Phase-locked physical impulse control device and control method for oscillators

Abstract

The oscillator is controlled with high precision. [Solution] By using the number of vibration detections as the measurement criterion rather than the passage of time, the measurement boundary and vibration are synchronized, and reset errors and periodic artifacts are structurally eliminated. Multi-stage statistical processing is performed on the acquired vibration period data, reducing short-term noise in the first stage and long-term fluctuations in the second stage, guaranteeing statistical accuracy based on the central limit theorem. Furthermore, by performing sparse intervention control, applying positive and negative symmetrical pulses to an even half-period phase window at a time ratio of less than half within the evaluation window, the rate is probabilistically corrected while keeping the average external force at zero.

Description

【Technical Field】 【0001】 The present invention relates to the step control of an oscillator, and particularly to a technique for realizing high-precision control while removing measurement artifacts by using the number of vibrations of the oscillator of a mechanical clock as a measurement reference and combining sparse control intervention and multi-stage statistical processing. 【0002】 This technology is widely applicable to periodic mechanical vibration systems represented by mechanical clocks, and relates to a physical control method based on stochastic resonance that maintains periodic stability using statistical methods, with the number of detections rather than time as the measurement reference. The principle related to this technology can be generalized to mechanical and electronic systems with periodic motion, such as clocks, MEMS resonators, precision machine tools, medical pumps, and rotating bodies. Therefore, the present invention also relates to technologies for realizing high-precision control of these. 【0003】 (Definition of Terms) The main technical terms used in this specification are defined as follows. Evaluation window T_eval: Refers to the reference time unit for performing statistical evaluation (corresponding to the update period of the first-stage average). Control cycle T_ctl: Refers to the period from one control decision to the next control decision. Observation interval: Refers to the natural response observation period without applying a control input. Intervention interval: Refers to the period during which a control input is applied. Detection reference measurement: Refers to a measurement that uses the number of vibration detections rather than the passage of time as the reference unit of measurement. Continuous measurement: Refers to a measurement with an update method (such as a circular buffer or sliding window) that does not involve resetting at the measurement boundary. First Swing Carryover Bug: Refers to a measurement contamination phenomenon that occurs immediately after reset in time-based measurement. Sparse intervention: Refers to intermittent control with a low control duty (typically less than 50%). SD spike (standard deviation spike): Refers to a phenomenon in which the standard deviation temporarily rises sharply in the time series or distribution of data. Measurement artifacts refer to periodic measurement contamination (e.g., First Swing Carryover or SD spikes) that occurs in time-referenced measurements when the periodic boundary is misaligned with the oscillation period. Vibration period data: This is a numerical representation of the time required for an oscillator to complete one reciprocating motion (i.e., one unit of vibration), and is the time difference (Δt) between the detection of the peak or zero-crossing point of the vibration signal continuously detected by the sensor. i This refers to data acquired as a vibration period. This vibration period data is updated sequentially based on the number of vibration detections, not on the passage of time, and represents the periodic fluctuation for each vibration unit. [Background technology] 【0004】 Conventional oscillator measurements were time-referenced measurements performed at fixed intervals. In time-referenced measurements, the oscillation period and measurement boundary became asynchronous due to resets at fixed time intervals, resulting in artifacts such as the First Swing Carryover Bug and periodic SD spikes. Furthermore, conventional oscillator control was high duty cycle control (80-100%), making it impossible to observe the oscillator's natural response and thus preventing the assurance of control stability. 