feedback calibration with pilot tones introduced in the proposed modematching technique easily leads to system instability. A stability analysis of the modematching control loop shown in Figure 2 is implemented. To simplify the complexity of the analysis, the calibration signal mixed with two different frequencies tones is replaced by one tone and the reference point of the PI controller is also changed into a specific parameter instead of zero correspondingly.
The modematching system can be expressed as where represents the integration gain of the PI controller, represents the output of LPF, represents the cutoff frequency of the lowpass filter. Dynamic equation (15) constitutes the frequency tuning system. To analyze system transient properties, suppose that and , where , , and represent the transient amplitude, transient phase angle, and phase, respectively. Thus can be derived as
Relative to , , and are slowly varying variables, assuming that
Then the secondorder derivative of voltage is
System transient characteristics can be derived from simultaneous equations (15)–(18) as
Equation (19) offers a transient relationship among the variables of amplitude , phase , LPF output , and tuning voltage . However, formulas in equation (19) are coupled reciprocally and nonlinear, which indicate the that characteristics of the system are implicit and difficult to analyze. Since phase angle is a quickly varying variable and , , , and are slowly varying variables, the averaging method can be applied. Setting the integral interval as , equation (19) can be rewritten as
The bar above the variables represents the average over one cycle. The average equation (20) is still complex and nonlinear. To further simplify the equation, local linearization can be done around its equilibrium point. Letting the formula on the right of equation (20) be zero, you can get the equilibrium points of each variables. Since four variables (, , , and ) are coupled to each other and the equations are complicated, it is difficult to directly obtain the analytical solutions of the equilibrium points. Therefore, we obtain the equilibrium points of the system by solving the numerical equations in the subsequent analysis. Numerical analysis shows that the equilibrium points are unique since the tuning voltage is always artificially set to positive, which can be further confirmed by subsequent numerical simulation results.
The relevant Jacobian matrix can be derived as where , ,, and are the only set of equilibrium points of each variables.
The characteristic equation of expression (21) is
The coefficients of characteristic variable are where represents the element of the Jacobian matrix and represents the coefficient of the characteristic variable. According to the RouthHurwitz stability criterion, the system is stable only when all characteristic roots have negative real components, that is, equation (24) must be satisfied. The impact of the key parameters on system stability, including , , , and , is indeterminate due to the coupled intricacy. By applying the control variable method, the relationships between analytical solutions of critical values , , , and are shown in Figure 9 according to stationary parameters shown in Table 2. Colored blocks in Figure 9 represent the stable area. Obviously, the critical value is in direct proportion to in general, but in inverse proportion to . And the change of the quality factor has no significant effect on the critical value, which confirms that the system has better redundancy for the change of quality factor, which is basically consistent with the previous theoretical analysis. The criterion inequality (24) only represents the stable condition near the equilibrium point. Since equilibrium points are unique, once the system is balanced, it must converge to the only equilibrium state, which will be confirmed in the subsequent system simulation.
(a)
(b)
3. Simulation
Figure 10 shows the block diagram of the closedloop modematching system with feedback calibration constructed in Simulink. The model is designed to observe the transient responses of the modematching system and verify the analysis of openloop system characteristics in the abovementioned chapter. The parameters of sense dynamics, pilot tones, and gain terms are kept the same as the openloop parameters summarized in Table 2, while and of the PI controller are 20 and 10000, respectively, which is set to satisfy the constraints of stability illustrated in Figure 9. The frequency split between the calibration tone and the drive mode is set to 60 Hz. Figure 11 illustrates the transient responses of the modematching system.