【0005】 Patent Document 1 discloses control using phase-locking and p / q injection, but the measurement method remains time-based. Patent document 2 discloses stochastic injection and duty cycle control, but does not mention the removal of measurement artifacts. Patent document 3 discloses an average external force of zero and a stopping mode, but does not include accuracy assurance through multi-stage statistical processing. [Prior art documents] [Patent Documents] 【0006】 [Patent Document 1] Japanese Patent Application No. 2025-125969 (Prior application by the present applicant) [Patent Document 2] Patent No. 7755098 [Patent Document 3] Patent No. 7755099 [Overview of the project] [Problems that the invention aims to solve] 【0007】 The present invention aims to solve the following problems. (1) Structural removal of artifacts caused by the measurement method. (2) Improved control stability by ensuring observability. (3) Achieving statistically guaranteed measurement accuracy. (4) Energy-saving operation with minimal intervention. [Means for solving the problem] 【0008】 According to a first aspect of the present invention, a control device for a vibrating body that performs periodic vibration or rotation, a) A sensor for detecting vibration or rotation of the vibrating body, b) A measurement means for collecting vibration period data using the number of detections as a measurement criterion, c) A calculation means for performing multi-stage statistical processing, d) Control means that control in less than half of the evaluation period, A control device is provided that includes the following. 【0009】 Here, the control device is defined as a rate control device for a mechanical clock, and the oscillating body is defined as the oscillator of a mechanical clock, and the control device is a rate control device for a mechanical clock, a) A sensor that detects vibrations of the oscillator, b) A measurement means that collects vibration period data based on the number of vibrations rather than the passage of time, using the number of detections by the sensor as the measurement criterion, c) A calculation means that performs at least two stages of statistical processing on vibration period data collected by the measurement means, which includes reducing sensor-derived noise by first-stage statistical processing and reducing oscillator-specific fluctuations by second-stage statistical processing. d) In the evaluation window T_eval, a control input is applied to the oscillator at a time ratio of less than half, and the natural response of the oscillator is observed at a time ratio of more than half, so as to separate control and observation temporally, and a control means; It may be characterized by comprising. 【0010】 (Application of Statistical Theory) Here, Based on the central limit theorem, the arithmetic means: In the first stage, N data are averaged to reduce the error to 1 / √N (the mathematical formula following the square root symbol "√" is enclosed by √. The same applies hereinafter), In the second stage, M first-stage average values are averaged to further reduce the error to 1 / √M, where N is selected from the range of 15 or more and 50 or less, and M is selected from the range of 256 or more and 2048 or less. 【0011】 (Requirement for Artifact Removal) Here, The measuring means performs continuous data update without a reset operation at the measurement boundary, and the continuous data update is realized by any one of a circular buffer, a sliding window, or a double buffer, and discontinuity at the measurement boundary and the First Swing Carryover phenomenon are structurally avoided. 【0012】 (Stepwise Limitation of Control Duty) Here, The time ratio of applying the control input of the control means is in the range of 10% or more and 40% or less, Preferably in the range of 20% or more and 35% or less, Most preferably, it is set near 30%. 【0013】 (Specific Configuration of Multilayer Control) Here, The control means A high-speed control layer that operates when the average of the immediately preceding L windows (L = 2 to 5) exceeds the first threshold T1, and A low-speed control layer that operates when the average of the most recent K windows (K = 30 to 60) exceeds a second threshold T2 (T2 < T1), It may be provided that the L, K, T1, and T2 can be adaptively updated according to the characteristics of the oscillator. 【0014】 (Average external force zero control) Here, By applying a pulse that is symmetric about zero in the even half-cycle phase window of the oscillator, the control means It may be possible to adjust the walking pace by probabilistic control according to the phase error while statistically maintaining the average external force at zero. 【0015】 (Utilization of negative correlation) Here, The arithmetic means calculates the lag-1 autocorrelation coefficient ρ of the second-stage average value, When ρ < 0 (negative correlation) is detected, by relaxing the convergence determination threshold or accelerating the convergence speed estimation, It may be possible to promote statistical convergence by utilizing periodic fluctuations such as beat errors. 【0016】 According to a second aspect of the present invention, there is provided a method for controlling a vibrating body that performs periodic vibration or rotation, A step of detecting the vibration of the oscillator and using the number of detections rather than the passage of time as a measurement reference, A step of performing multi-stage statistical processing on the collected vibration period data to reduce short-term noise in the first stage and long-term fluctuations in the second stage, A step of separating observation and control temporally by applying a control input at a time ratio of less than half and observing the natural response at a time ratio of more than half within an evaluation window T_eval, A control method characterized by including the above steps is provided. 【0017】 According to a third aspect of the present invention, a computer Means for collecting vibration period data using the number of vibration detections as a measurement reference, Means for gradually reducing noise at different time scales by multi-stage statistical processing, A means for separating control and observation based on the control duty cycle defined as a time ratio within the evaluation window T_eval, A program is provided to enable it to function as such. [Effects of the Invention] 【0018】 The present invention provides any of the following effects. (1) Artifacts caused by the measurement method are structurally removed. (2) Improved control stability can be achieved by ensuring observability. (3) Statistically guaranteed measurement accuracy is achieved. (4) Energy-saving operation with minimal intervention becomes possible. [Brief explanation of the drawing] 【0019】 [Figure 1] Figure 1 is a schematic diagram showing the control device of this embodiment incorporated into the mechanical watch body. [Figure 2] Figure 2 is a schematic diagram showing the configuration and fixed arrangement of the actuator. [Figure 3] Figure 3 is a functional block diagram showing the functional configuration of the control unit in this embodiment. [Modes for carrying out the invention] 【0020】 While this example illustrates a configuration for controlling the oscillator of a mechanical clock, its application to mechanical clocks is merely one example. The present invention is not limited to this application and can be applied to high-precision vibration control in a wide range of fields, from precision timekeeping devices to large-scale energy systems. 【0021】 The configurations, arrangements, element selections, and various values ​​shown in the following examples are merely examples and are not limited to these; they can be modified as appropriate. 【0022】 This invention integrates the technical concepts shown in A through D below. 【0023】 (Technical Concept A: Detection Criteria Measurement) By using the vibration frequency as the unit of measurement, measurement and vibration are essentially synchronized. This detection-based measurement concept fundamentally solves the asynchronous nature that is unavoidable with time-based methods. 【0024】 (Technical Concept B: Sparse Intervention Control) Within the evaluation window T_eval, the control input application time is limited to less than half (preferably 10-40%, optimally around 30%), and the natural response is observed for more than half of the time. This represents a sparse intervention control concept that temporally separates observation and control. 【0025】 (Technical Concept C: Multi-stage statistical processing) This is a multi-stage statistical processing concept that progressively separates and reduces noise of different origins and time scales, as shown in the first and second stages below. Stage 1: Average of N data points (Guidelines for selecting N will be explained later) Stage 2: Average the results of M items from Stage 1 (the criteria for selecting M items will be explained later). 【0026】 (Guidelines for selecting N) The determination of N (the number of samples in the first stage) is made by optimizing within the following range. Lower limit: The lower limit for which statistical significance is ensured (N≧15, the practical lower limit for applying the central limit theorem). Upper limit: The upper limit at which responsiveness is maintained (N ≤ 50, delay within 10-20 seconds). The optimization of N is determined by the following balance design formula, which takes a statistical and time-response balance by averaging multiple cycles (k times) of the "number of samples that the sensor can acquire during one vibration cycle" to suppress noise. N = fs / fv × k fs: Sensor sampling frequency fv: Vibrator fundamental frequency k: integer (usually 5-10) 【0027】 (Selection Guidelines for M) The determination of M (second-stage sample size) is optimized within the following range. The second-stage processing is performed to average out long-term fluctuations (temperature, attitude, friction fluctuations, etc.). Lower limit: The lower limit at which long-term fluctuations are averaged out (M≧256, the practical lower limit for applying the central limit theorem). Upper limit: The upper limit that satisfies the convergence time constraint (M ≤ 2048, within a few hours). Increasing M reduces the variance of the mean, but proportionally increases the "time required for evaluation". In the rate control of the mechanical clock in Example 1, it is desirable to converge (stabilize) within a few hours, so 2048 is set as the upper limit that can maintain an effective control cycle in real time. The optimization of M is based on the central limit theorem, which states that the standard deviation (statistical error) when averaging independent data decreases proportionally to 1 / √M with respect to the original variance σ0. If we let σ_req be the precision (target error) we ultimately want to achieve, then the M that satisfies it can be derived as follows. M = (σ_0 / σ_req) 2 σ_req:Required accuracy σ_0: initial variance 【0028】 (Technical Concept D: Detection and Utilization of Discorrelation) In the second stage of statistical processing, the statistical properties of the long-term mean value sequence {R[i]} are further analyzed, and its lag-1 autocorrelation coefficient (ρ) is calculated. The autocorrelation coefficient ρ is defined by Equation 1. 