(a)
(b)
(c)
(d)
(e)
(f)
(g)
As shown in Figure 11(a), the signal of the preamplifier is modulated by two calibration tones and mainly used to extract the excitation amplitude of the two calibration tones in the sense mode for subsequent evaluation of the tuning state. Simultaneously, the output of the LPF shown in Figure 11(b) is essentially zero after 0.5 s approximately. And the output of PI shown in Figure 11(c), that is, the tuning voltage, basically reaches the tuning point, which indicates that the amplitudes modulated by two calibration pilot tones in the sense mode are substantially equal and the mode matching is basically realized. We also alter the frequency split between the calibration pilot tone and the drive mode by adjusting the frequency of pilot tones, shown in Figure 11(d). Apparently, different frequency splits between the calibration tone and the drive mode lead to different frequency tuning errors. And the increase of frequency split enlarges the frequency tuning error, which is consistent with the analysis of equation (14). We compare the tuning effects at different quality factors, shown in Figure 11(e). Obviously, the frequency tuning voltage is basically the same under different quality factors. Finally, we analyze the frequency tuning error, shown in Figure 11(f). The frequency split between the calibration pilot tone and the drive mode is increased to 100 Hz in order to significantly demonstrate the effect of optimization. Simulation results show that a significant frequency tuning error of 3.29 Hz appears when the amplitude ratio of two pilot tones is 1. We use equation (14) to calibrate the amplitude ratio of two pilot tones. Then the frequency tuning error is significantly reduced to 0.14 Hz. Figures 11(d)–11(f) demonstrate that the closedloop characteristics of the modematching system are in consistency with openloop analysis results as shown in Figures 6–8. System stability comparison in different simulation parameters is shown in Figure 11(g). When simulation parameter , where the closedloop modematching system is in the stable region according to the theoretical curve of Figure 9(a), the tuning voltage quickly reaches the steady state. The stability simulation results show that the whole system is stable and the equilibrium state is unique and the system converges to the only equilibrium stable after stabilization, which verifies that the stability analysis in Section 2.4 is correct. However, when the simulation parameters exceed the stable region shown in Figure 9(a), the tuning voltage reaches saturation and the entire system loses stability control, which confirms that theoretical analysis of stability is basically credible.
4. Experiment
Experiments have been performed to verify the theoretical analysis and evaluate the characteristics of modematched MEMS gyroscopes compared to the former modemismatched state. Figure 12 shows the picture of the fabricated dualmass decoupled MEMS gyroscope with a frequency tuning mechanism.
(a)
(b)
In order to facilitate the realization of subsequent closedloop circuits, the openloop tuning capability of the tuning mechanism is first tested. Figure 13 shows the resonant frequency characteristics of the sense and drive modes. Preliminary openloop sweep experiments shown in Figures 13(a) and 13(b) indicate that the drive mode has an original resonant frequency of 3057.5 Hz and a quality factor of 15760, while the sense mode has an original resonant frequency of 3083.5 Hz and a quality factor of 12230. As the tuning voltage increases from 0 V to 8 V, the resonant frequency of the sense mode shown in Figure 13(c) decreases from 3083.65 Hz to 3026.15 Hz due to the electrostatic tuning capability, while the resonant frequency of the drive mode remains essentially constant at 3057.51 Hz. The perfect mode matching is implemented approximately with the tuning voltage of 5.41 V for the MEMS gyroscope shown in Figure 1.
(a)
(b)
(c)
Combined with classic closedloop selfexcitation drive, closedloop force feedback control, and quadrature error correction control, the entire control system scheme of the modematching MEMS gyroscope shown in Figure 14 is constructed according to the control scheme in Figure 10. All control loops are implemented in a FPGA device. The function and implementation of three classic control loops, which are indispensable in the control of the modematching MEMS gyroscope, are simplified as they have been expounded in our previous work [25]. The FPGA chip is EP3C55F484I7 from the Altera Corporation and has programmable logic units, RAM bits, 156 multipliers, and 4 PLL. The entire hardware platform of the FPGA is based on fixedpoint arithmetic processing with a main clock frequency of 12.29 MHz. A chip of AD7767, which is a highperformance, 24bit, oversampled SAR analogtodigital converter (ADC) with a lowtemperature drift and large dynamic range and input bandwidth, is utilized to implement analog to digital conversion with the sampling frequency of 48 kHz. A 24bit Stereo D/A Converter of CS4344 based on a fourthorder multibit deltasigma modulator with a linear analog lowpass filter is used to implement digital to analog conversion with the same sampling frequency of 48 kHz. The voltagecontrolled oscillator (VCO) in the modematching controller is controlled by a frequency control word generated from the drive mode controller. The frequencies of two excitation pilot tones are designed to change with the drive mode resonant frequency that does not remain unchanged but vary with the external environment, which could enhance the adaptability of the modematching system. Figure 15 shows the prototype of the modematching MEMS gyroscope.
(a)
(b)
The experiments are implemented to verify the theoretical analysis and simulation results about the characteristics of the modematching technique represented in this paper. The influences of the amplitude ratio of two pilot tones and the frequency split between pilot tones and the drive mode are analyzed. The openloop tuning experiment of the modematching system is shown in Figure 16.
(a)
(b)
Compared with the simulation results in Figure 5(a), the experiment results shown in Figure 16(a) indicate that the openloop LPF output curves with different tuning voltages share the same tendency with the simulated result in different frequencies of calibration pilot tones. This illustrates that the frequency variation of calibration pilot tones will not substantially alter the tuning process but results in different tuning errors, which is basically consistent with the simulation results. However, a zerocrossing point of tuning voltage 5.09 V in the calibration pilot tone with a 30 Hz frequency spilt with respect to the drive mode results in 3.01 Hz tuning error, while a zerocrossing point of tuning voltage 4.33 V in the calibration pilot tones with a 40 Hz frequency spilt results in 9.42 Hz tuning error. Compared with the simulation results, the tuning errors in the experimental results are obviously large due to the difference of the quality factor, the asymmetry of the resonance curve, and the reduction of the feedback gain due to the clamping of the power supply voltage. Simultaneously, the amplitude ratio between two calibration pilot tones is corrected as 1.43 to further suppress tuning errors shown in Figure 16(b). Experimental results confirm that the tuning error is reduced from 9.42 Hz to almost zero, which verifies the effectiveness of the method. Nevertheless, the actual correction coefficient of the amplitude ratio has some differences with the theoretical value shown in equation (14), which is also related to the asymmetry of the resonance curve in the sense mode.
In order to verify the effect of mode matching, the signal waveforms of the modematching control system are measured as shown in Figure 17. Figure 17(a) shows the feedback signal of calibration pilot tones, which is primarily used to modulate the sense mode to extract amplitudes in the frequencies of calibration pilot tones and provides reference for subsequent modematching control. The preamplifier signal waveforms of the sensitive interface shown in Figure 17(b) contain the calibration signal modulated by two pilot tones, quadrature error signal and offset error signal, which shows an atypical modulated wave.
(a)
(b)
(c)
(d)
In order to further analyze the signal components of the feedback calibration signal, the frequency spectrum of the feedback calibration signal is collected as shown in Figure 17(c). A significant amplitude difference in the frequencies of two pilot tones is observed before tuning. The main reason is that two pilot tones are located in an asymmetrical position on the openloop curve shown in Figure 3(b). Since the left pilot tone is far away from the resonance frequency of the sense mode, the amplitude response in the left pilot tone is larger than that in the right pilot tone. However, the amplitude responses in the frequencies of two pilot tones are basically equivalent after tuning, which is consistent with the basic principle of the proposed modematching scheme shown in Figure 3(b). Similarly, Figure 17(d) shows the transient response of preamplifier signal waveforms and tuning voltage. Tuning control results indicate that the tuning voltage and the preamplifier signal are stable after 0.7 s. Since the frequencies of calibration pilot tones are different from those of the quadrature error and the Coriolis signal, the tuning control loop does not interfere with the normal operation of the gyroscope. The transition time in the experiment is slightly longer than that in the simulation. The main reason is that the gain in the tuning loop is appropriately reduced due to the compatibility stability in other control loops and the limitation of the circuit supply voltage. The output of the tuning voltage indicates that the final control system implements a mode matching within 0.29 Hz. The above signal detection fully proves the effectiveness of the closedloop force feedback control, quadrature error correction control, and modematching control.