【0029】 【number】 【0030】 This calculation evaluates the correlation between consecutive second-stage mean values. Here, if ρ is negative, it means that periodic alternating fluctuations (such as beat errors) are occurring. In this embodiment, a significant negative correlation is determined when ρ < -0.2. 【0031】 Based on this determination result, the control unit dynamically performs the following corrections. In other words, if a negative correlation is detected, the threshold used for determining convergence is relaxed according to |ρ|, and the modified threshold T' is calculated using Equation 2. 【0032】 【number】 【0033】 Furthermore, in estimating the convergence rate, the effective sample size M_eff is increased to account for the effect of negative correlation, and corrected using the following equation 3. 【number】 【0034】 Furthermore, if ρ is negative, the central value is estimated from the alternating sign pattern of the continuous data, and the bias in the mean value caused by beat error is statistically corrected. This allows us to leverage periodic errors to accelerate statistical convergence, simultaneously improving the overall stability and responsiveness of rate control. [Examples] 【0035】 (Application to mechanical watches) Figure 1 is a schematic diagram showing the control device of this embodiment incorporated into the mechanical clock body 100. The control device of this embodiment is built into the mechanical clock body 100, detects the vibration state of the oscillator 110 of the mechanical clock, and statistically controls the rate based on the vibration period data. It consists of a sensor unit 120, an actuator 130, a control unit 140, an external reference clock 150, and a power supply unit 160. 【0036】 The oscillator 110 uses a balance wheel with a specification of 18,000 vibrations / hour (2.5 Hz) and stabilizes the rate through sparse intervention control (control duty cycle near 30%) within the evaluation window T_eval. 【0037】 The sensor unit 120 consists of an optical motion sensor (single CMOS optical sensor) that detects the reciprocating motion of the balance wheel. It correlates the scattered light pattern from the balance wheel's reflective surface with a light receiving array and generates a peak detection event corresponding to each reciprocating motion. This sensor output is measured based on the number of detections rather than the passage of time, and vibration period data between each peak detection is transmitted to the control unit 140. The sensor pulses a predetermined VCSEL (Vertical-Cavity Surface-Emitting Laser) and outputs vibration data, including amplitude, peak, and zero-crossing point, in real time. 【0038】 The control unit 140 consists of a dual-core microcontroller (e.g., RP2350) that continuously updates sensor data in a circular buffer and performs detection criterion measurements without time resets. The control unit performs multi-stage statistical processing in the first and second stages, gradually reducing noise and fluctuations based on the central limit theorem. Furthermore, within the evaluation window T_eval, control pulses are applied to the actuator 130 for less than half the time, and the natural response is observed for more than half the time, thereby temporally separating observation and control. 【0039】 The external reference clock 150 is 32.768 kHz (=2 15 It is a temperature-compensated oscillator (Hz, ±50 ppb) and functions as a period measurement reference for the control unit 140. By selecting this frequency, one cycle of the balance wheel (0.4 s) corresponds to approximately 13,107 clock cycles, enabling high-precision period measurement, frequency division, and statistical processing using integer clocks. 【0040】 The power supply unit 160 uses a low-power lithium battery or capacitor to periodically drive the sensor, control unit, and actuator. 【0041】 The actuator 130 is composed of a bipolar drivable multilayer PZT piezoelectric element arranged in a direction (X-axis) perpendicular to the main vibration axis (Z-axis) of the vibrator 110. A pair of piezoelectric elements are positioned opposite each other on both sides of the balance support or the jewel bearing 111, and are fixed so as to have position vectors r1 and r2, respectively. These two elements employ a cross-axis coupling configuration that generates opposing forces F1 and F2 in the X-axis direction, and generates a small rotational torque component around the main vibration axis by the cross product of their force vectors and position vectors. In an even half-period phase window, the control unit 140 applies positive and negative symmetrical voltage pulses (±V_p) to both actuators, performing stochastic pulse injection according to the phase error while statistically maintaining the average external force at zero. This achieves sparse intervention and zero-mean-force control, where the natural response is observed during odd half-periods, and phase correction is performed under conditions of zero mean-force during even half-periods. This minute energy modulation in the orthogonal direction is transferred to the main oscillation via mechanical cross-axis coupling, manifesting as angular velocity fluctuations in the balance wheel, and statistically stabilizing the rate. 