Finally, the performance experiments of the prototype are implemented. Figure 18 shows the inputoutput relationship curve. The prototype with modematching and modemismatching conditions has scale factors of −10.90 mV/°/s and 25.52 mV/°/s, respectively, under an input rate range of ±100^{o}/s. Nonlinear analysis shows that the nonlinearities of the prototype under the condition of mode matching and mode mismatching are 478.73 ppm and 669.81 ppm, respectively. Simultaneously, the drift characteristics of prototype are evaluated under modematching and modemismatching conditions. Figure 19 compares the Allan variance under the modematched and modemismatched conditions. Compared with those in the modemismatched state, the bias instabilities in the modematching state decrease from 2.40°/h to 0.63°/h and ARWs decrease from 0.0235°/h^{1/2} to 0.0056°/h^{1/2}, respectively. Experimental results demonstrate the bias instability and ARW performance are substantially improved by 3.81 times and 4.20 times, respectively, due to the modematched control. The Allan variance curve is not an ideal curve, especially when the Allan variance time is less than 10 s. The main reason is that some slowly variational interference signals with lowfrequency or tiny periodic signals, which come possibly from the unsatisfactory layout of the chip or interface circuit, fabrication error, and so on, are coupled into the circuit.
Indeed, the decrease of the frequency difference between the drive mode and sense mode by tuning means is beneficial in improving the mechanical sensitivity and bias stability or Allan deviation measurement. However, it will lead to an increase of nonlinearity due to the enhancement in the mechanical motion amplitude of the sense mode, which will decrease the dynamic range, since the closedloop detection technology used in the system will increase the dynamic range of the system and the entire detection range is not large. Therefore, the tuning technique does not have a significant impact on the measurement range.
5. Conclusions
This paper proposes a digital modematching control loop based on feedback calibration for a MEMS gyroscope. The structure of a dualmass decoupled MEMS gyroscope with integrated electrostatic frequency tuning mechanisms, quadrature correction electrode, and feedback electrode is presented briefly. A digital modematching control scheme, where two pilot tones are applied to actuate the sense mode and extract tuning voltage by the robust feedback controller, is designed. Compared with the previous modematching method of forward excitation calibration, the proposed modematching scheme based on feedback calibration has better adaptability to the variation in the frequency of calibration pilot tones and the quality factor of the sense mode. The influences of the calibration pilot tone frequency and amplitude ratio on tuning performance are studied in theory and simulation. Theoretical analysis and simulation results demonstrate that the tuning error increases with a frequency split between pilot tones and the drive mode, which is mainly due to the amplitude asymmetry of the sense mode. Simultaneously, the tuning error is significantly reduced by the amplitude correction technology of pilot tones. And the proposed modematching method has better robustness to changes in the quality factor of the sense mode. The relationship between the stability of a modematching control system and key parameters is deduced by using the average analysis method. The MATLAB simulation model of the modematching control system is constructed. The simulation results have a good consistency with theoretical analysis, which verifies the effectiveness of the closedloop modematching control system. The entire modematching control system combined with the closedloop selfexcitation drive, closedloop force feedback control, and quadrature error correction control is implemented in a FPGA device. Experimental results demonstrate that the modematching prototype has a bias instability of 0.63°/h and ARW of 0.0056°/h^{1/2}. Compared with those of the modemismatched MEMS gyroscope, the performance of bias instability and ARW of the modematched prototype are improved by 3.81 times and 4.20 times, respectively.
Data Availability
The data used to support the findings of this study are included within the article.
Conflicts of Interest
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgments
The authors wish to thank the support from the National Natural Science Foundation of China (Grant nos. 61571126 and 61874025), the Aviation Science Foundation (Grant no. 20150869005), Equipment Preresearch Field Foundation (Grant no. 6140517010316JW06001), the Fundamental Research Funds for the Central Universities (Grant no. 2242018k1G017), and the Eleventh Peak Talents Programme Foundation in the Six New Industry Areas.