【0042】 Figure 2 is a schematic diagram showing the configuration and fixed arrangement of the actuator. Two electronic actuators 201 and 202 are fixed to the borehole 111 of the vibrator 110 so as to sandwich both shoulders of the vibrator's main vibration axis 112, and the displacement axis of each electronic actuator is set to contact the shoulders of the main vibration axis 112 with a small preload at an inclination angle α (0° < α < 90°) with respect to the XY plane, which is the rotation plane of the balance wheel. At the point of contact with the main vibration axis 112, the electronic actuators 201 and 202 are displaced in the direction of their respective displacement axes 203 and 204, applying a pulsed external force. 【0043】 (Control unit) The control unit 140 of this embodiment is the core configuration for realizing the calculation means and control means of the present invention. This unit integrates the collection of vibration period data using the number of detections as a measurement criterion, the reduction of noise and fluctuations through first and second stage statistical processing (operation of the calculation means), and sparse intervention type rate correction based on the statistical results (operation of the control means). These processes are automatically executed within the microcontroller by a pre-installed control program (firmware). The control unit 140 is electrically connected to the sensor unit 120, actuator 130, external reference clock 150, etc., and works in cooperation with each component to form a rate control loop. 【0044】 The control unit 140 employs a dual-core microcontroller (MCU) with Core 1 and Core 0. Core1 is responsible for the functions of the measurement means (acquisition of sensor signals and generation of vibration period data), while Core0 is responsible for the functions of the calculation means (multi-stage statistical processing and control decision). The MCU measures the period based on an external reference clock of 150 (32.768 kHz, ±50 ppb), but its evaluation unit is the number of detections rather than the passage of time, and it performs continuous data updates synchronized with the oscillator's detection events. This allows the MCU to generate highly accurate oscillation period data with precisely measured time intervals by referencing an external reference clock 150. By using the number of detections rather than the passage of time as the measurement criterion, the vibration and the measurement boundary are synchronized, and measurement artifacts are structurally avoided. 【0045】 Figure 3 is a functional block diagram showing the functional configuration of the control unit in this embodiment. The control unit 140 consists of a sensor signal processing unit 301, a vibration period data update unit 302, a first-stage statistical processing unit 303, a second-stage statistical processing unit 304, an evaluation and control determination unit 305, and a pulse control unit 306. The first-stage statistical processing unit 303 averages N detected vibration period data to reduce sensor-derived noise, and the second-stage statistical processing unit 304 further averages M first-stage average values ​​to smooth out long-term oscillator fluctuations. As a result, the error is reduced inversely proportional to √(N×M) based on the central limit theorem. 【0046】 The sensor signal processing unit 301 processes the optical signal obtained from the sensor unit 120. The system detects vibration peaks and transmits the interval between each peak as vibration period data to the vibration period data update unit 302. This vibration period data is buffered based on the number of detections, not on time. 【0047】 The vibration period data update unit 302 continuously holds N of the latest data points using a circular buffer, enabling continuous measurement without resetting at measurement boundaries. This structurally eliminates measurement artifacts such as First Swing Carryover. 【0048】 The first-stage statistical processing unit 303 (the stage before the calculation means) calculates the average value and variance of N vibration period data and removes short-term noise. The second-stage statistical processing unit 304 (the downstream unit of the calculation means) averages the M first-stage results and smooths out long-term fluctuations such as temperature changes and attitude differences. 