References
 I.P. Prikhodko, S. Nadig, A.J. Gregory, A.W. Clark, and W.M. Judy, “Halfamonth stable 0.2 degreeperhour modematched MEMSgyroscope,” in 2017 IEEE International Symposium on Inertial Sensors and Systems (INERTIAL), pp. 1–4, Kauai, HI, USA, March 2017. View at: Publisher Site  Google Scholar
 T. Hiller, B. Kuhlmann, A. Buhmann, and H. Roth, “Noise contributions in a closedloop MEMS gyroscope for automotive applications,” in 2017 IEEE International Symposium on Inertial Sensors and Systems (INERTIAL), pp. 62–65, Kauai, HI, USA, March 2017. View at: Publisher Site  Google Scholar
 J. Giner, Y. Zhang, T. Shiota, D. Maeda, and K. Ono, “A concentrated springs architecture for singledigit frequency symmetry in Si MEMS gyroscope,” in 2016 IEEE SENSORS, pp. 1–3, Orlando, FL, USA, OctoberNovember 2016. View at: Publisher Site  Google Scholar
 T. Tsukamoto and S. Tanaka, “FM/rate integrating MEMS gyroscope using independently controlled CW/CCW mode oscillations on a single resonator,” in 2017 IEEE International Symposium on Inertial Sensors and Systems (INERTIAL), pp. 1–4, Kauai, HI, USA, March 2017. View at: Publisher Site  Google Scholar
 I.P. Prikhodko, J.A. Gregory, W.A. Clark et al., “Modematched MEMS Coriolis vibratory gyroscopes: myth or reality?” in 2016 IEEE/ION Position, Location and Navigation Symposium (PLANS), pp. 1–4, Savannah, GA, USA, April 2016. View at: Publisher Site  Google Scholar
 B. Johnson, K. Christ, D. Endean, B. Mohr, and R. Supino, “Tuning fork MEMS gyroscope for precision northfinding,” in 2015 DGON Inertial Sensors and Systems Symposium (ISS), pp. 1–10, Karlsruhe, Germany, September 2015. View at: Publisher Site  Google Scholar
 T.P. Tehrani, M. Kline, I. Izyumin, and B. Eminoglu, “Epitaxiallyencapsulated quad mass gyroscope with nonlinearity compensation,” in 2016 IEEE 29th International Conference on Micro Electro Mechanical Systems (MEMS), pp. 966–969, Shanghai, China, January 2016. View at: Publisher Site  Google Scholar
 A. Darvishian, B. Shiari, J.Y. Cho, T. Nagourney, and K. Najafi, “Anchor loss in hemispherical shell resonators,” Journal of Microelectromechanical Systems, vol. 26, no. 1, pp. 51–66, 2017. View at: Publisher Site  Google Scholar
 B. Sheng, F. Chen, C. Qian, D. Xu, S. Guo, and X. Li, “A dual quantization electromechanical sigmadelta modulator vibratory wheel gyroscope,” in 2016 IEEE SENSORS, pp. 1–3, Orlando, FL, USA, OctoberNovember 2016. View at: Publisher Site  Google Scholar
 R. Gando, H. Kubo, K. Masunishi et al., “A catchandrelease drive MEMS gyroscope with enhanced sensitivity by modematching,” in 2017 IEEE International Symposium on Inertial Sensors and Systems (INERTIAL), pp. 50–53, Kauai, HI, USA, March 2017. View at: Publisher Site  Google Scholar
 D. Joachim and L. Lin, “Characterization of selective polysilicon deposition for MEMS resonator tuning,” Journal of Microelectromechanical Systems, vol. 12, no. 2, pp. 193–200, 2003. View at: Publisher Site  Google Scholar
 M.A. Abdelmoneum, M.M. Demirci, L.Y. Wei, and C.T.C. Nguyen, “Locationdependent frequency tuning of vibrating micromechanical resonators via laser trimming,” in Proceedings of the 2004 IEEE International Frequency Control Symposium and Exposition, 2004, pp. 272–279, Montreal, Quebec, Canada, August 2004. View at: Publisher Site  Google Scholar
 T. Remtema and L. Lin, “Active frequency tuning for microresonators by localized thermal stressing effects,” Sensors and Actuators A: Physical, vol. 91, no. 3, pp. 326–332, 2001. View at: Publisher Site  Google Scholar