【0049】 The evaluation and control decision unit 305 calculates the rate error and the lag 1 autocorrelation coefficient ρ of the second stage data sequence based on the above statistical results. If ρ shows a negative correlation (ρ<0), the convergence threshold is relaxed or the convergence rate estimation is corrected to determine the statistically optimal control timing. If the evaluation result exceeds threshold T1 or T2, the sparse intervention control mode is activated. Here, thresholds T1 and T2 are defined as follows: Threshold T1 (first threshold): A reference value for activating the high-speed control layer when short-term rate fluctuations exceed the acceptable range. It responds to rapid changes and short-term error biases. Threshold T2 (second threshold): This is a reference value for correcting long-term rate drift (slow deviation), and is set to be smaller than T1. It is the threshold for activating the slow control layer when slow deviations accumulate statistically. In other words, T1 and T2 are statistical control initiation criteria for short-term and long-term rate errors, respectively, and the relationship T1 > T2 is a threshold for switching between multi-layer control (high-speed layer and low-speed layer). 【0050】 The pulse control unit 306 sets a control duty cycle (10% to 40%) based on the evaluation results and generates a control pulse that is positively and negatively symmetrical within an even half-period phase window. The control is performed probabilistically, and by combining sensor-referenced injection (5%) and timer-referenced injection based on a Poisson distribution (25%), phase errors are corrected while keeping the statistical mean external force at zero. 【0051】 The following is an example of pseudocode illustrating the main algorithms of the control unit 140. ```cpp / / Example implementation of detection count-based measurement and multi-stage statistical processing (calculation means) const int WINDOW_SIZE = 25; / / For first stage averaging float period_buffer[WINDOW_SIZE]; int buffer_index = 0; bool buffer_filled = false; / / Core1: Read detected events (fixed at 1kHz) void core1_sensor_task() { while(1) { int16_t delta_x = readSensor(); if (detectPeak(delta_x)) { float period = getPeriodMicros(); updateBuffer(period); } delay_us(1000); / / 1ms period } } / / Core0: Multi-stage statistical processing and control decision void updateBuffer(float period_us) { period_buffer[buffer_index] = period_us; buffer_index = (buffer_index + 1) % WINDOW_SIZE; if (buffer_index == 0 && buffer_filled) { / / First stage average (25 detections = approximately 10 seconds) float avg1 = calculateMean(period_buffer, WINDOW_SIZE); float std1 = calculateStd(period_buffer, WINDOW_SIZE); / / To the second stage average processSecondStage(avg1, std1); } if (buffer_index == 0) buffer_filled = true; } / / Sparse intervention control (30% duty cycle) void sparseControl() { / / Sensor-based injection: 5% probability if (random(1000) < 50) { schedulePulse(PHASE_0_427); } / / Timer-based injection based on Poisson distribution: 25% static uint32_t next_pulse = 0; if (millis() >= next_pulse) { firePulse(); / / Positive and negative symmetric pulse next_pulse = millis() + exponentialRandom(2667); / / Average interval of 2.67 seconds } } This program code is part of the control program stored in the microcontroller of the control device. 【0052】 In this pseudo-program, the updateBuffer() function corresponds to the first-stage statistical processing unit and the second-stage statistical processing unit (calculation means). Within this function, the part that averages the most recent N measurements stored in the vibration period data `period_buffer[]` is the first stage of statistical processing (short-term noise reduction). The averaging process calculatedMean() used here is based on the Central Limit Theorem and aims to reduce the error by a ratio of approximately 1 / √N. Within this function, processSecondStage() is called, and the second stage of statistical processing (long-term averaging) is performed. In the second stage, the M results from the first stage are further averaged, and the error is reduced by a ratio of 1 / √M. Thus, in terms of program structure, the calculation means performs "N×M step averaging" to suppress the period measurement error to 1 / √(N×M). 【0053】 (Guidelines for selecting N and M) The constant `const int WINDOW_SIZE = 25;` in the pseudo-program indicates the number of samples N for the first stage of statistical processing, forming an evaluation window of approximately 10 seconds for a frequency of 2.5 Hz. In the second stage of statistical processing, long-term fluctuations are removed by averaging over a 1024-window (M=1024) within processSecondStage(). Here, N and M are not limited to these values; rather, to balance responsiveness and convergence, N is set within the range of 15 to 50 and M within the range of 256 to 2048. This N·M design guideline is specifically implemented in the program through constant settings and loop counts. 【0054】 (Detection and utilization of negative correlation) The processSecondStage() function records the progression of the second-stage mean series {R[i]} and calculates its lag-1 autocorrelation coefficient ρ. If the control unit 140 detects ρ < 0 (negative correlation), it performs a process to relax the convergence threshold by a factor of (1 + |ρ| / 2) or correct the effective sample size M_eff = M × (1 + |ρ|). As a result, the program is optimized to dynamically increase the convergence speed by statistically utilizing the periodic fluctuations (beat error) of the oscillator. 【0055】 (Multilayer control structure) In the pseudo-program, the sparseControl() function corresponds to the control means of the present invention, and multiple control layers function within it. Specifically, sensor-synchronous injection (random(1000)<50 probability branch) is responsible for short-period error correction (high-speed control layer), while timer-referenced injection (exponentialRandom(2667)) based on a Poisson distribution is responsible for long-term average rate correction (low-speed control layer). These implementations correspond to "L high-speed control layers and K low-speed control layers," separating control decisions by referencing statistical results on different time scales within the program. 【0056】 (Average external force zero control) The pulse generated by schedulePulse(PHASE_0_427) is applied as a waveform that is symmetrical in positive and negative directions within an even half-period phase window. This corresponds to a configuration in which "positive and negative symmetrical pulses are applied within an even half-period phase window." When this function is called, symmetric pulses of +F / -F are automatically generated within the next half-period, and as a result, the mean external force is controlled to be zero. No pulses are generated in odd-numbered half-periods, and only the natural response is observed. This design ensures a zero mean external force condition at the program level and establishes temporal separation between observation and control. 【0057】 This pseudo-program has a set of statistical processing functions (calculateMean(), processSecondStage()), The control functions (sparseControl(), firePulse()) operate in an integrated manner using the evaluation window T_eval as the unit. The control unit 140 maintains the statistical mean external force at zero by applying positive and negative symmetrical pulses in an even half-period phase window, based on detection count-based measurement and multi-stage statistical processing. The control unit 140 uses the vibration period data obtained from the detection count-based measurement to determine the vibration period. The first and second stages of statistical processing are performed to reduce noise and fluctuations, and based on the results, sparse intervention and mean-zero external force control are implemented. This allows multi-stage statistical processing (operation of the calculation means) and sparse intervention control (operation of the control means) to work together in an integrated manner, achieving both the elimination of measurement artifacts and the stabilization of the rate. 【0058】 The above program configuration eliminates reset errors and First Swing Carryover, which are inherent to time-based measurement methods, and in principle, measurement artifacts are eliminated. In addition, by combining multi-stage statistical processing with stochastic sparse intervention control, we reduced rate variance and shortened the convergence time. [Industrial applicability] 【0059】 The multi-stage statistical and sparse intervention control method according to the present invention can be applied not only to mechanical clocks, but also to stabilizing MEMS oscillators, quartz oscillators, atomic clocks, and even periodic operating mechanisms such as turbines and pumps. [Explanation of Symbols] 【0060】 100 Mechanical watch bodies 110 transducer 120 Sensor Units 130 Actuator 140 Control Unit 150 External reference clock 160 Power supply section

Claims

【Claim 1】 A control device for a vibrating body that performs periodic vibration or rotation, comprising: a) a sensor that detects the vibration or rotation of the vibrating body; b) measuring means for collecting vibration period data using the number of detections of the vibration or rotation as a measurement reference; c) arithmetic means for performing two-stage averaging processing to reduce noise by first-stage averaging processing and reduce the inherent wobbling of the vibrating body by second-stage averaging processing on the vibration period data collected by the measuring means; d) control means for controlling the pacing error by applying a control input to the vibrating body within less than half of an evaluation period, which is a reference time unit for averaging the short-term and long-term pacing errors of the vibrating body; A control device comprising the above. 【Claim 2】 In Claim 1, the arithmetic means is based on the central limit theorem, in the first-stage averaging process, N data are averaged to reduce the error to 1 / √N (the number following the square root symbol "√" is enclosed by √. The same applies hereinafter), in the second-stage averaging process, M first-stage average values are averaged to further reduce the error to 1 / √M, where N is selected from the range of 15 or more and 50 or less, and M is selected from the range of 256 or more and 2048 or less. A control device characterized by this. 【Claim 3】 In Claim 1, the measuring means performs continuous data updating without a reset operation at the measurement boundary, the continuous data updating is realized by any one of a circular buffer, a sliding window, or a double buffer, and structurally avoids discontinuity at the measurement boundary and the First Swing Carryover phenomenon. A control device characterized by this. 