 S. Sonmezoglu, S.E. Alper, and T. Akin, “An automatically modematched MEMS gyroscope with wide and tunable bandwidth,” Journal of Microelectromechanical Systems, vol. 23, no. 2, pp. 284–297, 2014. View at: Publisher Site  Google Scholar
 D. Keymeulen, W. Fink, and M.I. Ferguson, “Evolutionary computation applied to the tuning of MEMS gyroscopes,” in Proceedings of the 2005 conference on Genetic and evolutionary computation  GECCO '05, pp. 927–932, Washington DC, USA, June 2005. View at: Publisher Site  Google Scholar
 D. Keymeulen, M.I. Ferguson, and L. Breuer, “Tuning of MEMS gyroscope using evolutionary algorithm and “switched driveangle” method,” in 2006 IEEE Aerospace Conference, pp. 1–8, Big Sky, MT, USA, March 2006. View at: Publisher Site  Google Scholar
 R. Antonello and R. Oboe, “Modematching in vibrating microgyros using an extremum seeking controller with switching logic,” in 2008 34th Annual Conference of IEEE Industrial Electronics, pp. 3425–3430, Orlando, FL, USA, November 2008. View at: Publisher Site  Google Scholar
 R. Antonello and R. Oboe, “Automatic mode matching in MEMS vibrating gyroscopes using extremumseeking control,” IEEE Transactions on Industrial Electronics, vol. 56, no. 10, pp. 3880–3891, 2009. View at: Publisher Site  Google Scholar
 M.F. Zaman, A. Sharma, and Z. Hao, “A modematched siliconyaw tuningfork gyroscope with subdegreeperhour Allan deviation bias instability,” Journal of Microelectromechanical Systems, vol. 17, no. 6, pp. 1526–1536, 2008. View at: Publisher Site  Google Scholar
 C.D. Ezekwe and B.E. Boser, “A modematchingΣΔclosedloop vibratory gyroscope readout interface with a 0.004°/s/√Hznoise floor over a 50 Hz band,” IEEE Journal of SolidState Circuits, vol. 43, no. 12, pp. 3039–3048, 2008. View at: Publisher Site  Google Scholar
 A. Thomae, A. Johannes, and R. Neul, “Method and device for tuning a first oscillator with a second oscillator and rotation rate sensor,” US Patent Us6654424B1, 2003. View at: Google Scholar
 H. Wu, System Architecture for ModeMatching a MEMS Gyroscope, [M.S. thesis], Massachusetts Institute of Technology, Cambridge, MA, USA, 2009.
 S. Sonmezoglu, S.E. Alper, and T. Akin, “A high performance automatic modematched MEMS gyroscope with an improved thermal stability of the scale factor,” in 2013 Transducers & Eurosensors XXVII: The 17th International Conference on SolidState Sensors, Actuators and Microsystems (TRANSDUCERS & EUROSENSORS XXVII), pp. 2519–2522, Barcelona, Spain, June 2013. View at: Publisher Site  Google Scholar
 S. Sung, W. Sung, and C. Kim, “On the modematched control of MEMS vibratory gyroscope via phasedomain analysis and design,” IEEE/ASME Transactions on Mechatronics, vol. 14, no. 4, pp. 446–455, 2009. View at: Publisher Site  Google Scholar
 B. Yang, X. Wang, Y. Deng, and D. Hu, “Mechanical coupling error suppression technology for an tmproved decoupled dualmass microgyroscope,” Sensors, vol. 16, no. 4, p. 503, 2016. View at: Publisher Site  Google Scholar
 L. Wu, B. Yang, and G. Wang, “Design of an automatic modematching system for MEMS gyroscope,” in The seventh international conference on Instrumentation & Measurement, computer, communication and control (IMCCC 2017), pp. 47–51, Changchun China, September 2017. View at: Google Scholar
Copyright
Copyright © 2019 Bo Yang et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.