【Claim 4】 In Claim 1, the control input application time ratio of the control means is set within the range of 10% or more and 40% or less. A control device characterized by this. 【Claim 5】 In Claim 1, the control input application time ratio of the control means is set within the range of 20% or more and 35% or less. A control device characterized by this. 【Claim 6】 In Claim 1, the control means includes a high-speed control layer that starts average control for short-term pacing error when the average of the most recent L (L = 2 to 5) window averages exceeds a first threshold T1, a low-speed control layer that starts average control for long-term pacing error when the average of the most recent K (K = 30 to 60) window averages exceeds a second threshold T2 (T2 < T1), and L, K, T1, and T2 are adaptively updatable according to the characteristics of the vibrating body. A control device characterized by this. [[ID=2⑨]]【Claim⑦】In Claim 1, The control means applies a positive-negative symmetric pulse in the even half-cycle phase window of the vibrating body, characterized in that while maintaining the average external force at zero on average, the walking pace is adjusted by probabilistic control according to the phase error. **Claim 8** In claim 2, the arithmetic means calculates the lag-1 autocorrelation coefficient ρ of the second-stage average value, when ρ < 0 (negative correlation) is detected, by relaxing the convergence determination threshold value or accelerating the convergence speed estimation, characterized in that the periodic variation of the beat error is utilized to promote average convergence. **Claim 9** A control method for a vibrating body that performs periodic vibration or rotation, comprising: detecting the vibration of the vibrating body and collecting vibration period data with the number of detections as the measurement reference instead of the passage of time; performing a two-stage averaging process on the collected vibration period data, reducing short-term noise in the first stage and long-term variations in the second stage; separating observation and control temporally by applying a control input at a time ratio of less than half and observing the natural response at a time ratio of more than half within the evaluation window T_eval; characterized by including the above steps. **Claim 10** In claim 9, in the step of performing the two-stage averaging process, based on the central limit theorem, averaging N data in the first stage to reduce the error to 1 / √N, averaging M first-stage average values in the second stage to further reduce the error to 1 / √M, where N is selected from the range of 15 or more and 50 or less, and M is selected from the range of 256 or more and 2048 or less. **Claim 11** In claim 9, the collection of the vibration period data performs continuous data update without a reset operation at the measurement boundary, the continuous data update is realized by any one of a circular buffer, a sliding window, or a double buffer, characterized by structurally avoiding discontinuities and the First Swing Carryover phenomenon at the measurement boundary. **Claim 12** In claim 9, the control includes high-speed control that operates when the average of the most recent L (L = 2 to 5) windows exceeds the first threshold value T1, and low-speed control that operates when the average of the most recent K (K = 30 to 60) windows exceeds the second threshold value T2 (T2 < T1), where L, K, T1, and T2 are adaptively updatable according to the characteristics of the vibrating body. **Claim 13** In claim 9, The control is performed by applying positive and negative symmetrical pulses in the even half-period phase window of the vibrating body, A control method characterized by adjusting the rate of motion through probabilistic control according to the phase error while maintaining the average external force at zero on average. [Claim 14] In claim 10, The process of performing the two-stage averaging process calculates the lag-1 autocorrelation coefficient ρ of the second-stage average value, A control method characterized by promoting average convergence by utilizing the periodic fluctuations of beat error, by relaxing the convergence criterion threshold or accelerating the convergence rate estimation when ρ < 0 (negative correlation) is detected. [Claim 15] Computers, A means for collecting vibration period data using the number of vibration detections of a vibrating body that performs periodic vibration or rotation as a measurement criterion. A means for progressively reducing noise on different time scales in the vibration period data by a two-stage averaging process in which short-term noise is reduced in the first stage and long-term fluctuations are reduced in the second stage. Means for separating the control and observation of the vibrating body based on a control duty cycle defined as a time ratio within an evaluation window T_eval where the control input is less than half of the control duty cycle, A program designed to function as such.

Images